International Energy Economics

INTERNATIONAL ENERGY ECONOMICS
INTERNATIONAL STUDIES IN ECONOMIC MODELLING
Series Editor
Homa Motamen-Scobie
Executive Director
European Economics and Financial Centre
London
Economic Modelling in the OECD Countries

H. Motamen-Scobie
Modelling the Labour Market

M. Beenstock
Input-Output Analysis
M. Ciaschini
Models of Disequilibrium and Shortage in Centrally Planned Economies

C. Davis and W. Charemza
Economic Modelling at the Bank of England

S. G. B. Henry and K. D. Patterson
Recent Modelling Approaches in Applied Energy Economics
O. Bjerkholt, O. Cblsen and J. Vislie
International Commodity Market Models

O. Giivenen, W. C. Labys and J. B. Lesourd
Dynamic Models for the Inter-relations of Real and Financial Growth
H. Ekstedt and L. Westberg
Economic Models of Trade Unions

P. Garonna, P. Mori and P. Tedeschi
International Energy Economics

Thomas Sterner
International Trade Modelling

M. G. Dagenais and P-A Muek
International Energy
Economics
Edited by
Thomas Sterner
Department of Economics, Gothenburg University,
Sweden
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.
1992 Springer Scien ce+Bu sine ss Medi a Dordrecht

Softcover reprint of the hardcover Ist edition 1992
OriginaUy published by Chapman & HaU in 1992
Chapters 2 and 4: The submitted manuscript has
been authored by a contractor of the U.S. Government under contract
No. W-31-109-ENG-38. Accordingly, the U.S. Government retains a nonexclusive, royalty-free license to publish or reproduce the published form of
this contribution, or allow others to do so, for U.S. Government purposes.
Typeset IOJI2pt Times by Interprint Ltd, Malta.
ISBN 978-94-010-5031-9
ISB N 978-94-011-2334-1 (eBook)
DOI 10.1007/9 78-94-011-2334-1
Apart from an y fair dealing for the purposes of research or private study , or
criticism or review, as permitted under the UK Copyright Designs and
Patents Act, 1988, this publication may not be reproduced, stored, or
transmitted, in any form or by any means , without the prior permission in
writing of the publishers, or in the case of reprographic reproduction only in
accordance with the terms of the Iicences issued by the Copyright Licensing
Agency in the UK, or in accordance with the terms of licences issued by the
appropriate Reproduction Right s Org anization outside the UK . Enquiries
concerning reproduction outside the terms stated here should be sent to the
publishers at the London address printed on this page.
The publisher makes no representation, express or implied, with regard to
the accuracy of the information contained in this book and cannot accept any
legal responsibility or Iiability for any errors or omissions that may be made.
A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication data
International energy economicsJedited by Thomas Sterner. - Ist ed.
p.
cm. - (International studic s in economic modelling)
Includes bibliographical references and index.
1. Energy policy.
2. Power rcsour ces-Forecasting.
Thomas, 1952- .
II. Series.
HD9502 .A215774 1992
333.79-dc20
I. Sterner,
91--42590
CIP
Contents
Contributors
Preface
1 Introduction
Thomas Sterner
2 Forecasting industrial energy use
Gale A. Boyd
3 Best-practice and average practice: technique choice and energy
demand in a vintage model
Lennart Hjalmarsson and Finn R. Forsund
4 The effects of changes in the economic structure on energy demand
in the Soviet Union and the United States of America
Yu. D. Kononov, H. G. Huntington, E. A. Medvedeva and G. A.
Boyd
5 Modelling transport fuel demand
Thomas Sterner and Carol A. Dahl
6 Modelling the long-run supply of coal
Ronald P. Steenblik
7 Global availability of natural gas: resources, requirements and
location
Daniel A. Dreyfus
8 Modelling oil exploration
Victor Rodriguez Padilla
9 Environmental cost functions: a comparison between general and
partial equilibrium
Lars Bergman
10 Energy policies in a macroeconomic model: an analysis of energy
taxes when oil prices decline
P. Capros, P. Karadeloglou and G. Mentzas
11 A comparison of energy-economy models: the French experience
Ghislaine Destais
VB
IX
11
31
47
65
81
105
117
141
157
185
Contents
vi
12 Models and projections of energy use in the Soviet Union

13
14
15
16
17
18
Yuri Sinyak
A detailed simulation approach to world energy modelling: the
SIBILIN and POLES experiences
Patrick Criqui
Inferred demand and supply elasticities from a comparison of
world oil models
Hillard G. Huntington
World oil market simulation
Nick Baldwin and Richard Prosser
International Energy Workshop projections
Alan S. Manne and Leo Schrattenholzer
Environmental regulations and innovation: a CGE approach for
analysing short-run and long-run effects
Gunther Stephan
CO 2 emission limits: an economic cost analysis for the United
States of America
Alan S. Manne and Richard G. Richels
Index
203
221
230
263
297
299
232
347
Contributors
Strategic Studies Department, PowerGen, UK

Stockholm School of Economics, Stockholm, Sweden
Argonne National Laboratory, Argonne, Illinois,
USA
National Technical University of Athens, Greece
P. Capros
Institut d'Economie et de Politique de l'Energie,
Patrick Criqui
Grenoble, France
Department of Mineral Economics, Colorado
Carol A. Dahl
School of Mines, USA
Institut d'Economie et de Politique de l'Energie,
Ghislaine Destais
Grenoble, France
Vice President, Strategic Planning and Analysis Gas
Daniel A. Dreyfus
Research Institute, Washington, DC
Department of Economics, Oslo University,
Finn R. F orsund
Norway
Lennart Hjalmarsson
Sweden
Energy Modeling Forum, Stanford University,
Stanford, California, USA
P. Karade1og10u
Siberian Energy Institute, Irkutsk, USSR
Yu. D. Kononov
Department of Operations Research,
Alan S. Manne
Stanford University, Stanford, California, USA
G. Mentzas
Siberian Energy Institute, Irkutsk, USSR
E. A. Medvedeva
National Power, UK
Richard Prosser
Victor Rodriguez Padilla Institut d'Economie et de Politique de l'Energie,
Grenoble, France
Richard G. Richels
Electric Power Research Institute, Palo Alto,
California, USA
International Institute for Applied Systems
Leo Schrattenholzer
Analysis, Laxenburg, Austria
Yuri Sin yak
International Institute for Applied Systems
Analysis, Laxenburg, Austria
Nick Baldwin
Lars Bergman
Gale A. Boyd
viii
Ronald P. Steenblik
Gunter Stephan
Thomas Sterner
Contributors
Organization for Economic Co-operation and
Development, Paris
Institute for Applied Micro-economics, University
of Berne, Switzerland
Sweden
Preface
The fact that editing a book is more work than you think it's going to be is a
'standard result' reported by most researchers. However many people forget to
tell you what a rewarding task it is. On occasion it has been a little hard on
my phone bill and even on my patience (the volume of E-mail, faxes and other
correspondence weighs several times more than the book) but the lasting
impression is that it has been a very stimulating experience to work together
with the authors of the individual chapters. I have learned a great deal and I
hope some of this is conveyed to the reader.
When Homa Motamen-Scobie first asked me if I wanted to edit a book on
energy modelling I must admit I was quite unsure how to go about the task.
I was sceptical of conference volumes or collections of papers where the
different chapters fail to interact and form a coherent book. My first idea was
therefore to draw up a rough structure and make a list of potential topics and
authors. The main challenge was to blow some life into the project and create
a dialogue between its different participants.
I am very grateful to Alan Manne who persuaded me that what I needed
was a 'critical mass' of authors who would all agree at the same time to get the
book going. Alan, together with Leo Schrattenholzer, also provided me with
an excellent forum: the International Energy Workshop (lEW) meeting at the
International Institute for Applied Systems Analysis (IIASA) in June 1989.
Roughly half the chapters of this book are in some way related to that very
stimulating meeting. Although only two or three of the chapters actually bear
any direct resemblance to papers presented at the conference, a contact was
established which allowed me to discuss with the various authors the kind of
contribution I was looking for. This, in turn, made it much easier to find other
authors to fill in the missing links.
My ambition has been to make this a broad survey of energy modelling in
the sense of both original research and survey articles, both empirical and
theoretical work. It covers demand, supply and energy-economy interaction;
models that are primarily short-run or primarily long-run, national or international, theoretical or oriented towards projections and policy analysis. The
book is also a reflection of the widely different academic and physical conditions in different countries. The authors come from a dozen countries and there
is a fairly even spread at least across the Atlantic. There are also two Soviet
Preface
contributions and there were to be a couple of papers from authors in the

Third World but unfortunately various factors intervened. The most worrying
case is of the Chinese author with whom I have been unable to re-establish
contact after he went back to China.
A considerable portion of energy economics research is undertaken at a
limited number of specialized institutions and this is clearly reflected in the
book. I have already mentioned Alan Manne who, together with Hillard
Huntington and others associated with the Energy Modelling Forum (EMF),
and their link with the IIASA through the lEW, have provided something of
a cornerstone for this work. Among European centres, I would particularly like
to mention, and thank, the Institut Economique et Politique de l'Energie
(IEPE) at Grenoble. While I was a guest at the IEPE in 1989, a significant part
of this volume was edited and three of its chapters (8, 11, 13) are written by
IEPE researchers. The IEPE was not only a very pleasant place to spend a
sabbatical but also a very appropriate one considering the prominent role
played by IEPE researchers such as Finon, Chateau, Lapillone and Criqui in
European energy-economy modelling (notably the MEDEE, EFOM, SIBILIN
and POLES models, see Chapter 13 and appendix to Chapter 10). Another
obvious centre of energy analysis is the International Energy Agency (lEA), one
of whose researchers has contributed with an article. I would like, in this
context, to thank both the lEA and also the Oxford Institute of Energy Studies
for hospitality and stimulating discussions related to this book.
With such a dispersed set of authors, my role as editor has necessarily been
quite active and I would like to thank all the authors for putting up with my
many suggestions and question marks. Where appropriate I have encouraged
the authors to use similar tables briefly describing the main features of their
models to facilitate comparisons (see Chapters 2, 9, 10, 11, 13 and 14). In order
to keep the usual academic level, I have tried to have all the papers reviewed
in some way. The trouble is that the incentive to review a chapter for a book
is much smaller than for a journal where you implicitly collect 'credits' for
future use. In several cases however I found that it was useful for the authors
to read each others' chapters in order to give more coherence to the book and thus, once again I would like to thank the authors to whom I have turned
in order to get comments on other chapters. In addition to these I would also
like to thank Dominique Anxo, Hans Bjurek, Per-Olov Johansson, Katarina
Katz, Catherine Locatelli, Geoffrey Harcourt, Homa Motamen-Scobie and
Peter Odell for constructive comments on various chapters or other help with
this book. I also want to thank Energy Economics, the OPEC review and the
Energy Journal for permission to reprint material in Chapters 15, 16 and 18
which are based on articles to which they hold copyright.
Thomas Sterner
Gothenburg
1
Introduction
Thomas Sterner
1.1 ENERGY MODELLING
As we enter the third decade of 'energy awareness' we discern both trends and
cyclical movements in the attention that different aspects of energy economics
receive. The political aspects of supply security were the hallmark of the 1970s:
this, in a way, was the period when energy economics was born, the International Energy Agency (lEA) and various national Departments of Energy were
established. These and other organizations, such as the International Institute
for Applied Systems Analysis (IIASA), embarked upon large-scale and somewhat grandiose modelling efforts motivated largely by the desire to ration
energy, optimize energy supply or trace the consequences of economic activity
on energy imports. They are epitomized by the US 'Project Independence'.
Their approach both to supply and demand was often technically biased and
much of the modelling was of the linear programming, input-output or
operations research type. This was the age of models such as EFOM and
MARKAL.
During the 1970s and early 1980s a number of impressive surveys of the
energy field were undertaken 1 and, as a result, there has come a gradual
awareness that energy as such is quite abundant albeit that various' costs (both
pecuniary and others) may be increasing. Naturally, supply and security
concerns regularly reappear during crises such as the latest one in the Gulf, but
it is still fair to say that interest has come to focus more on demand
management, including the role of prices and of environmental issues.
Modelling efforts appear to concentrate less on total optimization for the
whole system and more on detailed and in-depth understanding of incentives
and mechanisms of interplay between institutions, economics, energy and
environment. For the global models, demands have increased on their transIFor example, the IIASA study Energy in a Finite World (Hafele, 1981), Schurr et al. (1979) Energy
in America's Future; Landsberg et al. (1979) Energy: The Next Twenty Years, Brooks et al. (1979)
Energy in Transition, or Stobaugh and Yergin (1979) Energy Future. For an excellent survey of
early modelling see Manne et al. (1979).
Introduction
parency, comparability and the analysis of how results are conditioned by

model structure and assumptions. 1
The ambition of this book is to reflect, in a fairly broad fashion, the current
flavour of international energy modelling. We shall cover both the buildingblocks of energy demand and supply as well as different types of market model,
world or co~ntry-wide system models and projections. We concentrate on
formalized models but do not want thereby to exclude less formalized judgemental-political ones that may very well be appropriate to certain issues. The
approaches analysed include both 'bottom-up' and 'top-down', both process/econometric and recursive simulation or intertemporal optimization. The
purposes may be analysis, projections or application to other issues, such as
the environment.
Within a book of this format, one can, however, not possibly hope to do
justice to all the currents of energy economic modelling. For various reasons,
we have kept away from certain areas: there is little on the particular problems
of the heavily infrastructure-dominated energy carriers electricity and natural
gas. 2 Likewise we have stayed relatively clear of energy saving, demand
management and sustainability. As mentioned in the Preface, the Third World
has unfortunately not been covered in any way, and finally the focus is more
on the interaction between economics and energy modelling than on the
technical aspects of econometric modelling.
1.2 ENERGY DEMAND
The first section of this book deals with demand modelling, and the contributions have been chosen to represent not only distinctive types of energy
demand (industry, transport and general) but different methodologies. In
Chapter 2 Gale Boyd surveys a number of large-scale computerized industrial
energy demand models. Many of these are based on the academic 'KLEM'type econometric studies using flexible functional forms. In this applied type of
work, however, these econometric models face competition from process
models using engineering-type data. Although the latter may be theoretically
less satisfying, they do have a number of advantages. Their 'bottom-up'
approach allows for the incorporation of considerable amounts of actual
engineering and planning information. On the other hand, they can generally
not, unlike the econometric models, be expected to take price-induced effects
into consideration. A growing number of models appear to take a very
pragmatic view and incorporate elements of both econometric and process type
into a 'hybrid' model structure.
'The efforts of the Energy Modeling Forum and of the International Energy Workshop are
particularly instrumental to this process.
2See Recent Modelling Approaches in Applied Energy Economics ed. O. Bjerkholt et al. in this same
series of books.
Energy demand
One feature of reality that aggregate econometric models often gloss over is
that once technical choices are made and capital takes on an actual physical
configuration, then flexibility is greatly reduced. In many industries, technology
can best be described as 'putty-clay' and hence the vintage structure of the
plants comprising an industry is a vital characteristic determining their shortand even medium-run responsiveness to changes in market conditions. To deal
with this aspect Lennart Hjalmarsson and Finn F6rsund in Chapter 3 analyse
choice of technology in a vintage model for an industry with endogenous
investment. This framework is then used to analyse how fast average practice
catches up with best-practice under various assumptions. The authors are
somewhat sceptical about the optimism aroused by demonstrations of energyefficient best-practice technology, arguing that if capital turnover is slow, it
may take many years before average practice catches up with present bestpractice.
At the other end of the spectrum, we find that not only for an industry, but
for a whole economy, structure again plays a vital role in determining energy
consumption. Using Divisia indexes, a mixed US/Soviet quartet of economists
compare, in Chapter 4, the respective roles of technological versus structural
change in their two countries. Structural change, measured as change in output
composition, is, together with overall changes in average technology (in each
industry), expected to provide major opportunities for future Soviet energy
saving.
One rather striking feature of the Soviet economy as distinct from Western
economies is the very small role, in energy demand, played by the 'nonproduction sphere' - i.e. domestic consumption - and by transport. Even in
their forecasts for the year 2010 these sectors continue to be minor fractions in
total energy consumption. This runs very much counter to the result one would
expect with a free and unregulated market. In Chapter 5, Carol Dahl and I
discuss different ways of modelling transport fuel demand. Our main focus is
on the comparison between models, but practically all approaches find income
elasticities for transport fuels to be above one. Price elasticities by comparison
are often found to be below one, although it turns out to be difficult really to
estimate a true long run. The policy implications are anyhow clear, both for
our own OECD economies and for liberalizing Eastern European ones. If we
want to avoid the expected strong increase in consumer demand for transport
fuels, their relative price must rise faster than the rate of growth in income.
Without tough environmental policies, liberalization in the Eastern European
economies will probably imply a massive increase in transport fuel demand.
Energy demand is difficult to model for a number of reasons. One of these
is the complex pattern of substitutability and complementarity between energy
carriers and labour or capital. On the one hand, commercial sources of energy
have historically often been employed precisely to substitute for human labour.
In this task they have been complementary to certain pieces of capital
equipment. On the other hand, new vintages of capital or additional capital
equipment may well save energy, in which case capital (and sometimes labour)
Introduction
may be substitutes for energy. Either way the processes of substitution take a
long time since the demand for energy is a derived demand and its changes are
intimately tied to the structure and technology of society.
1.3 ENERGY SUPPLY AND THE ENVIRONMENT

On the supply side, the picture is no less complicated. New sources of energy
are often characterized by indivisibility and uncertainty. They generally require
heavy investments and often have a. long gestation lag from initial project to
actual production. The next section of the book deals with energy supply. The
appropriate modelling approach depends crucially on the type of supply under
consideration. For long-term supply of nuclear energy, a model built on
historical statistics clearly runs the risk of being irrelevant. Statistics are best
used to study the past behaviour of many independent agents under various
conditions. In this case the main agents are the state (and maybe public
opinion) and the future 6f the nuclear industry is obviously a political matter:
any attempt to make long-run projections must be based on a social, political
and maybe even sociopsychological analysis. This might thus be an area where
less formalized 'modelling' would be appropriate.
Although there is a common logic to supply modelling for all three fossil
fuels considered, there are also significant differences. Ronald Steenblik starts
in Chapter 6 by discussing the long-run supply of coal. Since coal is abundant,
exploration in the ordinary sense of the word is not strictly necessary at least
for the short and medium term. We know where the coal is, but as Steenblik
points out, coal is pretty heterogeneous and actual field development will
generally entail numerous surprises about quality, quantity, depth, overburden
and other factors that are decisive for costs. A particularly useful feature of this
chapter is that it specifically shows how a supply curve in principle should be
derived from the reserve and resource categories conventionally given in this
field. There are then numerous engineering cost and econometric cost estimation methods for producing such supply curves, but they all require detailed
geophysical/resource information and their performance is thus often impaired
by the deficient quality of available data.
Transport is another factor that is quite important for coal supply costs but
this factor turns out to be of even more importance to gas. For gas, exploration
does entail considerable uncertainty and, as Daniel Dreyfus points out in
Chapter 7, resource or supply modelling is difficult because much of the best
data is treated as proprietary information by the respective companies or
governments. Large parts of the globe are poorly explored and in fact likely to
remain so in the foreseeable future because the geographical mismatch between
supply and demand would make the transport costs prohibitive. The necessary
infrastructure investments of some projects are such that private enterprise
alone will find them difficult to finance. Future supply will thus concentrate
Energy and the economy
where governments are stable, and where innovative schemes for financing and
risk-management are found.
While transport costs are less important for oil we are once again concerned
with large investments (in exploration and development of fields) and thus
political stability is still very important. As economists we generally assume a
positively sloping supply curve. In fact, a positive supply response to higher
prices seems pretty obvious. However, as Victor Rodriguez Padilla shows in
Chapter 8, actual supply reactions can be quite different - particularly in the
Third World. Concentrating on the most fundamental, and the most long-run,
aspect of supply, namely exploration activities, Rodriguez shows that there is
in fact only a very tenuous link to oil prices. The complicated interplay between
host governments and oil companies pursuing different objectives under
different perceptions of risk and uncertainty implies that there may be very
long lags between rising prices and more exploration. In many countries no
such relation at all can be observed and some of the most crucial factors turns
out to be the political aspects of petroleum jurisdiction, the contractual
conditions and tax regimes of each respective country.
Another factor of increasing importance for the general costs of energy
supply is environmental effects. In Chapter 9 Lars Bergman illustrates the
differences between a partial and a general equilibrium approach to modelling
environmental control cost functions. The example chosen, that of sulphur
dioxide (S02), is of particular relevance for the competition between different
potential sources of energy supply. Bergman contrasts the 'true' (but hardto-calculate) general equilibrium cost, which takes into account all mechanisms
of adaptation within the economy, with various simpler cost functions. One
such function, that may lead to considerable overestimation of environmental
costs, assumes that energy demand is fixed at a certain level and then proceeds
to calculate the additional cost (for instance) of SOz reduction by fuel switching
without recognizing that higher energy costs should be allowed to induce a
lower overall level of energy use.
1.4 ENERGY AND THE ECONOMY
Recognizing that energy markets milst be analysed within the framework of a
complete model of energy-economy interaction leads us to a choice between
large and very complicated models on the one hand or interconnected groups
of models on the other. Chapter 10 by Capros, Karadeloglou and Mentzas
illustrates this point: in fact their contribution is almost two papers since the
Appendix is a very instructive survey of a number of energy and macroeconomic models used, in different combinations, by the European Community
(EC) to analyse economic aspects of environmental and energy issues.
The main body of the chapter itself is an application of the HERMES,
neo-Keynesian, multi sectoral model to the macro- and microeconomic policy
Introduction
issues related to falling oil prices in oil-importing countries. The main question
posed is whether domestic taxes should be increased to stabilize the domestic
product prices and, if so, how the resulting tax proceeds should be used. A
particularly attractive feature of this chapter is that it first explores the
analytical solutions of a simplified HERMES model and then goes on to present
the actual numerical scenario results using a full, empirical model for Greece.
HERMES is also one of the five major energy--economy models that has
been developed for France. In Chapter 11, Ghislaine Destais shows that
French modelling has a number of distinctive features: public sector involvement (as opposed to the significant proportion of consulting firm models in the
United States), and particular emphasis on the structure of the economy
(capacity utilization, post-Keynesian rather than neoclassical mechanisms, etc.).
Destais starts by introducing a useful typology with which she compares the
different models with respect to their level of aggregation, consumption and
production functions: 'bottom-up versus top-down', 'technical versus
econometric', supply functions and time horizon. She then points out that blind
reliance on anyone model is overly naive: many of their 'results' are merely the
transcription of the initial hypotheses adopted. A useful warning is provided
by an example in which three models analysing the same oil-shock give
completely different results.
Soviet energy planning in theory and in practice is the subject of Chapter 12
by Yuri Sin yak. Obviously this being the home of central planning, there was
originally considerable enthusiasm about the computerized planning of the
energy sector. As Sinyak writes, this sector was one of the first to be planned
in this way and the unique feature compared with Western modelling is the
web of models from perspective, federal, long-run projections, through the
national and sectoral 5-year plans and down to actual operative yearly or even
minute-to-minute load-management models for individual power stations. The
original belief that such complex systems could be optimized centrally has,
however, turned out to be an illusion. Interesting attempts at iterative solutions
of blockwise dis aggregated models are carried out but are hampered by the
poor quality, as usual, of data, the lack of workable prices and apparently by
the poor motivation of the researchers.
In the second half of his paper, Sin yak describes the contents of the current
Energy Plan and one is struck by the disparity between advanced methodology
on the one hand and poor quality of empirical evidence on the other.
Naturally, the usual difficulties are exacerbated by current uncertainties in the
Soviet Union. The forecasts are based on growth and system continuity (albeit
with big improvements in energy efficiency) but in reality the Soviet future will
depend decisively on the development of its economic system. Although much
mention is made of increased energy conservation through decreased materialintensity, the main focus of the Energy Plan is still supply. A supply, however,
that is coming up with many 'unplanned' surprises: scientific criticisms of
nuclear power, ecological barriers to hydro development, labour unrest in the
Forecasts and environment
coal fields, regional and transport infrastructure bottlenecks, etc. Thus the
actual future of, for instance, Soviet energy exports is at present very hard to
foresee.
1.5 INTERNATIONAL ENERGY MODELS, FORECASTS AND
ENVIRONMENT
Having started with individual market demand and supply models and then
looked at national energy-economy and forecasting models, we turn now to
world energy models and their applications. The next group of chapters, 13-16,
is intended to provide an overview of how world oil models work. Chapters 13
and 15 present two such models while 14 is a comparison of the elasticities
across models and 16 compares the latest set of forecasts using the models
participating in the 1989 International Energy Workshop. The final two
chapters of the book are applications of energy-economy models to environmental issues.
We start with Patrick Criqui who, in Chapter 13, presents two models from
the IEPE institute in Grenoble: the medium-term SIBILIN and the long-term
POLES currently being built. The SIBILIN is based on the simulation of
national energy balances (for all the major energy consumers - others being
represented in a simplified aggregate way). The models are resolved by
simulation, as opposed to optimization, and both have an ambitious degree of
disaggregation, by sector and geographically. Given the level of detail and the length of the time horizon - they are basically more 'bottom-up' than
'top-down' which does not, however, preclude a number of 'hybrid' traits in the
model, notably the use of econometric relationships and price elasticities in
many subsectors.
The philosophy behind these modelling choices is that the models are to be
used as a mental aid in scenario analysis: detailed balances force the researcher
to be consistent in his scenarios but the evaluation of the many decisive
political processes is too complex to be left to any optimization or other
algorithm. This is nicely illustrated by an analysis of possible OPEC strategies
of price defence or defence of market shares. The POLES model promises to
be an exciting extension since it will keep and even increase some aspects of
the demand-side detail of the SIBILIN while endogenizing prices and supply
and tuning up the model for long-term environmental analysis.
Most forecasters would agree that models cannot be allowed to do all the
forecasting for them - the individual researcher must assume his or her
responsibility. The unfortunate implication of this is that when forecasts are
different it is often hard to disentangle which of the differences depends on the
models used and which on different hypotheses, assumptions or other adjustments made by the researcher. Thus the efforts of the Energy Modeling Forum
are invaluable since they allow for systematic comparison using common
Introduction
hypotheses. In Chapter 14, Dr Huntington has found an extraordinarily

succinct way of summarizing the differences (as well as a certain degree of
consensus) between models by concentrating his presentation on demand and
supply elasticities.
Most of the models analysed by Hill Huntington are again recursive
simulation models in which three groups of agents - oil consumers, OPEC and
non-OPEC oil producers - have different response functions determining the
logic of scenario build-up. But two of them (DFI-CEC and ETA-MACRO, see
Chapter 18) are optimization models requiring some form of perfect foresight
by at least one agent.
All 11 models were represented 10 the Energy Modeling Forum (EMF-11)
comparison, and the elasticities were inferred from the differences in outcome
between high/average growth rate scenarios and between fiat/rising price
scenarios respectively. No changes in domestic taxes on petroleum products
were assumed (such as are being discussed with respect to environmental
effects). All the scenarios show increasing demand on OPEC oil though there
is quite a difference between the constant price/high growth and the rising price
(doubles by 2000) combined with low growth scenario.
The average results for demand were short-run price elasticities of -0.1 and
long-run elasticities of -0.4 which, considering the differences in crude and
product price, fit in reasonably well with what was found in Chapter 5 for
gasoline. Income elasticities averaged 0.8 but there were two separate groups
of models, one with values around 0.6 while many others had unitary income
elasticity. Generally the latter assumed some form of Autonomous Energy
Efficiency Increase (AEEI). Average supply elasticities started very low for the
short-run (0.03) but rose to around 0.4 in a long-run perspective.
Certain patterns of elasticities between models can be observed. Some
models have low elasticities of all types, some high. The World Oil Market
Simulation Model (WOMS) is an example in which both inferred price
elasticities of demand and supply are fairly low while the income elasticity of
demand is close to unity, thus implying that this model places some emphasis
on trends in autonomous technical development. In Chapter 15 WOMS is
described in detail, both the workings of the model and the estimation of its
main functions. The model is then used to carry out a fairly similar analysis of
the same question of OPEC strategies as that discussed by Criqui in Chapter
13.
In Chapter 16 Alan Manne and Leo Schrattenholzer summarize and discuss
the projections from the different models participating in the 1989 lEW. Over
the period 1985~2010 the median projections follow a 3-2-1 rule: economic
growth of 3% per annum, energy consumption increase of 2% and just 1% for
oil consumption. One way of explaining these figures is to say that there is
basically a unitary income elasticity of energy complemented by an AEEI of
1% per year. This explains the 2% energy growth, and finally a price-induced
substitution effect explains lower growth for oil. Median projections for oil
Forecasts and environment
price imply a yearly growth rate of 1.5%, suggesting a price elasticity of about
-0.67 which is on the high side but not wholly inconsistent with the values
found in Chapter 14. An interesting point to note is that while all forecasts for
oil consumption are fairly consistent, the price forecasts vary widely, which
shows that people still basically think in terms of physical trends and the role
of prices is secondary at best. Average price projections in the lEW polls show
that researchers seem to believe in rising real prices (by 1.5 or 2% per annum).
Whether this reflects expectations as to costs or other factors is not clear but
it may be yet another example of Hotelling's considerable influence on general
thinking in this area.
One of the gloomier aspects of these polls is their implicit median forecast
for global carbon emissions, which are expected to increase by more than 50%
by the year 2010. This is in sharp contrast to the rather drastic reduction
which, according to many scientists, would be necessary to moderate climate
change. In Chapter 17, Gunter Stephan illustrates how the level and distribution of economic costs of reducing emissions (either by emission charges or, in
his case, by imposing standards) may be calculated using a Computable
General Equilibrium (CGE) model. One distinguishing feature of this model is
its Neo-Austrian approach to capital formation and one of its results is similar
to the conclusions from Chapter 3: that in vintage type structures, the costs of
environmental improvements depend on the rate of turnover of capital equipment. Stephan also shows that even if total costs are not all that high, the
impacts on income distribution may be quite considerable.
Returning to the case of carbon in our final chapter, we do unfortunately
know that in this particular case emission reductions are expensive indeed. We
choose to conclude the book with this chapter not only because of the topical
nature and possible importance of the issue itself but also because its analysis
in a very challenging way brings together almost all the various aspects of
resource availability, supply cost, demap.d elasticities, model structure and
forecasting discussed in previous chapters.
In order to analyse the cost (in the United States) of a 20% reduction in
carbon emissions, Manne and Richels use the ETA-MACRO model. This, as
mentioned above, is not a recursive simulation model but one that builds on
intertemporal optimization. As such it is particularly appropriate for a discussion of long-run costs assuming a far-sighted choice of policy. This is just a
first, although very important step. As the authors point out, a worldwide CGE
model would open up also the possibilities of analysing the distribution of costs
among different nations of various policy instruments such as carbon quota
rights. In order actually to get a handle on costs, the authors make a series
of assumptions on the costs and carbon emissions associated with various
alternative methods of producing available energy. Similarly, assumptions are
made on growth in the economy and specifically in energy demand. In order
to minimize the risks (discussed by Bergman in Chapter 9) of using an all too
partial supply-oriented cost function, Manne and Richels analyse the cost of
10
Introduction
carbon emission restrictions under various scenarios. Making optimistic assumptions as to technical progress (AEEI of 1%) as well as optimistic
assumptions as to the cost and availability of alternatives (such as coal
gasification with CO 2 removal and deposition or advanced nuclear power) the
discounted cost from the year 1990 to 2100 would be $0.8 trillion. With the
most pessimistic assumptions this cost would be $3.6 trillion.
While this obviously is an enormous cost, it in no way implies a reduction
in present consumption levels. Only their rate of growth is affected. In the $3.6
trillion case, consumption from 2030 to 2100 would still grow but lie roughly
5% lower with the carbon limit than otherwise. The required policy can be
illustrated in terms of the necessary carbon tax of around $250 per tonne. It is
not easy to judge whether these costs would be strictly motivated by the
potential damage of global warming since the scientific knowledge on the latter
is still too limited. The costs could of course be seen as an insurance premium.
(particularly if we consider the opinion of those who argue that we actually
need even more drastic cuts in emissions). It is anyway clear from the
magnitude of these costs that we still need to further our understanding by
research into increased efficiency, both in the supply and demand technologies
of the energy sector and their interconnection with the rest of the economy.
REFERENCES
Brooks, H. et al. (1979) Energy in Transition:1985-20JO. Committee on Nuclear and
Alternative Energy Systems, National Academy of Sciences, Washington, DC.
Hafele, W. (ed.) (1981) Energy in a Finite World. IIASA, Laxemburg, Austria.
Landsberg, H. H. et al. (1979) Energy: The Next Twenty Years, Ballinger, Cambridge,
Mass.
Manne, A. S., Richels, R. G. and Weyant, J. P. (1979) Energy Policy Modelling: A
Survey. Operations Research, 27, 1.
Schurr, S. H., Netschart, B. c., Eliasberg, V. F. et al. (1979) Energy in America's Future.
Johns Hopkins University Press, Baltimore, Md.
Stobaugh, R. and Yergin, D. (1979) Energy Future. Random House, New York.
2
Forecasting industrial
energy use
Gale A. Boyd
2.1 INTRODUCTION
The term 'model' is a fragile one. To some it means a theoretical methodology
with equations and definitions, to others it is a simple mental reasoning
process, and to still others a computer with data bases and algorithms. In this
chapter we consider computer-based models that are used to forecast energy
use, but we will only describe their theoretical methods and underlying mental
reasoning processes. The technical descriptions of the algorithms and data
bases are left to their creators. We also consider only models whose primary
focus is forecasting. These models are often the arms of consulting firms and
government agencies. This paper draws much of its information base from the
Energy Modeling Forum Study 8 (EMF-8) on Industrial Energy Use and
Conservation. Some example forecasts from this study are presented as well.
Other forecasting models that are not included in EMF-8 have been added to
the list of EMF-8 participants for presentation and discussion. This paper is
not intended to be a comprehensive review and the author apologizes in
advance to those modellers whose modelling activities have been overlooked.
It may well be said that models, like certain mathematical objects, may be
completely described by a sequence. In this case the sequence is the questions
or problems that have been posed to the modeller. That sequence may
converge until the model's approach is clearly defined or the model may still
undergo evolutionary changes as new questions are posed and new information
in the form of data and methods is incorporated. The questions may determine
the number and type of industry sectors (i.e. aggregation level), the time
horizon (short, medium, or long), or the linkages the model has to energy
supply or economic feedback. The common, underlying question that all
industrial energy models face is how to treat the diversity that typifies this
sector; what aspects of industrial energy decision-making to abstract from;
which parts to overlook; and which parts to model in detail. All of the models
12
Forecasting industrial energy use
start with a few common elements, energy prices and industrial growth. From
that point on they diverge on the path finally to arrive at some outlook for the
amount and composition of energy consumption at some future date.
2.2 FACTORS THAT INDUSTRY CONSIDERS IN
MAKING ENERGY DECISIONS
Trends in industrial energy use are affected by more than just cost; this is
important to the behavior of computer models for industrial energy use. The
engineering, or 'process', models are often based on the life-cycle cost of a
generic application and include parameters to avoid extreme ('knife edge')
switching to the fuel that the model calculates as the least-cost. A market share
approach yields a distribution of different fuels, with the distribution sensitive
to relative costs. A dynamic approach is necessary to reflect momentum and
capital stock turnover rates in the process of adjusting to better fuel choice
options.
The following are some of the factors often cited by industry in the United
States as affecting fuel choices for new capacity or fuel-switching decisions for
existing capacity:
Fuel capability, condition and age of existing equipment (e.g. boilers,
furnaces, coal handling equipment).
Site-specific constraints such as availability of land for a coal pile and boiler
firing type. This was cited numerous times in the 1975 Major Fuel Burning
Installation (MFBI) survey (Alter, 1978).
Direction and magnitude of change in fuel prices, e.g. world oil prices,
OPEC's ability to stabilize prices, effects of natural gas deregulation and
effects on delivered coal prices of railroad deregulation under the Staggers
Act.
Availability of fuel.
Cost of capital (which is very important for capital-intensive investments and
for discounting future fuel savings), financial position of the firm, and
priorities for types of investments within the firm. The cost of capital for the
firm depends on the risk the firm faces in the marketplace.
Regulatory uncertainty (for example, related to the enforcement of the Fuel
Use Act 1978), future pollution regulations such as the Industrial New
Source Performance Standards for new boilers or acid rain controls, and the
future regulatory and institutional environment for cogeneration.
Reliability and performance of new technologies such as the atmospheric
fluidized bed, coal/water mixture (CWM), cogeneration and combined cycle.
Downtime to retrofit to coal or CWM, availability of replacement energy
during downtime, and resulting boiler derating.
Coal experience. (The relevance of coal experience is unclear for long-term
How models deal with fuel choice
13
forecasts, because lack of experience can eventually be overcome if coal is

fundamentally a better fuel option. Other 'hassle factors' are sometimes
mentioned in association with the use of coal.)
Scale of operation.
These uncertainties, as well as the volatility of energy markets and policy
over the past ten years, all contribute to a cautious, wait-and-see attitude by
industry. In a previous paper, the value of this attitude was estimated
quantitatively (Hanson and Bauer, 1981). Models can incorporate these factors
implicitly or explicitly.
2.3 HOW MODELS DEAL WITH FUEL CHOICE

Two basic model types are used to predict industrial fuel choice: process and
econometric. However, there is a grey area in modelling when both techniques
are applied. When some model uses a significant portion of both methods, the
term hybrid model is used. Models may use a bottom-up approach, computing
each component of energy demand (fuel choice for each sector or service
demand category) and adding them up to arrive at a total. An alternative
approach is 'top-down'. In this case a total demand is computed and then
shared into fuel types or other categories. Significant opportunities for a hybrid
approach are available here as well. For example, electricity demand may be
determined at a 'bottom-up' sectoral or service demand level, but the decision
to cogenerate the electricity may be based on some 'top-down' sharing
procedure. The characteristics, advantages and disadvantages of process and
econometric modelling are described below.
2.3.1 Process models
These models explicitly list and characterize (by cost, performance etc.) all the
alternatives in a fuel choice decision. Once each alternative is characterized, the
model decides which is most economically attractive. The selection criterion is
usually lowest levelized cost. For example, the model is told how many tons of
steel must be produced; it must then decide the most attractive process and
energy-using technologies to employ. This process is inherently 'bottom-up' in
its approach. The listing of all of a company's alternatives and all of the factors
affecting the decision is enormously complex. In this lies both a strength and
a weakness. The strength is that new and emerging technologies may be
included in the list of alternatives, so that the model can be 'forward looking'.
However, process models must also compromise since the selection of a single
'best' technology is not always the result that occurs in the real world. Process
models can use several techniques like:
Replacement of 'preliminary market shares' based on cost minimization with
14
'final market shares' that account for random (i.e., other unobserved) factors
through statistical sampling .
Behavioural lags used as surrogates for the many factors not considered
explicitly .
Random distributions for costs and other equipment performance characteristics.
Suppose a model estimates a 'preliminary market share' of 10% for a
technology or process. However, in 1985 this process may only have a 5%
market share. Without behavioural lags, the process model would predict
immediately (in its next forecasting year, say 1990) that the market share would
be 10%. Behavioural lags are an ad hoc method for adjusting the market share
from its current value (e.g. 5%) to its predicted long-run value (defined above
as 'final market share'), which is 10%. For example, with adjustments for
behavioural lags, the market share would gradually approach the predicted
10% share, but not until the year 2000. In the marketing literature these
behavioural lags are studied extensively for individual products. The length of
the behavioural lag varies considerably by product, but can be as long as 10-15
years.
A behavioural lag is defined as the time between the availability of a new
competitive process and its adoption. Behavioural lags are surrogates for many
undefined or unknown factors that affect choices of fuel type and technology,
such as some of those listed above. They also reflect the desire of most firms
to follow, rather than lead, in market innovations. No firm, unless it is the only
beneficiary, wants to be the first to work out the 'bugs' of a new technology.
Random distributions of various costs of energy-using equipment can be
used to determine market shares of technology choice, rather than a 'winner
take all' knife-edge type of decision. These market shares can be obtained by
convolution of the cost distributions or, more commonly, via Monte Carlo
type techniques.
Process models may also use econometric (statistical) techniques to project
energy choice, particularly fuel choice. The distinction between process and
econometric is very fine in this case. However, because the projection is made
at the 'process' level rather than the 'industry' level, the model is still termed
'process'.
2.3.2 Econometric models
Econometric models predict fuel choice by statistically analysing historical
relationships. They predict energy levels (or shares) from a given set of data on
energy, capital, labour prices and industrial output. Theoretically, econometric
models can capture the factors affecting fuel choice by assuming that those
factors continue into the future. That is, the models assume that our best guess
of the future is a continuation of past relationships. (Process models, in their
H ow models deal with fuel choice
15
attempts to represent the fuel choice explicitly, invariably omit factors and
hence produce biased results. However, process models may incorporate
econometric-like methods.) The US Department of Energy's Energy Information Administration feels that process models tend to be overly optimistic on
coal use and conservation (improvements in energy intensity).
Econometric models may be classified into two general categories; ad hoc
specifications and optimization-based. Ad hoc models use functional specifications that relate energy use (or shares) to a set of explanatory variables that
impact energy use patterns. These variables usually include energy prices,
industrial activity, as well as other variables. Some popular ad hoc specifications are linear, log linear and logit. Optimization-based models assume some
type of underlying production process and maximization behaviour and derive
the relationship between energy input demands and various prices and activity
variables. The basic form this approach takes is based on duality. Two specific
examples of the ad hoc vs. the optimization approach are examined in a
methodological experiment in Smith and Hill (1985). Examples of these
approaches and the background are presented below.
Ad hoc econometric models specify a simple functional form and estimate
the relationship between energy use (total level, shares or ratios) and relaxant
variables like energy prices, prices of competing energy forms, industrial
activity, specific technology variables and capacity utilization. Simple
econometric specifications may be used to derive short-run vs. long-run
relationships. Use of lagged variables, both dependent and independent, is a
possible specification. The emphasis is on the 'reasonableness' of the coefficients in terms of sign and magnitude and on goodness of fit. In other words,
'correct' signs for own-energy price coefficient (negative) and for cross-price
(competing) energy coefficients (positive) are the most important. Although the
functional form is usually linear in parameters for convenience, another
popular ad hoc approach is the use of the multinomial logit function.
The multinomiallogit function is based on a probabilistic choice model. In
other words, it assumes that the prices of energy and other institutional and
technological variables affect the decision-makers' probability of choosing an
energy form. This model can be used to derive a function that relates the ratio
of the share of two or more competing energy forms to the prices. Other
variables that affect the probability of choosing one energy form over another
can also be included in the model.
Optimization models assume a general form for an underlying production
process and that the decision-maker is cost-minimizing (or profit-maximizing).
Optimization models usually rely on duality theory and Shephard's lemma
(Shephard, 1970) that the optimal demand for an input (X) is the first
derivative of the cost function that is dual to the assumed production function.
Studies that examine total energy use (E) or total fuel and electricity use (EF)
often include capital, labour and materials (KLM) in prices included in the cost
function (C).
16

Total cost = qu, P K , PL , PE, PM)
or =
Demand for X =
qu, P K , PL , PE, P F , PM)

bC
uP x
The demand system may be of energy types only, by assuming the energy
inputs are weakly separable or the system may include labour, materials and
capital inputs. Studies that seek to relate the choice of energy form, e.g. natural
gas vs. fuel oil vs. coal, usually assume that the energy form is weakly separable,
for example some aggregator function (G) is chosen.
This is often done due to data limitations. Whether materials is treated as an
input depends on the assumption of separability of material from other inputs.
If materials is separable then value-added is used as the measure of industrial
output. If not, then total shipments may be used to measure industrial output.
Optimization models usually rely on an assumed form of the dual cost function
and derive a system of demand equations via differentiation. Common forms
of the cost or aggregator function are the so-called flexible functional forms,
which are based on second-order Taylor series expansions in output (U) and
prices (P). Specifying the function to be the natural log gives the TRANSLOG
form (1); specifying it
n
C=
L etJ(P i)+ L L
i= 1
+ buf(U) +
fJiJ(PJf(Pj)+"'hg(T)+
i=lj=l
n
L bJ(U)f(PJ
L M(T)f(Pd
i= 1
(1)
i= 1
to be the square root gives the Generalized Leontief form.l Shephard's lemma
applied to these forms yields cost share equations that are linear in the
parameters, hence estimatable with commonly available econometric packages. 2
Many empirical studies show that energy and capital in the aggregate have
tended to be complementary. When energy prices rise, there is less incentive to
automate, because automation requires energy-using capital. Similarly, when
capital prices fall we tend to invest in labour-saving devices, which increases
total energy use. There are specific conservation technologies for which capital
and energy are substitutes, but overall the complementary relationship (e.g.
automation) has dominanted the substitutability relationship (e.g. conservation). For example, see Solow (1987) for a discussion.
1 Many variations of these forms involving technology (time) trends, homotheticity assumptions etc.
are possible. Only the simplest example is shown.
2We will not discuss the additional economic or econometric applications in detail.
Selected models of industrial energy use
17
Econometric models may include a technology (time) trend to capture other

factors that increase energy use, such as continued electrification and the
long-run historical trend in this century toward automation and energyintensive capital stock, because these models naturally identify trends and
continue them. Similarly the trends that contribute to lower energy use may
be included in a technology trend as well. Optimization models that treat the
technology trends differently for different inputs are called factor-biased. For
example, the factor-biased technical change in electricity may be electricityusing while the technical change in labour or fossil fuels may be input-saving.
The performance of econometric models depends on the crucial assumption
that the analysis period represents the future. For example, if we ignore output
(U) and consider g(T) as a function of technology trends the signs of Yi
above determine the direction of technical change bias.
2.3.3 Similarity between process and econometric models
Both process and econometric models have the same purpose: prediction of
industrial fuel use, by fuel type. They both consider the same main factors:
Industrial growth (measured by shipments [gross output], value-added or
industrial production indexes).
Fuel prices.
The underlying production process.
Behaviour of the decision-maker.
2.4 SELECTED MODELS OF INDUSTRIAL ENERGY USE

The remainder of this chapater describes a number of industrial fuel-choice
models. Some of these models were participants in a study performed by the
Energy Modeling Forum. These models, by type, are:
Process models
ISTUM-l
ISTUM-2
PILOT
Econometric models
INFO RUM
PURHAPS
Hybrid models
ORIM
A few additional models that have come into use recently or that simply did
not participate in the EMF-8 study are:
18
Process models
ICE
Econometric models
AEO-PCIM
INDEPTH
HJGEM (Hudson/Jorgenson General Equilibrium Model)
HKGEM (Hazilla/Kopp General Equilibrium Model)
Hybrid models
INRAD
The Industrial Sector Technology Use Model (ISTUM-1) (US Department
of Energy, 1979) is a set of engineering and economic modules that integrate
information on industry characteristics, fuel prices, economic forecasts and
historical energy demands to develop predictions of industrial energy use. This
model is designed to project market penetration of energy technologies in the
industrial sector 20-30 years into the future.
The ISTUM-1 model is designed to test alternative scenarios and is used
frequently for various types of sensitivity analyses. The model can evaluate the
commercial viability of energy technologies in the industrial sector and test
their sensitivity to changes in macroeconomic activity, fuel price fluctuations
and new environmental regulations. It can analyse the fuel choices of the
industrial sector, including changes due to various economic and policy
scenarios.
The second Industrial Sector Technology Use Model (ISTUM-2) (US
Department of Energy, 1983) is the result of a 4-year effort managed by the
Energy Productivity Center at Mellon Institute in cooperation with Energy
and Environmental Analysis, Inc. An analytical framework was developed to
address a wide range of policy issues related to industrial energy use. This
model is primarily a tool for determining how to provide energy services to the
industrial customer who bases purchase decisions primarily on cost.
The ISTUM-2 model produces output that provides much information on
industrial fuel and technology use. The model provides:
Forecasts of industrial energy demand by service category, including industrial growth as well as changes in fuel mix.
Projected fuel demands by ten federal regions.
Projected market penetration for various technological options to improve
energy productivity.
Evaluation of the impact of government policy on industrial fuel use.
Behaviour analysis that identifies, on an industry-specific basis, the factors
most important to industrial decision-makers.
The ISTUM-PC model (Jaccard and Roop, 1990) is based on the ISTUM-2
code, but ported to the PC with data updated primarily for Canada. This
model has substantially the same structure, but is presumably more 'accessible'
than its mainframe predecessor.
19
The Oak Ridge Industrial Model (ORIM) (Reister et ai., 1980) was initially
developed at Oak Ridge National Laboratory to help the Energy Information
Administration prepare mid-term (5- to 20-year) projections of industrial fuel
demand. The model forecasts demand for six industrial fuels: distillate fuel oil,
residual fuel oil, liquefied petroleum gas (LPG), natural gas, coal and electricity. Demand is projected for 12 industrial sectors (nine manufacturing and
three other) in each of the ten federal regions in 5-year increments. The model
bases its projections on both statistical and engineering data. Types of forecasts
include energy demand by fuel, region, industry, energy service and vintage of
capital stock.
The following information is needed for input: (1) costs of capital, labour and
materials for each type of energy technology (for example, natural gas combustor, coal boiler, etc.) and (2) prices for the six fuels. Because fuel prices
sometimes change considerably from region to region, decisions based on fuel
cost are also simulated separately for each of the ten federal regions. Outputs
from the model are regional, industry-specific fuel and electricity projections
for each five-year period and for four energy services: heat, steam, mechanical
drive and uniquely electrical service.
The Stanford Planning Investment Levels Over Time (PILOT) Energy/Economic Model is a large~scale dynamic programming model that
calculates the time path of investments, production, consumption and imports
and exports so that the prices and quantities of all commodities make supply
and demand equal for the commodities. Policy issues addressed by PILOT
include the scheduling of various energy technologies to be built and used.
pollution abatement equipment to be installed and the nature and extent of
conversion to equipment types that use energy more efficiently. The model was
originally built to study interaction between the energy sector and the
macro economy.
Energy demands of the economy (industrial processing, consumers, exports,
government needs) are met in five energy forms: coal, crude oil, oil products,
natural gas products and electricity. The model contains detailed descriptions
of energy technologies; explicit descriptions of the depletion processes for oil,
gas and uranium; dynamics of capital formation and resource extraction;
accounting for trade with the rest of the world; and national consumption
trends.
The INFO RUM Long-term Interindustry Forecasting Tool Model is an
input-ouput model developed in the University of Maryland's project for
interindustry forecasting. The model determines constant-dollar output for 78
sectors of the economy, consistent with given levels of final demands. These
final demands are forecasts of INFO RUM's macroeconomic model. Inputs to
the macromodel include forecasts of population growth, government expenditure, money supply and primary energy and material prices. The INFORUM
model also estimates fuel use by industry. Unlike models that represent only
the industrial sector, INFO RUM ensures complete supply and demand balance in energy use.
20
The Energy Information Administration (EIA) within the US DOE created

the Purchased Heat and Power System (PURHAPS) (US Department of
Energy, 1983) to project industrial energy demand by year for six major fuels
(electricity, coal, natural gas, distillate oil, residual oil and liquefied petroleum
gas). The PURHAPS model is econometric - its estimates are based on
historical data.
Forecasts are made for 17 manufacturing industries and eight non-manufacturing sectors at national, federal region and state levels. Industrial energy
demand is derived as a function of industrial production, current fuel prices
and price response lags (due to slow turnover of capital stock and other sources
of momentum). Other variables accounted for in predicting energy demand are
prices of labour, capital and materials; effect of economic recession; time trends;
and difficulties in using coal in small-scale industries. The basic equation
structure is hierarchical. Total energy demand is determined first, then a series
of energy shares are computed based on logit function estimates, e.g. energy is
shared into electricity and fossil fuels, fossil fuels are shared into coal vs. oil/gas,
etc. This makes PURHAPS a top-down type of model. The PURHAPS model
has been used extensively as a component of the Intermediate Future Forecasting System (IFFS), the EIA model on which the Annual Energy Outlook (US
Department of Energy, annual) was based until 1988 when a new model
AEO-PCIM was developed.
At the core of PURHAPS are two models: the manufacturing model and the
agriculture, construction and mining model (henceforth called the ACM
model). The manufacturing model produces econometric forecasts of energy
demand for 17 manufacturing industries at the two-digit Standard Industrial
Classification (SIC) level (all two-digit SIC industries are included except for
refineries; chemicals and rubber are combined as are textiles and apparel). The
ACM model predicts demand for eight separate non-manufacturing sectors
that have accounted for the bulk of growth in industrial electricity for the
period 1974-81.
Among the major inputs for PURHAPS are prices for each fuel by state. For
each industrial sector the real gross output by state and for the United States
as a whole is input. Prices of capital, labour and intermediate materials are also
input for each industry nationwide. Major historical data sources for model
inputs are the Bureau of Labour Satistics' Time-Series Data for the InputOutput Industries (published annually) and the Annual Survey of Manufacturers
(Bureau of the Census, annual)
PURHAPS is no longer used by EIA for forecasting. In its place is the
Annual Energy Outlook PC Industrial model (AEO-PCIM). This model is
based on a similar level of sectoral aggregation as PURHAPS, but without the
state-level detail. The structure of the model is ad hoc. Individual demand
equations are estimated and forecast without statistical or structural interdependence. For example, the oil equation would include industrial growth, oil
prices and gas prices, but is forecast without regard to the forecast for gas
21
consumption. Total energy is bottom-up, rather than top-down. Unlike PURHAPS, no technology (time) trends are used in AEO-PCIM.
The Industrial Combustion Emissions (ICE) model (US Environmental
Protection Agency, 1988) was developed to forecast boiler fuel choice and
resulting pollution levels for the National Acid Precipitation Assessment
Program (NAPAP). While it does not cover the entire industrial sector and
has emission forecasting as its stated goal, the model uses several interesting features to forecast boiler fuel choice, i.e. coal vs. oil or gas. The model
incorporates process models features, but also uses econometric estimates. The
engineering factors that ICE considers are boiler size, utilization and the cost
of requisite pollution control devices. This data is used in addition to the level
of industrial boiler fuel demand by seven industry categories in each state and
the corresponding fuel prices.
The Industrial Regional and Energy Demand (INRAD) model (Boyd,
Kokkelenberg and Ross, 1990) was developed as part of the Integrated Model
Set for NAPAP. INRAD is a set of econometric equations that forecasts
electricity and fossil fuel demand. The model is based on the Generalized
Leontief functional form with modifications to allow for capacity utilization
effects in the energy equations. Model estimates are based on national level
data for eight industry groups, but are then benchmarked to state level energy
use and energy prices for 20 associated industries. The eight industries included
in INRAD are food, textiles, paper, chemicals, stone, clay and glass, steel,
aluminium, and all others. Forecasts of state level industrial activity and prices
indices for electricity, fossil fuel, labour, capital and materials are required to
drive INRAD. INRAD also incorporates penetration of two major electricityintensive technologies, thermomechanical pulping and electric arc furnaces in
its forecasts. This makes this model a basically econometric model that
incorporates some process model features.
The Industrial End-use Planning Methodology (INDEPTH) is a series of
energy models developed by the Electric Power Research Institute (EPRI,
1990). It incorporates three levels of design. The 'top' level is based on
econometric model estimates. The 'middle' level is a process engineering model.
The 'bottom' level is an energy services model that operates at the level of the
local utility service area. INDEPTH is estimated at the national level for
202-digit SIC industries using several different flexible functional forms that the
user may choose from. Specifications include Generalized Leontief and
TRANS LOG in both a variable cost and KLEFM specification. The default
choice of functional specification differs from industry to industry. This is
somewhat of a departure in econometric models, which usually use the same
specification for all industries. The energy forms the INDEPTH models use are
the same as INRAD; electricity and fossil fuel aggregates. Regional detail has
been obtained in INDEPTH application within the Electric and Gas Utility
Modelling System (US Environmental Protection Agency, 1990).
Two general equilibrium models, one based on the work of Hudson,
22
Jorgenson and others (Hudson and Jorgenson, 1974, Jorgenson and Wilcoxen,
1990) (KJGEM) and the other based on the work of Hazilla and Kopp (1990)
(KHGEM), both include within their structure energy demand equations based
on TRANSLOG cost-share functions. For each industry represented, a KLEM
cost function is estimated, E representing an energy aggregate. This energy
aggregate is broken into four energy forms, oil, gas, coal and electricity, by a
separable TRANSLOG aggregator subfunction. Unlike the other models
above, forecasting and analysis with these models allow for energy substitution
as well as feedback between the industry cost structure, energy prices and
industry growth. Both models represent 35 industries, corresponding to roughly the 2-digit SIC level of aggregation.
The key inputs and outputs of the models are summarized in Table 2.1. All
of the models require some degree of detail on industrial sector activity inputs
(with corresponding detail on the output side). The industrial detail is usually
at the 2-digit SIC level, but some models, ISTUM-2 and INRAD, require 3or 4-digit detail for some large energy-using sectors. ISTUM-1, PILOT,
INFORUM and AEO-PCIM are national-level models with varying degrees
of sectoral input detai1. Forecasts of national-level industrial activity, energy
price etc. are usually more readily available and more easily adaptable to
scenario analysis. The remaining models require regional detai1. INRAD
includes some simple sharing routines to go from the national level to the
regionalleve1. Regional forecasts are less available from typical macroeconmic
forecasting services. This makes scenario analysis more difficult and timeconsuming if consistent regional scenarios are to be constructed. The process
models require future outlooks for the performance of important process
equipment while the econometric models usually require forecast of capital,
labour and materials prices. Both of the inputs create their own problems in
forecasting. The ICE model accepts one input that is different from the other
models; environmental regulations and pollution control costs. As environmental controls become more important determinants of fuel and process
choice as well as overall energy use, more models will need to incorporate this
dimension.
Other characteristics of the models are summarized in Table 2.2. Almost all
of the models are long-run, with only two econometric models using ad hoc
treatment of short-run capacity utilization effects. The models are all FORTRAN code except ISTUM-1 and ISTUM-2, which are written in DYNAMO
and APL, respectively. None of the models include energy/economy interactions like the impact on higher energy prices on cost, hence industrial demand
or growth except PILOT, INFO RUM, HJGEM and HKGEM. PILOT is an
optimization model for the whole economy, including the energy sector, and
INFORUM is an 10 model of the economy with a separate 10 'skirt' table to
account for fuel choice in various energy services. HKGEM and HJGEM are
general equilibrium models, which solve for an internally consistent set of
market prices and production for all industrial outputs, energy products being
HKGEM
HJGEM
INRAD
ORIM
INDEPTH
AEO-PCIM
PURHAPS
INDEPTH
(Level 1)
INFORUM
PILOT
ICE
Regional industrial growth by sector, energy, capital, labour and

material prices
Regional industrial growth by sector, energy prices, energy equipment
performance
State level industrial growth by sector, energy, capital, labour and
material prices
Population growth, government and current accounts deficit, tax rate
(industrial growth and prices are endogenous)
Population growth, government and current accounts deficit, tax rate
(industrial growth and prices are endogenous)
State level industrial growth by sector, energy, capital, labour and

material prices
Industrial growth by sector, energy prices
Industrial growth by sector, energy prices, energy equipment

performance
Regional industrial growth by sector, energy prices, energy equipment
performance
Industrial growth by sector, energy equipment performance
State level industrial boiler fuel demand by sector, energy prices,
environmental regulations, pollution control costs and performance
Regional industrial growth by sector, energy, capital, labour and
material prices
GNP growth, money supply, other exogenous macroeconomic inputs
ISTUM-1
ISTUM-2
Inputs
Inputs and outputs of models
Model
Table 2.1
Regional energy demands and fuel choice

by industry
Energy demands and fuel choice by industry
State level energy demands and fuel choice
by industry
Regional energy service demands and fuel
choice by industry
State level electricity and fossil fuel
demands by industry
National level electricity, coal, oil and gas
demands by industry
National level electricity, coal, oil and gas
demands by industry
Federal region energy service demands and

fuel choice by industry
Energy service demands and fuel choice
State level boiler fuel choice by industry
Energy service demands and fuel choice
Outputs
Demand
Both
Fuel choice only
Yes
Yes
No
No
Yes
Yes
No
No
Yes
No
No
No
ISTUM-2
PILOT
ICE
INDEPTH
INFORUM
PURHAPS
AEO-PCIM
IN DEPTH
ORIM
INRAD
HJGEM
HKGEM
All 2-digit industries
All 2-digit manufacturing SIC

Process and energy service (for
major 2-digit sectors)
Major 2-digit and 3-digit manufacturing SIC
All 2-digit industries
Major 2- digit manufacturing SIC

All 2-digit SIC (mapped into BEA
10 codes)
Process and energy service (for

major 2-digit sectors)
Process and energy service
SIC 20, 22, 26, 28, 29, 33 and 'other'
Process and energy service
Disaggregation
Mansur and
Whalley (1984)
Hazilla and Kopp
(1986)
Census
Census ECDB,
MIPD
NEA and other
Census annual
survey
NEA
Other
NEA
ECDB
Other
ECDB and other
ECDB and other
Data base
'Usually run in some type of integration with other energy supply models.
blncludes treatment of short-run capacity utilization effects.
CThe current applications of the model do not include forecasting, but model would be capable of near- to long-run forecasts.
Both
Both
Demand"
Demand
Demand
Demand"
Demand"
None c
Long run2010
Long run2010
Long run
Long run2030
Short run
Long run2010
Long run b 2010
Long run2010
Long run
Long run2010
Long run b 2030
Long Run
Demand
Yes
ISTUM-1
Demand
Both
Time frame
Supply/demand
EMF-8
Model
Table 2.2 Other model characteristics
Yes
Yes
None
None
None
None
None
None
Yes
Yes
None
None
None
Economic
feedback
Conclusion
25
only one class of industrial outputs. These same four models include the supply
side explicitly in their forecasting approach, while PURHAPS, AEO-PCIM
and INRAD are usually run as part of some larger integrated system that
includes energy supply models.
Data sources for the major historical energy data bases are the Energy
Consumption Data Base (ECDB) (EEA Inc. 1977) for the process models and
either the Census Annual Survey, Fuels and Electric Energy Consumed
(CENSUS), or the National Energy Accounts (NEA). Some supplementary
sources and proprietary data-bases are also used. These are listed only as
'other'. They may be trade association data or results from other models. For
example ICE uses ISTUM-2 to calibrate its base year energy-use data which
is derived from the American Boiler Manufacturers Association boiler sales
data. INRAD, ISTUM-l and ISTUM-2 all use Steel and Paper Trade
association data as well as other sources.
2.5 CONCLUSION
As shown above, several of the models were involved in a series of workshops
conducted by the Stanford Energy Modelling Forum. EMF Study 8 was
conducted to exercise the variety of models that participated under alternative
scenarios to gain insight into their similarities and differences and to understand issues facing both the industrial energy decision-makers and industrial
energy modellers. The results of EMF-8 are documented in a series of working papers. The following observations on the participating models are from
the Final Summary Report for the study (EMF, 1987).
The models project that energy use per unit of industrial output will
continue to decline through 2010. This trend is apparent in the results
from all the models for all scenarios. Energy use per unit of output is
projected to decline by 0.5 to 1.5 percent per annum from 1985 to 2010.
The explanations for this trend are a shift towards the production of less
energy-intensive goods, the further penetration of new technologies that
are more productive in the use of all inputs, and a continuation of the
gradual adjustment to the rapid energy price increases of the past 15 years.
The much lower energy prices that emerged in 1986 may retard the last
trend, but will not significantly slow the other two.
The forecasted shift toward less energy-intensive goods is a function of the
macroeconomic inputs into the EMF-8 participants. This macro outlook was
another issue examined by EMF-8. The total growth of industry determines
the total growth in energy, conditional on the trend in energy intensity forecast
by the models
Despite this trend toward reduced energy consumption per unit output,
26
growth in industrial output over the next two decades is expected to result
in a modest increase in total industrial energy demand over that time
period. Energy use for heat and power in industry is projected to grow by
1 to 2 percent per year from 1985 to 2010, or by about 40 to 80 percent
of the rate growth of industrial output
The models were in reasonable agreement that industrial growth would
outstrip the various improvements in energy efficiency.
The rate of growth of industrial energy demand in the future depends as
much on the projected level of total industry output and the projected mix
of energy-intensive and non energy-intensive goods produced as on
projected energy prices. The adjustment in total energy use per unit of
output to changes in energy prices tends to evolve over a long period of
time. In addition, long-term trends towards less energy intensive products
and towards more efficient use of all input in producing those products
continue somewhat independently of changes in energy prices, up or
down.
A significant source of disagreement in the models comes when one examines
the treatment of electricity demand.
Electricity use by industry is projected to increase more rapidly than its
use of fossil fuels. However, the engineering process models project a
gradual decrease in electricity use per unit of economic output (about -1 %
per year), while the econometric models show a gradual increase (about
+ 1% per year). The econometric models assume that post-embargo trends
in the dependency of electricity use on fuel prices and output growth will
continue into the future. The process models explicitly represent individual
electricity-using technologies; thus, they can account for saturation effects
which could keep electricity growth rates below historical levels, but they
may not represent all future electricity-consuming technologies or subtle
process/product shifts towards greater or less electricity use.
This difference highlights the philosophical split between the models. INRAD,
which was not in EMF -8, incorporates explicit penetration of two electricityusing technologies in its econometric framework. This type of hybrid may be
required in the future to address these types of forecasting concerns. Another
concern related to electricity demand is where the electricity comes from,
purchased power or cogenerated
Further penetration of cogeneration in the industrial sector would result
in more electricity used than purchased by the industrial sector. While such
penetration depends on a host of regulatory, institutional, and business
strategy issues, scenarios involving increased cogeneration are possible.
Thus, purchased electricity per unit of industrial output will, in fact, be
less than electricity consumption in industrial processes. Similarly, if some
Conclusion
27
cogenerated electricity is sold to utilities, this extra source of electricity

could also augment central electric generation in the decades to come.
The treatment of interfuel competition is another area that highlights the
differences in process versus econometric models. Process models rely on data
on dual firing capability, while econometric approaches rely on market-sharing
equations, like LOGIT or TRANSLOG cost shares. One captures important
near-term structural effects, the other captures some structural effects, but also
behavioural ones.
In the near term, the most intense interfuel competition is between oil
and gas.
- Most existing dual-fired capacity is oil-gas, with relatively low associated capital costs. Currently, gas is in standard use.
- There is a potential for greatly increased oil use in the industrial sector
in the medium-to-Ionger term if relative oil prices are low, although
there are indications that many users may not switch from gas to oil
as rapidly as aggregate fuel price data indicates.
The models' marked lack of energy/economy interaction was noted strongly by
the EMF participants.
The models included in this EMF study cannot be used to study dramatic
jumps in energy prices without external information/analysis regarding the
effects of these shocks on the overall economy.
- The most important effects of energy price shocks may be their
impacts on savings rates, inflation and economic output, which are not
explicitly represented in these models.
- Some of the models use 5-year time periods which does not allow for
a detailed representation of the macro dynamics of energy shocks.
As one can see from the above comments of the EMF-8 study participants
regarding sectoral shift and macroeconomic shocks, the need to integrate
industrial energy demand into an overall framework is an important avenue of
research. The class of general equilibrium models accomplishes this integration
completely, while other models that are used within some partially integrated
framework do so only partly. One important question remains unanswered
about the importance of this integration. That is the degree to which the same
factors, e.g. energy price shocks, impact on both energy decisions and industry
growth. If these factors are different, then industrial energy forecasting can be
approached in a partial equilibrium framework, utilizing alternative economic
growth scenarios to measure the degree of uncertainty that the macro assumptions give to any single forecast. However, if the same factors that affect energy
decisions are closely linked to industrial investment and growth, then the
partial equilibrium class of model presented above will need to move to a more
integrated, general equilibrium approach typified by the Hudson/Jorgenson
family of economic models.
28

ACKNOWLEGDEMENTS
The author would like to acknowledge the support of the United States
Department of Energy, Assistant Secretary for Fossil Energy, Office of
Planning and Environment, under Contract W-31-109-Eng-38. This paper
would not have been possible without the input of all the participants of the
Energy Modeling Forum Study 8 on Industrial Energy Demand Modelling,
particularly the model developers who provided comments on an earlier draft
review paper of the EMF -8 models. I would like to thank ANL staff members
Ross Hemphill, Don Hanson and Don Jankowski for their input into that
earlier paper. Any omissions are entirely my responsibility.
REFERENCES
Alter, S. L. et al. (1978), Interim Validation Report: Major Fuel Burning Installation
System. Report prepared for US Department of Energy by Lawrence Berkeley
Laboratory, Berkeley, Ca.
Boyd, G., Kokkelenberg, E. and Ross, M. (1990) Sectoral Electricity and Fossil Fuel
Demand in US Manufacturing: Development of the Industrial Regional and Energy
Demand (lNRAD) Model. Argonne National Laboratory Report, ANLjEAISjTM-35,
Argonne.
Bureau of the Census (published annually) Annual Survey of Manufacturers. US
Department of Commerce, Washington, DC.
Bureau of Labor Statistics (published annually) Time-Series Data for the Input-Output
Industries US Department of Labor, Washington, DC.
EEA (Energy and Environmental Analysis) Inc. (1977) Energy Consumption Data Base,
3 vols. TID-27981/TID-27986, TID-27988/TID-27990, TID-27992, Arlington, Va.
EMF (Energy Modeling Forum) (1987) Final Summary Report for EMF-8: Industrial
Energy Demand, Working Paper. Stanford University Energy Modeling Forum,
Stanford, Ca.
Electric Power Research Institute (1990) Guide to the INDEP1H Level I Econometric
M ode/s: Final Report. EPRI Customer Systems Division, Palo Alto, Ca.
Hanson, D. and Bauer, P. (1981) 'Industrial Fuel Choice Under Uncertainty.' Presented
at ORSA-TIMS Conf.
Hazilla, M. and Kopp, R. (1986) Systematic Effects of Capital Service Price Definition
on Perceptions of Input Substitution. Journal of Business and Economic Statistics, 4,
209-24.
Hazilla, M. and Kopp, R. (1990) The Social Cost of Environmental Quality Regulation:
A General Equilibrium Analysis. Journal of Political Economy, 98(4), 853-73.
Hudson, E. A. and Jorgenson, D. (1974) US Energy Policy and Economic Growth,
1975-2000. Bell Journal of Economics and Management Science. 5, 461-514.
Jaccard, M. and Roop, J. (1990) The ISTUM-PC Model: Trial Application to the
British Columbia Pulp and Paper Industry. Energy Economics, pp. 185-96.
Jogenson, D. and Wilcoxen, P. (1990) 'Reducing US Carbon Dioxide Emissions: the
Cost of Different Goals. Draft report, Harvard University, Cambridge, Mass.
Mansur, A. and Whalley, 1. (1984) Numerical Specification of Applied General Equilibrium Models: Estimation, Calibration, and Data, in Applied General Equilibrium
Analysis (eds. H. Scarf and 1. Shoven), Cambridge University Press, Cambridge,
pp. 69-127.
References
29
Reister, D.B., Edmonds, J. A. and Barnes, R. W. (1980) The Oak Ridge Industrial Model:
vol. II, Model Description. Oak Ridge National Laboratory, Oak Ridge, Tennessee.
Shephard, R. W. (1970) Theory of Cost and Production Functions. Princeton University
Press, Princeton, NJ.
Smith, V. K. and Hill L. J. (1985) Validating Allocation Function in Energy Models: An
Experimental Methodology. Energy Journal, 16, 29-47.
Solow, 1. L. (1987) The Capital Energy Complementarity Debate Revisited. American
Economic Review, 77, 605-14.
US Department of Energy (1979) Industrial Sectors Technology Use Model (ISTUM),
vols 1-4, DOE/FE/2344-1 through DOE/FE/2344-4. Washington, DC.
US Department of Energy (1983) Industrial Energy Productivity Project: Final Report,
Vols 1-9, DOE/CS/40151-1 through DOE/CS/40151-9. Washington, DC.
US Department of Energy, Energy Information Administration (1983) A Statistical
Analysis of What Drives Industrial Energy Demand, DOE/EIA-0420/3. Washington,
DC.
US Department of Energy, Energy Information Administration (published annually)
Annual Energy Outlook, DOE/EIA-0383. Washington DC.
US Environmental Protection Agency (1988) Industrial Combustion Emissions Model
User's Manual, EPA-600/8-88-007a. Washington, DC.
US Environmental Protection Agency (1990) 'Electric and Gas Utility Modeling
System: Technical Documentation.' Draft report prepared by RCG/Hagler, Bailly,
Inc., Boulder, Col.
3
Best-practice and average practice:
technique choice and energy
demand in a vintage model
Lennart Hjalmarsson and Finn R. Forsund
3.1 INTRODUCTION
Comparisons between best-practice use of energy and average practice have
become quite popular in debates and scenarios about future need for energy,
and sometimes fairly strong conclusions are drawn about long-run energy
demand on the basis of such comparisons (see for example Goldemberg et ai.,
1988 and Johansson et aI., 1989). Since in most sectors of an economy there is
a substantial difference between the average and lowest energy input coefficients (i.e. the amount of energy used per unit output) an instantaneous
adoption of best-practice energy-using technology in all sectors would decrease
the use of energy radically.
However, most energy is consumed via capital goods and the longevity of
capital makes the transformation process from average practice to best-practice
an often slow and gradual one. Moreover, new and more efficient technologies
are embodied in new capital goods and the diffusion of new technologies
depends on the growth rates in different sectors. But a higher growth rate will
also increase energy demand. Therefore, the energy decreasing trend in the
input coefficients may be more than offset by the energy increasing impact of
a larger output volume so the net effect on energy demand cannot be
determined a priori. The purpose of this chapter is to study these offsetting
forces by analysing technology choice and energy demand in a vintage model.
The vintage approach is concerned with the dynamic process of structural
change in an industry producing a homogeneous good. Of particular interest
is the distance between best-practice and average productivity which is
determined by the development of the vintage structure over time through the
process of investments and the scrapping of old equipment.
Embodied technical progress yields a distribution of production units spread
32
Technique choice and energy demand in a vintage model
over different vintages, with later vintages being more efficient than older ones.
This pattern might be modified, at least for some inputs, by biased technical
change and non-uniform relative factor price development. Due to short-run
rigidity it makes sense to describe and analyse the structure on the basis of
fixed input coefficients and capacity.
What we have in mind is the following picture of a typical development of
a growing industry: for investments in new capacity there are economies of
scale. At the planning stage there is also a wide variation in conceivable designs
of a plant as regards technology: more or less capital, labour, energy and,
sometimes, raw material intensity. But after a plant is built capacity is given,
and the scope for substitution between inputs is very limited.
The firms in this industry expect a future growth in demand at some
percentage rate, and they also form expectations about the future development
of technique, output and factor prices. Typically, they may expect constant
capital costs, rising labour costs and falling or rising costs of energy and raw
material.
On the basis of the current state of the industry and the expectations about
the future, the firms make plans for future capacity expansion. At the outset
they have a certain market share (domestic or international) which they try to
keep in the future, if they can expect a 'reasonable' rate of return on their
investments in new capacity. Thus the industry consists of one or several firms,
all with their own expectations and plans for future capacity expansion. At the
industry level we may conceive of this as a centralized process of capacity
expansion or as a parallell process within several firms.
A cross-section of plants from such an industry will show a large variation
in technology between different plants and often a considerable distance
between best-practice and average productivity for current inputs. This chapter
will explore what conclusions about future factor demand may be drawn on
the basis of cross-section data revealing this kind of distance between bestpractice and average productivity for energy or labour. Thus, the main
question to which this chapter is devoted is the inferences about the future
development of input coefficients and factor demand that can be drawn from
a single cross-section for an industry.
The formal model is presented in section 3.2. This type of model is often
called Manne-type, since it was introduced by Manne (1961) who utilized a
capacity cost function in his model. Here we apply a full production function
model to enable investigation of factor demand and substitution between
inputs. This model was originally introduced in Hjalmarsson (1974) inspired by
Johansen (1972).
3.2 THE MODEL
The model may be described briefly as follows. Let us consider an industry
producing a homogeneous product. The set of techniques and technical
The model
33
progress is represented by a Cobb-Douglas ex ante production function from

which the firms choose their input coefficients and capacity when investing in
new capacity. This ex ante function has wide substitution possibilities between
different inputs and is characterized by economies of scale, i.e. the elasticity of
scale exceeds one. Technical progress is assumed to be embodied in new
capacity. The firms in this industry are assumed to supply, in a cost-minimizing
way, a gradually expanding market by an optimal timing of investments in new
capacity.
On the basis of this process of capacity expansion we may investigate the
consequences for the choice of input coefficients and factor demand of different
assumptions about technology, relative prices and demand growth, deriving
the distance between best-practice and average practice productivity. Though
this may be regarded as a highly stylized model, it highlights the problem we
are addressing in a satisfactory way without loss of generality.
We shall assume the following:
1. Demand grows at a constant exponential rate g.
2. Initially there is just enough capacity, denoted by y(O,O), to meet demand.
3. The ex ante function at the micro level exhibits increasing returns to scale
and is a Cobb-Douglas function with neutral technical progress, labour,
energy and capital equipment. This ex ante function now reads:
(3.1)
where
and
y(v, v), L(v, v), E(v, v) and K(v, v)
denote planned production at time v in vintage v, planned use of variable

inputs, labour, energy and planned capital investment respectively. Raw
materials are assumed proportional to output. b is the technical progress
parameter.
4. The following functions describe the change in the factor prices:
i=L, E, K
(3.2)
where qi (0) is the initial price.

5. Plant life and time horizon are infinite. This assumption makes the problem
more tractable without any significant loss of generality.
6. Capacity utilization in the latest plant grows at the same rate as demand
until the next investment point at which there is no un utilized capacity. This
assumption makes the model more tractable but may also be defended by
the following consideration: if the time period between two investments is
not too long we may regard it as a learning period. During this period
capacity utilization grows continuously.
34
7. Discrete time periods are assumed. To distinguish between the different

vintages, successive time points of investment are denoted by
n=O, 1,2, ...
These may generally differ from the real time index. It is assumed that the
first time an investment is made coincides with the starting point zero. Input
coefficients are fixed at the full capacity level independent of capacity
utilization. The assumptions (1}-(7) above imply the following 'constant
cycle time' theorem:
Theorem: An optimal policy consists of building successive plants at equidistant intervals of time. Proof: See Hjalmarsson (1974).
The time interval between two investment points is denoted by

T
and
Tn=nT,
n=O, 1,2, ...
The growth in demand during the interval
Tn
to
Tn + 1
is
y(O, O)e gtn + 1 - y(O, O)egtn = y(O, O)engt(egt -1)
This expression must be equal to the capacity installed at time
(3.3)
Tn
(3.4)
where the bars indicate full capacity values.

The full optimization problem consists of minimization of the discounted
stream of future capital and variable input costs under the constraint that the
increase in capacity equals demand growth. Depending on several parameters
there is a trade-off between the exploitation of scale economies and the cost of
excess capacity which determines the optimal length of time, denoted by T,
between two investments. A formal presentation of this problem and derived
expressions for factor demand functions and the development of input coefficients are given in the Appendix.
Since this model is not very tractable for analytical treatment, we will present
model results by numerical simulation examples.
Does this model provide a realistic description of a typical industry? There
is a lot of empirical evidence supporting this type of model, both with regard
to the underlying assumptions about technology and to the assumptions about
firm behaviour. This type of model has also been tested empirically with
promising results (see, for example, Peck, 1974; Hjalmarsson, 1976; Gilbert and
Harris, 1984; Eaton and Ware, 1987; Gilbert and Lieberman, 1987.)
3.3 SIMULATION RESULTS
The purpose of the numerical examples is to illustrate the influence of different
parameters on the choice of technique and particularly the development of
35
Simulation results
best-practice technology compared to the development of average practice.

Since we are mainly interested in the time paths, the parameters influencing
these are varied while other parameters are kept constant.
As a basis for the comparisons we proceed from a reference case (denoted
Ref) characterized by the following parameter values (r = real rate of interest,
I> = elasticity of scale):
A
Y(O,O)
e
1.2
%
j=L,E,K
100
0.08
0.05
aL
aE
aK
VL
VE
0.02
0.5
0.3
0.4
0.03
In addition to the reference case, ten other cases are presented in Table 3.1,
where B denotes best, A average, L labour and E energy. In Table 3.1 we show
the optimal time cycle T, choice of input coefficients for the first investment,
year 0, (BL-O and BE-O) and the level of input coefficients for the most modern
plant in operation in year 30 (BL-30 and BE-30) and the growth rates of the
best input coefficients around year 30 (BL-g and BE-g). Due to the constant
investment cycles this plant is usually taken into use one to three years earlier.
In addition, we also present the average input coefficients in year 30 (AL-30
and AE-30), their corresponding growth rates (AL-g and AE-g) and the ratio
between the best and the average input coefficients (BLjAL and BEjAE).
In the third and second to the last rows of Table 3.1 the change in input
coefficients is put on index form (Index = 100 year 0), and in the last row we
have also calculated the annual percentage rate of growth of demand for labour
and energy around year 30 (L-g%, E-g%).
3.3.1 Marginal and scale elasticities
The first two cases illustrate the effects of variations in marginal elasticities and
economies of scale. In Case 1 the elasticity of scale is increased from 1.2 to 1.3
by an increase in the marginal elasticity of capital. Compared to the reference
case a considerably larger plant covering the first 5.5 years of expected growth
of demand is erected.
The increase in scale economies makes it more profitable to build a larger
plant. Since there is no change in the price parameters the factor ratio is the
same in both cases in accordance with equation (3.21) (see Appendix). Larger
economies of scale, however, imply a more rapid decrease in both input
coefficients (BL-g, BE-g), and a slower growth of demand for both labour and
energy (L-g, E-g).
In Case 2 elasticity of scale is reduced to a very low level, 1.05. Capacity
expansion will now take place in small steps and small-scale economies imply
BL
AL
BE
AE
L-g %
E-g %
BEjAE
BL-O
BE-O
BL-30
BE-30
BL-g %
BE-g %
AL-30
AE-30
AL-g %
AE-g %
BLjAL
g
f>
e
aE
aK
VE
Case
1.43
1.35
0.41
0.88
-4.5
-1.6
0.69
1.03
-2.7
-1.0
0.60
0.86
29
48
65
76
3.0
4.9
2.39
2.26
0.74
1.58
-4.2
-1.3
1.20
1.82
-2.5
-0.8
0.61
0.88
31
50
70
81
3.4
5.2
4.13
2.61
1.40
2.12
-3.7
-0.7
2.20
2.29
-2.2
-0.5
0.64
0.92
34
53
81
88
4.0
6.0
1.1
5.5
0.5
1.05
0.2
0.35
1.3
Case
1
2.25
2.46
0.65
1.39
-4.6
-1.6
1.03
1.60
-2.9
-1.1
0.63
0.87
29
46
65
75
4.4
6.5
2.61
2.46
0.86
1.90
-3.9
-0.9
1.44
2.12
-2.1
-0.5
0.59
0.90
33
55
86
2.5
4.1
77
3.9
0.07
Case
4
4.1
0.03
Case
3
2.30
2.17
1.08
2.45
-2.6
0.4
1.44
2.35
-1.6
0.3
0.75
1.04
47
63
113
108
4.1
6.3
4.9
Case
5
-0.02
3.7
2.29
2.68
0.55
2.88
-4.8
0.2
1.00
2.81
-2.8
0.2
0.55
1.03
24
44
107
105
2.9
6.2
2.53
1.81
0.84
0.81
-3.8
-2.7
1.32
1.11
-2.2
-1.7
0.64
0.73
33
52
45
61
3.5
4.1
Case
7
0.02
4.2
Case
6
2.98
3.10
1.16
4.07
-4.1
0.9
1.95
3.69
-2.4
0.6
0.59
1.10
29
49
131
129
3.6
7.0
1.05
0.2
0.35
-0.Q2
1.1
Case
8
1.44
1.65
0.39
1.66
-5.0
0.0
0.67
1.65
-2.9
0.0
0.57
1.00
27
47
101
100
2.9
6.1
0.5
-0.02
5.3
1.3
Case
2.21
2.58
0.95
4.35
-3.1
1.9
1.33
3.66
-1.9
1.3
0.72
1.19
43
60
169
142
4.0
7.6
-0.02
4.5
10
Case
The level and rate of change of best (B) and average (A) input coefficients for labour (L) and energy (E)
0.05
0.02
1.2
0.3
0.4
0
4.0
Ref
Table 3.1
Simulation results
37
a less rapid decrease in input coefficients and a higher growth of labour and
energy use. This also yields a somewhat higher ratio between best and average
input coefficients in year 30.
3.3.2 Expected demand growth
Cases 3 and 4 take the effects of changes in the rate of capacity expansion into
consideration. It turns out that the length of the investment cycle is not very
sensitive to fairly large variations in g. On the other hand, a slower growth
yields a smaller total output in year 30 and consequently a slower growth in
labour and energy demand but higher input coefficients since scale economies
are less exploited.
A more rapid growth leads to a more rapid reduction of the input
coefficients compared to the reference case, but at the same time a higher
growth in factor demand.
3.3.3 Technical progress
The importance of technical progress is illustrated in Case 5. A slow rate of
technical progress yields larger steps in the investment cycle since the relative
importance of scale economies increases. If there is no time-dependent technical progress the only source of productivity growth is the exploitation of scale
economies. Compared to the reference case it now pays to increase the size of
the plants. This is the opposite of Case 2 where the most important source of
productivity growth is the rate of technical progress that made it profitable to
expand capacity in rather small steps, taking advantage of the continuous flow
of more efficient techniques. Due to the increasing relative price of labour the
input coefficient of energy increases.
Note that the drop in technical progress in Case 5 has about the same effect
on percentage change in factor demand as the drop in economies of scale in
Case 2.
3.3.4 Energy price changes
In Cases 6-10 the relative price of energy is varied. The most important effect
of an increase in the relative price of energy (Case 6) is that it retards the
decrease in the labour input coefficient and raises the rate of decrease in the
energy coefficient. Moreover, the labour-energy factor ratio still increases but
it is lower compared to the reference case. The demand for labour is somewhat
higher than in the reference case, but the growth in demand for energy falls by
more than one percentage point.
If, however, the relative price of energy is reduced, as in Cases 7-10, the
energy input coefficients may increase or decrease. In Case 7 the substitution
effect is stronger than the impact of scale economies and technical progress on
38
the development of the energy input coefficients, so the energy coefficient

increases. The price decrease has about the same effect on energy demand as
the drop in technical progress in Case 5.
The force towards substitution of energy for labour is still more pronounced
in Case 8 where the elasticity of scale is reduced to a low level concomitant
with a falling energy price. In this case we notice the same small-step capacity
expansion as in Case 2, compared to the reference case a somewhat lower
demand for labour and an increased energy demand.
With regard to the development of the energy input coefficient the substitution effect in Case 9 is exactly balanced by the impact of technical progress and
economies of scale. At the same time the decrease in the labour coefficient is
rather rapid, and as in Case 1 capacity expansion takes place in large steps.
With regard to the impact on factor demand, a comparison with Case 1 and
Case 7 reveals that labour demand is neither very sensitive to the change in
relative factor price nor to the change in elasticity of scale, while a drop in the
energy price growth from to - 2 per cent per year has a fairly large impact
on energy demand.
In Case 10, finally, the impact of zero technical progress is investigated again
(analogous to Case 5). In this case the substitution effect is rather strong and
not at all balanced by the elasticity of scale, yielding a rather strong increase
in the energy input coefficient, but a somewhat more rapid reduction of the
labour input coefficient compared to Case 5. Still compared to Case 5, there is
a small decrease in labour demand and a rather strong increase in energy
demand.
In none of the above cases have we assumed the price of energy to increase
faster than the price of labour. However, since labour and energy are treated
symmetrically in the model, we could analyse such a case by changing the
labels of the inputs without any change in the conclusions above.
3.3.5 Productivity structure

The structure of unit costs may be represented in a 'Salter like' diagram, called
a Heckscher diagram (Forsund and Hjalmarsson, 1987). Such a diagram also
yields the structure of quasi-rents if the price level of output is known. The
quasi-rent structure of Case 7 above is depicted in Figure 3.1.
The oldest plant is to the right and the latest built to the left. The cost share
of labour and energy is also indicated. (The energy cost share line in Figure 3.1
is not horizontal but slightly falling.) Let us assume that the first plant built is
on the zero quasi-rent margin. The surface between the rectangles and the price
line represents the total quasi-rent of the industry. Varying the price level or
the input prices shows which units will have negative quasi-rents.
A modern plant is characterized by a high quasi-rent share of the total unit
product price. Compared to the older plants, unit labour cost is strongly
decreasing, while energy unit cost is slightly increasing which means that the
39
Simulation results
Current
unit
costs
Uni t price
line
0.70
.--
0.60
Quasi-rent area
0.50
0.40
rl
0.30
0.20
----- ---
1---
- - --
--
Labou
cost
share
-- Energ
- y
cost
share
0.10
25
Fig. 3.1
50
75
100 % total capacity
The Heckscher diagram of case 7 in year 30.
energy cost share is strongly increasing. Thus, a ranking of units according to

quasi-rent shares or unit labour cost coincides with a ranking according to the
vintage of capital. This should also be the case in most industries where the
relative price between labour and energy has been increasing in the past.
Energy unit cost, on the other hand, may be positively or negatively correlated
with the vintage of equipment depending on the outcome of the trade-off
between economies of scale, technical progress and substitution.
An important question concerns the empirical realism of the vintage model.
In the stringent sense this model holds for an industry producing a homogeneous output. In practice, however, the model is also applied to industries at
more aggregated levels under headings like Salter analysis, productivity distribution analysis, gross profit share analysis etc.
Although it may be difficult to predict exactly which plants will be closed
down and when they will be closed down, there is usually a steady and gradual
change of the structure of an industry so that best-practice today may be the
average practice at some future time. What does the distance between best and
average productivity tell us about future input coefficients?
If we want to make predictions about future input demand both the
direction and speed of the process of structural change are crucial. Let us first
return to our vintage model and the numerical simulations above. We found
that technical progress 'breeds inefficiency' in the sense that the ratio between
best and average input coefficients increases when the rate of technical progress
increases. It also turned out that important factors behind the rate of change
40
of input coefficients were the rate of technical progress and the rate of change
of input prices. With regard to the direction of change of input coefficients, a
trade-off was noticed between substitution forces tending to move the choice
of technique along the isoquants in the cost-saving direction and the rate of
technical progress moving the isoquants towards the origin. The outcome of
this trade-off was dependent on the parameter values. Thus, a falling relative
price of energy in combination with slow technical progress and/or small
economies of scale should lead to an increase in the input coefficient of energy
and a decrease in that of labour. In such a case low input coefficients of labour
are correlated with high input coefficients of energy.
On the other hand, slightly falling energy prices in combination with a faster
technical progress and/or larger economies of scale should result in falling
input coefficients. In this case a ranking of plant input coefficients according
to labour intensity should give the same result as a ranking according to energy
intensity.
Even if the question of which development we should expect in a specific
industry is an empirical one, it is, however, reasonable to expect that the
substitution forces are stronger within the input aggregates of labour and
energy than between these aggregates. In Swedish industry there is a trend
towards higher input coefficients of electricity but lower input coefficients of
fuels within a decreasing aggregate input coefficient of energy (Bogren, 1984).
3.4 CONCLUSIONS
In this study we have analysed the choice of technique in a vintage model and
the potential usefulness of the vintage approach when forecasting future energy
demand. The result can be summarized as follows:
1. Information about the distance between best and average productivities or
a full cross-section of industrial statistics for an industry at a certain point

of time cannot be used in a meaningful way for predictions about future
input coefficients. To obtain forecasts we need historical time series too.
2. The ratio between best and average input coefficients is a variable~ in itself,
depending on a number of parameters. In the theoretical model it was
shown how the best-average ratio varied with important parameters. It
turned out that technical progress breeds inefficiency in the sense that a
more rapid technical progress increases the distance between best and
average input coefficients. A change in the relative price of labour and
energy had a large effect too. A decrease in this relative price reduced the
distance in the case of labour and increased the distance for the energy
input coefficients. A falling energy price even made the ratio between best
and average energy coefficients greater than one. Other parameter changes
were of less importance.
3. The rates of change of input coefficients and of input demand also depend
41
Appendix
on the same set of parameters in our model. In all of the examples in our
model we assume that the price of labour increases faster than or equal to
the price of energy. It turns out that labour demand always increases at a
slower rate than the growth of output, while the contrary holds for energy
except in two cases. Factor demand appears to be fairly sensitive to all
parameter changes.
4. If we expect the price of labour to increase faster than the price of energy
even in the future - which seems to be a reasonable assumption - modern
capital equipment should be characterized by a high labour productivity
while the energy productivity may be high or low in comparison with older
equipment.
The main conclusion of this study is that if we base our analysis on just a
cross-section of an industry, the predictions of future input demand must be
highly uncertain. Time series of cross-sections are necessary for this type of
analysis.
APPENDIX
In this Appendix we will present the vintage model in a compact form. For a
more detailed exposition, see Hjalmarsson and Eriksson (1985).
The cost of the plant to be constructed at time point Tn> discounted to year
0, is denoted by Crn and is given by the expression:
L
00
Crn =
t=t n
L
00
qdt)e-rtL(t,Tn)+
qE(t)e-rtE(t,T n)
t=tn
(3.5)
The expression is to be minimized under the constraint that the increase in

capacity equals demand growth, equations (3.3H3.4).
From the first order conditions the conditional factor demand functions are
derived:
L(Tn> Tn) =
[y(O, O)A -lat r;I aj-a'(1- e -T{ 1~~\Ol rJ-e

VK
r;I[1~j~~;-rJ(egr-1)J/e e~'v,a,-vLe-Hg)/e)m
(3.6)
42
K ('m 'n)= [Y(O, O)A -lair; aj-3 (1_e VK -r)"KqK(o)-e

i
n
(
qj(O)
1-ev1
)3lJl /e
r
i=L,E,K
(3.8)
For later use we define
i=L,E,K
(3.9)
and
LVjaj-VjI,-b+g
j
i,j=L,E,K
(3.10)
If equations (3.6H3.8) are inserted in equation (3.5), the following cost function
is obtained
(3.11)
where
(3.12)
and
()=-'-j----
-r i=L,E,K
(3.13)
Summation over all n yields the total cost function for the whole horizon as
a function of the time interval, to be denoted by q,). q,) includes the
discounted stream of construction costs as well as operation costs:
(3.14)
where ()<O, B>O

The optimal time interval is obtained by minimizing q,) with respect to ,.
Input per unit of output, the input coefficient, for vintage
is denoted by
~('n). It is a variable ex ante, but a fixed coefficient ex post. From equations
(3.6H3.8) and (3.3) one obtains
'n
43
Appendix
where
(3.16)
where
(3.17)
(3.18)
where
(3.20)
The development of the ratio between two input coefficients when new
capacity is built is given by
i, j=L,E,K
where
a q.(O)
[1-e J
Dij=~q:(O) l_e
Vi
Vj -
i,j=L,E
(3.21)
(3.22)
and
(3.23)
i.e. the development of the relative factor ratio is only governed by the
difference in factor price change between the two inputs. If the factor prices
change at the same rate, Vj=Vj, the factor ratio is constant.
The average input coefficients are obtained by the ratio of cumulated input
and cumulated output. From cumulated output at time r, Y(r) is obtained as
Y(rn) = y(O, O)eg(n + 1)<
(3.24)
and from (3.6) and (3.7) cumulated inputs at time rn, Vj(rn) are
(3.25)
44
where
Bj = y(O, 0)1/'H 1/'aj%(0) -1(egt -1)1/<(1- e
-r)
(3.26)
i=L,E
(3.27)
Vj
and Wi is given in (3.10).

Thus, the average input coefficient is obtained as
A . _ Vlrn) _~
(e",,(n+ 1)< -1)
V.- Y(rn) - y(0,0)(e"';t-1)(e g(n+1)t-1)
and the ratio between best-practice and average coefficients as
-iA.y(O, 0) (e (n+1)t_1)(e",;t_1)
e",;(n + 1)t -1
en'nt
g
BA Vi =
i = L, E
(3.28)
where
Wi-g
i=L, E, K
(3.29)
and
(3.30)
These expressions are fairly complicated and there seems to be no obvious
way to simplify. Depending on the sign and size of different parameters both
A Vi and BA Vi may decrease or increase over time, but not very much can be
said without information about the size of different parameters.
ACKNOWLEDGEMENTS
In the preparation of this paper valuable comments were received from
Thomas Sterner. Financial support from Jan Wallander's Research Foundation is gratefully acknowledged.
REFERENCES
Bogren, E. (1984) Energiproduktivitetens variationer inom industribranscher. Project
report from the Energy Research Board, Efn/AES, 1984:2.
Eaton, B. C. and Ware, R. (1987) A Theory of Market Structure with Sequential Entry.
RAND Journal of Economics, 18, 1-16.
Forsund, F. R. and Hjalmarsson, L. (1987) Analysis of Industrial Structure. A Production
Function Approach. Almqvist and Wiksell International, Stockholm.
Gilbert, R. J. and Harris, R. G. (1984) Competition with Lumpy Investment. RAND
Journal of Economics, 15, 197-212.
References
45
Gilbert, R. 1. and Lieberman, M. (1987) Investment and Coordination in Oligopolistic

Industries. RAND Journal of Economics, 18, 17-33.
Goldemberg, 1., Johansson, T. B., Reddy A. K. N. et al. (1988) Energy for a Sustainable
World. Wiley Eastern, New Delhi.
Heckscher, E. F. (1918) Svenska produktionsproblem. Bonniers, Stockholm.
Hjalmarsson, 1. (1974) The Size Distribution of Establishments and Firms Derived from
an Optimal Process of Capacity Expansion. European Economic Review, 5, 123-40.
Hjalmarsson, 1. (1976) Capacity Expansion and its Implications for the Size Distribution of Firms: Reply. European Economic Review, 7, 287-92.
Hjalmarsson, 1. and Eriksson, S-G. (1985) 'Choice of Technology and Energy Demand
in a Vintage Framework.' Unpublished.
Johansen,1. (1972) Production Functions. Amsterdam, North Holland.
Johansson, T. B., Bodlund, B. and Williams, R. H. (eds) (1989) Electricity - Efficient
End-Use and New Generation Technologies, and their Planning Implications. Lund
University Press, Lund.
Manne, A., (1961) Capacity Expansion and Probabilistic Growth. Econometrica, 29,
632-49.
Peck, S. C. (1974) Alternative Investment Models for Firms in the Electric Utilities
Industry, Bell Journal of Economics and Management Science, 5, 420-58.
4
The effects of changes in the
economic structure on energy
demand in the Soviet Union
and the United States of America*
Yu. D. Kononov, H. G. Huntington,
E. A. Medvedeva and G. A. Boyd
4.1 INTRODUCTION
It is widely recognized that an economy can reduce its energy intensity through
many channels. Improvements in the energy efficiencies of certain processes in

individual industries are important but capture only a part of the response.
Shifts in the structure of an economy's production for energy-intensive sectors
have been and will continue to be an important factor in lowering the energy
use per unit output in many economies (Maday, 1984). Analysts in both the
Soviet Union and the United States of America expect that the changing
economic structure will continue to reduce the aggregate energy intensity in
each country. There exists, however, considerable uncertainty about the rate of
decline in the energy intensity due to such shifts.
This paper focuses on studies conducted in both countries to measure the
relative importance of shifts within the economy on energy demand using the
same methodology - the Divisia index - for decomposing energy intensity
trends. Since conventions for measuring energy consumption and economic
output are different in the two countries, a comprehensive comparison of
international experiences was beyond the scope of our collaboration. A
rigorous comparison of past trends in the two countries, if it can even be done
reliably, would require extensive revisions and adjustments to the existing data.
Instead, our objective was more modest: to use the most recent data available
*This paper was originally presented at the joint American-Soviet National Academy of Sciences
Symposium on Energy Conservation Research and Development, Yalta, USSR, 6-12 October
1988.
48
Effects of changes in economic structure on energy demand
internally within each country to help identify and understand the important
energy intensity trends in the two countries. Thus, this chapter summarizes two
separate but concurrent studies on aggregate energy intensities.
The decomposition of energy intensity trends into those due to changing
economic structure and those due to other factors, new processes within an
industry for example, is an important first step in understanding energy use
patterns. For a market economy, energy intensity trends within an industry are
governed largely by the prices of energy and other inputs as well as industryspecific technological progress. These same factors, however, may not be as
critical for economy-wide or aggregate energy intensity trends, which will be
strongly influenced by shifts in the relative importance of different economic
sectors. For example, aggregate energy intensity is significantly influenced by
the widespread substitution of newer, more versatile materials for the traditional, energy-intensive raw materials, e.g. various plastics for primary metals.
Moreover, the indirect effects of energy price changes - the redistribution of
income and the shifting share of investment - are often at least as important
as the direct effect. Analyses that ignore these differences are likely to
misrepresent the factors determining aggregate energy-use trends in economies
that either now or will in the future depend upon markets.
Policymakers in the United States and USSR are increasingly interested in
strategies for reducing the aggregate energy intensity within each country. In
the Soviet Union energy production is often extremely capital-intensive. In the
absence of market prices, many analysts perceive an overinvestment in energy
supplies in that country which retards economic growth by misallocating
capital. Thus, declining energy intensity in the economy would release capital
to other sectors where it could be employed more productively. As the Soviet
Union moves towards a market economy, energy efficiency gains achieved
through shifts among sectors of the economy are likely to respond to different
factors than those gains realized from new processes and technology implemented within a sector (see further Chapter 12 by Yuri Sin yak).
In the United States and other developed market economies, some analysts
argue that market prices undervalue the social cost of using additional energy.
Incomplete information, public utility regulation favouring supply options and
other market imperfections could bias aggregate economic consumption towards too much energy use. Moreover, increased energy consumption means
increased production from and dependence upon the Persian Gulf, the source
of marginal supplies in world energy markets. Growing dependence upon
insecure energy supplies increases the vulnerability of the world market
economies to oil market disruptions and price shocks.
In recent years, policy concerns in both countries have been broadened to
include environmental problems such as pollution and possible climate change
through increasing concentrations of carbon dioxide and other greenhouse
gases. Much recent attention has been focused on the forces influencing energy
use and the relationship between energy use and environmental degradation.
Estimates of the effect of different carbon-reduction strategies, for example,
Methodology
49
depend critically upon how rapidly energy will grow in the absence of such
limits as well as upon the degree of substitution away from energy as energy
use is restricted.
The two-country studies reported in this paper were done to provide an
initial perspective on past USSR energy intensity trends and possible future
developments, within the context of the US experience. A clear understanding
of the different channels for reducing energy in the major economies is a
prerequisite for an informed analysis and discussion of the relative benefits and
costs of various strategies for reducing energy use.
The concurrent studies of the two countries were conducted just as new 1985
data were becoming available on the use of energy in both economies. The
research also commenced at a time when the Soviet Union was just beginning
to contemplate major changes in its economy and energy sector. Since that
time, the challenge of this transition has grown substantially, creating considerable uncertainty about future developments in both the Soviet economy and
its energy sector. Given these developments, one must view any attempt to
project future economic and energy trends as extremely risky, with large error
bands in either direction (a problem in any economy relying upon market
forces). Nonetheless, we view these projections and their decomposition into
sectoral shift and efficiency effects as a useful comparison of how these forces
could unfold in the two countries.
Our analysis reveals three key conclusions:
1. The use of a Divisia index for separating the shift effect from other factors
is a useful method when there are large or sudden changes either in energy
intensity within a sector or in the relative economic importance of a sector.
2. Shifts in the composition of economic output have an important effect on
the trend in aggregate energy intensity of the economy. However, shifts
within the industrial sector away from energy-intensive industries have a
more pronounced effect that do shifts among major macro sectors (e.g.
industry, transportation, agriculture).
3. Within industry, there has been a dramatic decline in aggregate fuel
intensity in both countries over the last 20 years. During the post-embargo
period (after 1973), however, this trend accelerated in the United States but
slowed in the Soviet Union, due largely to little improvement in energy
intensity within industries.
After a brief discussion of the methodology, we discuss the Soviet and US
results in separate sections.
4.2 METHODOLOGY
Variations in aggregate energy intensity were decomposed into changes in
energy intensity at the industry level (measured by real energy intensity) and
50
changes in the composition of output (i.e. sectoral shift). These components

were formally defined for the analysis as follows:
Real energy intensity: the (weighted) average of the energy intensity of
separate sectors, equivalent to the 'real' change in energy intensity that
remains when output mix effects are removed from the aggregate measure.
Given the level of aggregation in our studies, the measured change in real
energy intensity will be incorporating some shifts in the mix of output or
industry on a dis aggregated basis .
Sectoral shift: the total contribution of changes in output mix to changes in
aggregate energy intensity.
We decompose the aggregate energy intensity with the Divisia index, which
is constructed from growth rates using variable weights based upon total
energy use. This approach has been used to study energy trends in the United
States and other market economies (for example, see Boyd, McDonald et ai.,
1987); its advantages and limitations for energy use studies have been considered by Boyd et ai. (1988). The change in aggregate energy intensity is calculated
by the following formula:
~ln
e- L wj~ln ej + L Wj ~ln dj
where ej is the energy-output ratio in a particular sector, dj is the share of

output in that sector, Wj is the sector's share of total energy consumption 1 and
the subscript j refers to a sector.
The principal distinctions of this method are:
1. The effect of sectoral shift is estimated dynamically, year by year, rather
than by comparing the beginning and ending year in the period. This
procedure uses a logarithmic index representation that ensures additivity of
the results.
2. Sectoral shifts are weighted by the structure of energy demand changing
over time (in contrast to fixed-weight methods, which weigh the components by energy use in the base year).
4.3 THE SOVIET EXPERIENCE
Since energy trends in the USSR are not widely known, we initially discuss
some key aspects. We then analyse the contribution of sectoral shift to these
trends. The discussion and results are based upon historical Soviet data for the
period 1960-85 from Pavlenko and Nekrasov (1972), Aksutin et ai. (1981),
1 Btu rather than price weights have been used because prices have historically been a poor
indicator of relative value in the Soviet Union. Btu weights overstate the decline in energy intensity
for the United States because measured intensity includes substitution towards higher-quality
energy, e.g. electricity, as well as the shift away from energy (Huntington and Myers, 1987).
The Soviet experience
51
Melentiev and Makarov (1983), and Narodnoye (1985), and upon Siberian
Energy Institute projections presented by Anoshko et al. (1986). Some shares
and trends obtained from these data are described in the next section.
4.3.1 Key trends
The pattern of the energy-output ratio differs substantially depending on the
particular energy carrier (electricity, heat, fuels) and time period. Figure 4.1
presents the calculated indexes of electricity, heat, fuel and energy intensities of
economic output for 1985. 2 Total energy use is the sum of fuel, electricity and
heat use. The latter is thermal power produced and used in the form of steam,
hot water and space heat, primarily for district heating systems.
In 1960-85, the indexes of electricity and heat intensity were very similar;
annual increases in these intensities averaged 1%. The change of electricity and
heat intensity was caused mainly by the increasing contribution of industry to
aggregate output and its high energy intensity. Industry's energy intensity was
more than two times higher than for the country as a whole in 1960 and 1.3
times higher in 1985 (Melentiev and Makarov, 1983). Strong growth in the
electricity intensity of agricultural output (a tenfold increase for the period
reported by Melentiev and Makarov, 1983) also contributed to this trend.
However, in the future the electricity- and heat-output ratios may be expected
- - Energy
- - -
Heat
Fuel
,
II
Electricity
Ll)
CD
'",...
)(
Q)
"0
C
1960
Fig. 4.1
1970
1980 1985
Years
Index for energy-output ratio in the Soviet economy.
2Throughout this section, output refers to the net output, or value-added, in a sector. For the
aggregate economy, net material product (NMP) rather than gross national product (GNP) is used.
NMP excludes output in the service industries.
52
to decrease by 1 and 2% per year, respectively. This projected trend is caused

by the further decline of the electricity and heat intensity of industrial output;
the growth of industry's relative importance to the economy does not offset this
effect.
The rates of decline in the aggregate fuel-output ratio are much higher. They
average up to 2% per year in the historical period and the projection. This
trend is explained mainly by the low fuel intensity of industrial output, whose
value was 1.1 times lower than the fuel-output ratio for the economy in 1960
and will be 1.7 times lower in the future (Melentiev and Makarov, 1983). In
addition, the industry share of aggregate output grows continuously.
Changes in energy intensities equal changes in the combined intensities for
electricity, heat and fuels. The decline in the resulting aggregate energy-output
ratio was 1% per year in the historical period and is projected to be not less
than 2% in the future.
These trends show that the energy intensity of industrial output has a
decisive impact on the change in the aggregate energy-output ratio in the
1960-85 period as well as in the future because industry's relative importance
in the economy grows continuously. Figure 4.2 compares the changes in the
energy-indutrial output and the aggregate energy-output ratios. It also shows
the changes in energy intensity of agricultural output and transportation. The
indexes for the energy intensity of industrial output and of the economy are
similar.
Within the industrial sector, changes in the energy intensity will be governed
by trends in two sectors:
sector 1 (materials industries: ferrous and non-ferrous metallurgy, chemistry
and petroleum chemistry, construction materials, forest and pulp-paper,
fuel/energy complex);
sector 2 (manufacturing industries).
Total energy
ElectriCity
:c...
~
~ 2
/'
", "-
- - Soviet economy
- - - Industry
- . - Agriculture
...... Transport
8
..................
/'
.:~.
)
//.~
i
--1960
...../
1980
2000 Years
1960
1980
2000
Fig. 4.2 Trends in sectoral energy intensity.
Year~
53
The historical period is characterized by the high and increasing share of

sector 2 in industrial output. This trend reduces the aggregate energy intensity
within industry because the energy-output ratio for manufacturing industries
is much lower than for sector 1 (the materials-producing industries). The
electricity intensity of the output in sector 2, for example, is four to ten times
lower than it is in sector 1 (the first figure corresponds to 1960, the second one
to the end of the time period), and the fuel intensity is lower by a factor of
16-120 (calculated from the data of Pavlenko and Nekrasov, 1972; Aksutin et
at., 1981; Melentiev and Makarov, 1983). Such a large difference in the energy
intensity indexes for the two sectors causes changes in the relative shares of
industrial output in the two sectors to have a substantial effect on the changes
in the aggregate energy intensity of industrial output.
The contribution of the industries in sector 1 to energy demand within
industry is clearly seen from Figure 4.3, showing the trends in the sectoral
demand for electricity, heat, fuel and aggregate energy (calculated from the
same data). The shape of sector 1 in electricity demand by industry for the
1960-85 period decreases from 67 to 60%, in heat demand from 61 to 54%,
and in aggregate energy from 77 to 68%. This sector's share in fuel demand
increases from 89 to 94%.
Energy use by households and service industries (the so-called non-production sphere in Soviet statistics) increases in relative importance over the
1960-85 period. This sector's share of electricity demand for the whole period
increases from 11 to 16% and its share of heat energy rises from 10 to 18%,
reaching its maximum of 21 % in 1980. Its share of fuel demand decreases
somewhat. Agriculture's relative importance increases in total electricity use
but decreases in fuel and total energy use. Transportation's share falls for all
energy carriers during this period.
4.3.2 Contribution of sectoral shift to the change in energy intensity
The indexes calculated by the Divisia index approach for electricity and fuel
intensity of NMp 3 and industrial output are given in Figures 4.4 and 4.5.
Figure 4.4 shows the shifting trends in the electricity-NMP ratio during the
1960-85 period. In the period between 1960 and 1965 the index was growing
at a rate of 4% per year, during the following 7 years its growth rate was low
(0.7% per year), and later - from 1972 to 1985 and in the near future - the
electricity intensity is practically stabilized.
The effect of sectoral shift is particularly pronounced in the early 1960s when
its share is more than 90%. In the period 1965-85, its contribution to the
change of electricity-NMP ratio is in the range from 40 to 85%. In the future,
the electricity-NMP ratio is expected to fall. The role of sectoral shift will also
decrease, to 32% by the end of the period. The fuel-NMP ratio (see Figure
3Households and services are excluded from the decomposition of the energy-NMP trends.
54

(a)
,'," I' I''"11111

, lilli', ,'I'
100 % -rT"':"'T7"'r-:-r.,...,1"':","'T'-:-1,r:-r:-1,r:-r,-r-:""r.'"'1
I """""'"
100 % ".,iTT"I'TTi'TTTTT.......-r:-r.-r7'T"':'".,..,.-:'I
50%
1960
1985
1960
1985
1660
1985
(c)
1960
Agriculture
1985
Transport
Sector 1
industry
Sector 2
industry
Non-production
sphere
Fig. 4.3 Change in the structure of electricity (a), heat (b), fuel (c) and energy (d)
demand in the economy.
4.4b) decreases by 2.8% per year in 1960-85 and is estimated to decline by

2.5% per year in the projections. Its pattern is almost completely determined
by the decrease in real energy intensity; the share of sectoral shift is 13% in
1960 and drops to 9% at the end of the period.
The electricity intensity of industrial output (see Figure 4.5a) decreases by
1.4% per year in 1965-85 and is expected to fall by 1.6% per year in the future.
Sectoral shift plays an essential role in declining aggregate electricity intensity,
accounting for 62% in the 1960s and 68% at the end of the historical period.
It should be emphasized that during 1967-79 the real electricity intensity of
industry did not change and almost all of the decrease in aggregate electricity
intensity was due to sectoral shifts.
55

(a)
/" /Ds
1.1
....
C) 1.0
II
It)
eX)
OJ
0.9
I
-=
'.
1",,- -:.....=. .:....... '. -
-~
/"
/"
.--
_./
Q)
"0
./
0.8
Da
!J
.De
0.7
0.6
1960
Years
1985
(b)
1.8
C) 1.4
II
It)
eX)
OJ
,...
1.0
Q)
"0
-=
_ ____
---
--
Os
0.7
. ,oe
0.5
Da
0.3
1960
1985
Years
Fig. 4.4 Changes in electricity (a) and fuel (b) intensities of NMP (Da) in production
sphere. Ds, sectoral shifts; De, changes in real electricity and fuel intensities.
A similar situation is observed in fuel consumption. Figure 4.5b shows that

between 1973 and 1985 the real fuel intensity changed very little; therefore, the
decline in the aggregate index was caused principally by sectoral shifts. The
total decrease in the fuel-industrial output ratio amounted to about 4% per
year in 1960-85 and is estimated approximately at 3% in the future. The
contribution of sectoral shift increases from 40% at the early 1960s to 52% at
the end of the historical period.
56

(a)
1.3
//
1.2
~
II
II)
CD
1.1
-,"
,~
.....
:...:. ......
...
1.0
en
:::. 0.9
x
"-
Q)
"-
l:l
.E 0.8
"-
"-
... Oe
"-
"-
0.7
"- ,...os
0.6
1960
1985
Oa
Years
(b)
2.0
1.6
II
II)
CD
en
:::. 1.2
x
-- --
Q)
l:l
.E 1.0
0.8
"-
"-
"- ....
.......
- ...... Os
....
0.4
1960
1985
Years
Fig. 4.5 Changes in electricity (a) and fuel (b) intensities of industrial output (Da). Ds,
sectoral shifts; De, changes in real electricity and fuel intensities.
Changes in the composition of output within the industrial sector have a

greater effect on aggregate energy intensity of the economy than do shifts
among major macro sectors (e.g., industry, agriculture and transportation). As
revealed in Table 4.1, shifts among the major macro sectors actually increase
the economy's energy intensity during the 1965-85 period. In the projections,
the economy's energy intensity is expected to decline dramatically in the future,
with structural shift, both among the major macrosectors and within industry,
57
The US experience
Table 4.1
Impact of structural shifts and other factors on changes

energy-NMP ratio (%) in the Soviet Union
Structural shifts - total

among four macro sectors
among ten industries
Changes with industries
(composition of outputs,
energy conservation, ... )
Aggregate
1965-1985
1986-2010
- 3 (15)
6 (- 30)
-9 (45)
-17 (85)
- 29 (57)
- 6 (12)
- 23 (45)
-22(43)
-20(100%)
-51(100%)
III
the
Note: Percentage contribution to total effect appears in parenthesis. For ease of readability,
in Tables 4.1 and 4.2 percentage changes are reported in additive rather than multiplicative
form. Percentage charges within industries are equal to the difference between trends in
aggregate and sectoral shift intensities.
accounting for almost 60% of this decline. As the Soviet Union shifts towards
a market economy, it is likely that these shifts, if they are to occur, will be
induced by a set of factors different from those that determine trends in real
energy intensity.
Soviet studies in the coming years will examine in greater detail the effect of
sectoral shift on energy demand in the economy for the next 25 years. This
analysis is being based upon a set of simulation models described by Medvedeva (1986). It comprises models of the economy and energy demand as well
as models of energy demand by separate sectors. Each model is based on three
groups of relations: technological (intersect oral) flows of products, financial
flows and energy demand of a sector or of the economy as a whole.
4.4 THE US EXPERIENCE

In the United States energy use has been maintained practically level from 1973
to 1985, while economic activity (GNP) has increased about 30%. This trend
of declining use per dollar of output holds for total energy, for fossil fuels and
for electricity. In the case of fossil fuels, this decrease represented an acceleration of the pre-embargo trend of declining fuel intensity. For electricity, on
the other hand, the post-embargo trend represented a reversal of the increasing
electrification observed before 1973.
A significant portion of this decline was due to shifts among sectors of the
economy, among industries (measured at various levels of disaggregation) and
among products. We estimate that these changes in the composition of
economic output account for almost one-half of the decline in energy use per
unit of output in manufacturing since the embargo. These new estimates are
based upon recently available data from the US Bureau of the Census, updated
58
through 1985. Moreover, this may be a conservative estimate because even

greater effects can be found when feedstocks are included (Werbos, 1986).
This section analyses the impact of changes in the composition of economic
output (sectoral shift) on US energy demand in the manufacturing sector. We
confine the analysis to this particular sector owing to the lack of reliable time
series data covering all sectors of the economy. We initially discuss the
historical trends, followed by some projections of sectoral shift and US energy
demand.
Using the Divisia index, we constructed time series for output, energy use
and several measures of energy intensity and sectoral shift. Annual data on
energy use and real shipments were used for the 1958-85 period. Rates of
change were expressed in index number form with 1985 as the base year.
In calculating energy intensities, the measure used for output was real
shipments 4 (1982 dollars). The study decomposed manufacturing into 71
industries, roughly 3-digit SIC level for the energy-intensive sectors. Energy use
was taken from the National Energy Accounts (NEA) and reflects purchased
fuels and electricity as well as cogeneration for internal use. 5 The units of
measure were British thermal units (Btu) for fossil fuels and kilowatt-hours
(kWh) for electricity.
4.4.1 Historical trends in energy intensity

Electricity and purchased fossil fuel were treated separately in analysing the
effect of sectoral shift on energy intensity, because they exhibited quite different
trends. This separation avoids some of the biases introduced by using Btus
rather than the price weights as well as the issue of whether to value electricity
at the primary or secondary (including losses) level.
Fuel intensity
Figure 4.6 presents the indexes of aggregate and real energy intensity over time.
Real fuel intensity fell approximately 1.8% per year in the pre-embargo
(1958-73) period. A slight shift away from energy-intensive industries occurred
in 1964-7 and back in 1968-74. Part of the reason for this sectoral shift was
the strong growth in primary metals in the period 1967-74. During the 1974--5
recession, real energy intensity rose, due to a short-run capacity utilization
effect. That is, if output falls, but energy use does not fall proportionately, the
4The US Bureau of Labor Statistics, Productivity Measures for Selected Industries, and unpublished supporting data tape. Real shipments is a gross output concept that includes energy and
materials in addition to labour and capital (value-added). It is more appropriate to measure US
energy intensity with respect to an output variable that includes energy.
sJack Fawcett Associates (1984) National Energy Accounts 1958-1981, US Department of
Commerce, PB85-142024. Detailed data on byproduct fuel use in industry are not readily available.
59
The US experience
2.5
Aggregate
0'
0
2.0
II
co
Il)
(l)
:::.
1.5
- --
. !1 eal
-......
- - - --
......--
CD
Shift
--
"0
C
ra
.:;
1.0
CIl
0.5
O~------,-------,-------,-------~------~~
1960
1965
1975
1970
1980
1985
Year
Fig. 4.6 Fuel intensity decomposition for US manufacturing.

energy-output ratio rises. This effect is quite pronounced in energy-intensive
industries.
During 1974-81, the sectoral shift caused real energy intensity improvements
to be less than the aggregate index would suggest. While the real energy
intensity averaged - 3.0% per year, sectoral shift contributed an additional
- 1.2 % to the aggregate trend.
For purchased fossil fuels, the sectoral shift effect is largely captured by
disaggregation to only two sectors (materials and non-materials industry
groups). The change in aggregate energy intensity from 1974 to 1981 was
33.4%, of which 9.2% (equivalent to 27.6% of the change) can be accounted
for by sectoral shifts among 20 2-digit SIC industries. Of this 9.2%, 5.5% can
be attributed to shifts away from the materials sector. Thus, about half of this
sectoral shift is captured by a two-sector disaggregation. The remaining effect
is largely between the energy-intensive industries themselves.
Electricity intensity
The overall picture for electricity 6 is quite different from that for fuels (see
Figure 4.7). The 1960s were characterized by an aggregate 2.5% per year
electrification trend. This trend was composed of a 0.8% shift towards
electricity-intensive industries - a sectoral shift very similar to that for fuel 6
Includes purchased electricity and cogeneration for on-site use.
60

1.4
Shift
1.2
II
co
---"../-------"
Aggregate
1.0
10
C1l
..x
Q)
0.8
.' .
Real
"tJ
C
.!!!
(/)
.:;
Ci
0.6
0.4
0.2
O~------.-------r------.r------.-------r~
1960
1965
1975
1970
1980
1985
Year
Fig. 4.7 Electricity intensity decomposition for US manufacturing.

and 1.7% change in real energy intensity. After 1970 the aggregate electricity
intensity began to decline. This was the result of a fluctuating real electricity
intensity, characterized by large peaks in intensity as the result of capacity
utilization effects. However, the composition shift effect on electricity intensity
reversed itself, averaging -0.5% over the 1970-81 period, then accelerating
sharply to -2.8% from 1981-85. The sectoral shift away from electricityintensive industries is a marked reversal of the pre-1970 trend.
There is evidence of shifts among the materials and non-materials industry
groups during 1971-81. Hence, a two-sector disaggregation incorporates only
some of the total sectoral shifts effect. After 1981, however, the decline is largely
attributable (68%) to a two-sector disaggregation.
4.4.2 Projected trends in energy intensity

There exists no official forecast of the US economy and the associated energy
demands. Moreover, projections of changes in the structure of the US economy
can often differ by a substantial amount and these differences can result in
widely different outlooks for energy demand, as discussed in Boyd, Hochheiser
et al. (1987). Nevertheless, for the purposes of comparing with the Soviet
projections, we have selected a forecast and have shown it together with our
historical analysis in Table 4.2, much as we did for the USSR in Table 4.1. We
restrict Table 4.2 to manufacturing only, because reliable data on the other
61
The US experience
Table 4.2 Impact of structural shifts and other factors on changes in the energy-GNP
ratio (%) in the United States of America
Fuels
Structural shifts among industries
Changes within industries
Aggregate
Total electricity
Structural shifts among industries
Changes within industries
Aggregate
1965-1985
1979-1985
1985-2010
-36(47)
-40(53)
-76(100%)
-28 (70)
-12 (30)
-40(100%)
-9 (22)
-31 (78)
-40(100%)
-21
17
-4
-18
6
-12
-9
-2
-11
Notes: Percentage contribution to total effect appears in parenthesis. See also note to Table 4.l.
Total electricity includes self-generated power.
sectors are not available. In addition, we have separated the trends in fuels and
electricity because this distinction is critical for the US trends. And finally, we
have reported the 1965-85 as well as the 1979-85 period, because of the
dramatic acceleration in the sectoral shift effect during the latter interval.
Most economic projections show that the energy-intensive sectors do not
grow as rapidly as the rest of the economy. The projections from the Wharton
Annual Model, maintained by Wharton Econometric Forecasting Associates
(WEF A), are representative of such trends. These projections were used to
standardize the industrial energy demand projections of several models compared in a study by the Energy Modeling Forum (EMF, 1986).
They reveal a continuing trend towards less energy-intensive sectors, although at a slower rate than historically. Durable manufacturing, chemicals
and miscellaneous manufacturing grow more rapidly than aggregate industry
in these projections, while paper, petroleum refining and primary metals grow
more slowly. The slower decline in the shift away from energy-intensive sectors
is due to more widespread economic growth than during the past 12 years. This
growth stimulates the demand for capital goods production, which is very
energy-intensive relative to other manufacturing sectors.
Shifts among six major manufacturing sectors contribute about 22% of the
total decline in fossile fuel use per dollar of output in the reference case
projections reported for the ISTUM-II model in the EMF study.7 Aggregate
fossil fuels intensity falls by 2% per annum between 1985 and 2010 (Table 4.2).
Meanwhile, sectoral shift causes aggregate intensity to decline by 0.4% over
this same 25-year horizon. The rate of decline due to sectoral shift is about a
quarter of that observed for the historical period 1965-85. Table 4.2 also shows
that most of the historical shift effect occurred during the 1980s.
7 The fossil fuel trends for this model were representative of those for the other models in the EMF
study.
62
The dampening of sectoral shift contributes importantly to the slower

decline in aggregate energy intensity for all manufacturing. Fossil fuel use per
dollar of output declines by 36% less in the projections than in the historical
1965-85 period. The decline due to changes in fuel intensity within the major
sectors - shifts in product mix, new processes and improved operation of
existing facilities - appears to be somewhat slower in these projections than
during the historical period.
The ISTUM-II estimates for electricity show the electricity intensity decreasing slightly in the future. This trend was not robust across all models, some of
which projected electricity use rising faster than manufacturing output in the
future. 8
Electricity intensity in this model decreases by about 0.5% per annum over
the next 25 years. There is little change in the real intensity because the
increasing efficiency of electrical equipment essentially offsets the effect of new
applications within a sector. Aggregate intensity is reduced primarily by
shifting towards less electricity-intensive sectors.
4.5 CONCLUSION
Studies in both countries confirm the importance of structural economic
change on future energy demand patterns. This paper has provided an initial
perspective on past trends in the Soviet Union as well as one possible
transitional path in a highly uncertain future. Relevant US trends based upon
a separate US study have also been discussed in order to provide some context
for the Soviet trends.
Future work on this issue should focus on the various channels through
which changes in the economy's consumption and production patterns influence energy use and the causes for these shifts. The use of models of the
economy and its interactions with energy supply and demand will allow
researchers to understand these often complicated relationships better. Moreover, there exists a need for standardizing data on energy consumption and
economic output in the two countries.
ACKNOWLEDGEMENTS
H. G. Huntington wishes to acknowledge support from the International Fund
for the Survival and Development of Humanity for travel to the Soviet Union.
G. A. Boyd wishes to acknowledge support of the US Department of Energy,
Assistant Secretary for Fossil Energy, Office of Planning and Environment,
under contract W-31-109-Eng-38. The authors, however, retain the responsibility for the views expressed in this paper.
8These findings are discussed in Huntington and Myers (1987).
References
63
REFERENCES
Aksutin, P. K., Veretennikov, G. A. et al. (1981) Energetika SSSR v 1981-1985 godah
(Energy of USSR in 1981-1985), (eds. A. M. Nekrasov and A. A. Troitski). Energoizdat, Moscow, p. 352.
Anoshko, V. F., Borshchevski, M. Z., Krivorutski, L. D. et al. (1986) Sistemy energetiki:
upravlenie razvitiem i funktionirovaniem. (Modelling and system studies in energy
demand dynamics) T.2.-Irkutsk, SEI SO AN SSSR, pp. 41-7.
Bestuzhev-Lada, I. v., (ed.) (1982) Rabochaya kniga po prognozirovaniyu (Working
manual on forecasting). Mysl, Moscow, p. 430.
Boyd, G. A., Hanson, D. A. and Sterner, T. (1988) Decomposition of Changes in Energy
Intensity: A Comparison of the Divisia Index and Other Methods. Energy Economics,
October, pp. 309-12.
Boyd, G., Hochheiser, H. W., McDonald, 1. F. et al. (1987) Energy Intensity in
Manufacturing: A Comparison of Historical Results with Forecasts, in The Changing
World Energy Economy (ed. D. O. Wood), Massachusetts Institute of Technology,
Cambridge, Mass.
Boyd, G., McDonald, J. F., Ross, M. et al. (1987) Separating the Changing Composition
of US Manufacturing Production from Energy Efficiency Improvements: A Divisia
Index Approach. Energy Journal, 8, 77-96.
EMF (Energy Modeling Forum) (1986) Industrial Energy Demand. EMF-8 Summary
Report (December 1986). Stanford University Energy Modeling Forum, Stanford, Ca.
Huntington, H. G., and Myers, J. G. (1987) Sectoral Shift and Industrial Energy
Demand: What Have We Learned? EMF-8.3, Stanford University Energy Modeling
Forum, Stanford. Also in Forecasting Industrial Structural Change and Its Impact on
Electricity Consumption (ed. A. Faruqui and J. Broehl), Battelle Press, Colombus,
Ohio.
Marlay, R. (1984) Trends in Industrial Use of Energy, Science, 226, 1277.
Medvedeva, E. A. (1986) Avtomatizatsiya issledovani razvitiya energeticheskogo kompleksa (A System of simulation models to study energy demand). Irkutsk, SEI SO AN
SSSR, pp. 92-105.
Melentiev, L. A. and Makarov, A. A. (eds) (1983) Energetichesky kompleks SSSR
(Energy complex of the USSR). Ekonomika, Moscow, p. 264.
Narodnaye (1985) Narodnoye hozyaistvo SSSR v 1965-1985 godv (The USSR National
Economy in 1965-1985). Finansy i statistika, Moscow.
Pavlenko, A. S. and Nekrasov, A. M. (1972) Energetika SSSR, v. 1971-1975 godah
(Energy of the USSR in 1971-1975). Energiya, Moscow, p. 263.
Werbos, P. (1986) Industrial Structural Shift: Causes and Consequences for Electricity
Demand, in Forecasting Industrial Structural Change and Its Impact on Electricity
Consumption (eds A. Faruqui and J. Broehl), Battelle Press, Columbus, Ohio.
5
Modelling transport fuel demand
Thomas Sterner and Carol A. Dahl
5.1 INTRODUCTION
Transport fuels account for an increasing share of oil consumption, and savings
appear to be both technically and socially harder to achieve than in many
other sectors where substitutes are more easily available. Large sums of money
are invested in trying to improve efficiency of vehicles but the really most
relevant issue is that of whole transport systems. These systems cannot be
planned in detail, however, but are the result of many individual actions. It is
therefore of particular interest to study the economics of the transport fuel
market and thereby to evaluate the efficiency of the price mechanism as an
instrument of policy in this area. Taxes and hence domestic prices of transport
fuels vary considerably between countries (Sterner, 1989a,b; Angelier and
Sterner, 1990) and thus high demand elasticities would imply considerable
differences in consumption patterns.
A large number of different models have been conceived to explain how
gasoline demand 1 is related to price, income and other variables. This chapter
is a systematic review of these models and draws heavily on two earlier pieces
of work: Dahl and Sterner (1991), which is a complete survey of more than one
hundred published studies, and Sterner (1990) in which a large number of
models are tested and compared using one and the same data set for the
OECD. The results are summarized in Table 5.1, which gives average values
from the survey and average OECD (1960-85) values from our own estimations. We find that different models tend to give very different estimates,
showing the need for careful model specification and interpretation of elasticities.
lThe analysis in this chapter applies to highway transport fuels in general. Empirical work is,
however, nearly always focused on gasoline, presumably for reasons of better data quality.
66

Table 5.1 Summary of average elasticities by category
Price
elasticity
IR
Income
elasticity
SR
IR
LR
Model and data

type
Cat. SR
Panel data
-0.52
0.41
-0.53
-0.28
1.16
1.37
-1.01
-1.07
0.61
1.09
Static model TS data

Static model TS data
Static vehicle model
CS data
Static model CS data
Dynamic models
Lagged endogenous
Lagged endogenous
Polynomial distributed
lag
Inverted V lag (5.8.1)
Inverted V lag (5.8.2)
Vehicle models
Simple vehicle
Simple vehicle
Vehicle characteristics
Vehicle lagged
endogenous
Drollas model
A
A
LR
0.40
-0.24
-0.25
-0.80
-0.85
0.45
0.37
1.31
1.15
-0.20
-0.22
-0.21
-0.96
-0.94
-0.54
0.35
0.39
0.52
1.15
1.09
1.25
A
S
-0.31
-0.32
-0.16
A
S
S
S
Pooled estimators
Baltagi and Griffin (LE) S
Lagged endogenous
OECD (1960-85)
A
Hughes error
corrections
S
Vehicle
elasticity
0.52
0.55
0.29
0.53
0.43
0.48
-0.12
-0.41
-0.29
-0.77
0.38
0.42
0.60
1.11
-0.15
-0.80
0.14
0.74
-0.18
-1.35
0.10
0.73
-0.18
-1.42
0.33
0.33
0.19/0.32
S, Survey, average results from survey of earlier work; A, Own average (OECD) estimates taken from
Sterner (1990). TS, time series; CS, cross-section; SR, IR, LR, short-, intermediate- and long-run.
Model types and categories etc. are explained in the text. Figures are presented here in the order they
appear in the chapter.
5.2 THE DETERMINANTS OF INDIVIDUAL MOTORING

FUEL DEMAND
In principle the demand for any particular commodity is determined simultaneously with all the other variables of the economy. Assuming separability, the
demand for a product would just depend on its own price and the income of
the consumer. Naturally this is a very convenient hypothesis since it facilitates
estimation, interpretation and forecasting. The inclusion of additional explanatory variables will often increase a model's realism and performance but - at
least for the forecaster - this is of little avail if the values of these variables
The level of aggregation
67
themselves are hard to predict. For gasoline, separability is clearly a questionable hypothesis but it is still commonly used.
Let us start with the individual consumer: his demand for gasoline is
obviously a derived demand. The original demand is for transport together
with a number of other automobile-related characteristics such as comfort,
status etc. Thus gasoline demand is determined in a number of separate steps.
First there is the decision to buy a car (or how many to buy) and of what type
(in other words which efficiency); then come the decisions as to its utilization
(how often, how far and also in what way to drive).
Decisions relating to the stock of automobiles are much less frequent than
decisions on utilization. They may be affected to some degree by gasoline prices
(historical or expected) but also, and more strongly, by the price of automobiles, income and a whole range of other variables reflecting the habits, culture
and situation of the individual (or family) concerned. Among the most
important factors are the geographical location of residence in relation to
employment and other activities as well as to the availability of alternative
means of transport.
The daily decision on vehicle-utilization is obviously heavily constrained by
these factors but it is also likely to depend on gasoline price along with
innumerable other variables - such as the weather and various personal
variables such as health or mood, on which we can never hope to collect
information.
5.3 THE LEVEL OF AGGREGA nON
When we use econometric methods to estimate demand for a product such as
gasoline we must assume the existence of a demand function that is constant
over a certain group of consumers (at some level of aggregation) and for a
certain period of time. Thus the modeller has to decide what type of data to
use. Basically there are four different types of data: at the level of aggregate
statistics there are time series, cross-sections and pooled cross-section time
series data. For individual consumers there are panel data (which are a
cross-section of individuals).
From the standpoint of a utility function it is easier to conceive of a function
for an individual consumer. Archibald and Gillingham (1978, 1980) study the
influence of age, sex, race, martial status, education, area of residence and other
variables on individual gasoline consumption. In spite of all the variables
included they only explain a third of the variation in demand as measured by
R2. This does not necessarily decrease the value of their results. Our interest,
on the other hand, generally concerns total gasoline consumption in a region
and explanations at the individual level are not necessarily sufficient at the
aggregate level. Furthermore their explanatory variables are interrelated so
that the elasticities found are quite low (see first row of Table 5.1).
Aggregation introduces, as usual, a series of additional difficulties. Macro
68
parameters are not, in general, related to the corresponding micro parameters

in a simple way. To deduce macro from micro parameters we would need
detailed, a priori information on all the explanatory micro variables. New
variables such as income distribution, gasoline supply factors or levels of
highway and public transport congestion appear at the aggregate level that are
not in any simple way deducible from micro-level information. At the same
time the micro-level variables are all, in principle, still relevant. Thus if young
people tend to consume more gasoline and the composition of population
changes, then this will affect aggregate gasoline consumption. In practice
analysts of aggregate data find it necessary to disregard the majority of
micro-level factors. 2
5.4 THE CHOICE OF ECONOMIC MODEL: THE SIMPLE
STATIC MODEL
The simplest model of gasoline demand begins with the assumption that utility
depends on motor fuel demand (G) and an aggregate of other goods (0).
Consumers know the price of gasoline Pg and the price of the other good Po,
most commonly represented by the consumer price index, and choose G and
to maximize utility U(G, 0) subject to a budget constraint PgG+poO~y or
maximize equation (5.1).
U(G,O)+h(y-pgG-poO)
(h is a Lagrange multiplier)
(5.1)
With the usual assumptions regarding the utility function, this yields gasoline
demand as a function of gasoline price, other price and income. It is commonly
assumed that demand is homogeneous of degree zero in prices and income (so
that doubling nominal prices will leave demand unchanged).3 This allows
prices and income to be deflated into real terms yielding equation (5.2).
G=f(P, Y)
(5.2)
Assuming log-linearity gives the simple static model 4 as in equation (5.3.1):

LnG=c+oclnP+/Jln Y+<;
Where Y=Y/Po and Pg=pg/po. (Where there is

no risk of confusion P is used instead of Pg for
the real gasoline price.) <; is a random error term.
(5.3.1)
lOn the issue of aggregation see for instance Barker and Pesaran (\990) and Sterner (\990).
3 Hughes (1980), one of the very few who explicitly enters the rate of inflation along with nominal
prices instead of just using deflated prices, finds some evidence of money illusion on a quarterly
and monthly basis.
4This corresponds to G=cpyP. If simple linearity is assumed the equation would be
G=c+rxP+{JY. We found no systematic difference between average elasticities for these two
functional forms, see further Sterner (1990).
Dynamic models
69
As already mentioned, aggregation is problematic. We must, however,

consider at least the number of people we are aggregating over to distinguish
between the effects of a rise in GNP due to an increase in population (with
constant per capita income) and an increase in GNP/capita (with constant
population). Even if individual income elasticity exceeds unity (with respect to
per capita income) intuition suggests that increases in population should not
increase gasoline consumption more than proportionately (in fact the increase
should be less than proportional if there are any effects on congestion or on
public transport availability). 5
Ln( G/ N) = c + rJ. Ln P + f3 Ln (Y/ N) +
(5.3.2)
The most crucial problem for the simple static model is that there may not
be sufficient time for total adjustment to changes in price and income within
the unit time period of our data. This implies that the adjustment captured will
be less than total. At the top of Table 5.1 we see that this appears to be the
case for yearly time series data. Average price elasticities are - 0.5 in the survey
or -0.3 for Sterner's (1990) estimates. Income elasticities are, however, much
more like the long-run values found in other models. Monthly or quarterly
data estimates tend to give even lower values for both price and income (Dahl
and Sterner, 1991). Cross-sectional data do, however, give much higher values
(around -1), suggesting that these data reflect adjustment to long-run differences in price.
5.5 DYNAMIC MODELS
Dynamic models are essentially used to capture the fact that adaptation takes
time. If circumstances such as income or price change this year and the
consumer reacts by buying a bigger or smaller car or moving from one area of
residence to another, then this will continue affecting gasoline consumption for
many years to come. Hence we may argue that today's consumption is not only
a function of today's income and price structure but of earlier incomes and
prices as well. In this case the omission of these variables amounts to a
misspecification of the model explaining the intermediate range values found
(see Appendix).
One of the earliest and most widely used of the dynamic models was the
partial adjustment model which hypothesized that people adjust consumption
only partially to changes in price and income because of inflexibility in the
stock of consumer durables. Since information on this stock was often not
5The interpretation of equation (5.3.2) for instance assumes that (if per capita income is constant)
the population elasticity of gasoline consumption is 1. In some studies, population is added
separately. A better alternative would be explicitly to enter the characteristic of the population that
has changed such as age rather than total population but this is rarely done.
70
available, the model was devised with no stock specified. 6 In this model, a
desired consumption of gasoline G* is based on real income and real gasoline
price as in equation (5.3.3). Given habits, residential location, the durability of
the auto stock, etc. consumers may not be able or willing, when prices and
income change, immediately to adjust to their new level of desired consumption. Instead they adapt partially, by a fraction s as shown in equation (5.4).
G~ =c+rxPt +
f3Yt +/lt1
Gt-Gt- t =s(G~-Gt-d+/lt2
(5.3.3)
(5.4)
Combining (5.3.3) and (5.4) and rearranging gives the familiar. 7

G=sc+srxP +sf3Y +(1-s)Gt - t + Ilt
(5.5)
where
Ilt = S/lt1 + /lt2
The advantages of this formulation, which we will refer to as the Lagged

Endogenous (LE) model, are that it is quite undemanding in terms of data and
there is ease in interpreting the results - short-run elasticities are coefficients
(rxs, f3s) on the current values, while those in the long run are derived from
coefficients divided by s (rx and 13). In practice the lagged quantity by reflecting
the inherent inertia of custom tends to improve the statistical fit considerably. 8
Note that equation (5.5) is equivalent to (5.6) (compare Koyck, 1954) which
helps us understand the model more explicitly.
G=sc+srx[P t +(1-s)Pt - 1 +(1-S)2Pt _ 2+ ... J+
+sf3[Yt +(1-s)Yt - 1 +(1-S)2Yt _ 2 + ... J+'1t
(5.6)
where
'1t=llt+(1-s)llt-l +(1-S)2 Ilt _2 + ...
Thus the implied lags on the exogenous variables form a geometric series. For
instance, the lags on price are rxs, rxs(1-s), rxS(1-S)2 etc. A priori it seems quite
6For early energy models involving partial adjustment see, for instance, Houthakker and Taylor
(1966). See also Griliches (1967).
7Variables without subscript refer to time t. For reasons indicated above, it is preferable to use
gasoline and income per capita. It is also best to deflate and in general log-linear forms are used.
For reasons of simplicity, we will, however, here as well as in most of the equations that follow,
just write G. 'G' may thus generally be interpreted as In(GjN) and 'Y' as In(YjNjP).
8Similar models may also be derived from the so-called adaptive expectation model in which we
have for instance: G,=C+IXX;"+e, and X;"-X;"_I =(1-s)(X,-X~_I)' This gives (5b):
(5.5b)
This is similar to (5.5) but the error term is more complicated and OLS estimates turn out to be
inconsistent but instrumental variable or maximum likelihood estimates can be used. When using
this type of model, with a lagged endogenous variable, it is important to check for autocorrelation
with, for instance, a Durbin h test.
Other lag structures
71
plausible that the effects of a change in price or income have a gradually

decreasing effect through time. 9 Table 5.1 shows that average price elasticities
for this model are around -0.25 in the short and -0.8 in the long run. For
income the corresponding figures are 0.4 and 1.2 or 1.3.
5.6 OTHER LAG STRUCTURES
Naturally, there are other less restrictive ways of modelling the process of
adaptation, which we call Distributed Lag models. In general they can be
written as equation (5.7).
G=c+
f1
r2
i=O
i=O
L (Xi Pt-1 + L Pi Yt-
+Ct
(5.7)
This formulation again allows us to distinguish between the short-run elasticities (which are elasticities of G with respect to the current period's variables
only) and the long-run elasticities (which are the sum of the elasticities of G
with respect to each of the lagged values and the current value). A disadvantage
is that there are very many parameters to estimate and only a fairly limited
number of observations available but this can be partly remedied by imposing
restrictions on the structure of the lags. They could for instance be required to
follow a polynomial of a certain degree. In practice this type of model is not
so commonly used since it requires long series of data and because of
collinearity between the lagged values. Sterner's (1990) estimates with this
model gave similar values to the lagged endogenous (see Table 5.1).10
Another intuitively appealing lag structure is the inverted V, which implies
that adjustment is low in the earlier periods, increasing, and then decreasing.
The reasoning behind this lag structure is that adjustment is costly and takes
time and there may be a 'perception lag'; these factors account for a low
immediate rate of adaptation. Adjustment then picks up and reaches a
9We must, however, be aware that the assumption that they decline geometrically and identically
for income and price is quite a strong restriction. If the true lag structure is different then such
restrictions will obviously bias the results of both the short- and long-run elasticities. It is possible
to devise lagged endogenous models with different rates of geometric decline for the influence of
two (or more) exogenous variables. For instance
G, =c+cc1:s i P'-i + {J1:ri Y'-i +s,

can be transformed into
where
1J,=s,-(s+r)s'_1 +srs'_2
(see Kmenta, 1971, p. 478). We have, however, not seen any empirical work using such a model.
IONote that the geometric lag is, itself, a simple example of the polynomial distributed lag (PDL).
72
maximum after which it declines in the same way as with the geometric lag.
This type of lag structure can be modelled in a number of ways. One
alternative is shown in equation (5.8.1) which combines a lagged endogenous
term with one (or more) lags on the exogenous variables:
(5.8.1)
If we allow an extra lag (or several lags) on the price term then lags can first
increase for one (or several) period(s). After that the lag structure, however,
again becomes that of a geometrically declining series. 11 Another practical
approach to modelling an inverted V lag is to assume that the weights IX and
f3 in (5.7) follow a Pascal lag which can be used to derive an estimation
equation such as (5.8.2).12
(5.8.2)
Table 5.1 shows the average results found in the survey for studies using
(5.8.1) to be quite close to the usual dynamic model values. Sterner (1990) found
(5.8.1) could be rejected using statistical testing. Results for (5.8.2) are reported
in Table 5.1 and have a rather low long-run value for priceY
5.7 VEHICLE MODELS
The rationale behind the assumptions of various complicated lag structures is
the process of adaptation, which to a large extent is reflected in the stock of
vehicles. Another approach to modelling gasoline demand is thus to include
directly variables that reflect this stock (or certain characteristics of the stock).
In the simplest case we can assume that consumers purchase gasoline (G) and
automobile services (A) to produce transport miles - M = f(G, A) - thus
abstracting from other automobile-related characteristics such as comfort, as
well as the durability problems associated with automobiles. Now consumer
utility functions contain miles and another good which they maximize subject
to their income constraint. (Alternatively household production theory can be
applied.) Thus, they choose the 0, G and A that maximizes U{f(G, A), O} +
h(Y -POO-pgG-PaA).
The first-order conditions to this equation give the familiar consumer results
that marginal rates of substitution for miles and the other good must be equal
11 With h 'free' lags these will be written as Lr=os(l-s)iIXh-i' Note that these are unrestricted if
there is no restriction on the IXi' After h periods the lags will again decline geometrically so that
the lag at time t will be:
h
(l-s)('-h)
L s(l-s)i IXh _i
i=O
12Note that as usual with the Koyck transformation, serial correlation may be introduced into the
error term and this together with the overidentification of (5.8.2) can imply problems for
estimation. See, for instance, Kmenta (1971) p. 489 for details.
13Nested tests could not be used here given the parameter restrictions in (5.8.2).
Vehicle characteristics models
73
to the ratio of their relative costs and the familiar production results that the
marginal rate of transformation of gasoline and automobiles must be equal to
their relative prices. Solving and assuming homogeneity of degree zero in
income and price gives:
G = c + cx( pg/Po) + P(Y/Po) + r( P./Po) +t:
(S.9)
Aggregate formulations of equation (S.9) in either a static or dynamic framework have been used in a few studies. In practice, however, the most common
approach has been to assume, in the short run, that the stock of autos (A) is
fixed. Under this assumption (and assuming that prices can be deflated) we
arrive at the simple vehicle model (S.lO):
(S.10)
Extensive data on the stock of autos has made this formulation quite popular.
Estimated values are, however, small (Table S.1) and should be interpreted as
short-run.
In this context we should also mention those models that include the
availability or price of alternative modes of transport. These models are
generally derived as a reduced form in which the complete system of equations
reflects the demand for gasoline simultaneously with the demand for cars, other
transport, etc.
The advantage of these models is of course a degree of realism, which,
however, is achieved at a certain cost, including the difficulty of finding reliable
data. The principal problem however with the vehicle models lies in their
interpretation. They fail to capture the process of adaptation which takes place
through the replacement of vehicles. In models that hold fixed the number of
automobiles, we find that the price and income coefficients mainly pick up
short-term effects. Long-run decisions are embedded in the vehicle stock and
cannot be captured in such simple reduced-form equations. To capture
long-run elasticities in this context we would need a model with simultaneous
equations for gasoline and vehicle demand. 14
S.8
VEHICLE CHARACTERISTICS MODELS
The above formulations fail to take into consideration automobile size and
other characteristics. However, some studies of auto demand indicate that the
number of automobiles is not so sensitive to the price of gasoline, but rather
14Most conventional studies rest on the assumption that the supply curve shifts and the demand
curve is stable, no rationing occurs (sometimes rationing is accounted for by dummies) and thus
the observed data points will allow us to identify the demand curve. However demand may also
shift (due to changes in taste, technology or regulations - such as catalytic converters or mandatory
fuel efficiency standards) and supply-side effects may contaminate the demand elasticities estimated. The only way of tackling such problems is through the estimation of systems of equations
including both supply and demand for gasoline - and possibly equations for other petroleum
products, for vehicles, for other modes of transport etc.
74
people adjust through changes in the fuel consumption characteristics of the

vehicle stock. Holding the stock of vehicles constant may not be enough. If
there is some adjustment in vehicle characteristics in the observed time period,
the short-run elasticities may capture some of this effect as well as changes in
utilization of the stock. Some models therefore also include vehicle efficiency
(measured for instance by miles per gallon) or other measures such as the size
of the automobiles, the proportion of trucks, and so forth. This potentially
enables the researcher to include not only the adaptation of the vehicle stock
in terms of quantity but also of performance, thus leaving only the short-term
variation in utilization to be captured by the price and income terms. These
vehicle characteristic models capture an even less responsive short-run elasticity than the simple vehicle models (average price and income elasticities are
almost halved in Table 5.1), showing that all of the short-run adjustment in the
vehicle models cannot be due to changes in utilizationY
5.9 DYNAMIC VEHICLE MODELS
A number of researchers have estimated dynamic vehicle models including
both vehicles and lagged variables. A very general formulation allowing lags
on price and income is the vehicle-distributed lag model (5.11):
(5.11)
Sometimes the Koyck transformation is used analogously with equation (5.5)

to give a 'lagged endogenous-vehicle model' (5.12):
G=sc+saP+s!3Y+sDA+(I-s)G t -
+)1
(5.12)
The reader should note that equation (5.12) actually implies that vehicles (as
well as any other additional explanatory variables used) also have a geometrically declining influence on gasoline consumption (compare equations 5.5 and
5.6). This is a little hard to accept and the resulting long-run estimate of -0.29
in Table 5.1 should serve as a warning against simply 'sticking in' variables in
the equations to be estimated without properly considering the structural form
of the economic model.
5.10 SOME FURTHER EXTENSIONS OF GASOLINE
DEMAND MODELLING
There is a continuous flow of new articles in this area and it is hard to pinpoint
exactly anyone particular direction for research. However, more attention is
15Some studies model demand adjustment by estimating a miles travelled equation and miles per
gallon in place of or along with a demand for gasoline. For a discussion of some of the miles or
miles per gallon equations see Dahl (1986).
75
Use of pooled data
being paid in many studies to structural and functional form and at the same
time to lag structure and correct specification of the error structure.
Many models start by going back to fundamentals and by noting that
gasoline consumption G can be dis aggregated as follows:
G=A x (MjA) x (GjM)
where A is the stock of autos or vehicles, MjA is the miles driven per vehicle
and GjM is their average efficiency measured in gasoline consumption per mile.
To model this directly, one would need detailed data on the age structure of
cars and the efficiencies of different vintages. Another approach is taken by
Drollas (1984), who starts by dis aggregating gasoline use into U for utilization
(GjA) and A for the stock of autos. He then goes on to model these separately:
where
and
In G = In U + Ln A
(5.13.1)
In U = c 1 + Ct lIn P g + PlIn Y
(5.13.2)
In At -In At - 1 = s(ln At -In At - 1 )
(5.13.3)
In At =C2 + Ct21n(PgjPr) + p21n Y+ c:51n P a
(5.13.4)
s is rate of adjustment in the stock of vehicles;

P r is the real price of (non-auto) transport services;
P a is the real price of automobiles.
Combining these gives the model (5.13.5) to be estimated (Drollas also

estimates the model with a more complicated Pascal lag which gives additional
terms such as Gt - 2 in the equation estimated).
In G=(CI +C2S)+(Ctl + Ct2s)ln Pg
- Ct2s1n P r + (P 1 + P2 s)ln Y+ c:5s1n P a
- Ct1(1- s)ln P gt-1 - P1(1- s)ln 1';-1 + (1- s)ln Gt -
(5.13.5)
1
Because of the restrictions on the parameters of (5.13.5) (and the Pascal or

inverted V case), Full Information Maximum Likelihood estimates are used.
Unfortunately, this model needs the prices of autos and 'alternative modes
of transport', and particularly the latter is very hard to define properly.
Anyhow the estimates are not far from what we expect even if the long-run
price elasticity is a little low.
5.11 THE USE OF POOLED DATA AND THE

COINTEGRATION OF VARIABLES
As mentioned earlier, the use of cross-sectional data appears to give much
higher elasticity estimates. Baltagi and Griffin (1983) argue very strongly in
favour of the use of pooled cross-section time series data to gain efficiency.
76
They do, however, disregard adaptation of the number of vehicles to the price
of gasoline and model gasoline per vehicle GjA as utilization U (=MjA)
divided by efficiency E (= Mj A). Utilization is modelled by a formula analogous with what we have called the simple vehicle model (including income per
capita, gasoline price and the stock of vehicles per capita). Efficiency is
modelled as a function of income per capita and gasoline price but with distributed lags. In estimation both Koyck transformations and polynomial distributed lags are used. Table 5.1 shows their average 16 results with the lagged
endogenous model along with the values found by Sterner (1990) using a very
similar model. Sterner found that the inclusion of data for the 1980s and for a
larger variety of countries increased the estimated elasticities considerably. 1 7
Hughes (1988) also uses a somewhat similar model, but noting that the
gasoline per car series in most countries follow some sort of Random Walk, he
tests for non-co integration of the dependent and explanatory variables and
finds that this can in general be rejected. He therefore uses an error correction
model and again finds, for a pooled sample of OECD countries 1955-85, very
high long-run price elasticities. His model appears, however, to have trouble in
determining the long-run income elasticities.
5.13 CONCLUSION
It should be obvious from Table 5.1 that we are not out to find a unique set
of consensus values since elasticities surely vary between regions and time
periods. On the other hand, we do find some degree of consensus between a
number of models when applied to the same data and we also find models that
consistently seem to give odd or unsatisfactory results. We also find it natural
to interpret the results of some models as short- intermediate- or long-run.
We found that long-run elasticities can be estimated with either dynamic
models on ordinary time series data or with simple static (or vehicle) models
on cross-sectional data. The average elasticities for the dynamic models on time
series data generally fall in the interval -0.80 to -0.95 for price and 1.1 to 1.3
for income. Simple static models using cross data give roughly unitary
elasticities (+ j -1.0) for income and price. The income elasticity is lower in
vehicle models because of the inclusion of vehicles.
The same static models on time series data give intermediate values,
particularly for price elasticities. Other elasticities that may be considered
intermediate include the dynamic models estimated with monthly or quarterly
data (Dahl and Sterner, 1991).
Recent models using pooled data indicate that price elasticities may be even
16The average is used since the authors use five different GLS estimators with different results.
17 Sterner does not use gasoline per vehicle but total gasoline which is an additional but minor
explanation for higher estimates.
Appendix
77
as high as -1.3 or -1.4. This can be compared to the results found by Baltagi
and Griffin (1983) in their extensive analysis of pooled estimators for the
OECD. They concluded that, depending on the method used, long-run
estimates should lie between -O.SS and -0.9 (for gasoline per vehicle, which
naturally gives lower values).
The dynamic models mentioned above all give estimates of short-run in
addition to the long-run elasticities. These estimates, together with the results
from the vehicle and the vehicle characteristics models, generally fall in the
range -0.1 to -0.3 for price and O.1S and O.S for income. The estimates for
the rate of adjustment show that we are up against long-run changes: as much
as 10% of total adjustment may still be left after 10 years. IS
APPENDIX
Considering that there are so many alternative models estimated it is useful to
reflect on the effect of omitting a relevant explanatory factor. Suppose the true
model were (S.14.1) but we estimate (S.14.2) in which the variable X has been
omitted:
G=c+aP+ f3Y +rx
(S.14.1)
G=c+aP+bY
(S.14.2)
Suppose furthermore that X is correlated with Y but not with P. It can then
be shown 19 that the estimate b will be biased:
E(b) = f3 + PI>XY
(S.14.3)
Where <l>XY is the regression coefficient of the omitted variable X on Y.

In practice it is not always clear which is the 'correct' model (it depends on
the purpose of the exercise). It is, however, clear that the inclusion of additional
explanatory variables will systematically affect the elasticities depending on
how they are correlated with price and income. Thus the inclusion of vehicles
should reduce income and price elasticities.
A similar analysis can be applied to the omission of lagged values in the
simple static model (S.3.2). Since the income and price variables are positively
correlated with their own lagged values exclusion of the latter will tend to give
intermediate elasticities: suppose the correct (log-linear) model is (S.lS.1) and
181[ the coefficient on the lagged endogenous variable is 0.6 or 0.8 this implies 40% or 20%
adjustment the first year and (with a geometric lag) 99.4% or 89.3% adjustment in 10 years. Note
that the longest PDLs used are generally around 10 years.
19For detailed analysis see, for instance, Kmenta (1971) pp. 392-3. It should also be observed that
omission of relevant variables not only causes the estimates to be biased but also inconsistent and
their variance will be biased even if the omitted variable is not correlated with the other
explanatory variables.
78
the estimated model is (5.15.2):
G=C+ 1X 1Pt+ 1X 2 Pt-1 +f3Y
(5.15.1)
G=c+aP+bY
(5.15.2)
then the 'static' model's (5.15.2) price elasticity will be a. The 'dynamic' model
(5.15.1) has a short-run elasticity of 1X1 and a long-run elasticity of (1X1 + 1X2) and
since the regression coefficient <I> between P t and P t - 1 can be assumed to be
positive but smaller than 1 we find that, in analogy with (5.14.3), the estimated
elasticity a is positively biased compared to 1Xl> the short-run elasticity, but
smaller than the sum of the two elasticities (1X1 +1X2) that make up the long-run
elasticity:
(5.15.3)
ACKNOWLEDGEMENT
Thanks are due to the Swedish Transport Research Board for funding.
REFERENCES
Angelier, J. P. and Sterner, T. (1990) Tax Harmonization for Petroleum Products in the
Ee. Energy Policy, March.
Archibald, R. and Gillingham, R. (1978) 'Consumer Demand for Gasoline: Evidence
from Household Diary Data.' College of William and Mary and Bureau of Labor
Statistics (mimeo).
Archibald, R. and Gillingham, R. (1980) An Analysis of the Short-run Consumer
Demand for Gasoline Using Household Survey Data. Review of Economics and
Statistics, 62, 622-8.
Baltagi, Badi, H. and Griffin, James, M. (1983) Gasoline Demand in the OECD: An
Application of Pooling and Testing Procedures. European Economic Review, 22,
117-37.
Barker, T. S. and Pesaran, M. H. (eds) (1990) Disaggregation in Economic Modelling.
Routledge, London.
Dahl, Carol A. (1986) Gasoline Demand Survey. Energy Journal, 7, 67-82.
Dahl, Carol A. and Sterner, T. (1991) Analyzing Gasoline Demand Elasticities. Energy
Economics, April.
Drollas, L. P. (1984) The Demand for Gasoline: Further Evidence. Energy Economics,
6, January, pp. 71-82.
Griliches, Z. (1967) Distributed Lags: A Survey. Econometrica, 35, pp. 16--49.
Houthakker, Hendrik, S. and Taylor, Lester D. (1966) Consumer Demand in the United
States, 1929-1970. Harvard University Press, Cambridge, Mass., p. 116.
Hughes, Warren R. (1980) Price and Income Elasticities of Demand for Motor Gasoline
in New Zealand, University of Waikato, Hamilton, New Zealand (mimeo).
Hughes, G. A. (1988) 'On the Consistency of Short and Long Run Models of Gasoline
Demand', paper given at the Scottish Economic Society Conference, St Andrews, 9
April 1988.
References
79
Kmenta, Jan (1971) Elements of Econometrics. Macmillan, London.

Koyck, L. M. (1954) Distributed Lags and Investment Analysis. North Holland,
Amsterdam.
Sterner, T. (1989a) The Politics of Energy Pricing: Oil Products in L. America. Energy
Journal, 10, 25-45.
Sterner, T. (1989b) Le Prix des Produits petroliers en Afrique. Revue de I'Energie, no.
415, November, pp. 3-11.
Sterner, T. (1990) The Pricing of and Demand for Gasoline. Swedish Transport Research
Board, Stockholm (ISBN 91-87246-57-0).
6
Modelling the long-run supply
of coal
Ronald P. Steenblik
6.1 INTRODUCTION
There are many issues facing policy-makers in the fields of energy and the
environment that require knowledge of coal supply and cost. Such questions
arise in relation to decisions concerning, for example, the discontinuation of
subsidies, or the effects of new environmental laws.
The very complexity of these questions makes them suitable for analysis by
models. Indeed, models have been used for analysing the behaviour of coal
markets and the effects of public policies on them for many years. For
estimating short-term responses econometric models are the most suitable (see,
e.g., Labys and Shahrokh 1981; US EIA, 1986). For estimating the supply of
coal over the longer term, however - i.e., coal that would come from mines as
yet not developed - depeletion has to be taken into account. Underlying the
normal supply curve relating cost to the rate of production is a curve that
increases with cumulative production - what mineral economists refer to as the
potential supply curve. To derive such a curve requires at some point in the
analysis using process-oriented modelling techniques.
Because coal supply curves can convey so succinctly information about the
resource's long-run supply potential and costs, they have been influential in
several major public debates on energy policy (see, for example, Layfield, 1987).
And, within the coal industry itself, they have proved to be powerful tools for
undertaking market research and long-range planning. The purpose of this
chapter is to describe in brief the various approaches that have been used to
model long-run coal supply, to highlight their strengths, and to identify areas
in which further progress is needed. The chapter starts with a review of
concepts and terminology, in order to provide a framework for discussing the
actual models.
82
Modelling the long-run supply of coal

6.2 POTENTIAL SUPPLY: A DIAGRAMMATIC EXPOSITION
Harris and Skinner (1982) state that the ideal model of long-term mineral
supply would consist of three separate but interconnected modules: an endowment inventory module, an exploration module and an exploitation module.
In the inventory module each deposit would be described by key attributes,
including location, quality, and in situ geological characteristics, in a manner
whereby tonnage estimates could readily be aggregated according to desired
combinations of these attributes. The exploration module would then act upon
this inventory, examining each deposit from the standpoint of discoverability.
For most mineral products, such as oil, the modelling of exploration is
imperative (Chapter 8). In the case of coal, however, the bulk of the endowment
is already 'discovered'; hence the task is usually limited to estimating the
amount of effort, adjusted for uncertainty, that would be required (within the
reference period) to evaluate and develop various deposits or portions of
deposits. Finally, the exploitation module would simulate development decisions in these deposits, evaluating them for that optimal configuration of size,
extraction method, etc., that would maximize discounted profits. The ultimate
product would be a potential supply curve, or schedule, showing the amount
of coal that could be produced at different mine-mouth prices.
Before discussing the individual modules, an explanation of terms is in order.
An endowment inventory is essentially an aggregation of estimated quantities
contained in deposits that meet certain geotechnical limits with respect to
grade, thickness of seam, burial depth, size and suchlike. The classification of
endowment is analogous to the classification of reserves and resources, except
that in place of an economic criterion (cost per tonne), geotechnical criteria are
used. These geotechnical criteria are usually important determinants of mining
costs, but rarely are their effects on costs linear. Figure 6.1 shows how
endowment categories might be defined using a single geotechnical criterion,
say overburden ratio (i.e., the ratio between the thickness of the rock and soil
between the surface and the top of the seam, and the thickness of the seam). In
classifying coal endowment, geologists often use two limits, one defining the
reserve base (limit 'A' in the figure) and one defining the rest of the total
endowment (limit 'B'). All other accumulations of material not satisfying these
inclusion criteria are contained in the remainder of the resource base. The
boundary values for excluding coal occurrences from the endowment should
be set below the values that would apply to resources, and near the foreseeable
technological limits of mining. The ideal endowment inventory would also be
probabilistic, describing the statistical distribution of key geological attributes
(rather than average values).
Mineral endowment is a relatively new term, and arose out of the need to
distinguish the geologist's concept of resources as a stock inventory from the
resource economist's concept of resources as a flow (Harris and Skinner, 1982).
Dorian and Zwartendyk (1984: p. 660) define it succinctly as 'the physical
83
Potential supply
Geotechnical
limit 'B'
REMAINDER OF THE RESOURCE BASE
L-----J..
~I:
Geotechnical
limit 'A'
I
~
IDENTIFIED
MINERAL
ENDOWMENT
.. .
UNDISCOVERED
MINERAL
ENDOWMENT
RESERVE
BASE
: Measured: Indicated:
'
I
.. --- - - - - -- -- - -- -- - - -, Inferred : Hypothetical: Speculative
Demonstrated:
:
:
:
~-------------------~---------~----------------------
:L _____________________________
IDENTIFIED
: ______________________
UNDISCOVERED _
~
Geological assurance categories
Fig. 6.1 Mineral endowment classification scheme.

(Source: Steenblik (1986), Figure 2.4.)
envelope beyond which our technological capabilities do not reach, neither
now nor in the foreseeable future'. An important difference between an estimate
of mineral endowment and an estimate of resources is that an endowment
estimate is immutable with respect to changes in prices or technology. By
contrast, the term reserve should only be used in reference to 'those accumulations of a mineral that are known and have been explored to the extent that
there is a reasonable assurance that the mineral could be produced from them
economically' (Harris, 1984, p. 13) - that is, that it can be produced at current
prices and using currently available technologies. Reserves are a sub-set of
resources.
Potential supply differs from resources, reserves and endowment in that, by
definition, it considers explicitly the trade-off between the costs of producing
from known deposits and the (possibly lower) costs of finding, appraising, and
developing less well-identified deposits. And it is explicitly defined with respect
to time. Formally, potential supply represents 'that part of resources that
would be discovered by an optimum amount of exploration, given the specified
economics and unconstrained markets' (Harris, 1984, p. 14).
While it is possible - indeed, preferable - to obtain an estimate of potential
supply from an endowment inventory, it is easiest to demonstrate the concept
graphically with reference to reserves and resources. Figure 6.2 shows a
hypothetical set of mineral deposits, classified according to degree of geologic
assurance and economic feasibility - as determined at the date of classification.
84

Cost of
extraction
~ ~i"ro
Me"U'"/Ubom,,:::::::,:,
O..c
/
z: ~
+H
0 :":.:" ....... .
~:.~
01
:a
iii
ca
.,
().
Ul::
............
oz
o()
eo
...... ~ 90 per cent

.
confidence limit
::J:"
~
.!!
u
30-year potential supply
()
+.:
UJ
""""" . I - - - - - r - - - - - , - - - - - . - - - - - - - ; r - - -...
: Measured: Indicated:
'
,
1---------->----------, Inferred :Hypothetical: Speculative
Demonstrated:
,
,
:
r--------------------~---------~-----------l----------_
IDENTIFIED
UNDISCOVERED
__________________ :
L ______________________
_
~------------
Geological assurance categories
Fig. 6.2 Relationship between resources and potential supply.

(Source: Adapted from Steenblik (1986), Figure 2.5a.)
Using the scheme suggested by McKelvey (1972), deposits A and B would be
classified as measured economic reserves, deposit C would be an indicated
economic reserve, deposit D would be a para-marginal (or subeconomic)
measured resource, and so on. Assuming rational economic behaviour, the
least-costly and most thoroughly appraised deposits will be developed first.
The trade-off between mining cost and the costs of exploration (the identification of deposits not yet discovered) and appraisal (the detailed measurement
and assessment of the characteristics of an identified deposit) will vary from
mineral to mineral and from place to place.
An estimate of potential supply makes sense only if it is defined with respect
to a stated period of time and corresponding assumptions about technology,
prices and exploration effort. In Figure 6.2, an estimate of 30-year potential
supply might include deposits A through F, but exclude deposits G and H. The
cost of developing and mining deposit H would just be too high given the
assumptions about prices and technology; similarly, because of the uncertainty
surrounding the existence and characteristics of deposit G, this deposit is
unlikely to be developed within the specified timeframe.
No estimate of long-run potential supply can be expected to be precise.
Conceptually, this uncertainty can be depicted by enclosing the boundary of
potential supply within its associated confidence limits (the dotted curve in
Potential supply: a diagrammatic exposition
85
Figure 6.2). One would also expect that the dispersion of uncertainty as to
where the boundary lay would be greater for the less-well identified portions
of the resource - the right-hand area of the diagram - since the geological
uncertainties add to the uncertainties already accompanying the technical and
economic variables.
Figure 6.3 shows schematically how a potential supply curve might be
constructed for the set of hypothetical mineral deposits used for illustrative
purposes in the previous figures. Taking the six deposits, A through F, the task
is to reduce the three dimensions by which they are classified - cost of
extraction, amount in place and the degree of assurance about the amount in
place - and compress them into two: cost and quantity. This is done by adding
to the expected extraction costs the costs of exploring and examining each
deposit to the point where the amount of information available about it
becomes sufficient to take a decision on development. Obviously, the less is
currently known about a deposit - that is, the further right it lies on the
geological assurance axis - the greater the expenditure (and risk) associated
with bringing it into production.
The final step is to arrange each deposit in increasing order of its total
expected unit costs. The recoverable quantity producible from each deposit at
an estimated cost forms a segment of width q and unit cost p that, when
arranged in increasing order, forms a stepped function, the potential supply
curve. The procedures required to translate coal endowment information into
an estimate of potential supply are more complex than when using resources
as the starting point, but because the assumptions and criteria used in
compiling a mineral endowment are apt to be less subjective and mutable, the
empirical results are likely to be superior. The major additional step is the
conversion of the geotechnical data into an estimate of extraction cost.
.2
ii
c
.. 0
Cumulative 3D-year
potential supply curve
0-
-a.ca
IIC::::
GI 0
-'Q.
_ GI
III
O'Q
1.1 C
01 as
_c as
III III
as .-
GI ~
.. Q.
1.1 Q.
.5 as
Cumulative production
Fig. 6.3 Idealized cumulative potential supply curve.

(Source: Steenblik (1986), Figure 2.6.)
86

6.3 DISTINGUISHING CHARACTERISTICS OF COAL
While similar in many respects to other hard-rock minerals, coal has several
distinguishing characteristics that make the task of modelling its supply easier
than that of some other minerals. First, the structure of the coal supply
industry has generally been considered to be workably competitive. Second,
user costs are not considered to have an important bearing on price (Zimmerman, 1981; Adelman et at., 1990). By not taking this element into account, the
computational aspects of estimating the supply response are greatly simplified.
Coal supply modelling is also facilitated by the existence of relatively
abundant data on the location and geological disposition of coal. Enough is
known about the whereabouts of coal in most regions that new discoveries play
a relatively small role in its supply. But though the location and approximate
amount of coal in the ground in many areas is well assured, this does not mean
that no geological risk is associated with mining: for coal, the uncertainty lies
in the detail. Especially in underground mining, the expensive surprises are
often not discovered until the mining machinery is already in place.
Decades of mining experience, confirmed by engineering-cost analyses, have
shown that any from a wide range of geometric, geomechanical and geochemical characteristics can profoundly affect the costs of mining (see, e.g., Barnett,
1980; Klein and Meany, 1984). These include the size of the deposit; the
thickness of the coal seam; the depth of the rock and soil overlying the seam
(the overburden); the amount of tectonic disturbance; the angle from horizontal
in which the seam occurs; the friability of the overburden; and the rate of inflow
of water and methane into the mining section.
Two characteristics of coal complicate matters: its heterogeneity and its
bulkiness. Grade in the case of metallic ores is typically a one-dimensional
consideration - ultimately, the metal is refined and used in its pure state. While
it is possible to extract pure carbon from coal (as is done in small quantities
to produce carbon rods, for example), such processing is not yet economical
for producing fuel, coal's principal use. Generally, coal mined in its raw state
contains a number of undesirable impurities, such as ash and compounds of
sulphur or chlorine, the concentrations of which can be reduced at the
processing stage, but which cannot easily be eliminated completely. Finally,
coals differ considerably in their thermal content. These considerations mean
that, to be useful, coal supply curves must be distinguished according to quality
characteristics that affect the coal's market value.
The task of describing the distribution of the coal endowment is further
complicated by what might be loosely described as its fractal-like properties:
namely, that the variation of coal tonnages by certain quality or geotechnical
characteristics appears sometimes to be independent of the scale of the area
over which it is measured. The variation in coal seam thickness, for example,
can be as great within seams, and even within individual mines, as it is among
different seams (Newcomb and Fan, 1980). A coalfield with numerous mineable
Models of coal exploitation
87
deposits, each exhibiting large spatial variation in seam thickness, would have
quite different mining costs than one with the same variation overall but
wherein each deposit was homogeneous with respect to the thickness of its coal
seam. Moreover, predicting costs and the sequence of mine development is
much easier in the latter case than it is in the former.
Finally, because coal is bulky, and therefore transport costs weigh so heavily
in its final price, supply functions are meaningful only if they are specified for
fairly limited geographical areas.
6.4 MODELS OF COAL EXPLOITATION
Estimating a supply curve for a region involves three steps: (i) the development
of relationships between the physical conditions of mining and costs; (ii)
defining that portion of the coal endowment that is developable and of
potential economic interest over the period covered by the analysis; and (iii)
using the cost relationships developed in step (i) to transform the geological
data into a potential supply schedule. The term 'exploitation model' refers to
all the procedures used to examine the deposits described in the coal endowment inventory and to calculate their expected costs of development and
production. Because the simulation of development decisions in the coal
endowment necessarily involves translating in situ into saleable quantities, it
also encompasses procedures to take into account losses associated with the
recovery and benefication of coal.
6.4.1 Modelling depletion
The focus of this section is on formal mine-costing models - that is, those that
use objective computational procedures for estimating long-run average coal
mining costs as functions of geotechnical and engineering factors. It may be
observed, though, that many worthwhile empirical studies of coal-supply costs
have been published over the last 5 years that have been based on discounted
cash-flow models of individual or 'representative' model mines (see, e.g, Long,
1986; Jamieson, 1990; US Department of Commerce, 1990). Generally, the
analysts undertaking these studies have relied primarily and directly on the
informed judgement of mining engineers (often from the mining companies
themselves) in order to estimate the values of cost elements influenced by
geological variables. The results of these exercises may be accurate, but their
value is limited because they are difficult to replicate or to recalculate using
different assumptions. Some of the most interesting studies in the public
domain have been those undertaken for the purpose of comparing coal mining
costs among different countries. But too often the critical data inputs to the
models, especially those that would enable other analysts to regenerate the
supply functions using common assumptions, are not reported; this diminishes
88
their usefulness, for example, as inputs to models of world coal trade.

Quebral (1990) has reviewed a number of one-off empirical studies that have
used econometric techniques to explain the effects of changing technical
conditions and factor prices in particular coal industries. The discussion here
is limited to models that have been developed and maintained for extensive and
repeated use.
Much developmental work on formal models of long-run coal supply has
focused on the relationship between costs or productivity and in situ geological
conditions, in order to measure the changes that may be expected in costs as
these conditions deteriorate. Over the past 15 years, numerous mining-process
models of varying degrees of sophistication have been developed, particularly
in North America. There are over 3000 coal mines operating in the United
States alone - an unwieldy number to cost on a mine-by-mine basis. More to
the point, the United States boasts a demonstrated coal reserve base of some
226,000 million tonnes of bituminous coal, against an output of around 600
million tonnes annually. But a large part of this reserve base includes coal that
is not yet - indeed may never be - profitable to extract. It is thus only with the
help of models that sense can begin to be made of the endowment figures.
Beginning in the early 1980s analysts began to apply these models to other
countries, often basing their analyses on US experiences. Notable examples
include Ellis's (1979) study of coal mining costs in South Africa, and ICF
Incorporated's study of Australia, Canada and South Africa (ICF, 1980a).
Basically, two approaches can be distinguished, though some models combine features of both: the engineering-cost approach and the statistical!
econometric approach. The former is exemplified by the coal supply costing
programme, initially developed by the US Federal Energy Administration (now
the Energy Information Administration), and since modified and enhanced by
the consulting firm ICF Incorporated as a component of their Coal and
Electric Utilities Model (see ICF, 1977), and by the EIA for their Resource
Allocation and Mine Costing (RAMC) Model (see SAIC, 1986, 1988a). Although the two versions of the model differ in detail, they are similar in their
approaches. A quite detailed process-evaluation model was developed by NUS
Corporation for the Electric Power Research Institute (NUS, 1981; 1984),
though the application of this model in potential coal supply analysis has been
largely restricted to generating 'pseudo-data' for use in other, regional-scale
models. Finally, Skelly and Loy (CRA, 1986) have developed a model that uses
a subjective point system to evaluate individual deposits, and then calibrates
the results against the point rating of mines of known cost.
The second approach is exemplified by the models pioneered by M. B.
Zimmerman (1977, 1981), and now used primarily by the consulting group
Data Resources Incorporated (DRI). The basic mine-costing relationships
derived by the Zimmerman model are also incorporated into a model that was
developed by another Cambridge (Mass.)-based consulting group, Charles
River Associates (see CRA, 1982). A slightly different mining-cost model, also
89
employing a log-linear functional form, has been developed by Newcomb and

Fan (1980).
The early versions of these models have already been reviewed by Gordon
(1979), and have been the subjects of detailed comparative assessments by the
Energy Modeling Forum (EMF 1978), and by many others, including the
present author (Steenblik, 1985). The ICF/EIA coal supply models - mainly by
virtue of their central roles in US coal policy formation and analysis - have
also received special critical attention by a number of economists and model
analysts (e.g., see Vogely, 1979; Goldman and Gruhl, 1980; Wood and Mason,
1982). What follows below is therefore limited to a very brief description of
their salient features.
Engineering-cost models
In the ICF and EIA mine-costing frameworks, major cost elements (initial
capital, deferred capital and certain elements of annual operating costs), which
are presumed to be functions of the physical conditions of mining, are
estimated individually on the basis of relationships derived from engineeringcost models of representative mines. These model mines are hypothetical
constructs, distinguished by size, mining method (e.g., surface or underground),
and seam conditions. Rules are then developed to vary costs with variations in
mining conditions from those assumed for the base-case model mines; output
is assumed to adjust optimally so that average costs are minimized. Finally the
cost functions are used to evaluate discrete, pre-configured coal deposits.
Statistical-econometric
The statistical approach, as developed initially by Zimmerman and refined by

Barrett (1982), rather than estimating costs from physical conditions directly,
starts instead by defining the relationship between the productivity of the
relevant producing units comprising a mine and the mine's in situ seam
characteristics and output. In deep mining, a production unit is defined as a
mining machine and its complement of miners, i.e., a mining section. The
corresponding measure in the surface mining model is the capacity of the
overburden-removing equipment: the maximum usefulness factor (MUF) of
draglines (defined as the product of the volumetric capacity of the dragline's
bucket and the length of its dumping reach) for area, or strip mining; and the
shovel bucket capacity (SBC) for open-pit mines. Then, using either pseudo
data (e.g., from detailed engineering-cost models) or, preferably, data for actual
operating mines, an equation relating output per production unit as a function
of mine size, and its geological and physical dimensions, is estimated. After
further manipulations, the model eventually yields an expression for minimum
average cost at some optimal level of output. For the sake of brevity, only the
90
equations for an underground mining model are described here; similar

procedures are used in the models for surface mining.
In developing his original model, Zimmerman had data on only three
observable characteristics: number of mining sections, seam thickness and the
number of mine openings. Using public data from 1975 on 244 underground
mines in the United States, each of which produced at least 100,000 short
tonnes in that year, he estimated the following regression:
(6.1)
where q = average mining section productivity; Q= the total annual output of

a mine; S = the number of mining sections operating at the mine; Th is the coal
seam thickness; Op is the number of openings (or shafts) to the mine; and B is
a random disturbance term for the unobservable geotechnical characteristics.
Using the productivity equation as a base, Zimmerman next estimated a
long-run cost function. He showed that necessary expenditures on labour,
operating supplies and capital could be approximated as a function of the
required number of mining sections. Zimmerman used engineering estimates of
expenditure categories for hypothetical mines, but it could also be fitted to
actual mine data, if available. Combining the cost equations yields an equation
for the average cost as a function of measurable geological parameters and
mine output.
On the assumption that, in the long run, the marginal cost of producing coal
will equal the minimum average cost of a mine, the final task is to establish
the minimum efficient scale (Q*) of a mine and then to evaluate the average
per-tonne cost at that point. (Somewhat different means are used to determine
(Q*) for underground and surface mines.) Ultimately, the procedure yields an
equation for the minimum average cost at some optimal level of output.
Following Church (1981), the resulting long-run marginal cost function may be
generalized as:
(6.2)
where Km is a constant corresponding to a particular mining technology.
A major result (and limitation) of the early modelling work by Zimmerman
was the large unexplained residual in the productivity equation. Zimmerman
(1983, p. 308) concluded that 'if only the seam thickness for a given mine is
known, a 90 percent confidence interval includes productivity levels almost
seven times greater than those predicted by' the equation. Barrett (1982) has
shown that considerable improvements in the regression results can be obtained by performing the anaysis on a smaller, regional basis, using empirical
data from mines specific to that region. This makes sense, for as Brooks (1976,
p. 168) observes: 'in one [group of deposits or a district] ... many of the
economic variables will hold constant. That is, though the relationship between
cost and certain physical characteristics may not be simple, it should at least
be stable for all of the deposits under consideration'.
91
Another of Barrett's enhancements was to expand the analysis to include

several technologies. Whereas Zimmerman assumed one type of mining technology, continuous mining (whereby coal is ripped from the coal face and
loaded onto a conveyor belt in one continuous operation), Barrett looked at
three. The results (Figure 6.4) are interesting because they indicate a moderating, through technological substitution, of the long-run effects of depletion.
Thus, in the example, as thick seams are depleted, the preferred mining method
may be expected to change, from the use of continuous techniques to the use
of conventional mining techniques (wherein the blasting, extraction and transportation of coal are handled by different machines). In other coal fields, the
relationship of the curves may be different, with longwall mining playing a part
at some stage of the field's development.
6.4.2 RecoverabiIity and yield

Endowment refers to quantities in situ, but potential supply is concerned
ultimately with producible or saleable amounts. To translate from one to
another one has to take into account all losses or reservations of coal due to
geological, technological, economic and legal factors. Current potential supply
curve procedures treat these losses exogenously rather than incorporating them
as endogenous variables.
40
QI
35
.2.. 30
QI
Q.
~ 25
::;)
CI
00
0)
20
a;
0
.2
U
:::I
"tI
Longwall mining
------
15
mining
......
10
'"
...........
'.,
Q.
Continuous mining
50
75
100
125
150
175
200
225
250
275
300
Coal seam thickness (cm)
Fig. 6.4 Minimum average cost of underground mining by mining method in the
midwest United States as a function of coal seam thickness. The shapes of the curves
are approximations of the originals. (Source: Adapted from Barrett (1982).)
92
Losses and reservations occur at all geographic levels and at every stage of
production. At the broad regional level, portions of the coal endowment will
at any given time be unavailable where beds are overlain by urban structures,
parks and similar surface features with which mining would conflict. At the
mine-property level, coal will often be left unmined and therefore lost where it
underlies mine roads and buildings, or abuts other properties. Where underground mining methods are being used, some coal situated above or below the
bed being worked may be rendered unmineable and thereby lost. Losses
inevitably occur during extraction as well, as a result of spillage, accidents, fires
and so forth. Finally, a portion of the coal is lost or discarded whenever it
undergoes cleaning or beneficiation.
These concepts can be understood with the help of Figure 6.5, which places
these losses in a sequential hierarchy. Implied by the reducing lengths of the
bars in Figure 6.5 is that at each stage the decisions or actions that affect a
Lo.... end
.'eg. of Co./ recoll.ry
Planning
losses
Legally accessible
and mineable coal
II
Regional and
urban planning
Layout
losses
Technically mineable
coal
Original
coal In
place
Lo..g retlng
ectill/ty
Mine
layout
Winning
losses
Cleaning or
.beneficiation, ----,
losses
~
Economically
mineable coal
Extraction
(winning)
Run-of-mine
coal
Cleaning or
beneficiation
Saleable
coal
Fig. 6.5 Relationships of concepts and terminology for designating losses and recoverability of coal. Relative sizes of losses are only indicative. (Source: Steenblik (1986),
Figure 6.2.)
93
coal deposit will reduce the quantity available at the next stage. Viewed over
a finite period - that is, several years - the losses from one stage to another
are likely to be cumulative. Over the longer term, however, certain losses may
at least be partially reversible. In a physical sense no losses (with the exception
of coal lost in fires) are permanent; i.e., material is not destroyed. It may,
however, be so costly to recover that in a practical, economic sense it is lost
forever. Much of the losses that occur during the extraction process fall into
this category. On the other hand, certain types of planning losses may be
regarded as temporary reservations - for example, coal underlying railroad
rights-of-way. Such coal may not be available to contribute to 20-year
potential supply, but some of it could very well contribute to 30-year supply
should the rights-of-way be sold or abandoned.
Between these two extremes may be found a large number of factors
influencing recovery of varying and changing degrees of irreversibility: time,
technology and prices interact in ways that are complex and largely indeterminate. As Harris and Skinner (1982, p. 306) explain, a deposit developed later
when prices for the mineral are high is likely to be mined more intensively than
an identical deposit developed earlier when prices are low. In some cases the
material left behind may be unprofitable to produce under any imaginable
price. In other cases, however, as the deposit is worked, higher prices or lower
production costs may justify expanding the initial mine in order to extract
previously excluded material. Similarly, a portion of the coal rejected as waste
during the screening and washing process and dumped in culms may become
profitable to mine at a later date. Such secondary recovery from culms is
currenty taking place, for example, in eastern Pennsylvania and in Belgium.
Although improvements in technology generally increase the rate of recovery
from a deposit, changes in technology and factor costs can affect the level of
recovery in counteracting ways. As Fettweis (1983) and others have pointed
out, the level of recovery with respect to the coal endowment as a whole
decreased with the introduction of mechanized room-and-pillar mining: seams
that previously would have been mineable using picks and shovels (at high risk
of injury to miners), became unmineable using mechanical cutters. Hence the
'recoverable' portion of the endowment was reduced. More recently, the
introduction of longwall mining machines has increased the percentage of coal
that can be mined profitably from many seams, as compared with recovery
using room-and-pillar methods; but because the method requires more favourable geologic conditions, its increasing use may have a neutral or even negative
effect on overall, ultimate recoverability.
Thus the relationship between the level of recovery and the coal endowment
is largely an economic one. Under conditions of constant technology and factor
prices, mining costs are an increasing function of both deteriorating geological
conditions and increasing rate of recovery (Fettweis, 1983). From the static
perspective, the amount of coal that can be recovered at any particular cost
can come from different amounts in place. But given the partial irreversibility
94
of mining with respect to time, the decision of where to operate on the cost
surface at any moment will require choosing, consciously of unconsciously,
between accepting a low rate of recovery now and faster depletion, or higher
recovery now and slower depletion. This implies also that the level of recovery
is unlikely to remain constant through time.
Considering the complex nature of recovery, it is understandable that
applications of mine-costing models have generally ignored the economic and
time-dependent dimensions of recovery, treating it purely and simply as a
technology-determined matter. None the less, it is important to move away
from assuming global, and rather arbitrary recovery rates; wherever possible
one should instead use empirically derived estimates. There appears to be
strong evidence to suggest that the recovery rates commonly applied to
aggregate estimates of the coal endowment - 50% for deposits mineable by
underground methods, 80% for deposits mineable by surface methods (e.g.,
Averitt, 1975) - are over-optimistic by a factor of two (Schmidt, 1979).
Beneficiation, the removal of unwanted mineral matter (dirt and sulphur in
particular), invevitably results in some loss of material. Again, beneficiation
losses are a function of a number of technological and economic factors,
including coal quality (especially ash content) before and after cleaning, method
of mining, available beneficiation technology and existing cost-price conditions. Accordingly, a wide range of washery losses are possible from any
particular amount of coal in place - anywhere from a few percentage points to
over 50%. In general, the further the coal has to be transported, the more likely
it will undergo cleaning. And coal used for steel-making is normally cleaned to
a higher degree than coal used for power generation.
6.5 THE COAL ENDOWMENT INVENTORY
The preceding section described models for estimating the production costs of
mines. This section examines the data on geological variables that are required
for these models, and the problems involved when the data are incomplete.
An analyst seeking to model coal supply would ideally already have
available an inventory of the region's coal endowment, containing data
describing the characteristics of the coal in situ, by seam, in a manner whereby
tonnage estimates could readily be aggregated according to desired combinations of geographical units, geological attributes, quality attributes and so
forth. Such an inventory would enable the analyst to describe not only the
inter-seam distribution of these characteristics, but the intra-seam distribution
as well. It would also identify the surface land-use around the deposit,
indicating any structures that would inhibit mining (such as oil wells), so that
a judgement could be made about the accessibility of each deposit.
To date, however, detailed inventories of coal endowment have been
produced for only a few small areas of the world. This is not surprising: many
The coal endowment inventory
95
coalfields have not been thoroughly explored; others are controlled by state
monopolies that treat their geological data as commercial secrets. But even
where a large amount of data has been collected and made available to the
public, it has been rare that any attempt has been made to organize the
information systematically. And information that would allow the correlation
of data from different samples is often missing. Only with the advent of
computers have governments considered undertaking the expensive task of
compiling the data in a way that it may be used to construct detailed,
three-dimensional maps of their countries' coal endowments.
The US Geological Survey's (USGS) effort to inventory the coal endowment
of the United States is perhaps the largest programme of this kind. Since 1977,
the USGS, in cooperation with state geological bureaux, has been digitizing
data relating to the geophysical and quality characteristics of the nation's coal
deposits, along with information on land use and cultural features; to complete
the task, several hundreds of thousands of observations will have to be
compiled (Carter et al., 1981; Gluskoter, 1991). A network of microcomputers
and software, known as the National Coal Resource Data System (NCRDS),
has been devised to process the data and, where the density of observations is
sufficient, to generate isoline maps or to calculate quantities of coal in relation
to seam thickness, depth of seam, and various quality attributes. The USGS's
goal is to set up the inventory and supporting software so that it can be
integrated with mining-cost models, such as those under development at the
US Bureau of Mines, to produce production-cost curves at the desired level of
aggregation.
The problem for the moment is that the coverage of the NCRDS is still
rather limited: only around 15 quadrangles (each measuring about 250 km 2 ) in
the central Appalachian coal province, out of an eventual total of 450, will be
completed by the end of 1991. Work is continuing on mapping other areas of
the country as well; but, for the time being, any analyst seeking a comprehensive coal endowment inventory of the United States (or of other countries)
must make do with pre-aggregated tonnage estimates, distributed typically by
administrative units, coal rank, and by a small number of seam-thickness,
depth, and sulphur-content classes.
For the United States, at least, efforts have been made to assemble these
pre-aggregated inventories into a form that can be used as an input to
coal-supply models. This work has been carried out over the last 15 years by
or for the US Department of Energy and its predecessor agencies in an effort
to upgrade the Demonstrated Reserve Base (DRB) of coal, originally compiled
by the US Bureau of Mines in 1975 (for a description of this work see US EIA,
1989, pp. 4-14). The method used has involved, essentially, partitioning each
region's coal endowment inventory into geographical sub-units of known or
calculable quantities and ascribing average values for the relevant physical and
quality characteristics to those sub-units. The quantities of coal identified
within each defined category (representing a range of values for any number of
96
physical and coal-quality characteristics) have then been aggregated across the
region.
Early efforts, limited by poor data and short deadlines, used rather simplistic
procedures for distributing the coal endowment by geotechnical and quality
characteristics ('coal type'). (Documentation of these efforts is to be found in
ICF Incorporated, 1977, 1980b; US EIA, 1982). Tonnages were assumed, for
example, to be distributed uniformally across a wide range of seam thickness
classes, rather than being skewed towards the thinner seams, which studies of
actual coalfields would suggest is the case. Over the last decade, however, both
the endowment data and the methodologies used to distribute it by coal type
have been steadily improved.
One important enhancement has been to introduce principles drawn from
geostatistics into the endowment assessments. Geostatistics differs from classical statistics in that the latter assumes that observations have no spatial
relationship between them; the former recognizes the spatial correlation of
geological phenomena and provides a coherent set of probabilistic techniques
to characterize the degree of continuity (Sani, 1979). Matheronian geostatistics
(Kriging), an interpolative method that gives a 'best' (i.e., the least biased and
with a minimum estimation error) estimate of an unknown spatial variable
(such as seam thickness), has been applied successfully at the level of an
individual deposit or coal bed to quantify the extent and variability of in-seam
coal characteristics, (e.g., Kim et ai., 1980; Pauncz and Nixon, 1980; Tewalt et
al., 1983). But such methods require that the data points be relatively close
together and evenly spaced. Moreover, regional inventories can only be compiled by aggregating the statistics of individually Kriged deposits.
Science Applications International Corporation, a consultant to the US EIA,
has evaluated the DRB and other available data sources to determine whether
geostatistical techniques - what they refer to as 'the optimal approach' - could
be used effectively to re-calculate the distribution of the DRB tonnages by
quality attributes (SAIC, 1986, 1988b). Their conclusion was that 'the data
needed to support such an approach are generally not available' (SAIC, 1988b
4.4). However, SAIC did change the method of weighting the individual
observations - i.e., the 'area of influence' over which the value associated with
a data point could be projected - to more closely approximate the way the data
would be handled using a formal geostatistical procedure. Testing this revised
procedure on the data for four regions (eastern and western Kentucky, and
northern and southern West Virginia), SAIC found that the distribution of the
DRB by thermal value differed significantly from that obtained using earlier
methods.
It seems likely that significant advancements in detailed, aggregative procedures for estimating the distribution of regional coal endowments by important in-seam characteristics must await the further development of computerized data bases (such as the USGS's Coal Resource Data System,
described above). The goal of developing a complete, detailed, computerized
Exploration, discovery and appraisal
97
inventory of the United State's coal endowment, much less the coal endowments of other countries, is many years away from realization. Such a detailed
mapping of the world's 'significant coalfields' was intended when lEA Coal
Research first set up its Coal Resources and Reserves Data Bank Service in the
late 1970s (Gregory, 1979; lEA Coal Research, 1983), but after several years the
enormity of the task was acknowledged and the project was abandoned.
In view of these problems other investigators have counselled using an
alternative, statistical approach. This entails obtaining detailed observations on
a well-explored area within a region, assuming (or, better yet, empirically
deriving) a functional form to represent the distribution of coal tonnages
according to each modelled geophysical and quality variable within the sample
area, and then deriving statistically the parameters of the (joint) distributions.
The parameters so derived would then be used to represent the distribution of
these variables across the whole region.
This approach has been used in modelling the mineral endowment of a
number of hard-rock minerals in the earth's crust (see discussion in Kaufman,
1983, pp. 223-32), but only rarely for coal. The most notable examples are the
studies performed by Zimmerman (1977, 1981). On the basis of detailed
information on the coal endowment of several selected major coal-producing
counties in the eastern and midwestern United States, Zimmerman determined
that the distribution of coal by seam thickness (and also by overburden ratio)
was skewed and could be approximated 'fairly well' by log-normal or displaced
log-normal distributions. His major assumption was that the variance of this
distribution could be applied to describe all coal deposits in the United States.
Other observers (see, especially, ICF, 1980b), questioned the validity of describing coal deposits across a country as large as the United States on the basis
of just one 'representative' area. However, this problem can presumably be
overcome by increasing the number of areas' sampled in detail within the region
under study.
6.6 THE MISSING LINK: EXPLORA nON,
DISCOVERY AND APPRAISAL
In most mineral extraction industries, the search for new and richer deposits,
and the discoveries that are the fruit of such activities, play an important role
in the long-run supply response (see Chapter 8). The simulation of this
exploration and discovery process therefore forms an integral link between the
cost modules and the endowment inventories in models of potential supply. In
the Harris and Skinner scheme, exploration refers to the activities involved in
searching for deposits where the presence of the target mineral is only
suspected. By this definition little exploration is carried out by the coal
industry: because of coal's shallow occurrence, its ease of identification and
information provided from the search for other minerals and water, knowledge
98
about the existence and rough magnitude of coal occurrences is already

extensive. In well-explored areas, such as Europe and the United States,
'exploration' activities carried out by coal industries are mainly aimed at
determining the grade and dimensions of discovered deposits - what would be
called appraisal and development activities in other extractive industries
(Zimmerman, 1983). Coal supply modellers have therefore treated exploration
as inconsequential: potential coal supply has been derived from demonstrated
(or proven plus probable) deposits, and mining company decisions have been
assumed to proceed as if knowledge about the extent and characteristics of all
the coal within this portion of the endowment was perfect and uniform.
How reasonable are these assumptions? To answer this question requires an
understanding of how geological assurance is defined. The practice in most
countries is to define geological assurance categories for coal according to one
simple measure: distance from an observation, generally either a core-hole
sample or a surface out-cropping (Todd, 1982). For example, the US Geological Survey defines measured coal to include all coal within a radius of 0.4 km
of a data point; indicated coal extends in a band between 0.4 and 1.2 km; and
inferred coal extends outwards to a radius of 4.8 km (Wood, et al., 1983).
Beyond 4.8 km from an observation, the presence of coal can only be
hypothesized. The category 'undiscovered' includes both this hypothetical coal
and coal that is only speculated to be present. The dividing line between
inferred and undiscovered coal is thus somewhat qualitative.
As Zimmerman (1983) points out, the measured and indicated coal endowment categories may be treated as equivalent, without introducing a substantial
bias: the difference in knowledge is represented by five extra core-holes per
square mile. The extra cost of developing indicated coal is thus simply the cost
of drilling enough additional core-holes to attain the same data point density.
In 1983 he calculated that, on a levelized-cost basis, the added costs would be
negligible - around $0.10/tonne for a bed of 1 m thickness, or only 0.5-2% of
the coal's final sales value.
The number of extra core-holes that have to be drilled in order to impound
coal from the inferred into the measured assurance category is many more than
the number required to go from indicated to measured. Also, the risk that the
actual tonnage is less than expected is higher. Zimmerman (1983, p. 304)
suggests that this geological risk can be diversified away by drilling enough
separate parcels: 'unless the estimation process were biased, the "expected"
amount of coal would be found'. But the estimation process is biased. One
dimension, seam thickness, may turn out to be smaller or larger than expected.
But the aerial extent of coal within the delineated area can only be as expected
or smaller, since the calculation procedure normally assumes the presence of
coal over the whole area defined by the assurance category boundaries. Hence
all else equal, a mine developer will discount the amount of coal in an inferred
area compared with the amount that would be estimated if it were classified as
indicated or measured.
Summary
99
That coal classified as inferred is more risky and probably less economically
attractive than measured or indicated coal does not necessarily mean that it
should be ignored. The decision whether to include it in the analysis should be
made on a case-by-case basis. For most of the well-explored coalfields of the
world, the portions of the coal endowment classified as inferred tend to lie in
beds that are thinner, deeper or more remote than those in the demonstrated
(measured plus inferred) categories. This sort of skewing in the data one would
expect: once enough information has been gathered to suggest that a particular
coalbed would be of substantially higher cost to develop than others in the
area, there remains no rational economic incentive to explore it further. In such
mature coalfields, therefore, one would probably not understate 20- or 30-year
potential coal supply by ignoring the inferred portions of the endowment.
But the same cannot be said of newly developing coalfields. Although one
would expect some upward bias in costs towards the less well-defined deposits,
the bias in many cases may not be large. To have assessed the 30-year potential
coal supply from the Peace River coalfield in British Columbia, or the EI
Cerrejon coalfield in Colombia, on the basis of the tonnages reported as
demonstrated in 1970 - 15 years before these fields were developed - would
likely have been to understate considerably the long-run potential supply from
these areas. If examination of the data suggests that there is likely to be a
substantial amount of good-quality coal that can be mined profitably in the
inferred portion of the reserve base, then that coal should be included in any
analysis of long-run potential supply. Clearly, however, some adjustment must
be made to the modelling procedure to reflect the higher appraisal costs that
would be incurred by mining companies in the process of developing these
deposits. One unsophisticated way to handle the problem might be to add an
exploration charge to the development cost estimates, based on an estimation
of the extra core samples that would have to be obtained in order to bring the
degree of geological assurance up to that of coal in the indicated or measured
reserve base.
SUMMARY
This chapter has provided an overview of the main approaches to modelling
the long-run supply of coal. The components of these modelling systems that
deal, respectively, with the estimation of mining costs, the description of the
coal endowment and the simulation of appraisal and development decisions in
the identified deposits, have been examined within the potential supply analysis
framework of Harris and Skinner (1982).
Strictly speaking, the systems for modelling long-run coal supply that have
been developed to date do not model potential supply, because they lack any
mechanisms for simulating exploration or appraisal, either at the intensive or
at the extensive margin. Rather, they treat one portion of the coal endowment
100
as well-enough appraised (usually the demonstrated reserve base), and ignore

the rest. Mining is then assumed to proceed sequentially through the reserve
base strictly in order of increasing mining costs. While ignoring the inferred
portion of the endowment may be reasonable when analysing medium-term
supply from a highly-developed coalfield, it may be inappropriate for lessdeveloped fields and longer time horizons. Further research on the simulation
of exploration and appraisal decisions in coal mining might shed more light on
this problem.
The main challenge tackled by model developers so far has been to improve
the ability of models to predict costs of extracting coal given information on
variables known to affect mining. Undoubtably, the ability of the current
generation of models to estimate mining costs could still be improved. The
modelling of production and beneficiation losses, for instance, could be made
more sensitive to economic conditions and methods of mining. More generally,
dynamic feedback effects of demand and prices on the coal industry's willingness and ability to search for new deposits or to invest in better exploration or
extraction technologies - i.e., not only on its willingness to invest in new
capacity - need to be integrated into the modelling process. The principal
features of a model that incorporates such feedback loops have been described
in general terms by Lee (1984), but there remain many obstacles to implementing the model for a specific mineral, such as coal.
Because of data limitations, the usual procedure for describing the coal
endowment has been to make inferences about the distribution of key geological and quality characteristics from pre-aggregated inventory data that have
been developed for the purposes of calculating the size of a region's 'reserves',
not for the analysis of potential supply. The efforts of the US Geological
Survey (section 6.5 above) demonstrate what improvements can be made in the
collection and reporting of coal endowment data. Greater standardization
among countries of geotechnical criteria and terminology would also benefit
potential coal supply analysis. But so far, few comprehensive, detailed and
internationally comparable coal endowment inventories have yet been prepared of the world's most important coalfields. For this reason, there is an
urgent need for more research into the use of less data-intensive, statisticalinference methodologies for describing coal endowment.
ACKNOWLEDGEMENT
I would like to thank Thomas Sterner for his patience and his helpful
suggestions. Any errors or omissions, however, are my fault alone.
REFERENCES
Adelman, M. A., De Silva, H. and Koehn, M. F. (1990) 'User Cost in Oil Production'.
MIT CEPR Working Paper 90-020, Cambridge, Mass.
References
101
Averitt, P. (1975) Coal Resources of the United States, January 1, 1967. US Geological
Survey Bulletin 1275. US GPO, Washington, DC.
Barnett, D. W. (1980) How Large are Our Coal Reserves? A Look at Australia and the
USA. Materials and Society, 4, 225-38.
Barrett, S. A. (1982) 'Modeling Coal Supply: An Econometric Analysis of Competing
Mining Methods'. Unpublished manuscript prepared by Data Resources, Inc. for the
Electric Power Research Institute.
Brooks, D. B. (1976) Mineral Supply as a Stock, in Economics of the Mineral Industries,
3rd edn. (eds W. A. Vogely and H. E. Risser), American Institute of Mining,
Metallurgical, and Petroleum Engineers, Inc., New York, pp. 127-207.
Carter, M. D., Medlin, A. L. and Krohn, K. K. (1981) The National Coal Resources
Data System: a Status Report. Geological Society of America Bulletin, 92, 563-73.
CRA (Charles River Associates, Inc.) (1982) CRAjEPRI Coal Market Analysis System.
EPRI Report EA-907 (in 4 vols). Electric Power Research Institute, Palo Alto, Ca.
CRA (Charles River Associates, Inc.) (1986) Reserves and Potential Supply of Low-Sulfur
Appalachian Coal. EPRI Report EA-471O. Electric Power Research Institute, Palo
Alto, Ca.
Church, A.M. (1981) Taxation of Nonrenewable Resources. D. C. Heath, Lexington,
Mass.
Dorian, 1. P. and Zwartendyk, J. (1984) Resource Assessment Methodologies and
Applications. Materials and Society, 8, 659-79.
Ellis, P. A. (1979) 'An Economic Analysis of the South African Coal Industry with a
Focus on Exports'. Unpublished Master's thesis, Sloan School of Management, MIT,
Cambridge, Mass.
EMF (Energy Modeling Forum) (1978) Coal in Transition: 1980-2000, vols 1 and 2,
Stanford University Energy Modeling Forum, Stanford, Ca.
Fettweis, G. B. (1983) 'Considerations on Coal Deposits as Basis of Coal Production.'
Paper presented to Symposium 5 (Economic Geology: Coal Resources and Coal
Exploration) of the 10th International Congress of Carboniferous Stratigraphy and
Geology, Madrid, Spain, 12-17 September 1983, pp. 93-110.
Gluskoter, H. (1991) US Geological Survey, Reston, Virginia. Personal communication
with author, 22 January 1991.
Goldman, N. L., and Gruhl, 1. (1980) Assessing the ICF Coal and Electric Utilities
Model, in Validation and Assessment Issues of Energy Models (ed. S. I. Gass),
National Bureau of Standards Special Publication 569, US GPO, Washington, DC.
Gordon, R. L. (1979) Economic Analysis of Coal Supply: An Assessment of Existing
Studies. EPRI Report EA-496 (3 vols). Electric Power Research Institute, Palo Alto,
Ca.
Gregory, K. (1979) The Approach of lEA Coal Research to World Coal Resources and
Reserves, in Future Coal Supply for the World Energy Balance (ed. M. Grenon)
Pergamon Press for IIASA (Proceedings Series, vol. 6), Oxford, pp. 114-22.
Harris, D. V. P. (1984) Mineral Resources Appraisal: Mineral Endowment, Resources, and
Potential Supply: Concepts, Methods, and Cases. Clarendon Press, Oxford Geological
Science Series, Oxford.
Harris, D. V. P. and Skinner, B. 1. (1982) The Assessment of Long-term Supplies of
Minerals, in Explorations in Natural Resource Economics (eds V. K. Smith and 1. V.
Krutilla) Johns Hopkins University Press for Resources for the Future, Inc., Baltimore, Md, pp. 247-326.
ICF Incorporated (1977) Coal and Electric Utilities Model Documentation, 2nd edn.
ICF, Inc., Washington, DC.
ICF Incorporated (1980a) Coal Supply Curves for Australia, Canada, and South Africa.
Prepared for the US Department of Energy, Energy Information Administration
(Contract EI-78-C-01-63334). National Technical Information Service, Springfield,
Va.
102
ICF Incorporated (1980b) An Examination of the Lognormality of Coal Seam Thickness.

ICF Inc., Washington, DC.
lEA Coal Research, World Coal Resources and Reserves Data Bank Service (1983)
Concise Guide to World Coalfields. lEA Coal Research, London.
Jamieson, E. D. (1990) Current and Prospective Australian Coal Supply. Report No.
ICEACR/20. lEA Coal Research, London.
Kaufman, G., Adelman, M. A., Houghton, 1. C. et al. Oil and Gas, in Energy Resources
in an Uncertain Future (eds M. A. Adelman et al.) Ballinger, Cambridge, Mass., pp.
83~294.
c., Martino, F. and Chopra, I. K. (1980) Application of Geostatistics in a Coal

Deposit. Preprint No. 80-90. Society of Mining Engineers of AIME, Littleton, Ca.
Klein, D. E. and Meany, D. P. (1984) Effects of Resource Depletion on Future Coal
Prices. EPRI Report No. EA-3733. Prepared by ICF Incorporated Washington, DC
Kim, Y.
for Electric Power Research Institute, Palo Alto, Ca.

Labys, W. C. and Shahrokh, F. (1981) 'A Dynamic Forecasting Model of the United
States Coal Market.' Working Paper No. 12. Mineral and Energy Resources, West
Virginia University, Morgantown, wv.
Layfield, Sir Frank (1987) Sizewell 'B' Public Inquiry. HMSO, London.
Lee, R. (1984) A Model for Minerals Activity and Decision Simulation. Materials and
Society, 8,
551~66.
Long, R. (1986) The Availability of Cost of Coal in South Africa. Report No. ICEAS/C6.
lEA Coal Research, London.
McKelvey, W. (1972) Mineral Resources Estimates and Public Policy. American
Scientist, 60,
32~40.
Newcomb, R. T. (1981) The Geostatistical Estimation of Unexplored Resources for

National Coal Modeling, in The Economics of Exploration for Energy Resources (ed.
1. B. Ramsey), JAI Press, London, pp. 271~311.
Newcomb, R. T. and Fan, 1. (1980) Coal Market Analysis Issues. EPRI Report No.
EA-1575. Electric Power Research Institute, Palo Alto, Ca.
NUS Corporation (1981) Coal Mining Cost Model. EPRI Report EA-1273 (in 4 vols).
Electric Power Research Institute, Palo Alto, Ca.
NUS Corporation (1984) Coal Mining Cost Model (MOD-3). EPRI Computer Code
Manual, EQ-3526-CCM. Electric Power Research Institute, Palo Alto, Ca.
Pauncz, I. and Nixon, T. R. (1980) Application of Geostatistics for a More Precise
Statement of Coal Reserves. Published Report 80-3. Australian Coal Industry
Research Laboratory Ltd, North Ryde, NSW, Australia.
Quebral, M. C. (1990) 'Estimating the Extraction Technology for an Exhaustible
Resource.' Unpublished manuscript, Institute for Environmental Studies, University
of Illinois, Urbana, Ill.
Sani, E. (1979) The Kriging Method versus Classical Statistics in Estimating Mineral
Reserves. Materials and Society, 3, 81~98.
Schmidt, R. A. (1979) Coal in America: An Encyclopedia of Reserves, Production and Use.
McGraw-Hill, New York.
SAIC (Science Applications International Corp.) (1986) 'Development of Revised
Reserve Inputs to the Resource Allocation and Mine Costing Model.' Report
prepared for US Energy Information Administration, Washington, DC.
SAIC (Science Applications International Corp.) (1988a) 'Selected Modifications to the
Resource Allocation and Mine Costing (RAMC) Model'. Report prepared for US
Energy Information Administration, Washington, DC.
SAIC (Science Applications International Corp.) (1988b) 'Development of a Revised
Methodology to Estimate the Distribution of the Demonstrated Reserve Base (DRB)
by Quality.' Report prepared for US Energy Information Administration, Washington, DC.
References
103
Steenblik, R. P. (1985) 'Issues in Modelling International Coal Supply.' EURICES

Paper No. 85-2. Centre for International Energy Studies, Erasmus Universiteit,
Rotterdam.
Steenblik, R. (1986) 'International Coal Resource Assessment'. Working Paper No. 73.
lEA Coal Research, London.
Tewalt, S. 1., Bauer, M. A., Mathew, D., et al. (1983) Estimation of Uncertainty in Coal
Resources. EPRI Report No. EA-3133. Electric Power Research Institute, Palo Alto,
Ca.
Todd, A. 1. H. (1982) Lexicon of Terms Relating to the Assessment and Classification of
Coal Resources. Graham & Trotman for lEA Coal Research London.
US Department Commerce, International Trade Administration and US Department
of the Interior, US Bureau of Mines (1990) A Cost Comparison of Selected U.S. and
South African Coal Mines. Appendix D: Detailed [Mine Costing] Methodology.
Washington, DC. (Available from International Trade Administration by request.)
US EIA (US Department of Energy, Energy Information Administration) (1982)
Documentation of the Reserve-Related Data Inputs to the Resource Allocation and
Mine-Costing Model. Report No. DOEjNBB-0026. Prepared by Synergic Resources
Corp. US GPO, Washington, DC.
Documentation of the Integrating Module and Stock Module of the Intermediate Future
Forecasting System. US GPO, Washington, DC.
Estimation of u.s. Coal Reserves by Coal Type. Report No. DOE/EIA-0529. US
GPO, Wasington, DC.
Vogely, W. A. (1979) Analysis of Bechtel's Regional Energy System for Planning and
Optimization of National Scenarios (RESPONS), and ICF's National Coal Trade
Model (NCM). EPRI Report No. EA-1133. Electric Power Research Institute, Palo
Alto, Ca.
Wood, D.O., and Mason, M. J. (1982) Analysis of the Energy Information Administration's Coal "Supply Model, in Coal Models and Their Use in Government Planning (eds
J. Quirk, K. Terasawa and D. Whipple), Praeger, New York, pp. 37-57.
Wood, G. H., Kehn, T. M., Carter, M. D. et al. (1983) Coal Resource Classification
System of the U.S. Geological Survey, US Geological Survey Circular 891. USGS,
Alexandria, Va.
Zimmerman, M. B. (1977) Modeling Depletion in a Mineral Industry: the Case of Coal.
Bell J. Economics, 8, pp. 41-65.
Zimmerman, M. B. (1981) The U.S. Coal Industry: The Economics of Policy Choice.
MIT Press, Cambridge, Mass.
Zimmerman, M. B. (1983) Coal, part IV in Energy Resources in an Uncertain Future
(eds M. A. Adelman et al.), Ballinger, Cambridge, Mass., pp. 295-332.
7
Global availability of natural gas:
resources, requirements
and location
Daniel A. Dreyfus
7.1 INTRODUCTION
In the past few years, the enormous significance of the largely untapped global
natural gas resource has received increasing attention. Initially, gas was viewed
as one component of a strategy to moderate the world's increasing dependence
upon liquid petroleum. More recently, the low air pollutant emissions and the
relatively low contribution to the so-called 'greenhouse' gases which result
from the combustion of natural gas compared to those of other fossil fuels,
have given the gas option greater emphasis.
Today, discussions of acid rain controls and of more comprehensive
measures to moderate man-made contributions to global warming usually
include some consideration of natural gas. Greater reliance upon gas is viewed
as an interim measure to bridge the gap between business as usual and a future
energy balance which minimizes fossil fuel dependence.
The natural gas resource base is quite large in relation to the current levels
of demand upon it. Proved reserves of natural gas, those amounts which are
reasonably well known based upon drilling information, are estimated to be
the energy equivalent of about 740 billion barrels of crude oil. This approaches
the size -of global proved reserves of oil (896 billion bbls), but the annual
production of natural gas is only about half that of petroleum. The global
reserve to production ratio for natural gas, a measure which is often used as
an indication of near-term supply capability, is, therefore, about 60 to 1. In
comparison the ratio for petroleum is 40 to 1.
The estimated total remaining worldwide gas resource base which could be
economically recoverable with current technology is much larger than the
proved reserves. Estimates of total remaining recoverable reserves, which
include undiscovered portions of the resource base, are on the order of 1.5
106
Global availability of natural gas
trillion barrels of oil equivalent. Clearly, the natural gas resource is capable of
supporting much greater levels of production. It will, undoubtedly, continue to
increase in its significance as an energy source in the coming decades.
The availability of natural gas to serve the requirements of global society,
however, involves more than the simple geological existence of the resource.
The geographical distribution relative to probable demand centres is also a
factor, particularly because transportation and storage are especially costly
where gaseous fuel is concerned.
7.2 THE NATURE OF RESOURCE ESTIMATES

Estimation of natural gas resources is done for a variety of purposes and at
widely varying levels of precision and detail. At one end of the spectrum lie the
resource estimates associated with reservoir engineering. They are based upon
considerable knowledge of the subsurface situation through detailed, on-site,
geological exploration, seismic investigation of the subsurface formations,
confirmatory information from the drilling of wells, and sometimes actual
production experience with the particular reservoir. These estimates usually are
made to support the technical and economic operation of the reservoir. The
extreme alternative is the appraisal of the remaining discovered and undiscovered recoverable resources which might exist in areas of national, regional,
or even global scope. Studies of the latter type are done for more abstract
economic motives, to support public policy decisions, or for general scientific
and academic interest. While they also rely upon a history of the exploration
and production experience with developed gas resources and upon specific
geological, seismic and drilling data where they are available, they also involve
a more generalized inference of the probable occurrence of producible gas
based upon broad geological conditions.
Authoritative estimates of the natural gas remaining to be produced are
generally characterized either as proved reserves or recoverable resources. It is
important to differentiate between the two classes of estimates.
Proved reserves are the recoverable quantities of gas which are estimated to
remain in known oil and gas reservoirs. Usually, these estimates are based
upon actual production experience or drilling data. Reserve estimates, therefore, are reasonably reliable and represent an inventory of discovered gas.
Proved reserves, however, do not necessarily reflect the overall size or geographical distribution of the global gas resource base. Exploration for gas has
been heavily concentrated in a few areas. More than 90% of all hydrocarbon
drilling experience has occurred in the United States and Canada alone. Low
estimates of proved reserves in some areas may indicate limited exploration
rather than limited geological potential.
107
Estimates of conventional gas resources

7.3 CONTEMPORARY ESTIMATES OF CONVENTIONAL GAS
RESOURCES
Table 7.1 summarizes some of the principal current estimates of proved gas
reserves and estimated resources. As might be expected, agreement among
estimates of proved reserves is quite good. From the point of view of the
longer-term global energy strategy, however, estimates of the remaining recoverable resources, whether discovered or not, are more pertinent. Here a
wider degree of uncertainty concerning the undiscovered portion of the
resource base complicates the picture.
Estimates of undiscovered resources encompass some areas which are, as yet,
undrilled extensions of formations that are known to be productive and some
areas which are postulated to contain economically producible resources based
only upon judgements made by analogy to geological conditions that have
been productive elsewhere.
There are two major methodological techniques commonly used for the
estimation or appraisal of undiscovered gas resources. They can be characterized as historical approaches and geological approaches (Office of Technology
Assessment, 1983). Both approaches ultimately rely upon an extrapolation of
the knowledge of the resource base which has been garnered from past
exploration and development experience. They differ, however, with regard to
the historical data they emphasize and in the way in which the data are used.
Historical approaches extrapolate past trends in gas discoveries and production relative to the effort exerted by the producers. They rest upon an
Table 7.1
Estimates of worldwide natural gas resources (trillion cubic feet)

Proven reserves
OGJ
WO
Cedi
Remaining recoverable
resources
Masters
IGT
Western hemisphere
Western Europe
Middle East
Africa
Asia Pacific
CPE
518
200
1182
253
240
1561
516
227
1167
206
245
1502
535
192
1065
257
327
1532
1498
423
2126
570
630
2807
1902
377-395
1903
453-653
671-833
1966--2807
World total
3955
3862
3909
8107 b
7352-7830b
% OPEC
%CPE
40%
39%
37%
39%
37%
39%
"Includes proven reserves.

bSeparately estimated.
References: OGJ, 'Worldwide Report', 1988; WO, '43rd Annual International Outlook', 1988;
Cedi., CEDIGAZ (1988); Masters, Masters et aI., 1987; IGT, Institute of Gas Technology, 1986.
108
assumption that the nature of the resource base is the predominant factor
controlling discovery and production success. The historical trends in discovery, therefore, are strong evidence of the resource characteristics. Geological approaches rely on data and assumptions about the physical size and
richness of the resource itself. In general, historical approaches tend to be
constrained by the quality and availability of relevant operational data and are
weak in dealing with frontier resources or changes in technology, economics,
or industry practices. Geological approaches are heavily dependent upon the
expert judgement of the estimators and are difficult to validate.
The historical approach is characterized by the estimates of the US oil
resource base done in the 1950s by Hubbert (1967) of the United States
Geological Survey. Hubbert used historical trends in the drilling effort required
to discover new resources to produce decline curves that predicted the overall
extent of the resource yet to be discovered.
The approach rests upon the assumptions that (i) there is a continuous
relationship between the effort put into discovery and the amount of hydrocarbons found; (ii) the discovery of an incremental unit of resource requires
increasing effort over time; and (iii) the ultimate resource is finite. These
assumptions can be represented by a continuous differentiable function relating
the quantity (Q) to effort (E). The first derivative is positive dQ/dE) >0) and a
function of effort. The second derivative is negative.
These equations are used to develop 'find rate models', using mathematical
forms that conform to the basic assumptions, such as the logistical function or
some quadratic equations. Find rate equations or models are commonly used
to generate resource discovery success predictions. The PROLOG model used
by the US Energy Information Administration to project onshore US oil and
gas resource discoveries is one such model.
Such models are obviously critically dependent upon their fundamental
assumption of declining success per unit of effort over time. They do not
represent the potential impact of changes in the rate of technological advances,
potential additions to the resource base consisting of large new increments such
as frontier discoveries and unconventional resources, or changes in the economic parameters of development.
Another mathematical approach to modelling the historical resource discovery experience is the Arps-Roberts model (Arps et at., 1971). The ArpsRoberts approach, known as the discovery process approach, is more directly
associated with the physical nature of the resource base. It assumes that the
probability of discovering a field with an exploratory well is directly proportional to the size of the field and inversely proportional to the extent of the
entire resource. Based upon exploration to date, it is possible to fit a
mathematical form to the historical data and compute the probability of
finding another field of a given size. The equations are of the form:
V =F/(1-(1-A/B)W)
Estimates of conventional gas resources
109
where
F = number of fields found after 'W' wells have been drilled,
U = the ultimate number of fields to found in the region,
A = the geometric extent of a representative field, and
= total region being explored.
The equation may be modified by a 'fitting parameter' which modifies 'W'

based upon exploratory experience with a particular region. The estimation of
this fitting parameter represents the critical problem in developing ArpsRoberts equations. As the region under consideration is disaggregated, the
relevance of the available data to the more intensively explored areas of the
resource becomes more powerful, but the sparsity of data in underexplored
areas reduces the approach to one of analogy from similar areas elsewhere. An
advanced application of this modelling approach is exemplified by the highly
dis aggregated GRI Hydrocarbon Model of the US oil and gas resource base
(Energy and Environmental Analyses Inc., 1990).
In contrast to the historical approaches which rely heavily upon the
statistical relationships between past discovery efforts and success, the geological approach focuses upon a physical description of the undiscovered resource
base derived from scientific geological theory and whatever measured evidence
is available.
Parameter estimation methods strive to estimate the extent of potentially
productive formations, such as sedimentary basins, and the concentration of
oil or gas in place based upon surface and available drilling evidence (National
Petroleum Council, 1980). Structure count uses seismic data to evaluate
potentially productive formations (Energy Information Administration, 1973).
Analogy is used to impute potential resource availability of unexplored areas
reasoning from the observed geologic similarity to known producing areas.
The geological approach is commonly used by organizations; such as the US
Geological Service, that have extensive in-house geological expertise and
groups of experts assembled for the purpose of resource estimation. The
biennial estimates of the US natural gas resource base made by the Potential
Gas Committee are based upon the geological approach using teams of expert
geologists who have specific knowledge of the regions being evaluated. While
the geological approach has the flexibility to encompass notions of technological and economic changes, it is highly subjective and dependent upon the
capabilities and the attitudes of the estimators.
Compilations of the aggregate global remaining recoverable resources are
drawn from a broad range of disparate sources. Such sources vary greatly in
regard to the amount and reliability of geological data available, the definitions
used to decide upon the scope of resource types that are included in the
estimate, and, of course, the judgement and motivations of the estimators.
It should be evident that any effort to estimate undiscovered resources will
110
afford imprecise results, but strong agreement remains that the resource base
is double the proved reserves or larger.
In approaching estimates of the global gas resource, it is important to
remember that the US hydrocarbon resource has been more intensely explored
and drilled than any other in the world. Outside of the United States there is
much less historical experience, exploration drilling and production data are
scarce, and in many areas that data which does exist may be difficult to obtain,
of questionable validity, or both. Because aggregate global compilations are
drawn from diverse sources, consistency among the approaches must be
expected to be far less than among US estimates.
Some general factors that will affect the validity of all resource estimates are
the following:
1. Data: Each of the estimates relies on the data available to the estimators;
there is not a single, authoritative and comprehensive data base used by all
estimators even in the well-explored and well-documented areas of the
mature US domestic gas producing regions. Significant discrepancies in the
publicly available data must be resolved by each estimator, and some have
access to exclusive, proprietary data unavailable to others.
2. Definitions and scope: Each estimator has chosen to include certain components of the resource base. Variations particularly are found in the
treatment of the less conventional resource components (coal bed methane,
Devonian shale, tight formations, etc.). The economic and technology
considerations affecting unconventional resources are more speculative, and
some estimates simply exclude some resource categories even where there is
actual production experience. There are variations as well concerning the
limit on the depth of resource base considered, the assumed limitation of
water depth on the deep offshore potential, attitudes about the ultimate
accessibility of resources in remote and hostile regions, etc.
3. Judgemental attitudes: Estimators differ in their judgements even about the
potential for relatively well-known conventional resources. These
judgemental attitudes, in addition to differences in professional geological
interpretations, are coloured by attitudes about future economic parameters
(sales prices of gas and costs of factors of production) and relative optimism
about the future pace of innovation in exploration and production technologies and their application in the field. While these differences are
particularly critical in the more judgemental geological approaches, they
can also affect the ways in which mathematical forms are chosen to fit
historical data series.
Conventional resource estimates are customarily constrained to include only
methane contained in deposits that might be amenable to discovery and
production using current conventional practice and reasonably foreseeable
technical and economic extensions of that practice. This limitation excludes
significant portions of the longer-range resource potential from consideration.
Such gas resource estimates are reasonably comparable to similar conven-
Distribution of supply relative to demand
111
tional estimates of alternative energy forms, such as the quantities of coal and
petroleum that could be provided using conventional practice. They are
appropriate estimates, therefore, for discussions of energy supply issues in the
near term, that is, for two or three decades.
For discussions of the energy situation beyond the middle of the next
century, however, especially when the potential for gas is being compared with
significant advances in the technologies of alternative sources, such as commercial electric power from fusion energy reactions, power storage in superconductive devices, and advanced solar concepts, the conventional gas resource
estimates are somewhat limiting. For example, substantial amounts of methane
are known to be present locked in hydrate form in permanently frozen arctic
areas and in the deep ocean floor. Large sedimentary basins are known to exist
in such circumstances and estimates of the methane present are enormous,
ranging as high as 3000 trillion cubic feet. Technologies required to exploit this
resource are currently beyond reach, but cannot be entirely dismissed as
long-term possibilities.
Resource estimates usually also omit from consideration those quantities of
gas that would require the investment of energy for production in amounts
approaching the energy value of the gas produced. To some extent, this
constraint also might be subject to revision should major breakthroughs in
production technology occur.
Even within the more conventional resource situations, definitions of recoverable resources always include some notion of technological feasibility and
economic attractiveness. Although the precise criterion of technical recoverability is seldom explicitly stated by the estimator, statistical approaches as well
as geological judgements generally capture the current, or conventional, technical practice of the industry along with such evolutionary improvements in
technology as a knowledgeable practitioner can foresee. These limitations are
revised from time to time to reflect advances in practice and theory that are
evident in the trends, but the potential for scientific breakthroughs, or even
major innovations in engineering not yet apparent in practice, are usually
omitted.
Criteria of economic feasibility are even less explicit. It seems clear, however,
that most estimators, in determining the recoverable portions of the gas in
place, have in mind the cost and price regimen that exists at the time the
estimate is made. Obviously, these assumptions might not be appropriate to a
future situation and might unduly constrain the estimate of recoverable
resources.
7.4 GEOGRAPHICAL DISTRIBUTION OF SUPPLY
RELATIVE TO DEMAND
Natural gas resources, until very recently, were discovered as a coincidence of
the worldwide search for liquid petroleum and were developed only when they
112
occurred in near proximity to energy demand centres. The aggressive exploration for additional gas reserves then tended to be concentrated in areas that
were convenient to an established market and transportation system.
Table 7.2 compares the current regional gas consumption patterns with the
outlook for the year 2010. The 2010 scenario is adopted from the Global
Outlook for End-Use Energy Requirements (GOSSER II). Projections of
energy supply, demand and price are most commonly based upon historical
experience with the reactions of supply and demand to price changes. The
future requirements for energy are projected in terms of historical uses and
price elasticities. This approach captures the observed behavioural responses
that have influenced past consumption patterns, but it may not recognize the
constraints imposed upon the use of particular energy sources by infrastructure
which have not been tested historically. It is also unlikely to anticipate evolving
technologies or structural shifts in the nature of energy uses.
Supply reactions, similarly, are predicted using assumed revenue optimization strategies and supply curves based upon the calculated costs of production and descriptions of the resource base previously discussed. As in the
demand analysis, this approach does not explicitly address the technical
limitations upon the choice among energy sources except to the extent they
have constrained historical activity.
An alternative approach to evaluating future energy demand, which conTable 7.2 Outlook for global consumption of natural gas (trillion cubic feet)
Region
Consumption
1987" 2010 b
Proved
reserves c
United States
Total, North America
Total, Western Europe
Total, Australasia
Japan
Total, Pacific Rime
Total, Latin America
Total, Middle East
Total, Africa
Total, China
Total, Soviet Bloc
(17.1)
19.4
8.5
0.7
(1.5)
2.9
3.0
2.1
1.3
0.5
25.3
(19.4)
22.9
14.6
1.6
(3.8)
6.9
6.7
9.7
2.7
4.8
51.4
(186.7)
284.7
218.8
23.8
(1.0)
201.0
226.6
1084.0
248.6
30.7
1479.3
Total, world
63.7
121.3
3797.5
Interregional trade
Exports ( + )/
imports (-)
1987 d
2010
( -0.9)
0.0
-2.4
( -1.4)
-0.1
+0.0
+0.1
+0.9
0.0
+1.5
(-2.6)
-1.0
-6.3
+0.7
(-3.8)
-2.1
+0.6
+2.8
+ 1.8
0.0
+3.5
2.2
9.4
aBP Statistical Review of World Energy, June 1988.

bDrawn from Ashby and Dreyfus (1988).
'End of 1987.
dAdopted from CEDIGAZ (1987).
eIncludes Japan, South and Southeast Asia and Pacific nations, but excludes Australia and China.
Distribution of supply relative to demand
113
siders the requirements for energy services by end-use sectors (transportation,

industry, electric utilities, residential and commercial), was used to produce the
GOSSER projection. End-use energy service requirements are the direct
consequences of the economic and demographic trends in societies. They also
reflect cultural variations in energy use patterns. The future requirement for
energy in each sector can be projected based upon historical use patterns and
upon estimates of future trends in GDP, population, industrial growth or
restructuring, energy intensity and urbanization. Energy requirements are then
allocated among the alternative energy sources that are capable of providing
the particular service. The choice of an energy source for each service is a
function of the combined fuel and capital costs of providing the service,
considering the available and probable future technological options, and the
availability of indigenous resources. The latter approach can more readily
incorporate trends in the technical and physical constraints upon the choices
among energy sources which are often omitted from specific consideration
when historical price elasticity responses dominate the methodology.
There are several observations to be made about the geographical distribution of natural gas demand. Historical use in 1987 reflects the abundance of
resources in the Soviet Union and North America coupled with long histories
of petroleum development and large energy consumption in both areas. The
intensive energy markets in Western Europe have already attracted imported
gas. Elsewhere in the world current uses of gas are small.
Although significant growth in gas consumption should be anticipated, the
greatest near-term increase is likely to take place where transmission and
distribution infrastructure already exists and where gas is already established
as a part of the historic energy mix. Of the nearly 60 trillion cubic feet of
incremental gas demand anticipated by the projection, about 36 will be
accounted for by North America, Western Europe and, especially, the Soviet
Union and its Eastern European trading partners (see Chapter 12).
The distribution of total estimated recoverable resources (as currently
estimated) is roughly proportional to the reserve figures. Although new
discoveries may alter the estimated resource distributions over time, existing
discovered reserves will be a more potent influence in the use of gas over the
next two decades. Existing discovered reserves represent investments or sunk
'inventory costs' that are already actively in search of markets.
Most international trade in gas currently takes place within the major
continental regions. The total international gas trade in 1987 is estimated to
have been 9.0 trillion cubic feet, of which 7.0 were by pipeline and 2.0 by
liquefied natural gas (LNG) shipment. The largest components of this trade are
among the (formerly) centrally planned economies, within Western Europe and
between Canada and the United States.
Achievement of the projected consumption levels for 2010, however, will
require much larger global trade between those regions that have large reserve
to production ratios and large remaining recoverable resources relative to their
114
indigenous energy requirements and those regions that face growing needs for
natural gas relative to their resource bases.
The GOSSER projection already has been constrained in its forecast of fuel
choices to reflect the regional availability of natural gas and the large capital
costs associated with long-distance transportation facilities. The projected
regional requirements for the year 2010, therefore, are an indication of the
relative attractiveness of gas in competition with alternative energy sources
despite the costs of interregional transportation which will require additional
pipeline and LNG transportation facilities.
The projection indicates that the interregional trade in natural gas must
increase threefold by the year 2010 to accommodate the patterns of use that
are emerging. From only about 2.4 trillion cubic feet in 1987, interregional
transfers would increase to 9.4 trillion cubic feet in 2010. As much as 6.0 of this
amount would be intercontinental LNG shipments. The general patterns of
international transfers are shown in Figure 7.1. The arrows in the diagram are
directional only and do not imply any scale of magnitude.
7.5 CONCLUSIONS
The importance of interregional transportation as a limiting factor on global
gas consumption cannot be ignored.
For the foreseeable future, geological resources of natural gas that can be
produced at prices competitive with alternative energy forms are more than
ample to meet the probable market requirements. In the longer term, or in a
world that has decided for policy reasons to curtail another major energy
supply source, overall energy prices may be expected to be higher. Higher
prices will certainly expand the economic natural gas resource base beyond the
levels in contemporary estimates. Therefore, in a future in which more gas is
called for, more probably can be provided.
The greater uncertainty is the capability of global markets to develop
rational approaches to the international gas trade and to underwrite the large,
long-term investments in transportation facilities that will be necessary to
moderate the geographical imbalances between supply and demand. Especially
in the developing countries, which will be the driving factor of future energy
demand growth, capital will be a scarce resource. If even indigenous gas
resources are to play a role in the energy supply mix, investment in costly
transportation systems must somehow be accommodated.
Intercontinental trade in gas, by long-distance pipeline and by LNG, will
depend upon innovative trade arrangements and investment schemes. The
focus of discussions about natural gas supply availability to meet global
requirements should include these infrastructure factors along with resource
estimates.
Fig. 7.1
Interregional gas trade, year 2010.
116

BIBLIOGRAPHY
'43rd Annual International Outlook' (1988) World Oil, 2, 31-126.

'Worldwide Report' (1988) Oil and Gas Journal, 52, 43-119.
Arps,1. 1., Mortada, M. and Smith, A. E. (1971) Relationship between Proved Reserves
and Exploratory Efforts. Journal of Petroleum Technology, June, pp. 671-5.
Arps,1. 1. Mortada, M. and Smith, A. E. (1971) Relationship between Proved Reserves
and Exploratory Efforts. Journal of Petroleum Technology, June, pp. 671-5.
Ashby, A. B. and Dreyfus, D. A. (1988) Global Outlook for Service Sector Energy
Requirements (GOSSER II): Gas Research Institute, Washington, DC.
CEDIGAZ (1988) Natural Gas in the World in 1987. Rueil Malmaison, Cedex, France.
Energy and Environmental Analysis, Inc. (1990) Guide to the Hydrocarbon Supply Model
1990 Update. EEA, Arlington, V.
Energy Information Administration (1983) OCS Oil and Gas Supply Model. EIA,
Washington, DC.
Geological Survey of Canada (1984) Oil and Natural Gas Resources of Canada, 1983.
Energy, Mines and Resources Canada, Canadian Government Publication Center
Supply and Services, Ottawa, Canada.
Institute of Gas Technology (1986) IGT World Reserves Survey as of December 31, 1984.
Chicago, Ill.
King, Hubbert M. (1967) Degree of Advancement of Petroleum Exploration in the
United States. American Association of Petroleum Geologists Bulletin, 51, 2207-27.
Masters, C. D., Attanasi, E., Dietzman, W. et al. (1987) World Resources of Crude Oil,
Natural Gas, Natural Bitumen and Shale Oil. Twelfth World Petroleum Conference,
Houston, Texas.
National Energy Board (1988) Canadian Energy Supply and Demand 1987-2005.
Minister of Supply and Services Canada, Ottawa.
National Petroleum Council (1980) Unconventional Gas Source, vol. 5: Tight Gas
Reservoirs, Parts I and II. NPC, Washington, DC.
Office of Technology Assessment. (1983) US Natural Gas Availability. Congress of the
United States, Washington, DC. pp. 31-66.
Organization of the Petroleum Exporting Countries. (1987) OPEC Annual Statistical
Bulletin. OPEC, Vienna.
Potential Gas Committee. (1987) Potential Supply of Natural Gas in the United States.
Potential Gas Agency, Mineral Resources Institute, Colorado School of Mines
Foundation, Golden, Col.
University of Texas at Austin, Bureau of Economic Geology; ICF-Lewin Energy
Division, ICF Inc.; and Argonne National Laboratory (1988) An Assessment of the
Natural Gas Resource Base of the United States. Washington, DC.
US Department of Interior, US Geological Survey; and Minerals Management Service
(1988) 'National Assessment of Undiscovered Conventional Oil and Gas Resources.'
USGS-MMS Working Paper, Open-File Report 88-373.
World Energy Conference (1986) 1986 Survey of Energy Resources. Oxford, Holywell
Press.
8
Modelling oil exploration
Victor Rodriguez Padilla
8.1 INTRODUCTION
Most oil supply models refer exclusively to North America. Models covering
other regions or the world situation as a whole are much rarer. In addition,
their forecasts of oil supply tend to be based on estimates of final resources and
cumulative production trends rather than on a sequential representation of
exploration, development and production.
Why should modelling be limited in this way? The answer is twofold. First,
outside the United States and Canada (and to a lesser extent the North Sea),
data concerning oil activities are sorely lacking, especially for exploration.
Secondly, the phenomena involved are much more complex outside North
America. Since the early 1970s the exploration market has split in two: on the
one hand the so-called, 'reliable' oil-producing regions in the industrialized
countries and, on the other, the less developed countries (LDC).
In the first market there is a large number of active operators They have
virtually free access to the territories involved and, by law as well as in practice,
whatever hydrocarbons are discovered belong to whoever discovers them. In
addition, the tax situation is favourable, state intervention is relatively limited,
there are virtually no political risks, the oil market is well-developed and
operators have access to advanced technologies. In contrast, access to territories in the second market is in general highly restricted. The owner of the
resources, the State, regulates the flow of companies seeking access to them.
The number of protagonists is thus limited and the authorities only have to
deal with a small number of international companies. In addition, because of
its lack of technology, skills and knowledge of the oil business, the State, or
national oil company, is in general unable to carry out exploration without the
cooperation of foreign companies. On the other hand, state intervention in oil
activities can be felt at all levels. Finally, there is a twofold competition within
the market: among states possessing oil resources and seeking to attract
companies to invest in the search for oil, and among oil companies wishing to
118
gain access to zones which are promising in geological or contractual

terms.
Actual exploration diverges from the optimal exploration path described by
theoretical studies. Institutional factors, such as the contractual framework,
taxation and political risks, alter oil operators' expectations. Operators
have to take such factors into account with the result that their behaviour is
different from what it would have been in the absence of such 'non-market'
factors.
Any extension to LDCs of the models constructed for developed countries
is thus limited because of this twin difficulty. The mathematical representation
of upstream oil activities in such models is very simple and their results not
very encouraging because of the generally poor quality of the available data
which the models have to use in their attempt to account for real processes
which are by nature extremely varied and complex.
Given the difference between oil-producing regions, several questions arise.
Can the same modelling approach be adopted in both cases? Are the factors
determining oil exploration the same? And, more specifically, is the price of oil
still the most appropriate factor when analysing oil exploration as theory
suggests?
The role of prices in driving upstream investment in the oil industry is
complex. Profit expectations are related to future prices, but companies' cashflow (and hence investment capacity) is affected by current and past prices. It
should not be forgotten that the search for oil ~ a risky, capital-intensive
business ~ is in most cases self-financed.
There can be no doubt that total exploration investment is somehow linked
with the price of oil. However, the price of oil is certainly not the main factor
determining the distribution of exploration investment among producing
regions. There may be a correlation between prices and exploration for some
regions, but in most cases the correlation is neither obvious nor immediate.
The impact of price changes on exploration activities varies from country to
country and from company to company. Institutional factors delay the
reactions of the various protagonists involved, to the point even that they have
been known to deny the existence of any simple relation between prices,
exploration, reserves and production. Such delays may be the dominant factor
in creating tensions within the oil market.
Our analysis will be developed in four parts. We first describe the way in
which exploration has been dealt with in oil supply models. We then look at
five recent models, paying particular attention to the explanatory variables
used when modelling exploration activities. This will be followed by a discussion of the factors which have been shown by several empirical studies to
determine exploration in LDCs. Finally, we analyse the interdependence
between institutional factors, oil prices and exploration effort with a view to
drawing conclusions for modelling in the future.
The key factor determining oil supply
119
8.2 EXPLORATION: THE KEY FACTOR DETERMINING

OIL SUPPLY
Various approaches have been adopted to forecast the volume of undiscovered
oil. Some modellers use exclusively economic hypotheses, others base their
approach on oilfield engineering or a combination of the two. Kaufman (1979,
1983) distinguishes six main approaches. Several, such as those based on
life-cycles, effort rates, geological-volumetric methods, subjective probability or
econometric methods, can also be used to make supply forecasts.
Since each approach has some advantage, models often employ more than
one technique. Numerous comparative analyses have been made of the various
models, for example Adelman et al. (1983) provide a global analysis, as do
Clark et al. (1981), comparing ten models used for oil supply in the United
States.
Oil supply models generally cover three steps: first the global estimation of
undiscovered resources; secondly, the conversion of resources into reserves; and
thirdly, the elaboration of scenarios of how new discoveries are brought into
production.
Some studies eliminate the exploration phase and go directly from final
resources to production. Production and discoveries are taken to be a function
of time only, thus reducing the task to that of adjusting a mathematical
equation to the cumulative production curve or the curve of cumulated
discoveries. This is the case with very general models in which the introduction
of exploration would increase operational difficulty (see, for example, Baldwin
and Prosser's oil market model in Chapter 15). This is also true of countrybased studies in which the exploration process is rendered irrelevant by the fact
that the proven reserves are extremely large (see, for example, the US
Department of Energy's estimates of oil potential in Nigeria (US DOE 1979)
and the Middle East (US DOE, 1981), see also Chapter 6 for the case of coal).
From a conceptual point of view, the most satisfactory models are those that
attempt to reproduce the relations between the various links in the 'chain' of
upstream oil activities: exploration, development and production (MacA voy
and Pindyck, 1973; US DOE, 1978; Adelman and Jacoby, 1979; Woods and
Vidas, 1983; CERI, 1988; and Choucri et aI., 1990).
Roughly speaking, the approach adopted by these models is as follows: an
oil operator decides to prospect in a given region in the light of geological and
economic data; exploration pays off with the discovery of an accumulation of
hydrocarbons; the operator then decides whether or not to go ahead and
develop the discovery; if the operation goes ahead, the discovery produces a
certain volume of oil and gas each year depending on the extraction rate. This
approach presupposes that the region or regions involved are homogeneous in
geological terms with the result that the operator does not have to arbitrate
between intensive and extensive exploration.
120
One hypothesis currently adopted is that for each year the oil discovered is
proportional to the exploration effort. The key factor in this type of approach
is the determination of the exploration effort. One of the principal tasks of oil
supply models attempting to reproduce the logic of upstream activities is thus
to determine which factors explain the level of exploration and to link them
together in a suitable manner.
There are two approaches to the construction of exploration models: one
based on optimization, the other on econometrics. The former, e.g. the
Department of Energy (US DOE, 1978) and Gas Research Institute (Woods
and Vidas, 1983) models, maximize complex objective function in order to
determine the way in which exploration responds to changes in prices, costs
and taxation. 1 These models are not particularly relevant to the present article
since they are not interested in establishing decision-making rules which fit
with historical data. Econometric methods are better suited to explaining the
successive phases in upstream oil activities and changes in exploration in
particular. These models are all more or less related to Hotelling's analysis
despite the numerous weaknesses which can be seen in this theory and its
extensions (see for example, Bohi and Toman's critical analysis (1986) of this
theory and its application to oil supply models).
8.3 FROM THEORY TO EMPIRICAL MODELLING
The rapid increase in the number of empirical models of oil supply during the
1970s was accompanied by a similar development on the theoretical side.
Improvements in Hotelling's model include the exploration process - for
mining basins or entire regions (Cairns, 1986).
Most theoretical models approach the problem of oil supply, including
exploration, either as one of constrained intertemporal maximization (cf.
Pakravan, 1977; Peterson, 1978; Pindyck, 1978), or as a problem of stochastic
optimization under uncertainty (cf. Gilbert, 1979; Pindyck, 1980; Arrow and
Chang, 1982).
Pindyck (1978), for example, treats the fundamental problem of the firm
within a deterministic situation as one of maximizing the profit function given
the constraint imposed by the exhaustion of reserves (R) and that imposed by
the discovery of new reserves. The profit function is defined as the value of
'The United States Department of Energy model first of all calculates the quantity of economically
exploitable resources in each of 12 oil regions in the United States. It then generates regional curves
for 'latent' or 'desirable' demand for exploration drilling at each price level. The relation between
latent exploration demand and price is unique for each region. Total demand is the sum of the
regional latent demands and falls with the reduction in the quality of discovered reservoirs and
cost increases. The model then looks at total desirable drilling in the light of several constraints
such as availability, capacity and the economic life of drilling equipment in order to obtain the
total quantity of 'realizable' exploration drilling. The model finally shares out the total effort
required to achieve the realizable drilling among the regions.
From theory to empirical modelling
121
output (PQ) minus production C 1(R) and exploration C 2 (E) costs. Variations
in the levels of reserves (R) are a function of the rate of addition to reserves (X)
and the extraction rate. The rate of addition to reserves is a function of the
cumulative reserves (x) and the effort expended on exploration (E). In mathematical terms this gives:
Max n= foo [PQ - C 1(R)Q- C 2 (E)] e- ol dt
Q,E
R=x-Q
x= f(E, x)
R, Q, E, x ~O
(8.1)
The actual objective function to be maximized depends on several factors.

For example, does the firm undertake exploration in order to reduce uncertainty or rather to accumulate reserves; is the firm in a competitive market or
in a monopoly position; does it exploit one or several fields, etc.
In a competitive market exploration follows the rent in situ, the scarcity
value of the resource underground. The rent is the shadow value of one
additional unit of the resource. The marginal profit on the production of one
additional unit of reserves is the shadow price of the reserve. Thus the rent is
a measure of the real sacrifice the oil company is prepared to make to obtain
the marginal unit of the non-exploited resource. However, it is usually
impossible to observe this rent. How then can it be measured? In the ideal case,
costs are zero and the rent equals the price of the resource on the market.
Guadet and Hung (1986) point out that in reality the price is only an
approximate indicator of the rent because of the imperfections of the market
and because the price is a reflection both of changes in the extraction cost and
in the option cost which constitutes the value of the resource underground.
Another indicator is the marginal cost to extract the resource, or, even better,
the price at which known reserves are traded in the market. Their use is,
however limited by the availability of statistical data. Devarajan and Fisher
[1982] point out that in a world without any uncertainties, the marginal
discovery cost is an appropriate estimate of the rent in situ. With uncertainty,
marginal costs differ from the rent. The difference depends on the extent of
uncertainty in exploration.
A large number of econometric models have been built in parallel with the
lessons learned from theory. These can be divided into two groups. The first
group takes the problem of inter-temporal maximization as its starting point,
i.e. it starts off from an equation similar to (8.1) (cf. Epple, 1975, 1985; Cox and
Wright, 1976; Nielssen and Nystand, 1986) in order to construct an exploration
function.
The second group combines the variable suggested by theory - for which
there exist historical series of data - with other technical and economic
variables recognized by the oil industry as capable of modifying explorationrelated investment decisions. A linear or logarithmic equation is usually used
to establish an expected profitability function for exploration based on 'ra-
122
tional expectations'. Assuming that the estimated coefficients for the past
remain unchanged, the equations can be used to forecast future exploration
levels. The expected profitability function is estimated in several ways. The
'Canadian school' for example (cf. Uhler, 1983; Scarfe and Rilkoff, 1984; Ryan
and Livernois, 1985; Bing, 1987) use the 'reserves price'; Moroney and Bremmer (1987) and Desbarats (1989) use 'netbacks'; Attanasi (1984) uses the value
of undiscovered deposits; and MacAvoy and Pindyck (1973) use an index of
average deflated revenue.
8.3.1 Models based on the price of reserves
Several Canadian economists have attempted to construct statistical series of

'reserves price' in order to get as close as possible to theoretical (but unobservable) shadow prices. The reserves price has been defined as the price of
discovered but undeveloped reserves. It is supposed to take overall account of
geological, price, cost and taxation effects. This concept is directly derived from
Epple's 'price of a discovery' [1975] which is equal to the discounted net profit
from a given discovery, divided by the amount of oil discovered. Uhler and
Eglington (1983) and Ryan and Livernois (1985) have built data series for
Canadian reserves prices. These are based on several hypotheses concerning
the operators' outlook in terms of production from given discoveries, the
evolution of oil and gas prices, development and operational costs, taxation
and inflation. Reserves price series thus vary from one author to another.
Several empirical models of exploration behaviour in Canada have sucessfully used reserves prices. In Scarfe and Rilkoff (1984) the level of exploration
(E) is a function of reserves price and output. In the light of the uncertain
nature of oil prospection, the explanatory variables used by the model are: a
weighted average of oil and gas undeveloped reserves price (P) where the
weights are given by the proportion of exploratory wells intended to uncover
oil and gas; a weighted average of oil and gas production volumes (Q).
Exploration effort is measured through exploration expenditure. Desbarats
(1989) has carried out several estimations on this model, with the following
results:
Equation 1
Equation 2
LnQ
0.249
-0.019
LnP
0.128
0.024
LnEt 0.733
0.928
R2
0.94
0.90
Price elasticities
Short-run
Long-run
0.13
0.02
0.48
0.33
The results for equation 1 were obtained using Uhler and Eglington's (1984)
reserves prices for the period 1980-91. The results for equation 2 were obtained
using Ryan and Livernois's (1985) reserves prices for 1951-81. The results show
that Scarfe and Rilkofs approach is encouraging but that Ryan and Livernois's
reserves prices series do not significantly contribute to explaining investments
in exploration.
123
Bing's model [1987] is a variation of the previous one. The explanatory

variables are: expected profit on exploration, hydrocarbon production and
land expenditures. The profitability of exploration is no longer measured in
terms of expected profit per unit of reserves, but rather by the expected profit
per unit of production. This concept makes it possible to take into account the
time necessary to produce reserves in the ground. Profit is calculated using
after tax development and operational costs only. The finding cost and any tax
incentives likely to reduce the ex post exploration costs are excluded. In this
way the author puts the emphasis on 'half-cycle profits' instead of 'full-cycle
profits'. The expected half-cycle profit per unit of exploration (UP) is calculated
using data on the size of discoveries, the probability of a discovery of a given
size, costs, prices, taxes, royalities and estimates concerning the evolution of
prices and inflation:
(8.2)
where RVj(t), CJt) and Qi(t) are the present value of income, costs and
production related to discovery i, and nt) is the conditional probability of
developing a reservoir of size i given that the discovery has been made. The
model estimates exploration activity (E) for each of the four provinces of
Western Canada over the period 1962-85. Exploration activity is measured by
land expenditures to begin with, and then by the number of metres drilled.
Straightforward regression is used to estimate the coefficients of the following
equation:
(8.3)
Q is the output in million cubic metres of oil equivalent; noil and n gas are the
expected half-cycle profit on oil and gas respectively; DU is a dummy variable

introduced to cover the effect of the National Energy Programme between
1981 and 1988. The expected profit is estimated as a weighted average for the
current year and the previous year using an Almon lag procedure. The model
fits reasonably well with historical data, but the explanatory power of expected
profit is low in most of the provinces studied.
8.3.2 Netback-based models
The net price for the producer or 'netback' is defined as the wellhead price of
oil (or gas) less production costs, taxes and royalties. In the absence of a
credible explicit measure of profits in the oil industry, the (observable)
producer's net price is a good approximation of companies' cash flow available
for future investment. It also provides a proxy cost of acquisition of reserves.
This indicator has the advantage that it does not require hypotheses concerning operators' (unobservable) outlook concerning the future of prices, costs,
taxation etc.
124
Moroney and Bremmer (1987) develop a complete model of oil supply for
Texas which can be used to forecast additions to reserves, the number of
exploration and development wells, the drilling success rate, hydrocarbon
reserves and output. In the part dealing with exploration, three econometric
equations are used to calculate the number of exploration wells, the success
rate and new discoveries. According to the authors, the factor driving new wells
for exploration and development is the discounted current value of expected
profit from these wells. They take net backs as a starting point for the
construction of a variable which they call 'profit', which is intended to sum up
exploration profitability. Thus, the producer's net prices for oil and gas are
calculated first before going on to estimate the production profile of the
discovery made in year i. Multiplying the netback for the year of the discovery
by the volume of hydrocarbons extracted each year provides an estimate of the
annual revenue over the life of the discovery. Finally, future revenue is added
up and discounted in order to obtain a discounted net operating profit per
barrel of oil or million cubic feet of gas. In order to take geological factors into
account, the success rate and discovery rates (volume of oil and gas discovered
per successful exploration well) are introduced into the explanatory variable
'profit'.
(8.4)
where Pt is discounted net price, S successful oil (gas) wells, N total exploratory
wells, R volume of oil and gas discovered and DC average after-tax drilling cost
per well. The tested equation is log-linear. The expected profit and exploration
drilling (N) for the preceeding year are explanatory variables. The following
results were obtained:
LnN t = -0.091 +0.824LnN t _ 1 +0.134LnIl t (-0.124) (9.190)
(2.930)
Estimation: OLS (1959-83)
R2 =0.86
DW=2.02
(8.5)
t-statistic in parentheses
The author thus finds that the short-run elasticity of exploration with respect
to profits is only 0.13, but the long-run response would be much higher:
0.13/(1---D,824)=0.76.
Desbarats [1989] puts forward a model for exploration expenditure in
Alberta in which exploration activities are taken to be an investment in new
production capacity. Additions to reserves are handled as analogous to
additions to capital stock in an investment model. A long-term equilibrium
relation is the starting point of a general self-regressive-distributed-Iag model
of exploration expenditure. The equation is simplified in order to take into
account reserves prices and several restrictions:
125
Ln Et = -2.27 +0.60 Ln Et - 1 + 1.24 P~8-0.09 Pt-l +0.003 X t - 1 +0.0021';-2

(- 6.S2) (IS.03)
(14.67)
(-9.17) (9.56)
(S.84)
+0.13 Q~~ 1 -0.21 PW t - l -0.87 dQog-0.20 dPW t - l

(4.79)
(-2.41)
(-S.2S)
(-1.87)
R2 = 0.99
D W= 2.40
(8.6)
t-statistic in parentheses
E is real exploration expenditure, pog is the 2-year average of the sum of oil
and gas netbacks, P is the rate of inflation, X (= DO /Q o + Dg/Qg) is the sum of
the ratio of additions to reserves (D) to production (Q) for oil and for gas,
Y( =D8/Q8) is the same ratio for gas, Q0 8 is the sum of oil and gas production
and Pw is the difference between domestic and world oil prices.
The model responds well to the statistical tests. To our knowledge, this is
the best empirical model of exploration behaviour to have been constructed for
a given oil province. Desbarats does not include expectations regarding
economic or geological factors, nor taxation. The model places emphasis on
feedback processes rather than forward-looking factors. This approach is
consistent with the point of view according to which firms' exploration
programmes are based on past and present data concerning capital costs,
output, netbacks and certain (e.g. financial) constraints.
8.3.3 Analyses by basin or region
We shall now examine two models with very different aggregation levels: that
of Attanasi (1984), which deals with the American Permian Basin, and the
Candian Research Institute's model which deals with exploration in large
geographical regions.
According to Attanasi (1984), the decision to invest in exploration in the
short term depends on the field manager responsible, whose profit evaluation
will vary according to the size of the deposits discovered in the zone and to
drilling costs. In the long run the value of deposits still to be discovered in the
province is compared with that of other regions. The model assumes that the
exploration effort (E) is a function of the expected short-term return (II) as well
as of the expected value of deposits remaining to be discovered (0). II is
calculated as the value of the hydrocarbons discovered by successive exploration wells less the cost of the wells, multiplied by the exploration success rate.
In calculating cumulative discoveries X, Attanasi used a model of the discovery
process which predicts the number of oil and gas deposits to be discovered for
different sizes and at different depths. On the basis of prices for year t and
drilling costs, the value of each deposit can be estimated. There is a one-year
lag in the discovery process model's forecasts in order to take account of the
time which the operator needs to estimate correctly the size of new discoveries.
Exploration activity is measured first by the number of wells, then by
126
expenditure:
II
Equation 1 (wildcats)
Equation 2 (expenditures)
0.1005
(2.1)
0.337
(2.1)
t-statistic
R2
0.0579
0.75
(15.2)
5.805
0.50
(16.8)
in parentheses
DW
1.75
2.44
Although the fit is not particularly good, Attanasi concludes that American
firms have been guided by the history of past discoveries in their activity in the
Permian Basin. It should be noted that this model uses variables which
measure operators' expectations in terms of the quantity and size of deposits
only. It includes variables relating to other economic factors. The model
calculates the value of the deposits instead of expected discounted profit. Thus
it only uses observable variables.
One of the few models which examines each oil activity explicitly when
dealing with world oil supply and demand is the Canadian Energy Research
Institute's World Oil Market Model [CERI, 1988]. The supply part is divided
into four sub-models: drilling, additions to reserves, production and costs. The
drilling sub-model calculates the number of metres which will be drilled
annually in each of 16 regions worldwide. The model does not distinguish
between exploration drilling and development drilling, nor does it distinguish
between oil and gas prospecting. Within the model, the drilling effort (E) is a
function of profitability (11), the cost of capital (C) and the utilization rate of
reserves (U R). Since net backs are non-existent for each region, profitability is
estimated simply through prices of representative crude.
(8.7)
The results in Table 8.1 show that it is not easy to reconcile the equation
with historical data. The authors of the model explain this difficulty by the lack
of information concerning real profitability in each region and on the decisionmaking process as regards drilling in LDCs.
8.3.4 Explanatory variables of empirical models

The ways in which models express firms' decision-making vary but the same
factors are nearly always used to explain the level of exploration.
Economic factors
Wellhead prices and technical costs are unavoidable. Some authors also
include other variables, such as drilling activity opportunity costs, operators'
debt servicing costs, the cost of reserves, the inflation rate etc.
127

Table 8.1
OLS estimates for the total drilling, 1960-1984
Supply region
II
UR
Canada
1080.1
(2.33)
13701.5
(5.21)
95.9
(1.73)
44.6
(2.02)
248.9
(3.98)
48.6
(2.79)
29.6
(6.23)
16.3
(2.94)
13.8
(4.63)
27.1
(5.00)
35.8
(1.91)
10.7
(2.45)
81.3
(1.52)
47.2
(7.05)
85.4
(4.78)
231
(6.59)
-123.9
( -1.62)
18989.77
( -2.99)
-286.3
( -2,05)
0
47.5
(1.31)
0.69
-829.7
( -3.38)
-128.1
( -2.07)
0
United Sates
Mexico
Brazil
Argentina
Colombia
Peru
Trinidad
Egypt
Angola
India
Malaysia
Australia
Other Middle East
North Sea
Western Europe
0.71
0
0.4
0.54
-98.3
( -1.82)
0
0.58
-32.3
( -2.90)
0
0.72
0.68
0.60
0.53
-29.6
(- 2.01)
0
0.68
0.31
0.62
0.69
0.86
-905.9
( -7.38)
0.72
-113.8
( -5.27)
0
0.33
Source: Modified from CERI (1988).

OLS modified where necessary for serial correlation.
t-statistic in parentheses.
Geological and technical factors
Exploration results are directly dependent on the volume of the resource

underground. Variables have to express not only the efficiency of exploration,
but also exhaustion of the resource. The drilling success rate, the discovery rate
and the cumulated drilling effort are commonly used. In some cases, global
128
growth in oil activities is brought in through technological development (e.g.

by introducing an improvement in the recovery or success rate.
Taxation
Perfectly competitive markets do not exist in reality. In particular the state's

influence can be very significant. Firms have to pay taxes and rights and are
subject to various rules and regulations (extraction rate, accounting rules, etc).
Firms can also benefit from tax incentives such as a provision for the
reconstitution of a field.
Can these variables explain exploration trends in LDCs? The answer to this
has to be no. The use of these variables is a necessary condition but not a
sufficient one. In the models covering developed countries' oil regions, institutional factors are represented by taxation alone. This is indeed a major
explanatory variable, but it is not the only one. Political risks, national
companies, non tax-related clauses in contracts, lack of infrastructure, etc., can
all affect exploration, often introducing strong discontinuities in the search
activities. The weaker the oil potential and/or the stronger state interventionism, the more discontinuous the exploration effort.
8.4 PARTICULARITIES OF EXPLORATION IN LDCs
Oil exploration in LDCs displays certain particularities due to institutional
and political factors. By institutional and political factors we mean respectively
those factors that relate to state management of natural resources and those
factors capable of altering or even upsetting this management. The former
depend on the legislative, contractual and tax framework governing exploration-production activities and the latter are the 'components' of political risk.
The heterogeneous nature of states implies a complex contractual and tax
framework, a diversity of management approaches and the permanent presence
of political risks.
8.4.1 A complex contractual and tax framework
The first feature of exploration in LDCs is that the state usually owns the
natural resources and foreign companies cannot engage in oil activities unless
they have either a licence or a contract with the authorities. Within the
contracts category, we can distinguish among other production-sharing contracts and risk-service contracts. There are also forms which combine the two
types, especially when it comes to taxation mechanisms.
Under licences, production belongs to the concessionaire who pays a
proportional mining royalty and a profit tax. In general, the company has to
go through one or more intermediate titles before being granted a full
Particularities of exploration in LDCs
129
concession. Under the terms of the licence, the company takes responsibility
for all exploration risks and development, and supplies all the necessary finance
for operations. The oil extracted belongs to the company, though sometimes
the crude legally belongs to the state. In practice the company acts as if the
crude was its own. The national company often has the right to take a stake
in the concession, especially when a commercial discovery has been made. Even
if this mining title hardly resembles the extremely advantageous concessions
seen in the past and is less and less used in LDCs, it remains very widespread
in developed countries.
In production-sharing contracts 2 , the discovered oil belongs to the state.
Production is divided in two: one part (the cost oil) is used to recover costs;
the other (profit oil) is given over in part to the contractor as remuneration for
services rendered. The rest belongs to the state. The company normally pays a
profits tax. Some contracts require payment of the royalty in cash as a
precondition. The company is responsible for all the necessary financing. The
national company can also take a stake.
In risk-service contracts, output remains the property of the state which then
repays and rewards the contractor for its services out of income gained from
the discovery. The 'risk capital' required for exploration is provided by the
company. Development expenditure is also the company's responsibility or is
shared with the state. The contractor pays a tax related to the payments made
by the state. The contractor has the right to buy a share of output. Such
contracts began to appear initially in Brazil, around 1976-77, and developed
rapidly in other South American countries.
In association contracts (or joint operating ventures), the state company
joins up with a foreign company to prospect, and sometimes produce as well,
within a defined zone. The two partners are joint holders of a mining title
issued by the state, usually in the form of a licence. The national company is
treated as a private company with all the rights and obligations which this
entails. Output is shared between the partners according to predefined rules.
On the whole, those countries which already have an established oil industry
prefer to use risk-service contracts and joint operating ventures. Countries
which have proven oil potential or exsiting oil production use productionsharing contracts. Small or little-explored countries use concessions. This is not
a general rule, however. Each country choses a contractual system and tax
regulations closest to the aims of its oil policy in the light of technico-economic
limitations
The main feature of an oil contract is its tax aspect. Under the concession
2 Among the various production-sharing contracts should be noted the system based on profit
sharing (rate of return based profit sharing contract) which was introduced at the end of the 1970s
backed by the World Bank as a suitable system for both underdeveloped and developed countries.
Under this contract, the basic taxation mechanism is the resource rent tax or additional profit tax
(ATP). This tax essentially guarantees the investor a minimum profitability on his investment
before paying any tax. Profitability is calculated on the basis of the project's current cash flow.
130
system, the basic taxation mechanism is the royalty and profit tax. Within
production-sharing contracts, the rules governing the share out of profit oil
take precedence over other aspects. In service contracts, on the other hand, it
is factors such as the indemnity per barrel output which take precedence. Tax
mechanisms usually associated with one form of agreement are sometimes used
in other types of contract. In particular, some production-sharing contracts
include royalty clauses. Oil companies react to the whole tax structure not only
to the level of taxation but also the source of the taxation (Figure 8.1). If, for
t.,.
".~IiI(II.1I
'n~,.ltlf) sJt.rilt,
IL PR
~
c:::J
IO
o"U'IC'IOI"
ROYllli.,
01'
.Jh.,.t or nn:.nd"~
lue"
Profir "./
Yurt
no
Fig. 8.1
H RI G
0 TRA T
Particularities of exploration in LDCs
131
example, the company can recover the same profit through either a concession
or a production-sharing contract, the financial indicators used (e.g. rate of
return, payback period, profit to investment ratio, etc.) are likely to vary
according to the particular situation.
As well as fixing the rules governing the sharing of profits between the state
and the company, contracts lay the foundations on which exploration activities
are based. They determine what work is to be carried out, the minimum
investment required on the part of the company, and the timetable for work
and expenditure. They specify the terms and conditions of renewal of the initial
exploration period and the zones which the company has to give back to the
state as exploration progresses. Several stipulations concerning monetary
transfers become effective during the exploration phase, e.g. bonus payments at
the time of signature and discovery, rights on the granting of a permit, leases.
etc. Some obligations imposed on the company have financial implications
which can discourage or encourage companies from committing themselves to
exploration within a given country. This depends essentially on three factors:
the extent of the financial burden implied by these obligations, the possibility
of recovering this initial financing once production has started, and the rigidity
of the timetable governing payments. Companies consider financial obligations
particularly burdensome during the exploration phase since they have to be
paid before production starts. In addition, such an investment does not bear
any real relation to profits. Considerable sums of money are tied up over long
periods of time and if exploration turns out to be fruitless, the investment is
lost. Financial burdens imposed on companies can render a project unprofitable and block the development of marginal regions.
Contracts containing clauses covering operational restrictions and non-taxrelated obligations can also reduce the overall profitability of a project. These
include the choice of the operator, the way in which operations are carried out,
checks during exploration activity, production targets, a preference for the use
of nationally/produced goods and services, exchange controls, marketing of
output, the sale of oil on the national market at reduced prices, the evaluation
of the oil, the use of natural gas, imports and exports of goods and services, etc.
To sum up the situation from the contractual point of view, the rhythm and
intensity of exploration in LDCs depends on (i) the structure of the tax system
and the level of taxation; (ii) clauses concerning the exploration phase,
particularly those that have financial implications for the company; (iii)
operational restrictions and non-tax-related obligations.
8.4.2 Diversity of management methods in the mining domain
The management of mining operations in LDCs bears hardly any resemblance

to that in developed countries. The state's monopoly power over resources
manifests itself in various ways. Some countries are closed to private exploration (e.g. Mexico, Iran, Iraq and more recently, Brazil). Some countries (e.g.
India, Venezuela, China) keep the best zones for state companies. In these
132
countries, most of the exploration effort is undertaken by the national

company. Investment criteria as well as the financial resources given over to
national companies by the state for oil exploration are such that the rhythm
and intensity of exploration follow quite specific trends, sometimes in opposition to the general trend. For example, during the 1970s and early 1980s
exploration declined considerably in Algeria, Ecuador, Mexico and Nigeria, in
spite of the two oil shocks.
In countries where companies are invited to take part in exploration, the
authorities choose the company according to several (generally secret) criteria
which are heavily influenced by political factors. The competitive bidding
system as practised in developed countries is not used in LDCs. Competitive
bidding is limited to certain contractual clauses, notably the signature bonus
and the minimum level of operation to be respected. In addition, the mining
domain is only accessible when the authorities choose to let it be so. Some
countries offer several exploration permits at a time, others, such as Angola,
prefer to divide the domain into blocks and open them up one by one.
Poor management in the mining domain sometimes leads to monopolization
of zones. This happened in Africa between 1975 and 1979. The phenomenal
extension of mining operations in Africa has not led to a proportional increase
in exploration activities (see Rodriquez Padilla, 1990). The continent did not
profit from the first oil shock. Oil companies acquired exploration zones with
the basic aim of getting a foothold in the territory in order to get a lead on
their competitors or to conserve their zones as an insurance policy or a
speculation for the future. Everywhere but in Africa, price rises and threats to
oil supply have given considerable impetus to hydrocarbon exploration.
Summing up from the point of view of the management of mining operations, the rhythm and intensty of activities depend on:
1. the extent of activities developed by the national company;
2. the speed at which the authorities offer exploration to private companies
and the number of permits;
3. the degree of monopolization of the mining domain by the national
company or by foreign companies.
Ever-present political factors

If political risks are very slight in developed countries, this is not at all the case
in LDCs. Indeed this is one of the main reasons why international oil
companies retreated to their national base after the first oil shock. From an
exploration point of view, political factors can be divided into two subgroups:
those concerning changes in the legal and taxation aspects governing oil
companies' activities, and those which stem from political instability.
The first group of political factors covers total or partial nationalizations,
violations or unilateral renunciation of contracts, arbitrary tax increases/changes in tax mechanisms or suffocating bureaucracy leading to an
increase in costs. Companies pay a great deal of attention to what could be
Factors determining exploration in LDCs
133
called the 'fiscal risk' (i.e. the evaluation of the probability of changes in the
rules of the game) when deciding to commit themselves in a developing
country. The second group of political risks covers political instability due to
ethnic or religious rivalries, civil wars, frontier problems and hostile activities
towards foreigners (sabotage, bombings, etc.).
Most specialists give greater importance to the first set of factors, those
concerning changes in the rules of the game. Political stability is above all a
matter of stability in political behaviour, i.e. a matter of continuity in attitudes
towards foreigners rather than government stability.
It would be impossible to explain changes in exploration activities in West
Africa, for example, without taking into consideration the impact of political
factors (Rodriguez Padilla, 1990). These factors include: partial nationalizations
(Nigeria, Gabon, Angola, Cameroon, Congo, Zaire, Ivory Coast and Ghana);
compulsory changes in contract concessions (Cameroon); unilateral rupture of
agreements (Angola and Benin); constant leadership changes (Nigeria, Ghana);
sabotage and bombings of oil installations (Chad and Angola); civil wars
(Nigeria, Angola); and disagreements concerning the technical performance of
companies (Congo, Ivory Coast and Benin). In the case of the Angolan war of
independence and of the litigation between the Congo government and Elf
concerning the development of the Emerald oil field, exploration was totally
abandoned for between one and two years.
However, it has to be recognized that political factors do not always have a
negative effect, as can be seen in the case of Angola. The Angolan government's
success in maintaining a high level of exploration was based not only on a
flexible approach to taxation but also on a stable attitude with respect to
foreign investors. The government's marxist-leninist ideology did not get in the
way of business relations with the oil companies. Economic relations with the
Luanda government have been unanimously described as very satisfactory.
Finally, it has be said that external political factors have also altered
exploration behaviour in LDCs. One example is provided by the policy of
'diplomatic drilling' developed by French oil companies in several countries in
the Gulf of Guinea which accompanied French policy in these countries at the
end of the 1970s. A second example is provided by the anti-Angolan campaign
orchestrated by the Reagan administration in the mid-1980s which was aimed
at forcing the withdrawal of Cuban forces from the country and a general
weakening of the country's position within the region, especially with respect
to South Africa. Chevron, Texaco and Mobil had to leave the country as a
result of this policy.
8.5 FACTORS DETERMINING EXPLORA nON IN LDCs
In order to discover the relative importance of the factors determining oil
exploration activities in LDCs, pooled cross-section time series analyses have
often been used (see Siddayao, 1980; Broadman, 1985).
134
In her study of the renewal of reserves in South East Asia, Siddayao suggests
the following explanatory variables for exploration drilling (E): international
oil price (P), drilling costs (C) approximated by the US cost index on drilling
machinery equipment and supplies, success rate (SR), size of discoveries by type
of field (T), average depth of wells (DP) and institutional factors, dummy
variable (D U):
The index j refers to the countries - Brunei, Burma, Indonesia, Kampuchea,

Malaysia, Philippines, Thailand, Vietnam - and the index i to the four
categories of field - very large, large, medium and small. The results (see Table
8.2) suggest that the size of discoveries (approximated for a priori expected
return) and 'probably' the institutional factors are the principal explanatory
variables for exploratory drilling in this part of the world.
Using a larger and more varied sample of countries, Broadman comes to
similar conclusions. He suggests that the exploration effort undertaken in
developing countries is a function of geological potential, infrastructure type of
oil contract, taxation system, and of the degree of political risk. The final
specification of the model is as follows:
Eit=al +a2 SR it-l +a3 Xit-l +a4INFi+asCSHit+a6JVSit+a7TXit
+asPRKit+a9DPit+alOPt-l +allAi+a12NOCit+eit
(8.9)
The number of wells (E) is again used for exploration. The geological potential
is approximated by the success rate (SR) and the rate of addition to reserves
(x) (both calculated as running three-year averages). Infrastructure development (INF) is approximated using the contribution of manufacturing to GDP.
The contractual framework is approximated by three dichotomous variables
each of which represents one of the currently used oil contracts: concessions
and production-sharing contracts (CSH), and risk-service contracts, nonrisk-service contracts and joint ventures (JVS). These variables take the value
1 when the corresponding agreement is the dominant contract in year t and 0
when this is not the case. The impact of the taxation system on exploration
(TX) is approximated using the oil revenue taxation rate and the impact of
political risks (PRK) by the operation risk index proposed by the American
firm BERI Inc. Broadman also introduces four other explanatory variables,
two of which are intended to represent economic factors and two to express
certain particularities of LDCs: the depth of wells (DP) provides an approximation of drilling costs; international oil price, calculated as a simple average
over the previous 2- and 3-year periods (P2 and P3 respectively); the size of the
country (A); and a dichotomous variable (NOC) indicating the existence of a
national oil company.
Factors determining exploration in LDCs
135
Table 8.2 Cross-country time series regression for exploration wells

drilled in South-East Asia, 1967-1977
Independant variables
(lagged)
Equation I
(auto)'
Equation II
(auto),
Wildcat
0.54
(5.74)
0.89
(0.68)
-0.15
( -1.11)'
0.12
(0.71)
24.22
(7.14)
4.89
(0.79)
3.16
(0.77)
5.06
(4.21) ..
2.32
(0.48)
-6.66
( -1.56)"
0.98
0.92
Price
Cost
Success ratio
Giant discovery
Large discovery
Medium discovery
Small discovery
Dummyb
Dummy (not lagged)
R2=
0.82
(1.33)"
N
N
29.38
(5.76)
19.1
(2.30)'"
6.71
(1.51)'
10.42
(7.42)
9.82
(1.89)
N
Source: Adapted from Siddayao (1980).

Least squares estimation by Cochrane-Orcutt type procedure (convergence = 0.001).
"Dummy for institutional change.
N=not used .
.... Significant at 0.005 level.
... Significant at 0.05 level.
"Significant at 0.10 level.
'Significant at 0.20 level.
a
Having analysed 47 LDCs over the period 1970--82 (Table 8.3), the author
concludes that the exploration market in LDCs appears to be segmented. The
factors which determine exploration in producer countries are not the same as
in non-producing countries. In the former, institutional factors (contractual
regime, taxation system, existence of a national company) and political risks
have a greater influence than factors related to the physical resource itself. The
opposite is the case in non-producing countries where geological factors play
a greater role.
136
Table 8.3 Cross-country time series regression models for exploration

wells drilled in developing countries, 1973-82
Oil
producers
Dependent variable
Success ratio of wildcats
Discovery size
Infrastructure development
Use of risk contracts or
joint ventures
Use of production
sharing contracts
Income tax rate
4.81
(2.77) ....
2.30E-06
(1.04)
1.20
(3.97) ..
-7.14
(- 3.47) ....
-3.68
( -1.72)'
-0.18
( -1.77)0.26
(1.07)
7.30E-05
(0.69)
0.36
(3.78) ....
Political risk
Average well depth
World oil price 3"
World oil price 2b
Area
Presence of a national
oil compagnie
R
Non-oil
producers
1.15
(2.56)'"
-3.70E-03
( -0.10)
0.63
(1.26)
-1.80E-03
( -0.11)
4.00E-03
(0.62)
-1.00E-06
(- 3.37) ....
0.Q2
(3.62) ....
6.07
(5.12) ....
-4.60E-03
( -1.04)
1.00E-05
(0.423)
0.32
(0.63)
0.62
37.54 ....
250.00
0.03
1.664
220.00
Source: Adapted from Broadman (1985).

P3(t-l)= [P(t-l)+P(t-2)+P(t-3)] -;. 3
-1) = [P(t) + P(t -1)] -;. 2
.Significant at 0.10 level.
.. Significant at 0.05 level.
... Significant at 0.02 level.
.... Significant at 0.01 level.
b P2(t
8.6 CONCLUSION
This chapter has shown the complexity of the determinants underlying oil
exploration in various countries. Particularly for developing countries there
has been relatively little systematic research and results are certainly discourag-
References
137
ing for those who would be in search of a simple 'supply (or exploration) curve' ..
The specific conditions are so heterogeneous because they reflect a multitude
of distinct historical, economic, legal, political, geological, cultural and other
characteristics. Clearly the type and relative importance of the factors determining exploration activities depend on the particular situation of each
country. However, the results of econometric analyses show that non-market
factors playa very important role in explaining exploration behaviour. In some
cases, their importance is such that even the price of oil has a negligible
influence. Siddayao, for example, finds that prices do not have a significant
statistical influence on exploration drilling in South-East Asian countries. And
Broadman concludes that prices go some way in explaining the extent of
exploration effort in oil producing developing countries but that the correlation between exploration effort and price is negative in the case of non oilproducing countries. The price elasticity of exploration is only 0.2 in the first
study and actually -0.1 in the second.
This is not so surprising since even in a simple case we would expect
exploration to depend not only on expected (future) price but also on past price
(since they generate the necesary cash flow). With lags in behaviour and
discrepancies between different perceptions of the future, it is quite natural to
find a weak correlation between price and drilling.
Finally we should make two points on oil supply. First drilling and supply
elasticities are not the same: on the one hand short-run supply can be increased
or decreased by variations in stock or production level that are independent of
exploration. On the other hand, long-run increases in supply do require
increased exploration but here - once again - there are considerable lags and
uncertainties before an exploration effort actually materializes as increased
supply. Secondly we should bear in mind that supply as well as exploration
should not in general be thought of in terms of small individual producers
maximizing expected profits. Rather we have a complicated oligopolistic game
in which a few dominant players make up strategic plans (calculating the effect
of their own actions on the whole market and coordinating their plans with
others) and this introduces additional difficulties in trying to estimate price
elasticities.
REFERENCES
Adelman, M. A., Houghton, 1.e., Kaufman, G. et al. (1983) Energy Resources in an
Uncertain Future, Ballier, Cambridge Mass., p. 434.
Adelman, M. A. and Jacoby, A. D. (1979) Alternative Methods of Oil Supply Forecasting, in Advances in the Economics of Energy and Resources (ed. R. Pindyck, vol. 2,
JAI Press, Greenwich, Connecticut. pp. 1-38.
Arrow, K. 1. and Chang, S. (1982) Optimal Pricing, Use and Exploration of Uncertain
Nature Resource Stocks. Journal of Environmental Economics and Management, 9,
1-10.
138
Attanasi E. D. (1984) A Note on Petroleum Industry Exploration Efficiency. Energy

Journal, 5, 133-85.
Bing, P.e. (1987) A Model of Exploratory Drilling in Western Canada, IAEE, Ninth
International Conference, Calgary, 6-8 July, p. 707-17.
Bohi D. R. and Toman, M. A. (1986) Analyzing Nonrenewable Resource Supply.
Research Project 1220--1. Electric Power Research Institute, Palo Alto, Ca.
Broadman, H. G. (1985) Incentives and Constraints on Exploratory Drilling for
Petroleum in Developing Countries, Annual Revue Energy, no. 10, pp. 217-49.
Cairns, R. D. (1986) 'The Economics of Energy and Mineral Exploration: a Survey.'
IIASA, Laxenburg, Austria (mimeo).
CERI (Canadian Energy Research Institute) (1988) World Oil Market Model; Documentation and Operating Procedures. CERI, Calgary, December, p. 65.
Choucri, N. Heye, Ch. and Lynch, M. (1990) Analyzing Oil Production in Developing
Countries: a Case Study of Egypt, Energy Journal, 1,91-115.
Clark, P. Coene, P. and Logan, D. (1981) A Comparison of Ten U.S. Oil and Gas
Supply Models, Resource and Energy, 3, 297-335.
Cox, J. e. and Wright, A. W. (1976) The Determinants of Investments in Petroleum
Reserves and their Implications for Public Policy, American Economic Review, 66
153-67.
Desbarats, e. M. (1989) Empirical Modelling of Canadian Petroleum Exploration

Activity. OIES Papers on Energy Economics, Oxford Institute for Energy Studies,
June, p. 81.
Devarajan, S. and Fisher, A. e. (1982) Exploration and Scarcity Journal of Political
Economy, 90(6).
Epple, D. N. (1975) Petroleum Discoveries and Government Policy: an Econometric

Study of Supply. Ballinger, Cambridge, Mass., p. 139.
Epple, D. N. (1985) The Econometrics of Exhaustible Resource Supply: a Theory and
Application, in Energy, Foresight and Strategy (ed. Thomas 1. Sargent). Resources for
the Future, Washington, De.
Gaudet, G. and Hung, N. M. (1986) Theorie economique des ressources non renouvel abies: quelques elements de synthese, Quebec, Universite Laval.
Gilbert, R. (1979) Search Strategies and Private Incentives for Resource Exploration, in
Advances in the Economics of Energy and Resources (ed. R. Pindyck). vol 2, JAI
Press, pp. 149-70.
Kaufman, G. M. (1979) Models and Methods for Estimating Undiscovered Oil and Gas;
What they Do and Do Not Do, in Methods and Models for Assessing Energy
Resources (First IIASA Conference on Energy Resources, 20--21 May 1975) (ed.
Michel Grenon), Pergamon Press, Oxford pp. 173-85.
Kaufman, G. M. (1983) Oil and Gas Estimation of Undiscovered Resources, in Energy
Resources in an Uncertain Future; Coal, Gas, Oil and Uranium Supply Forecasting (eds
M. A. Adelman et al.). Ballinger, Cambridge, Mass.
Macavoy, P. and Pindyck, R. Alternative Regulatory Policies for Dealing with the
Natural Gas Shortage. Bell Journal of Economics and Management Science, 4, 454-98.
Moroney, 1. and Bremmer, D. (1987) An Integrated Regional Petroleum Model, in
Advances in the Economics of Energy Resources (ed. 1. Moroney), vol. 6, JAI Press,
Greenwich, Connecticut, pp. 187-220.
Nielssen, T. and Nystand, A. (1986) Optimum Exploration and Extraction in a
Petroleum Basin. Resources and Energy, no. 8, pp. 219-30.
Pakravan, K. (1977) A Model of Oil Production Development and Exploration. Journal
of Energy and Development, 3, 143-53.
Peterson, F. (1978) A Model of Mining and Exploring for Exhaustible Resource. Journal
of Environmental Economics and Management, 5, 236--51.
References
139
Pindyck, R. (1978) The Optimal Exploration and Production of Nonrenewable Resouces. Journal of Political Economy, 86, 841-61.
Pindyck R. (1980) Uncertainty and Exhaustible Resources Markets. Journal of Political
Economy, 88, 1203-25.
Rodriguez Padilla, V. (1990) 'L'impact de la fiscalite sur l'effort d'exploration-production de petrole; Ie cas des pays producteurs d'Africa de l'Ouest. PhD thesis, Institut
d'Economie et de Polique de l'Energie, Universite de Sciences Sociales de Grenoble,
p.520.
Ryan, D. and Livernois, 1. (1985) 'Testing for non-lointness in Oil and Gas Exploration:
A Variable Profit Function Approach? Discussion Paper No. 85-6, University of
Calgary.
Scarfe B. L. and Rilkoff, E. (1984) Financing Oil and Gas Exploration and Development
Activity. Economic Council of Canada, Discussion Paper No. 274.
Siddayao, C. M. (1980) The Supply of Petroleum Reserves in South-east Asia; Economic
Implications of Evolving Property Rights Arrangements. Oxford University Press,
Oxford, p. 240.
Uhler, R. S. and Eglington, P. (1983) The Potential Supply of Crude Oil and Natural
Gas Reserves in the Alberta Basin. Monograph prepared for the Economic Council
of Canada, Ottawa.
US Department of Energy (DOE) (1979) Nigeria - An Assessment of Crude Potential.
Analysis Report DOEjEIA-0184jI4, Washington, DC, p. 26.
US Department of Energy (DOE) (1981) Middle East-Crude Oil Potential from Know
Deposit. Analysis Report DOEjEIA-0289, Washington, DC, p. 143.
US Department of Energy (DOE) (1978) Midterm Oil and Gas Supply Modeling
System Methodology Description. Technical Memorandum TMjESj79-05. Washington, DC p. 43.
Woods, T. 1. and Vidas, H. (1983) Projecting Hydrocarbon Production and Cost: an
Integrated Approach. Journal of Energy and Development, Spring, pp. 267-82.
9
Environmental cost functions:
a comparison between general and
partial equilibrium analysis
Lars Bergman
9.1 INTRODUCTION
An emission control cost function, in the following denoted ECC function, is
a condensed representation of the minimum costs associated with emission
control within a well-defined system such as a plant, a region or a country. In
models of trans boundary pollution, for instance IIASA's RAINS model (see
Alcamo et al., 1987) of the so-called 'acid rain' problem, the ECC functions are
national cost functions. Thus the relevant cost functions are aggregate representations of the costs of reducing emissions from a large number of micro
units within each country.
The estimation of such functions involves two major steps. The first is to
identify emission control measures at the micro level and to estimate the costs
associated with these measures. In practice this amounts to combining engineering and economics. The issues raised in this work have attracted a lot of
attention in several studies during the past few years. Rentz et al. (1988) give a
thorough exposition of issues related to emission control costs in the energy
system. Johnsson et al. (1988) deal with related issues in local energy systems.
The papers in UN-ECE (1988) summarize the 'state of the art'.
H reliable micro data are available, however, it still remains to aggregate this
information in a consistent way. Thus the second step in the estimation of a
national ECC function is to aggregate the micro information. Moreover, the
results should be presented in a closed form, suitable for direct integration in
decision-support models such as RAINS.
The natural point of departure when micro data are to be aggregated into
sectoral or economy-wide ECC functions is to use a general equilibrium
perspective. The reason is that emission control measures, at least when
ambitious emission reduction goals are to be attained, might affect prices and
142
Environmental cost functions
the allocation of resources in the economy as a whole, i.e. have general

equilibrium effects. The purpose of this study is to discuss the aggregation of
micro cost information into macro cost functions against the background of
the possibility of general equilibrium effects of emission control measures.
9.2 THE DEFINITION OF ECC FUNCTIONS l
In general an ECC function is a minimum value function, i.e. the solution to a
cost minimization problem. It follows that the nature of the ECC function
depends on the specification of the underlying cost minimization problem. In
this section a simple taxonomy for national ECC functions will be introduced.
For simplicity it is assumed that the minimum cost of attaining any emission
reduction at each micro unit is known. It thus remains to aggregate the micro
data.
The aggregate opportunity cost, C, for a country of reducing its SOx
emissions can be written as a function of the parameters of the underlying cost
minimization problem. In the most general case the resulting macro cost
function can be written
C=C(E;Z,X)
(9.1)
where E is the emission level, Z is a vector of policy instrument values and X

is a vector of exogenously given economic variables.
The function C( ... ) is decreasing in E and attains the value zero for some
finite value of E. As the definition of the cost function implies that any given
emission level is attained in a cost-efficient way, Z, i.e. the vector of policy
instrument values, does not include the policy instruments that are used to
affect the emission level.
Within this very general framework it is possible to discuss several different
emission control cost concepts. Any choice of variables to include in the vector
X implies a specific partitioning of the economy into one part which is affected
by the emission control measures and one part which is not. In other words,
the specific partitioning adopted essentially determines the nature of the cost
minimization problem and thus what kind of model is needed for the
estimation of the national ECC function. In the following, three possible
alternatives will be briefly discussed. For simplicity it is assumed that all
emissions originate in the energy sector.
The general equilibrium perspective amounts to assuming that emission
control measures may affect prices and the allocation of resources in the entire
economy. This implies that the vector X in (9.1) only includes measures of the
economy's initial resource endowments and world market prices of traded
goods. In the following we will refer to this type of cost function as C l, and
1 The
exposition in section 9.2 is to a large extent based on Carlsson (1988).
The definition of ECC functions

the corresponding X and Z vectors as X
C1
143
and Z 1. That is
= C 1 (E;X\Zl)
(9.2)
However, if variations of E within the relevant range do not affect the prices
of non-tradable goods and factors, these variables can be taken as exogenously
given and included in the vector X. This is the case if emission control
measures within the energy sector have small or negligible effects on other
prices in the economy. As a result a modified cost function, C 2 , is obtained.
This function can be written
(9.3)
with X 2 and Z 2 being obvious modifications of X 1 and Z 1. It is important to

note that as C 1 implies a wider range of adjustment possibilities than C 2 , it
must hold that C 2 is greater than or equal to C 1 for any given level of E and
compatible X and Z. This result follows from the so-called 'Ie Chatelier
principle' (see for instance Varian, 1984).
The cost function C 2 implies that any emission reduction is the result of the
combined effect of a number of adjustment mechanisms in the energyproducing sector as well as in the energy-consuming sectors of the economy.
More precisely, the emission level would be affected through direct measures
and/or structural changes such as:
1. Installation of technical devices reducing the emissions from a given set of

energy production units using given types of fuels.
2. Implementation of fuel switching measures and/or installation of new
production units in the energy sector.
3. Changes in the level and compositon of energy consumption.
The ECC function C 2 includes all the adjustment mechanisms (1) to (3) in
the list above. It is implicitly based on some kind of partial equilibrium energy
market model. Thus C 2 is a measure of the loss of consumer and producer
surpluses resulting from emission control measures in the energy sector. As all
other prices by assumption are constant, there are no losses of consumer
and/or producer surpluses on other markets than the energy market. Consequently C 2 is a measure of the cost of emission control for the economy as a
whole.
However, if the cost-efficient combination of emission reduction measures
does not affect the consumption of energy, we can envisage a third, and from
an economic point of view even more narrow, type of ECC function, C\
defined by
(9.4)
where X 3 includes exogenously given energy consumption levels and Z3 is an

appropriately modified version of Z2. In terms of adjustment mechanisms this
cost function includes (1) and (2) in the list above. By the same reasoning as
144
above it holds that C 3 is larger than or equal to C 2 for any given E and
compatible X and Z.
Obviously one can go on excluding additional adjustment mechanisms and
thus define even more narrow cost concepts. Thus a C 4 cost function would
only include type (1) adjustments only, i.e. installation of technical devices such
as flue gas desulphurization, fuel cleaning etc. As a result any specific cost
estimate based on the C 4 concept would be greater than or equal to the
corresponding C 3 estimate.
Cost estimates based on a narrow cost concept such as C 4 or C 3 can be
obtained from a simple model, but may be biased upwards. The question then
is to what extent narrow cost concepts quantitatively affect the estimates of
emission control costs in practically relevant cases. Since some of the major
studies of sulphur emission control costs are based on narrow cost concepts, 2
this is not an unimportant issue.
In this study the evaluation of general equilibrium effects of sulphur emission
control and the estimation of C 1 ECC functions has been carried out by means
Table 9.1
Character
Periodicity
Time
Use
Economic agents
Sectors of
economic
activity
Data base
Endogenous
Exogenous
Size
Software
Brief model descriptions
CGE
ENMARK
Computable general
equilibrium
Yearly
Long-term
Projections
Households, firms
Partial equilibrium
Forest industry
Steel industry
Manufacturing industry
Construction
Services
Transport and communications
Public sector
National accounts
Factor and product
prices and quantities
Factor supplies and
world market demand
and prices
230
GAMSjMINOS
Yearly
Long-term
Projections
Energy producers and
consumers
Central electricity
District heating
Specific data
Energy production,
consumption and prices
Nuclear and fossil fuel
prices, GDP, political
constraints
475
GAMSjMINOS
2The cost estimates generated within IIASA's RAINS model are based on the C 4 cost concept,
while the cost estimates generated within the BICRAM study are of the C 3 type. The IEA/ETSAP
study, which is based on several national variants of the so-called MARKAL model, is aimed at
estimating national cost functions of a type somewhere between C 3 and C 2.
Equilibrium effects of emission control
145
of a separate CGE model of the Swedish economy. For the comparison of C 2

and C 3 an energy market model called ENMARK, implemented on Swedish
data, has been used (see Table 9.1 and Carlsson, 1988).
9.3 GENERAL EQUILIBRIUM EFFECTS OF EMISSION CONTROL
The general equilibrium effects of sulphur emission control are evaluated by
means of a computable general equilibrium (CGE) model of the Swedish
economy. A complete description of the model is given in Bergman (1989) and
Bergman (1990). As this study mainly is concerned with the choice of ECC
function concept, only a brief description of the model will be given here.
9.3.1 General features of the CGE model
The model pictures an economy with three tradables-producing and five
non-tradables-producing sectors. Power production is one of the nontradables-producing sectors. Two of the tradables-producing sectors are pricetakers on international product markets, while the third faces a downward
sloping, price-dependent export demand function. The economy's resource
endowment consists of capital, labour and electricity from existing hydro and
nuclear power plants, and the supply of these resources is completely inelastic.
Both production and consumption of goods and services lead to emissions
of SOx, NO x and CO 2 In particular both of the price-taking sectors, the forest
industry and the iron and steel industry, happen to be relatively energyintensive and produce a relatively large amount of SOx emissions per unit of
output. Environmental policy takes the form of national maximum total
emissions of each one of these pollutants. The environmental policy objectives
are implemented by means of a system of tradable emission permits. In order
to comply with the regulations implied by the limited supply of emission
permits producers can change their technology, i.e. the proportions in which
various inputs are used, or implement direct abatement measures. However,
the emissions of CO 2 cannot be affected through direct abatement measures.
Consumers can change their pattern of consumption of final goods, and
thereby affect the pattern of production. In addition the real costs of emission
control tend to reduce disposable income and thus the level of final goods
consumption.
As emission permits are assumed to be tradable, the equilibrium prices of
emission permits will be equal to the marginal cost of emission control. 3 The
cost reflects the direct abatement cost functions, the substitutability of various
3 An alternative interpretation is that the calculated equilibrium emission permits prices represent
the emission taxes that would have to be implemented in order to comply with the given total
emission constraint.
146
inputs in the production processes and the substitutability of various goods

and services in final consumption. Thus the national ECC function for SOx can
be estimated on the basis of the relation between national emission levels and
equilibrium SOx emission permit prices.
9.3.2 Some simulation results
The basic issue is whether contemplated measures to reduce emissions of SOx

are likely to produce significant general equilibrium effects. If they do, there is
a case for C 1 ECC functions. Otherwise the estimation of aggregate ECC
functions can be based on partial equilibrium sector models. In order to
elucidate this issue the CGE model of the Swedish economy was used to
simulate conditional general equilibrium resource allocations in the Swedish
economy in 1995.
Assumptions about the trends in exogenous variables such as the accumulation of capital, the growth of the labour force and world market conditions are
essentially the same as in the latest long-term economic survey published by
the Ministry of Finance. However, it was assumed that all nuclear power plants
currently in operation in Sweden will be in operation also in 1995. 4 One
consequence of this assumption is that there is no need for new power capacity
before 1995. Thus, as in the current situation, 5 electricity generation produces
no SOx emissions in 1995.
The basic issue with regard to this report is to what extent more stringent
constraints on SOx emissions affect the equilibrium factor prices. In order to
highlight this issue a series of simulations, where everything except the
constraint on SOx was kept constant, were carried out. The simulation results
are summarized in Table 9.2.
Table 9.2 Calculated marginal cost of SOx emission control and equilibrium
factor prices at various SOx emission constraint levels in 1995
Constraint"
324
300
275
250
225
200
175
Price index C of
Marginal ECC b
Capital
9.2
17.1
29.2
42.7
57.2
72.3
0.999
1.015
1.007
0.985
0.960
0.934
0.908
Labour
Electricity
1.321
1.305
1.274
1.254
1.233
1.210
1.186
1.580
1.619
1.668
1.720
1.778
1.841
1.906
"SO. emissions in 1000 tonnes.

bIn SEK/kg.
C 1985 = 1.000.
4 According
to a recent decision two nuclear reactors should be phased out around 1995.
Sweden the bulk of the electricity generated comes, in roughly equal proportions, from hydro
and nuclear power plants, i.e. from processes that do not emit any SOx'
5 In
Equilibrium effects of emission control
147
As a guide for the reader, it can be mentioned that the total SOx emissions
in Sweden in 1980, measured as S02, were 483 000 tonnes. Thus a reduction of
30%, a number often referred to in relation to the so-called '30% Club', would
mean a reduction to annual emissions equal to 338 000 tonnes. The first row
of Table 9.2 indicates the emissions when there are no additional constraints
on SOx emissions. As can be seen the emission level in that case represents a
greater than 35% reduction compared to the level in 1980.
It is easy to see that increasing stringency of the SOx emission constraint
does affect factor prices. The question is how strong the impact needs to be
before one can say that the impact is significant. One arbitrary limit would be
a 10% change in relation to the no constraint case. On the basis of this norm
the emissions of SOx can be reduced by almost 40% without 'significant'
impact on the prices of capital and labour. If the limit is set at 5%, the
corresponding emission reduction is around 25%.
As the Swedish economy to a relatively large extent is specialized on energyand SOx emission-intensive production, the results are sensitive to the assumptions about the market power of Swedish producers on international markets.
As was mentioned above, the model is based on the assumption that the
Swedish producers in the energy- and emission-intensive industries are pricetakers on international markets. If they, instead, do have a certain degree of
market power, this would reduce the general equilibrium effects of SOx
emission constraints.
It would have been desirable to be able to generate a C 2 ECC function that
was directly comparable to the C 1 ECC function implied by the results in
Table 9.2. Unfortunately that is not possible. However, Table 9.3 displays the
result of calculations based on two simplifying assumptions. These assumptions
are that the marginal emission control costs reported in Table 9.2 only consist
of capital and labour costs and that the proportions of these cost items are
9 to 1. 6
Table 9.3
Calculated C 1 and C 2 ECC functions for Sweden
SOx emission
constraint"
Marginal cost of emission control b

C1
C2
C 2 jC 1
300
275
250
225
200
175
9.2
17.1
29.2
42.7
57.2
72.3
9.3
17.3
30.2
45.2
62.2
80.8
1.011
1.012
1.034
1.059
1.087
1.118
"1000 tonnes.
bSEKjkg.
6 Although these assumptions are quite arbitrary, they seem to be well in line with available micro
data on emission control costs.
148
It is clear from these resuits that the C 2 ECC function tends to overestimate
the emission control costs. In this particular case, however, the overestimation
is not that serious for relatively small emission reductions.
The next step is to compare C 2 and C 3. Estimation of such functions can be
based on a partial equilibrium model of a sector of the economy. In this study
the sectoral cost analysis is focused on the energy sector, i.e. the use of primary
energy for production of electricity and heat. The model used in the analysis is
called ENMARK, the ENergy MARKet mode1.
9.4 THE ENMARK MODEL 7
ENMARK is designed as a static, partial equilibrium model of a set of
interdependent energy markets. On the demand side a number of demand
categories are distinguished. For each demand category a linear demand
function in energy prices and real income (or level of real output in the case of
industrial energy demand) is specified. The numerical implementation of the
demand functions is based on available econometric results on the price- and
income responsiveness of the electricity and heat demand by different consumer
categories.
On the supply side each production activity is assumed to exhibit constant
returns to scale within given capacity limits. Thus the supply functions for
electricity and district heat generated by ENMARK are increasing and
step-wise linear functions. The numerical specification of the supply functions
is based on engineering cost estimates for various electricity and heat production technologies, as well as on data on installed capacities. 8
A detailed description of ENMARK is given in Carlsson (1988).
BIn brief the GAMS-modelling of the supply side of ENMARK is done in the following way: a
number of technically feasible production activities are defined and each one of these is defined in
a three-dimensional fashion. This means that the GAMS representation of an ENMARK
production activity variable is written X(T, F, A), where the T-dimension refers to 'Technology',
the F-dimension to 'Fuel' and the A-dimension to 'Abatement level'. By defining the vector of
technologies, the vector of fuels and the vector of abatement levels the complete set of production
activities is defined. Thus, in the Swedish ENMARK model the T-vector is defined as
7
T=(COND, COMB, CHP)

where COND means condensing power, COMB heat production by ordinary combustion of fuels
and CHP combined heat and power production. The F-vector is defined as
F=(NUC, COAL, OIL, GAS)
where NUC is nuclear fuel while the other elements have obvious interpretations. The A-vector,
finally, is defined as
A=(NO, FGD)
where NO means that there is no abatement technology installed, while FGD means that flue gas
desulphurization is installed.
The T-, F- and A-vectors, in conjunction with some constraints ruling out unreasonable
Comparison between C 2 and C 3 cost concepts
149
The energy markets are modelled as competitive markets, and all energy
demand functions are linear. In view of the linearity of the demand functions
and standard economic theory ENMARK is specified as a quadratic programming model in which the sum of consumer and producer surpluses across
energy markets are maximized subject to market clearing and capacity constraints. Numerical solutions are obtained by means of GAMS/MINOS.
The core of the model is a national market for electricity and a set of regional
markets for district heat. ENMARK is driven by exogenous assumptions about
real incomes and the prices of fuels and other inputs. It endogenously
determines the prices of electricity and district heat, the consumption of
electricity, district heat and various fuels. It also determines the emissions of
SOx, NO x and CO 2 originating in the energy and heating sectors.
The general structure of ENMARK implemented on Swedish data and the
corresponding energy flows in 1985 are shown in Figure 9.1. The demand
variables V, Hand U represent the electricity demand for direct electric heating
systems, the electricity demand for water-borne heating systems and all other
types of electricity demand, respectively. In other applications the demand
variable U is further subdivided. In analyses of energy taxation, for instance, it
is necessary to distinguish the electricity demand by the so-called electricityintensive industries, i.e. the industrial electricity consumers for whom a special
reduction rule applies. Other symbols in Figure 9.1 should be self-explanatory.
If an upper limit on the total emissions of, say, SO., is imposed, the resulting
loss of consumer and producer surplus can be calculated by means of
EN MARK. The outcome of the calculation is a C 2 estimate of the cost of
emission control. If, in addition, the energy demand functions are replaced by
a vector of fixed energy consumption levels, the emission limit generates a C 3
ECC estimate. By comparing the two cost estimates a ceteris paribus comparison of the two cost concepts is obtained.
9.5 A COMPARISON BETWEEN THE C 2 AND C 3 COST CONCEPTS
The purpose of this section is to use ENMARK for an explicit comparison
between a C 3 and a C 2 cost function for the Swedish energy markets. The
alternatives, define the production alternatives in the model. Thus conventional nuclear power is
the activity X(COND, NUC, NO), while combined heat and power production in coal fired plants
with flue gas desulphurization is represented by the activity X(CHP, COAL, FGD). It also follows
that the set of production activities can easily be changed and/or extended by changes in or
additions to the T-, F- and A-vectors.
Each one of the production activities is represented by an output vector, an input vector and an
emission vector. In addition there is a capacity limit that can be extended through investments.
The investment criterion is based on a comparison between scarcity rents on existing capacity and
annualized capital costs of new capacity. Thus in a simulation of energy market conditions at some
future point in time new capacity is added to existing capacity as long as the scarcity rents in
question exceed the annualized capital costs of new capacity.
150
-0. ._-Light oil market
Production of electricity
Inputs
Electricity market
{=
F=F(PF,PH, Y)
H= H(PF' PH' Y)
t + - - - V= V(Pv, Y)
' - - - - U=U(Pu,PB , Y)
Elec. for heat

pumps 0,
He':O~'k"
Elec. for
elec. boilers 0,
Heat
Production of heat, market 1

Inputs {
1+---- W, = W, (Pw" Y)
Heat market 2
Inputs {
....- - - W,= W,(Pw" Y)

Heat market 3
Inputs {
Oil markets
(GWh)
Consumption
Light oil
Heavy oil
53800
32900
Electricity
market
Heat markets
(GWh)
Consumption
Heat p
Boiler
Export
16100
10300
88590
1175
3600
1400
Total
121165
V
U
(GWh)
Production
Hydro power
Nuclear
Cogeneration
Backpressure
Condense
Gas turbines
Fig. 9.1
Consumption
Production
64220 W1
51200 W2
2590 W3
2330
770
55
20100
1700
7800
121165
35600
Cogeneration
Heat planes
Heat pumps
Elec boilers
Industry
6940
19900
3200
3000
2560
35600
The structure of the ENMARK model.
critical difference between the two cost concepts is that the level of energy
consumption is endogenous in C 2 but exogenous in C 3 The specific assumptions about exogenous variables in ENMARK are described in Carlsson
(1988).
Comparison between C 2 and C 3 cost concepts
151
The experiment was designed as a comparison between two projections for

1995. In one of these the 1995 sulphur emissions of the Swedish energy sector
were constrained to be at least 30% lower than the corresponding emissions
in 1980, while there was no constraint on such emissions in the other projection.
As the sulphur emissions of this sector were 154000 tonnes in 1980, the 30%
reduction implied an upper limit for 1995 equal to 107800 tonnes in the
experiment. However, owing to endogenous emission reductions the forced
emission reduction implied by the constraint was only 19905 tonnes. Using
ENMARK a C 2 estimate of the cost of complying with this emission constraint
was made.
A critical assumption behind the simulation is that new investments in
nuclear power are not allowed, and that investments in natural gas or
biomass-based electricity production are more costly alternatives than investments in coal power. In Table 9.4 we have summarized the contribution of
various adjustment mechanisms to the overall emission difference between the
two projections.
These results clearly suggest that emission reductions in the energy sector
are likely to lead to higher consumer prices of electricity and district heat, and
thus to consumption reductions. In this particular case the emission constraint
produced a 19% increase (in relation to base case projection for 1995) of the
marginal cost of high voltage electricity, which roughly corresponds to the
producer prices of electricity. As a consequence total electricity consumption
was reduced by 3.5% (in relation to base case projection for 1995). This
consumption reduction was sufficient to reduce the need for additional capacity in coal fired condensing plants by approximately 45%, which, as can be
seen in Table 9.4, implied a considerable saving in terms of sulphur emissions.
The results in Table 9.4 suggest that a more narrow cost concept than C 2
would, by definition, exclude important adjustment mechanisms and thus
overestimate the emission control cost. In this particular case it turned out that
the corresponding C 3 cost estimate was approximately 25% higher. Obviously
there are realistic cases where emission reduction cost functions of the C 3 type
seriously overestimate the true cost. We conclude that higher order cost
functions than C 2 should be used only when there is strong a priori evidence
that demand responses are likely to be insignificant.
Table 9.4
rences
Allocation of projected emission diffe-
Emission difference (in tonnes) due to

Difference due to fuel substitution
Difference due to abatement measures
Difference in electricity consumption
-30
-5510
-14365
Total difference in sulphur emissions
-19905
152

9.6 ECC FUNCTIONS IN CLOSED FORM
In addition to the conceptual issues dealt with so far, there are some practical
issues related to the use of ECC functions. ENMARK is, like most energy
systems models, a relatively big model. 9 Thus it cannot easily be integrated in
a strategy model such as RAINS. The question then is whether the cost
function implied by a set of ENMARK solutions could be represented by a
continuous function of the type
(9.5)
In order to represent the cost functions generated by ENMARK, the model

has been used to generate a set of 'observations' on emission control costs at
various maximum levels of SOx emissions from the energy sector. Then each
1.2
1.1
1.0
0.9
0.8
:.::
LU
en
c Cil
0.7
=
E
0.6
0-0
- C
-
(1j
I/)
::J
"'-
0
()
0.5
0.4
0.3
0.2
0.1
0
5 10
15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100105110115
Emissions (ktonne)
Real
+ Estimated
Fig. 9.2 The ECC function, Sweden 1.

9The Swedish version consists of around 150 equations and inequalities.
153
ECC functions in closed form

2
1.9
1.8
1.7
1.6
1.5
1.4
1.3
C/J_
1.2
c '"
1.1
0
_ "
c
ro
'"
:::J
1.0
0.9
Cil- 0.8
0
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100 105 110 115 120 125 130 135
Emissions (ktonne)
OReal
+ Estimated
Fig. 9.3 The ECC function, Sweden 2.

of the cost functions was estimated by means of conventional econometric
techniques on the basis of these 'observations'.
Two examples of cost functions estimated in this way on Swedish data are
depicted in Figures 9.2 and 9.3. The cost function in Figure 9.3, 'Sweden l' is
based on the assumption that 12 nuclear reactors are in operation, while the
cost function 'Sweden 2' in Figure 9.3 is based on the assumption that ten
nuclear reactors are in operation. For both of these cost functions it is assumed
that the maximum sulphur content in heavy fuel oil is 0.4%, and that natural
bas is available in the production of district heat.
As can be seen in Figures 9.2 and 9.3, the cost functions tend to have a
quadratic form. The estimated equations are the following (t-values in parentheses):
Sweden 1 (Figure 9.2)

C=1235.371
(59.773)
20.912*E
( -14.462)
0.091*E2;
(2.995)
R2=0.999
154
Sweden 2 (Figure 9.3)
C=2088.952
(32.542)
43.625*E
( -11.260)
0.349*E 2;
(5.274)
R2=0.996
In these ECC functions the cost is expressed in MSEK and the emissions in
1000 tonnes S02. The marginal cost of emission control is the negative of the
derivative of C with respect to E. As the estimated ECC functions are quadratic
it follows that the level of the marginal cost depends on the level of emission.
Thus, at the emission level 50000 tonnes the marginal cost of emission
reduction would be around 11 Swedish kronor (SEK) per kg S02. The
corresponding figure for Sweden 2 is 9 SEK per kg S02.
9.7 CONCLUDING REMARKS

The conclusions of this analysis can be summarized in a few points:
1. ECC functions that only incorporate direct abatement measures can lead to
seriously upward biased cost estimates.
2. In some cases it is necessary to take general equilibrium effects into account
when ECC functions are to be estimated.
3. In a wide class of cases, however, it seems that C 2 is a sufficiently elaborated
cost concept.
4. It is a relatively straightforward task to estimate C 2 ECC functions and to
represent the results in a closed form, suitable for integration in strategy
models such as RAINS.
ACKNOWLEDGEMENT
The author is grateful to the Nordic Council of Ministers for financial support
and to Bo Andersson and Anders Carlsson for research assistance.
REFERENCES
A1camo, 1., Amann, M., Hetterlich, 1. P. et al. (1987) Acidification in Europe. AMBIO,
16, no. 5.
Bergman, L. (1989) 'Tillvaxt och miljo - en studie av mAlkonflikter'. (Economic growth
and the environment - a study of goal conflicts). Bilaga 9, LU90, Ministry of Finance.
Bergman, L. (1990) 'General Equilibrium Effects of Environmental Policy: A CGEModelling Approach.' Research Report, EFI, Stockholm School of Economics.
Carlsson, A. (1988) 'Estimates of the Costs of Emission Control in the Swedish Energy
Sector.' Research Report 273, EFI, Stockholm School of Economics.
Johnsson, 1., Bjorkqvist, 0., Larsson, T., et al. (1988) 'LAngsiktig kommunal energi-
References
155
och miljoplanering i Uppsala kommun'. (Long range energy and environmental

planning in Uppsala). Rapport A 88-173, Institutionen for Energiteknik, Chalmers
Tekniska Hogskola, Gothenburg.
Rentz, 0., Haasis, H. D., Morgenstern, T., et al. (1988) 'Energy and Environment:
Optimal Control Strategies for Reducing Emissions from Energy Conversion and
Energy Use.' Working Paper, Institute for Industrial Production, University of
Karlsruhe.
UN-ECE (1988) 'Air Pollution in Europe: Environmental Effects, Control Strategies
and Policy Options.' Discussion document from a conference in Norrtalje, Sweden.
Varian, H. R. (1984) Microeconomic Analysis, Norton, New York.
---10--Energy policies in a
macroeconomic model:
an analysis of energy taxes
when oil prices decline
P. Capros, P. Karadeloglou, G. Mentzas
10.1 INTRODUCTION
Economic planners and policy-makers have been acquainted, in the last fifteen
years, with unanticipated oil price increases. The energy economics literature
is abundant with research exploiting alternative macroeconomic policy rules
that would mitigate the negative impacts of oil price shocks. However, the
unprecedented nosedive of oil prices in 1986 poses some new policy questions,
beside serving as a reminder about the volatility of the world oil market.
According to the literature the recession observed in the 1975-85 period and
the high inflation figures are well tied to the oil price shocks of 1973-74 and
1979-80 (see for instance Eckstein, 1978; Gordon, 1975, Hamilton, 1983; the
contributions in Mork, 1981; Bruno and Sachs, 1982, 1985). It is generally
admitted that oil price shocks impose a macroeconomic cost on oil-importing
economies through various mechanisms: reduction in real national income;
increase in the marginal cost of production; drop of labour and/or capital
productivity through factor substitutions; etc. These effects have been quantified for many oil-importing countries by means of macroeconomic models (e.g.
Hickman, Huntington and Sweeney, 1987).
In order to provide a means of mitigating the negative impacts of unanticipated oil price increases, various taxation instruments have been examined;
e.g. consumption taxes, import tariffs, subsidies, etc. (see Hudson and Jorgenson, 1974); Gilbert and Mork, 1984, 1986; Pindyck, 1980 and Mork, 1985).
Most authors conclude that establishing accommodating policies seems more
efficient than varying tax rates on the consumer price of oil; such accommodating policies include monetary accommodation, income tax reduction,
supply-side credits, etc.
If, however, the developed theories were applied in reverse, the result of an
158
Energy policies in a macroeconomic model
oil price decline should be a substantial stimulus to real activity and a drop of
inflation. Macroeconomic models support such a conclusion, although one
should be aware of the fact that most of them experience an almost linear
symmetry around their solution point. lOne might then suggest that maximum
efficiency will be obtained by letting the market forces free to act. However,
significant policy questions are still put forward:
1. Given that the tax rates on oil products were defined during a period of
upwartl trends of oil prices, one policy dilemma is: should one change these
tax rates and if so, in which direction?
2. A number of countries facing significant public deficits need to maintain or
even increase the public revenues that stem from the taxation of oil
products. In the case of an oil price decline such countries tend to increase
tax rates on oil products. What is the macroeconomic impact of such a
policy?
3. Low oil product prices for domestic consumption may not comply with
long-term objectives, mainly with the reduction of the dependence on oil
imports. In fact low prices may delay or even cancel energy conservation
programmes or energy restructuring projects that aim at substituting oil.
What is the macroeconomic cost of using tax rates on oil products in order
to avoid a shift of energy dependence on imported oil?
4. If, for the above reasons, an increased taxation on oil products is decided,
what is the best accommodating policy, i.e. where should the increased
public revenues be allocated? The importance of this last question is related
to the policy response of the public sector when facing a variety of options
aiming at accelerating growth. Should the public sector try to boost the
economy by increasing demand (e.g. public investment increase) or should
it follow a supply side oriented policy? Should the public sector reduce its
borrowing requirements (PSBR) or should it keep the PSBR at its initial
level and spend the additional resources for development purposes?
These issues have not been sufficiently analysed in the literature, although
most governments have changed their oil products pricing policies after the oil
price decline of 1986. For instance, in most European countries taxes on oil
products have been used to increase public revenues while energy restructuring
funding has been retarded.
In this paper we attempt an analysis of these issues within the context of
macroeconomic modelling. We retain a medium-term perspective and start by
assuming that the economy experiences unemployment and excess capacity
when the price decline occurs. The analysis excludes any response elements that
refer to long-term equilibria, optimum allocation of resources or welfare
1 See e.g. pages 42-5 of Hickman, Huntington and Sweeney (1987) who report results of 14 models
for the US economy. This symmetry, however, was not experienced in practice and has recently
been criticized by a number of research studies, although no concrete theoretical explanation of an
asymmetry has yet been offered; see Davis (1987), Loungani (1986) and Mork (1989).
Introduction
159
characterization of results. Such issues should be analysed within the context

of (price-adjusted) general equilibrium models 2 rather than within the stream
of (quantity adjusted) neo-Keynesian macroeconomic models. 3 The latter
approach is adopted in the present paper. It focuses on the short- to mediumterm maladjustments which can be better treated under the assumptions of
imperfect competition and temporary disequilibrium. However long term
results (which go to the year 2000) are indicative of the long-term trend of the
impacts.
The scope of the paper is not only to analyse energy and macroeconomic
policy issues within the context of decreasing oil prices, but also to inquire into
the properties of macroeconometric models currently used by the Commission
of the European Communities (CEC), since the model used is one of the family
of the HERMES macroeconometric models [see Appendix for further description of CEC energy-economy modelling].
The analysis is carried out with the use of the HGRV (Hermes Greece
Reduced Version) model which is a large-scale neo-Keynesian multi sectoral
macroeconometric model of the Greek economy.4 The economy represented is
thus demand driven and excess supply in all markets is assumed. Potential
supply and derived factor demand is determined by a neoclassical production
function. Disequilibrium between potential and effective production influences
price formation. Demand determines total production which in turn evaluates
the quantity of labour required. Labour supply is determined by households,
while employment is determined as the minimum of supply and demand.
Factor demand permits factor substitution to take place. The share of each
production input is determined by relative factor prices. Imports and exports
depend on the level of demand and on competitiveness. Supply, demand and
inter-industry transactions (Input/Output) are determined at a four sector level.
Domestic prices are influenced by the unit cost of production, the profit rate
and disequilibrium in the goods market. The wage rate is determined by a
Phillips curve formulation, in which the rate of productivity also plays a role.
A complete presentation of the model can be found in Capros et al. (1988).5
The paper adopts a dual approach by solving analytically an aggregate
conceptual model and simulating a large-scale one. The next section of the
paper provides a conceptual analysis with the use of a strongly simplified
aggregate energy-economy model that replicates the properties of current
2There exist quite a lot of general equilibrium models used for energy-economy analysis; see for
instance Hudson and Jorgenson (1974), Dantzig et al. (1978), Manne (1978), Longva and Olsen
(1983), Borges and Goulder (1984), etc. Bergman (1988) provides a survey.
3 Within the neo- Keynesian stream of energy--economy modelling, one can include Wharton EF A
(1979), Brillet et al. (1982), d'A1cantara and Italianer (1982), Capros et al. (1984), the models used
in Hickman, Huntington and Sweeney (1987), Helliwell et al. (1987), Patterson et al. (1987), Capros,
Karadeloglou, Mentzas and Valette (1989), etc.
4For detailed descriptions of the model equations and variants see Capros, Karadeloglou, Mentzas
(1988, 1989a and 1990) and Karadeloglou (1989).
sSee also Valette and Zagame (1990).
160
large-scale neo-Keynesian models. The purpose of this construction IS to

attempt a standardization of models' properties with respect to oil price
variations and to illustrate the analysis based on large-scale modelling. The use
of the large-scale model in several policy scenarios is presented in Section 10.3.
Section lOA highlights the conclusions as well as the limitations of our study.
10.2 CONCEPTUAL ANALYSIS
The conceptual model reflects the view that macroeconomic equilibrium is
obtained in the short- to medium-term via quantity- rather than priceadjustments, by assuming that the economy experiences excess supply and
imperfect competition in all markets. The properties of all similar neo-Keynesian models may be attributed directly to their basic structure: demand
drives production which in turn determines labour and capital requirements;
the latter are evaluated by means of factor demand relations derived from an
underlying production function (or cost minimization behaviour) that permits
factor substitutions. Labour supply is set by the households and the labour and
capital markets are quantity adjusted by assuming excess supply; within
imperfectly competitive markets, commodity prices are determined from production costs, but also influenced by demand-supply depending factors which
express disequilibrium pressures on markets. Labour demand, together with
wages adjusted for inflation, drives private income that mainly forms demand;
the economy as a whole is not financially constrained in foreign markets in the
sense that permanent deficits in its current account may be experienced.
In order to emphasize the role of energy, a tripartite representation of energy
is included. Energy enters the underlying production function as a production
factor, creates economic activity as a producing sector and contributes to
private consumption and to inflation as a final commodity.
The conceptual model represents an open economy with two composite
goods, energy and non-energy, two sectors, energy and non-energy, as well as
three economic agents; namely a consumer, a producer and the government.
The non-energy good is produced by means of three production factors:
capital, labour and energy. Both goods are domestically produced, imported
and exported. The public sector (consumption and tax rates) is exogenous. The
monetary sector is neglected. However, note that the assumption of exogenously defined interest and exchange rates is in line with the common hypothesis
of a small open economy with fixed exchange rates and free international
capital movements. The equations of the model and the definition of symbols
are given in Tables 10.1 and 10.2.6
The model includes a production possibility frontier for the non-energy
6 Recall that all specifications and type of determination of variables (i.e. endogenous or exogenous)
are valid also for the large-scale model used in the paper.
Table 10.1 The conceptual aggregate model

(1)
(2)
(3)
(4)
(5)
(6)
~=<1>
[t ~ PEJ
dQc
I N ' C' C
(7)
dL N = <1>
dQc
LN'
(8)
dEN =<1>
dQc
EN'
(9)
K N=(l-o N)(K N)_l +IN
[t ~ PEJ
C' C
[t ~ PEJ
C' C
(10)
LN=dLN+ DdLN)-i
(11)
EN=dEN+ DdEN)-i
(12)
QC=.f(KN, L N, EN, t)
(13)
K E=(l-o E)(K E)_1 +IE
(14)
(15)
(16)
(17)
LE=IEQE
(18)
ME=EN+CE-QE+XE
(19)
162
Table 10.1 - contd.

(20)
_QN
UcQc
(21)
(22)
(23)
(24)
w= [(1 +tv)p)]oU~
(25)
c=p(l +bN+r-w)
(26)
QN = C N+ G + IN + IE + X N- M N
(27)
Y=CN+CE+G+IN+IE+XN+XE-MN-ME
=QN-EN+QE
(28)
SG=twWL+tvPCN+tXEPE(CEEN)
+tk[P(QN- EN)+ PEPQE-wL] - pG
sector. Given the short/medium-term orientation of the model, we adopt a

putty-day formulation for the behaviour of firms. Hence the representative
producer decides only for the contribution of factors in the last vintage of
production inputs, while the remaining production factor stock remains rigid;
see Artus (1983), Helliwell et al. (1986), Struck meier (1987) and Hogan (1989).
The cost minimization behaviour of the producer leads to a system of vintage
factor share equations, a simple form of which is presented in equations (6) to
(8). Factor stocks are determined following accumulation mechanisms; see
equations (9) to (11). The ex-ante vintage production function is used for the
determination of production capacity; see (19); while the rate of utilization of
production capacities is determined as a ratio by (20).
By assuming imperfectly competitive markets, we adopt a mark-up rule on
production costs for the determination of non-energy market prices. Production costs depend on factor prices and productivities. Non-energy prices bear
the influence of disequilibrium pressures in the goods market and of foreign
prices; see equation (23). The consumer price of non-energy goods is augmented
by the value added tax.
The supply of labour is determined exogenously. Labour demand is determined from (21) as the sum of sectoral labour demand, which comes from (7)
and (17). A Phillips curve formulation is employed for the determination of
Conceptual analysis
163
Table 10.2 Symbol definitions

ENDOGENOUS VARIABLES
C
private consumption
priv. cons. of non-energy goods
priv. consumption of energy goods
imports of non-energy goods
exports of non-energy goods
KE
QE
XE
PEP
LE
IN
investment in the non-energy

sector
labour demo non-energy sector
energy demo non-energy sector
capital stock non-energy sector
potential output non-energy
sector
market price of non-energy goods
capital cost
Gross Domestic Product
current account current prices
ME
capital stock in the energy sector

production of the energy sector
exports of energy
production price of energy
labour demand by the energy
sector
imports of energy
PE
Uc
L
UN
consumer price of energy

rate of capacity utilization
total labour demand
rate of occupation
wage rate
production non-energy sector
public balance current prices
CN
CE
MN
XN
LN
EN
KN
Qc
c
y
QN
SG
EXOGENOUS VARIABLES
Rf
net income from abroad
Q~
P*
foreign prices of non-energy goods

replacement rates
energy reserves
foreign demand for energy
total labour supply
public investment and
consumption
(j
RE
Q:
Lo
G
IE
PEM
IE
r
foreign demand non-energy

goods
exchange rate
investment in the energy sector
foreign prices of energy goods
technical coefficient
interest rate
TAX RATES
tax on salary income
value added tax
tax on non-salary income

tax on energy consumption
MISCELLANEOUS
<f),f
- i
functional forms
lagged variable
=I.lx/x rate of variation of
variable x
t
I.l
time
first difference
wages; see equation (24). The model thus assumes a flexible nominal wage rate.
The wage rate is however sticky because estimated long-term wage indexation
on prices is not different from one.? The rate of labour force occupation (one
minus the rate of unemployment) is derived as the difference between labour
7Long-terrn elasticity equals 1.1.
164
supply and demand, while excess supply of labour is assumed to prevail (22);
this rate influences wage rates in (24). The cost of capital depends on the price
of non-energy (capital) goods and on the difference between the interest rate
and the anticipated growth rate of wages; see Ando, Modigliani, Rasche and
Turnovsky (1974).
Private consumption is a function of real income (eq. (1)). In the large-scale
model, the effects of the real interest rate and the unemployment rate are also
included. Total consumption is further allocated into energy and non-energy
goods via a typical expenditure allocation system which is driven by the effects
of relative prices (eq. (2) and (3)). Under the assumption of a small open
economy, imports and exports depend on the level of demand, domestic and
foreign respectively, and on the impacts of relative prices as well as on
short-term implications of demand pressures; the latter are proxied by the rate
of capacity utilization; see (4) and (5).
Total demand, that is the sum of private consumption, (exogenous) public
expenditures, demand for capital goods by investments, and net exports,
determines the domestic production of non-energy goods, as in equation (26).
The gross domestic product, the public balance and the current account are
determined as an output from the model but do not have any feedback effect.
The energy sector extracts, and supplies primary energy (e.g. crude oil,
natural gas at the well-head, hydroelectricity, etc.). Thus, in the input/output
representation of our conceptual model we include only net energy production
(that is total energy treated minus energy conversion). Domestic energy
production depends on the capital stock in place (investments in the energy
sector are exogenously determined) and on available energy reserves.
Four tax instruments are included: a value added tax on non-energy goods,
two income taxes and a tax on energy consumption. The latter is applicable to
both households and firms. The government receives revenues from taxes and
allocates them to consumption and investment. Permanent deficits in the
public balance are allowed. The financing of such deficits, however, should
influence the interest rate if we assume neutrality of monetary policy. This
effect is neglected in the model but has been considered in one of the scenarios
with the large-scale model that are presented in the following section.
Below we apply a linearization of the conceptual model by evaluating the
growth rates of the variables. 8 Hence a comparative statics analysis is achieved
with respect to the steady-state solution of the model; see Capros, Karadeloglou and Mentzas (1989) for details on the analytical solutions of the linearized
form of the model.
The direct effect of a 10% fall of the crude oil price (in terms of the model's
nomenclature this refers to the foreign price of energy goods) is a decrease in
8 In fact we evaluate the first difference of the logarithm of both sides of each equation of the model.
Thus, the linearization operator is as follows: ~[log(x)] ',>;; ~x/x. By applying this operator to the
product of two variables one obtains the sum of their growth rates, while by applying it to the
sum of two variables one obtains the sum of their growth rates weighted by their mean shares.
Conceptual analysis
165
the domestic price of energy, which is estimated to be about 7 to 8%. If tax

rates on energy remain constant, this decrease is transmitted to the consumer
price of energy. Production costs are then diminished by a rate of about
-0.8%. By assuming a full indexation of wage rate to prices, the wage rate is
also decreased by 0.8%. The cost of capital will also fall but less rapidly. Thus,
the direct effect of the crude oil price decline is deflationary.
Various kinds of substitution occur between production factors on the
supply side. These generally lead to increased productivity of labour which
implies a further reduction of prices by about -0.1 % in the short-term. Even
if one assumes that there is no change in the foreign trade, this situation leads
to an increase in demand, especially through private consumption and investment. Thus, gains in both the non-energy output and GDP growth rates are
obtained. These lead, however, to increased disequilibrium pressures coming
from the rate of capacity utilization and the unemployment rate that are both
rising. The latter contributes negatively to the deflationary process. In fact
demand pressures in the goods market are proxied by the capacity utilization
rate and have a positive influence on price formation. A high capacity
utilization ratio indicates the existence of backlogs in the goods market and a
short-term reduction of total demand by increasing prices is obtained. However backlogs in the goods market increase investments in the medium-term
and the productive capacity of the economy.
Finally another effect is concerned with the increase in income due to
cheaper imports. In fact an energy price decline implies that the amount of
income transferred to abroad is reduced. The GDP growth which would result
from domestic price decline and factor substitution, will thus be strengthened
by this second effect.
If the foreign prices are not affected by the drop of crude oil prices, or if they
are decreasing less than the fall of domestic prices, the economy experiences
gains in competitiveness. These have strong positive implications for real
output, employment and national income. Through the multiplier accelerator
mechanism of the neo-Keynesian model [see for instance equations (6) and (7)]
the positive growth effects are consolidated.
In the case that the energy tax rate remains fixed, the government bears a
reduction of real revenues from energy taxation of about -7%. Depending on
the structure of the tax revenues, this may lead to a degradation of the public
budget in constant monetary units, in spite of the increased revenues due to
the rising activity and to an improvement of the current account. This occurred
in our particular numerical example of the large-scale econometric model,
calibrated to the Greek figures. One might assume that the reduction of tax
income as a result of the fall in the oil price would be partially counterbalanced
by increasing energy consumption. This is however hardly the case for the
Greek economy because the price elasticities are estimated to be relatively low
compared to other European economies. In countries where energy consumption is more price sensitive, the overall negative effect on public budget could
166
be less important but on the other hand the positive effects on the current
account balance should also be less important.
If the government wishes to maintain the level of real revenues from energy
taxation, the tax rate on energy consumption should be augmented. Through
equation (19), this leads to a less important reduction of consumer energy
prices. This is transmitted to production costs and prices, and thus affects
competitiveness. Output gains and GDP growth rates become lower. Such a
situation, however, complies with energy policy objectives, since it leads to
lower levels of energy consumption.
The magnitude of the above interactions depends on the openness of the
economy and on the relative importance of the energy sector in the economy.
10.3 EMPIRICAL ANALYSIS
10.3.1 Scenario definition
The empirical analysis is carried out with the HGRV, which has about 370
equations and performs annual time-forward simulations. It includes four
production sectors (agriculture, energy, industry and services), four private
consumption categories and a full account of costs, incomes, tax revenues and
balances in the economy. From a methodological point of view, the HGRV
model is quite similar to the conceptual aggregate model presented in Section
10.2.
The HGRV model is used for building policy scenarios and deriving
conclusions by examining the deviations from a baseline projection. For each
scenario the model runs dynamically for 14 time periods (years). In all policy
scenarios we assume that the dollar price of the imported crude oil decreases
by 10% throughout the simulation period. Energy taxation and other accommodating policies are, defined exogenously for each scenario, as follows:
SCENARIO A: There is no change in any taxation or other exogenously
defined policy. Thus, this scenario provides numerical indications about
the impacts of the crude oil price drop (see Table 10.3). Clearly the
objective of this scenario is to analyse the direct and indirect, first round
and induced effects of the energy price decline. All behavioural equations
of the model remain unchanged while the only variable which is
exogenously modified concerns the crude oil price in dollars.
SCENARIO B: The drop in the crude oil price is linked to an augmentation of the tax rate on energy consumption. It is assumed that the
government increases the tax rate in order to compensate the loss of real
revenues (see Table lOA). In this scenario the tax rate on energy products
is exogenously increased by 9.2% in the beginning and 7.6% in the end of
the period while all other exogenous variables of the model remain the
same. The objective of this scenario is to quantify the effect of this type of
policy response on GDP growth and the public deficit.
167
Empirical analysis
Table 10.3 Scenario A: 10% decrease of the crude oil price 2
accomm. policy: none

Private Consumption
Investment (industry)
Labour demand (industry)
Energy demand (industry)
Total energy demand
Total imports
Total exports
Imports of energy
GDP Deflator
Wage rate
Cost of capital
Energy tax rate
Public investments
Interest rate
Budget def./GDP (abs. dill)
Current acc./GDP (abs. dill)
2
1987
1988
1990
1995
2000
0.73
0.51
-0.14
0.42
0.86
0.76
-0.02
1.69
0.95
-3.00
-3.00
6.10
0.00
0.00
0.00
-0.09
0.72
0.92
0.59
0.59
0.47
1.06
0.93
-0.22
1.79
1.06
-3.50
-3.25
1.40
0.00
0.00
0.00
-0.08
0.82
1.13
0.72
0.42
0.66
1.65
1.30
-0.12
2.17
1.39
-3.95
-3.46
-3.30
0.00
0.00
0.00
0.03
1.00
1.93
1.24
1.39
0.83
2.63
2.13
0.11
2.50
2.18
-5.00
-4.11
-4.76
0.00
0.00
0.00
0.13
1.16
2.65
2.06
1.68
0.82
3.22
2.76
0.85
2.35
2.81
-5.50
-4.60
-5.40
0.00
0.00
0.00
0.24
1.04
All results are reported as percentage deviations from the baseline solution.
Table 10.4 Scenario B: 10% decrease of the crude oil price

accomm. policy: tax rates on energy consumption (9.2-7.6%)

Private Consumption
Total energy demand
Total imports
Total exports
Imports of energy
GDP Deflator
Wage rate
Cost of capital
Energy tax rate
Public investments
Interest rate
Budget def./GDP (abs. diff.)
Current acc./GDP (abs. diff.)
1987
1988
1990
1995
2000
0.63
0.43
-0.25
0.36
0.74
0.65
0.03
1.56
0.81
-2.61
-2.80
5.49
9.20
0.00
0.00
0.04
0.70
0.77
0.49
0.41
0.40
0.91
0.79
-0.15
1.64
0.91
-3.04
-2.94
1.29
9.05
0.00
0.00
0.04
0.80
0.94
0.58
0.26
0.55
1.41
1.11
-0.01
1.98
1.19
-3.49
-3.14
-3.00
8.80
0.00
0.00
0.14
0.97
1.64
1.00
1.17
0.69
2.29
1.84
0.15
2.29
1.88
-4.46
-3.74
-4.38
8.20
0.00
0.00
0.23
1.15
2.25
1.68
1.43
0.65
2.82
2.39
0.77
2.18
2.43
-4.96
-4.18
-4.96
7.60
0.00
0.00
0.32
1.09
SCENARIO C: This scenario combines the drop in oil price with an

increased tax rate on energy, as above. The increased government revenues
serve to finance the additional public investment, at a level such that the
public deficit is not significantly lower than in scenario A (see Table 10.5).
168
Table 10.5 Scenario C: 10% decrease of the crude oil price

accomm. policy: tax rates on energy consumption (9.2-7.6%) and increased
public investment (3.8-2.5%)

Private Consumption
Total energy demand
Total imports
Total exports
Imports of energy
GDP Deflator
Wage rate
Cost of capital
Energy tax rate
Public investments
Interest rate
1987
1988
1990
1995
2000
0.81
0.50
-0.19
0.43
0.83
0.78
0.18
1.51
0.95
-2.72
-2.87
5.80
9.20
3.80
0.00
-0.10
0.66
0.98
0.58
0.63
0.47
0.99
0.93
-0.01
1.65
1.04
-3.18
-3.03
1.34
9.05
3.60
0.00
-0.09
0.76
1.14
0.69
0.40
0.64
1.52
1.25
0.15
1.99
1.33
-3.61
3.22
-3.09
8.80
3.40
0.00
0.01
0.92
1.90
1.18
1.31
0.81
2.44
2.02
0.32
2.29
2.06
-4.61
3.83
-4.48
8.20
2.90
0.00
0.11
1.09
2.57
1.94
1.59
0.81
3.01
2.61
0.97
2.17
2.65
-5.15
-4.31
-5.07
7.60
2.50
0.00
0.20
1.00
The aggregate public investment expenditure is exogenously determined in

the model used. Disaggregation of the total amount is undertaken in the
model using fixed technical coefficients. Total investments are disaggregated to industrial and service goods while the agricultural and energy
sectors are not affected.
SCENARIO D: Here, the drop in oil price is again combined with an
increased tax rate on energy, but the additional funds are used to finance
the public deficit. It is assumed that, through an appropriate credit policy,
the credit supply constraints are relaxed in the capital market. The latter
leads to a downward shift of the interest rate (see Table 10.6). The purpose
of this scenario is to explore further the positive effects of the oil price
decline assuming that the financing of the public sectors borrowing
requirements (PSBR) by increased taxation will decrease interest rate by
2%.
In designing the scenarios no supplementary deficit in the public budget or in
the current account has been provoked, when results are compared to those
obtained by scenario A. All scenarios assume that the exogenous variables
related to foreign markets, other than the oil market, remain unchanged.
10.3.2 Analysis of results
The macroeconomic implications of the crude oil price decline are quite positive,
in all respects. A 10% fall of oil prices leads to additional real output of 0.73%
Empirical analysis
169
Table 10.6 Scenario D: 10% decrease of the crude oil price

accomm. policy: tax rates on energy consumption (9.2-7.6%) and decreased
interest rate by 2%

Private Consumption
Total energy demand
Total imports
Total exports
Imports of energy
GDP Deflator
Wage rate
Cost of capital
Energy tax rate
Public investments
Interest rate
1987
1988
1990
1995
2000
0.65
0.43
0.40
0.29
0.82
0.69
0.09
1.54
0.91
-2.65
-2.76
3.68
9.20
0.00
-2.00
0.05
0.69
0.80
0.50
0.91
0.29
0.95
0.83
-0.09
1.69
0.95
-2.71
-2.94
-1.11
9.05
0.00
-2.00
0.05
0.80
0.96
0.58
0.56
0.47
1.45
1.14
-0.01
1.99
1.21
-3.51
-3.16
-4.69
8.80
0.00
-2.00
0.14
0.97
1.67
0.98
1.32
0.48
2.28
1.84
0.09
2.31
1.88
-4.52
-3.76
-5.53
8.20
0.00
-2.00
0.23
1.17
2.28
1.64
1.53
0.36
2.81
2.38
0.66
2.25
2.45
-4.96
-4.23
-5.51
7.60
0.00
-2.00
0.31
1.11
in the short term while medium- and long-term increase is respectively 1.13 and
2.65%. Both output and lower prices raise energy demand by 0.8%, 1.3% and
2.7% in the short-, medium- and long-term. This leads to an equal rise of
energy imports. The deflationary spiral is important, leading to a decrease in
prices ranging from -3% to -5.5%. Both the energy tax revenues and the
public deficit are degraded in real terms. However, the current account is
improved. If the effects of oil price decrease on competitive countries are taken
into account the overall results on the national economy are rather limited.
This is confirmed by the results of a scenario (not presented here) in which
foreign prices are decreasing equally to the domestic ones. In this case, positive
growth effects are experienced during only four years after the oil price shock
and then vanish. This is mainly due to the fact that competitiveness gains,
which are a very important factor, disappear.
Scenario A illustrates numerically that although the economy profits from
growth and deflation, public revenues may diminish and energy dependence
may be aggravated. As can be seen in Table 10.3, a worsening of the PSBR by
-0.09 points is observed in the beginning of the simulation period. This
negative effect is however reduced in the medium-term because of increased
revenues coming from the boosting of the economy. In the medium-term this
improvement reaches 0.03 points of GDP.
For these reasons increase in the tax rate on energy consumption is worth
discussing. This is examined in scenario B. Results show that positive growth
effects are not reduced to zero in this case. The economy achieves 2.26%
170
additional output with positive effects on the public deficit, and on the current
account as percentages of GDP. This is accomplished mainly at the expense of
deflation which is less important than previously (-4.89% in the long-term).
Households seem to bear most of the indirect impacts of the increased taxation,
since they gain significantly less in real income and consumption, compared to
scenario A. However no significant decrease in energy demand is obtained with
respect to scenario A. This is totally due to low price elasticity for Greek data.
Thus scenario B can only be justified for public budget policy reasons. 9
Following scenario B, overall government gains are estimated to reach 0.04 as
percentage of GDP.
By comparing the results of scenarios A and B, one can evaluate what is the
cost of reducing the budget deficit through increased expenditures. Clearly the
reduction of the public deficit (result of scenario B) would reduce GDP growth,
employment and increase the general level of prices. The reduction of the
public budget deficit would clearly have depressive effects on the economic
activity (potential positive financial effects are not examined at this stage).
The question which then arises is the following: should the public sector use
the increased funds from additional energy taxation to reduce its borrowing
requirements or should these funds be used for development purposes? Thus,
it may be interesting to investigate if, by using these increased funds, the
economy would achieve better growth results than in scenario A, while not
degrading the public deficit, compared to this scenario.
In fact the increased funds can be used by the public sector in different ways.
Government may increase public investments or consumption. Furthermore
subsidies and/or grants can help private firms to finance their investments.
Moreover increased resources can allow the reduction of income and/or
corporate tax rates. On the other hand the reduction of the PSBR can have
positive impacts on the interest rate. All the above policy options may have
positive effects on overall economic activity. The scenarios that follow examine
this issue. Scenario C uses the additional funds for financing public investment,
while in scenario D funds are used for relaxing credit supply constraints in the
capital market which leads to diminishing interest rates.
The additional public investments assumed within scenario C have positive
multiplier effects in the short-run. The gains in output, employment and real
income, compared to scenario A, however vanish in the medium-term. The
main reason is that increased public investments create demand-pull inflation.
The direct impact of a lesser decline of the general price level is a decrease of
competitiveness gains obtained in scenario A. Following this result, exports
perform less well than in scenario A. Moreover imports increase and GDP
9We also examined the case of diminishing the tax rate on energy consumption. This is justifiable
the case of decreasing foreign prices of non-energy goods due to the possible gains in
competitiveness. In this case the economy experiences improved growth rates and employment, but
the public deficit is seriously aggravated.
III
Conclusions
171
growth is less than in the non accommodating policy scenario. Even if increased
public investments boost economic activity in the short-term, the results are
different in the medium- and long-term. The beneficial effects cannot be
maintained because of inflationary pressures. However the results obtained in
this scenario, i.e. the case in which additional energy taxes are spent by the
public sector for investments, are more beneficial when compared to scenario
B where additional expenses are used to finance the public deficit. 10
It was shown above (scenario B) that if the energy price decline is followed
by an energy tax increase, the public deficit is significantly improved. However
a different option in macroeconomic policy would be to maintain the public
investments. The macroeconomic results of this scenario come from the
demand side. In this scenario the supply-side effects transmitted through
financial channels are examined. The reduction in public sector borrowing
requirements will decrease credit demand in the economy. Assuming that the
interest rate is determined by market forces according to demand and supply,
a fall in credit demand will equilibrate the market at a lower level. Interest rates
may thus be reduced and have supply-side macroeconomic impacts on economic activity. We implicitly assume that the decrease in the interest rate is 2%,
which corresponds to a decrease of 0.36 and 0.1 points in the short- and
long-term respectively. The supply-side effects consist of factor substitution
effects: both energy and labour are substituted by capital, so investments
increase even in the first year. This complies with energy objectives but leads
to significantly less gains in real income for households. The results are globally
positive but the efficiency of the scenario is lower when compared to the case
where public investments are increased but higher than in the case where no
accommodating policy is undertaken. 11
lOA CONCLUSIONS
Referring to the policy questions put forward in the introduction, we conclude

as follows:
1. If the economy experiences excess supply when an oil price decline occurs,
the implications will be significantly positive for real output and employment. Depending on the degree of openness of the economy, these results
will be maintained if the economy is able to transmit the competitiveness
gains on the domestic to foreign relative prices. This depends to a large
extent on wave indexation. If wage indexation is less than unity and
assuming a symmetric behaviour, real wage rate increases will result. If
however wage indexation is larger than unity, more competitiveness gains
'OThis is due to the absence of an explicit financial sector in the model variant used.
result is related to the structure of the neo-Keynesian model and may be different if
price-induced equilibrium structure is adopted.
11 This
172
should be expected. Following the above linkage and if additional growth

is expected from foreign trade, the tax rate on energy consumption must
remain unchanged or even decreased. This would lead to a further price and
wage decline and extra competitiveness gains. However, such a decision has
significant consequences on the public deficit. If on the other hand, domestic
income effects constitute a constraint to the growth process, income policy
(structural change on wage indexation) and income distribution should be
considered together with tax policy.
2. The macroeconomic cost of augmenting the energy tax rate does not seem
important. The growth of real output is maintained, although households
bear the main part of the indirect macroeconomic cost. Thus positive effects
on growth are maintained and improvement of the public deficit is
achieved.
3. The evolutions in the energy sector following an oil price decline clearly do
not comply with the long-term objectives of energy policy as usually
formulated in oil-importing countries. Increased taxation at the level
examined here does not seem able to restrain the growth of energy demand.
It should however be mentioned again that this result is mainly due to the
fact that the estimated energy price elasticities for Greece are relatively low
compared to other countries. On the contrary, income elasticities are the
most important explanatory variables for energy demand. If price elasticities were higher, energy taxation should be more efficient.
4. Policies that impose additional energy taxes and spend the increased funds
in public investment or in financing the public debt do not seem to perform
better than the 'doing nothing' case (i.e. scenario A) at a macroeconomic
level. However, a clear trade-off emerges between the demand and supply
sectors as receivers of the increased funds. Nevertheless, due to its underlying methodology, our model is not able to capture the full range of effects
of policies using the additional funds for financing past public debts.
The policy maker and the reader should be aware of two potential limitations
in the above analysis. First because the study focuses on the medium-term
perspective, any general equilibrium considerations are excluded. This is connected to the importance of neo-Keynesian macroeconomic modelling tools in
the EEC, which we preferred to use in the analysis; the need for an in-depth
inquiry into the properties of such tools; and the necessity to standardize their
behaviour. It is clear that the analysis should be complemented by a general
equilibrium model.
The second potential limitation is related to the sectoral disaggregation and
micro economic aspects. The macroeconomic models currently used (ours
included) exhibit a symmetry in their behaviour concerning variations of the
energy price. This symmetry has been challenged theoretically and should be
examined with scrutiny.
Beside the above limitations, it is important to quantify the margin within
which the impact of a certain policy is included. The modelling approach can
References
173
give all necessary indications for the quantitative evaluation of different

policies and can give a numerical evaluation of the different options. The
analysis undertaken above can evaluate the effectiveness of different energy
related instruments or policies; for this purpose macroeconomic effectiveness,
static and dynamic effects as well as the fairness and the acceptability of a
policy can be measured.
In summary, the analysis of the impacts of a decline in the crude oil price
seems to be a very stimulating policy question. The points of interest could be
more specifically concentrated on two specific subjects: energy taxation and the
exploitation of positive opportunities arising out of such an event. Doing
nothing and following up the evolutions of the energy price seems quite
efficient for employment, output and current account related objectives. However the crucial problem is the choice of the most efficient accommodating
policy. It was thus concluded that if problems with public deficits dominate, an
increase in the tax rate is needed. If on the other hand additional growth and
reduction of unemployment prevail as the targets for the policy maker, the
spending of additional resources to finance public investments is recommended.
REFERENCES
d'Alcantara, G. and Italianer, A. (1982) European project for a Multinational Macrosectoral Model, CEC, MSll, XII/759/82.
Ando, A., Modigliani, F., Rasche, R. and Turnovsky, S. (1974) On the Role of
Expectations of Price and Technological Change in an Investment Function, International Economic Review, 15(2).
Artus, P. (1983) Capital, Energy and Labour Substitution: The supply block in OECD
Medium-term Models, OECD/ESD Working Paper No 2.
Bergman, L. (1988) Energy Policy Modeling: A Survey of General Equilibrium Approaches, Journal of Policy Modeling, 10(3), 377-99.
Borges, A. M. and Goulder, L. H. (1984) Decomposing the Impact of Higher Energy
Prices on Long-term Growth, in Applied General Equilibrium Analysis, (eds.) H. Scarf
and Shoven, Cambridge University Press.
Brillet, L. 1. et al. (1982) Energie et economie: Ie modele mini-DMS energie, Economie
et Statistique, No 146.
Bruno, M. and Sachs, J. (1982) Input Price Shocks and the Slow-down in Economic
Growth: The Case of UK Manufacturing, Review of Economic Studies, XLIX, (5).
Bruno, M. and Sachs, 1. (1985) Economics of World-wide Stagflation, Basil Blackwell,
Oxford, England.
Capros, P. et al. (1984) Le modele macroeconomique MIEL, Revue Generale Nucleaire,
No 2, Mars-Avril, p. 166-82.
Capros, P., Karadeloglou, P. and Mentzas, G. (1988) Structure and Properties of the
GEM-NTUA Annual Econometric Model for Greece, presented at the First Cotiference of the Greek Economic Association, 14-16 December, Athens.
Capros, P., Karadeloglou, P. and Mentzas, G. (1989a) Neo-Keynesian and Applied
General Equilibrium Models: New Developments for Greece, paper presented at the
International Federation of Automatic control, IFAC Symposium Modelling and
Control of National Economies, 27-29 June, Edinburgh, Scotland.
174
Capros, P., Karadeloglou, P. and Mentzas, G. (1989b) Energy Policy Extensions of

KLE-based Models, Journal of Policy Modeling, 11(4), 507-30.
Capros, P., Karadeloglou, P., Mentzas, G. and Valette, P. (1989) A New Modeling
Framework for Medium-Term Energy-Economy Analysis in Europe, The Energy
Journal, 10(4).
Dantzig, G. et al. (1978) Stmiford Pilot Energy/Economy Model, EPRI EA-626, May.
Davis, S. J. (1987) Allocative Disturbances and Specific Capital in Real Business Cycle
Theories, American Economic Review, Papers and Proceedings, 77, 326-32.
Eckstein, O. (1978) The Great Recession, with a Postscript on Stagflation, Amsterdam,
North Holland.
Gilbert, R. 1. and Mork, K. A. (1984) Will Oil Markets Tighten Again? A Survey of
Policies to Manage Possible Oil Supply Disruptions, Journal of Policy Modeling,
6(1),111-42.
Gilbert, R. J. and Mork, K. A. (1986) Efficient Pricing During Oil Supply Disruptions,
The Energy Journal, 7(2) 51-68.
Gordon, R. 1. (1975) Alternative Responses of Policy to External Supply Shocks,
Brookings Papers on Economic Activity, 1, 183-206.
Hamilton, 1. D. (1983) Oil and the Macroeconomy since World War II, Journal of
Political Economy, 91, 228-48.
Helliwell, J. F., Sturm, P., Jarrett, P. and Salou, G. (1986) The Supply Side in the
OECD's Macroeconomic Model, OEeD Economic Studies, No 6, Spring, pp. 76-129.
Helliwell, 1. F., MacGregor, M., MacRae, R., et al. (1987) Supply Oriented Macroeconomics: the MACE model of Canada, Economic Modelling, 3(4), 318-40.
Hickman, B. G., Huntington, H. G. and Sweeney, J. L. (1987) Macroeconomic Impacts
of Energy Shocks, North Holland, Amsterdam.
Hogan, W. (1989) A Dynamic Putty-Semi-Putty Model of Aggregate Energy Demand,
Energy Economics, January, pp. 53-69.
Hudson, E. and Jorgenson, D. (1974) U.S. Energy Policy and Economic Growth,
1975-2000, The Bell Journal of Economics, 5(2).
Longva, S. and Olsen, O. (1983) Price Sensitivity of Energy Demand in Norwegian
Industries, Scandinavian Journal of Economics, 85, 17-36.
Loungani, P. (1986) Oil Price Shocks and the Dispersion Hypothesis, Review of
Economics of Statistics, 68, 536-9.
Manne, A. S. (1978) Energy-Economy Interactions: An Overview of the ETA-Macro
Model, in Energy Modeling and New Energy Analysis, Symposium, Chicago, pp.
341-51.
Mork, K. A. (ed.), (1981) Energy Prices, Inflation and Economic Activity, Ballinger Pub!.
Co., pp. 43-63, Cambridge, Mass.
Mork, K. A. (1985) Taxation as a Protection Against the Effects of Price Fluctuations:
the Case of Oil, The Energy Journal, 6, Special Tax Issue, pp. 73-87.
Mork, K. A. (1989) Oil and the Macroeconomy when Prices Go Up and Down: An
Extension of Hamilton's Results, Journal of Political Economy, 97, 740-4.
Paterson, K., Harnett, I., Robinson, G. and Ryding, J. (1987) The Bank of England
quarterly model of the UK economy, Economic Modelling, 4(4), 398-529.
Pindyck, R. S. (1980) Energy Price Increases and Macroeconomic Policy, The Energy
Journal, 1(4), 1-20.
Struckmeier, C. S. (1987) The Putty-Clay Perspective on the Capital-Energy Complementarity Debate, Review of Economics and Statistics, 69(2), May, pp. 320-6.
Wharton EFA Inc (1979) The Wharton Annual Energy Model: Development and
Simulation Results, EPRI, EA-1115.
Valette, P. and Zagame, P., (eds.) (1989) HERMES Model, Commission of the European
Communities, DG/XII, North Holland.
Appendix
175
APPENDIX: ENERGY-RELATED MODELS OF THE

EUROPEAN COMMUNITIES
INTRODUCTION
In this note the energy-related models constructed by the DG/XII of the
Commission of the European Communities are described. Moreover, examples
of concrete applications of these models are presented together with a table
presenting the models' identity in a condensed way
The models were constructed with the financial support of CEC-DGXII but
their use has been extended to other DGs of the European Communities. The
HERMES model has been used for the evaluation of macroeconomic effects of
the completion of the European Internal Market in the Cecchini Report;
HERMES, EFOM and MEDEE have been used in the European Communities Task Force report on 1992 and the Environment. Actually MIDAS is the
official model of DG/XVII for forecasting and analysing energy policy.
In the first section of this note a description of different models is presented.
The description includes the presentation of the models' main features, together
with references and a brief analysis of their functioning properties. In the
second section a resume of two examples of real-world use of these models is
presented.
DESCRIPTION OF MODELS
HERMES! is a European-wide large-scale neo-Keynesian multi sectoral
macroeconometric model. Its general features have been described in the text
and are recapitulated in Table lOA.l.
In the manufacturing sector a 'putty-clay' production function of the CES~
Cobb-Douglas type is adopted, with three production factors: capital, labour
and energy (KLE). In the services sector a 'clay-clay' approach is used, with
two production factors, capital and labour (KL). A Phillips curve (unemployment rate and inflation rate) is employed for the estimation of the wage rate.
Prices are evaluated as a mark-up of labour cost, intermediary demand cost
and capital cost, and affected by the evolution of the utilization rate.
The model is annual and dynamic, while the forecasting period is up to ten
years. All economic sectors are described in detail and all macroeconomic
1 HERMES stands for Harmonized European Research for a Multinational Economic and Energy
System. It was initiated by d'A1cantara and Italianer (1982). The model was constructed by several
national European teams under the coordination of DG/XII of the Commission of the European
Communities. For a detailed presentation of the model see d'A1cantara and Italianer (1982),
Italianer (1986) and Valette and Zagame (1991).
Size (number of equations)

Software
Exogenous
Data base
Endogenous
Sectors of economic
activity
800
TROLL
Specific data
Energy demand
Macroeconomic
Income, activity,
variables, international
energy shares,
energy prices, energy
Iifemodes, technical
reserves, capacity
parameters.
expansion plans
n.a.
1000
TROLL
Specific
Eurostat
Energy demand
Energy supply
Energy prices
10 000 variables
MPSX
Energy demand
technological
coefficients
Specific data
Energy production
and investment
Central Electr.
Self Electr.
Coal
Gas
Refineries
Industries
5 years
Long term
Optimization, scenario
analysis
Energy subsystems
Yearly
Long term
Projections, scenario
analysis
Energy consumption
end-uses
Detailed disaggregation
of end-uses of energy
Yearly
Short-medium term
Projections, scenario,
analysis
Households, firms,
public sector
Agriculture
Energy
Cons. goods
Interm. goods
Equip. goods
Trans. Telecom.
Build. Constr.
Services
Non-market
National accounts
Demand
Supply
Prices
For. trade
Employment
International
environment
demographic variables,
public sector behaviour
Periodicity
Time
Use
Economic agents
Linear programming
Technical
Econometric and
technical relations
Yearly
Short-medium term
Projections, scenario
analysis
Energy consumers
and producers
6 Industries
3 Residential uses
4 Transport means
6 Supply sectors
24 Energy products
Econometric
Character
EFOM
MEDEE
MIDAS
Main features of the models
HERMES
Table IOA.I
Appendix
177
mechanisms are represented in the model. The latter and their interactions are
activated in a simultaneous way during model resolution and simulation.
MIDAS is a medium-term annual energy model, combining econometric
with process analysis formulations. It simulates the evolution of energy
demand, energy supply and prices. Energy is measured in physical units on the
basis of the energy balance system.
Energy demand is dis aggregated by sector, i.e. industry, households, transport etc. Demand is evaluated per sector, use and energy product and is a
function of real disposable income and the relative prices of energy products.
Energy demand as a production factor is influenced by relative prices (i.e. wage
rate and cost of capital) and the volume of production.
On the supply side the electricity production module formulates a linearized
load duration curve which is constructed by using the energy demand results.
Electricity plants are allocated within the load curve following cost minimization while the marginal cost is approximated for each consumption category,
and is introduced in the energy prices module.
The refining sector module uses an aggregate representation of a typical
average refinery, including distillation, cracking and reforming. Desired production of each refining unit is compared to existing capacities in order to
obtain estimations of the rate of capacity utilization and of production cost.
Refinery throughput and production flows are estimated from supply-oriented
econometric formulations.
The production of coal is formulated by means of supply curves (involving
reserves), which also serve in the evaluation of mining profitability and coal
pnces.
Finally MIDAS evaluates energy prices by using the dollar price of crude oil
as the main exogenous variable. Most of the equations follow a rate of
variation formulation, connecting an energy product price with the price of a
product which is considered as a leading price and national and international
labour and capital factors. Concerning electricity pricing, the model adopts a
special algorithm which corresponds to marginal cost tariffication for each
consumer type. Taxes are then added to these prices, in order to obtain consumer
prices. 2
The MIDAS energy model project was initiated by N. Kouvaritakis (see
ECOSIM, 1986) and supported by the Commission of the European Communities. At that early stage, development was concentrated on the energy
demand module. Recently, Detemmerman et al. (1988) re-formulated MIDAS
energy demand module and re-estimated econometric equations for four
European countries. Finally energy supply and pricing modules and the
integration of the demand and supply MIDAS modules into a single model has
been effected by Capros, et al. (1990b). The complete MIDAS model is
2For more information see Capros et al. (J990b).
178
available for the United Kingdom, Italy, France, Germany, Greece, Holland
and Belgium.
MEDEE is a disaggregated technoeconomic energy demand mode1. 3 The
main energy demand variables in MEDEE concern sectoral demand by industry,
residential and tertiary, passengers and goods transport. These sectors are
further disaggregated in many sub sectors that are homogeneous with respect
to energy consumption patterns. The variables determining sectoral energy
demand in MEDEE are linked with sectoral activity, GDP, disposable income
while relative and/or absolute price effects do not have any impact. The
methodology used in MEDEE in general follows three steps: evaluation of the
end-use energy requirements for each homogeneous consumption subsector;
determination of the energy technology and energy form mix for the satisfaction of energy needs with predetermined penetration rates of energy forms; and
evaluation of the final energy demand by the use of the efficiency coefficients
of energy appliances.
EFOM is an optimization model based on the energy flow representation of
the energy supply system of a country.4 The model provides a convenient
graphical portrayal of the flows of energy through the economy, with an
explicit accounting of energy losses at each stage of conversion, distribution
and end-use. The model aims at defining the structures of energy activities that
MEDEE
MIDAS
EFOM
~
I FRET
~'--D-E-R-E~
HERMES
Fig. lOA.!
Information flow between models.
3 See Chateau and Lapillone (1978).

4For details on EFOM see Finon (1979) and Van der Voort (1982).
Appendix
179
permit the energy needs of a country to be satisfied at minimum cost. The

environmental module of energy supply on air pollution emission and more
specifically on S02 and NO x, EFOM can also produce the effects of emission
reduction on the energy system and propose solutions based on cost efficiency.
APPLICATION EXAMPLES
Introduction of a carbon tax in the UK for CO 2 reduction
The carbon tax has been put forward as a policy instrument to induce
substitutions and fossil energy savings aiming at reducing CO 2 overall
emissions. The carbon tax consists in applying taxes to fossil energy products
at a rate proportional to their CO 2 unit emission.
The work with MIDAS in the UK for carbon tax policy analysis comprised
the following tasks:
construction and incorporation into MIDAS of a 'C0 2 impact evaluation'
module;
definition of the carbon tax and adaptation of the corresponding consumer
energy price equations of MIDAS.
The use of the HERMES/MIDAS/Linked System introduces important mechanisms in the evaluation of both the potential of CO 2 emission reduction and
the impacts induced by the carbon tax. The major additional mechanisms refer
to substitutions between energy and non-energy goods, services and production factors. If this mechanism permits further energy savings, the CO 2
emission reduction potential may increase.
The MIDAS energy model is particularly suitable for energy system insights
associated with the carbon tax analysis. This is due to the very nature of the
model, as it fully represents, in the heart of its formulation, price-induced
interactions in energy demand, energy supply and energy demand-supply
adjustment. Energy prices are not shadow prices as in other models, while its
formulation and use as simulation tool allows for a straightforward treatment
of carbon tax-related issues. MIDAS' econometric behavioural equations
permit the full endogenization of carbon tax-induced substitutions, while its
fixed process analysis and econometric formulations ensure engineering orientated reliability of results. Moreover, MIDAS is a global energy system model
in the sense that it covers all energy subsystems, interactions and mechanisms
within the energy sector. MIDAS provides estimates of country energy balances, costs, prices and capacities through dynamic annual simulations for
10-15 years.
The MIDAS-HERMES fully integrated linkage is a particularly suitable
tool for the assessment of carbon tax implications through the energyeconomy interactions. The impacts of energy prices on the economy, factor
180
substitutions, induced non-energy substitutions etc. are fully covered by the

MIDAS-HERMES linkage. In particular, in estimating the carbon tax CO 2
reduction potential, this tool ensures a global evaluation, in the sense of
accounting for all types of substitutions, interindustrial (I/O) exchanges and
supply contractionary effects. Because of its global character, this approach is
clearly more reliable than partial evaluations, for instance sectoral KLEMbased estimates.
The HERMES/MIDAS/Linked System has been used to evaluate the
macroeconomic implications and served for the re-estimation of the expected
CO 2 reduction potential from carbon tax. 5 The main conclusions drawn are
summarized in the following:
1. The stabilization of CO 2 emission is very difficult to attain by means of
2.
3.
4.
5.
6.
7.
8.
a policy based solely on carbon-tax.

If only a carbon tax is used, stabilization of CO 2 emissions may be
obtained with carbon tax rates implying a mean energy price as much as
two to three times higher.
The reduction of CO 2 emissions is obtained mainly through the decrease
of total energy demand and secondarily from inter-fuel substitutions.
The elasticity of CO 2 emission reduction with respect to a unit increase
in the mean energy price (induced by the carbon tax) is very small,
ranging from -0.2 to -0.1; moreover, the elasticity is diminishing when
the carbon tax rate increases.
The role of structural adjustment of power generation, consisting in accelerating the combined cycle and natural gas programmes, is important in
CO 2 emission reduction. If such a structural adjustment is combined
with carbon tax policy, CO 2 emission reductions are substantial.
Within the energy system, the carbon tax induces substitutions that are
favourable to electricity and to a lesser degree natural gas; the consumption of liquid fuels is substantially reduced; if no structural adjustment
occurs in power generation, the share of solid fuels is maintained, with
adverse effects to CO 2 emissions.
The additive carbon tax seems significantly more efficient than the
multiplicative carbon tax.
The macroeconomic implications of the carbon tax are generally negative; if the carbon tax is applied at an international level, results may
differ substantially.
The effects of the introduction of a CO 2 tax in the UK have been extended

to three other European tax countries, namely France, Germany and Italy.
Moreover, technological aspects are introduced and comparison of energy
versus CO 2 tax has been effected. The conclusions and numerical results are,
however, provisional and are not presented here. The whole work has been
5See Capros et al. (1990a).
Appendix
181
effected by the National Technical University of Athens and is a part of the

'Models for Energy and Environment' section of the Joule Programme financed by the Commission of the European Communities - DG/XII.
Environmental policy in the European internal market

A second application is concerned with the evaluation of the environmental
impacts of the completion of the European Internal Market. Environmental
costs and benefits were evaluated using three European-wide models: the
macro econometric model HERMES, the energy demand model MEDEE and
the energy supply and environmental model EFOM-Environment (hereafter
EFOM-E).
The analysis follows a global equilibrium approach and excludes partial
equilibrium estimates. First the macroeconomic impacts of the completion of
the Internal Market and the implications on energy demand and supply as well
as on production levels and consequently on the environmental system are
studied. Second, a variety of policy measures necessary to finance emission
abatement technologies are examined. The study focuses on the macro level of
analysis and uses the European-wide macro econometric model HERMES.
Such an approach is more accurate than a mere accounting of costs and
benefits since it takes into account the interdependences in national economies
which might reveal unexpected 'costs' and 'benefits'. The steps followed and the
models used are presented below:
1. The first step deals with the evaluation of the macroeconomic impacts of
the completion of the Internal Market for Greece with the use of the
HERMES model.
2. The second step analyses the environmental impacts of the completion of
the Internal Market. The macroeconomic estimations of the first step are
used to evaluate the impact of the Internal Market on energy demand with
the MEDEE model. This information is then used to evaluate the impact
on energy supply and the environment by means of the EFOM-E model.
The quantitative results concern the environmental impacts on S02 and
NO x , and the cost of policies aiming at the reduction of S02 and NO x
emission. Moreover, different cost efficiency scenarios for the reduction of
the level of the emission are studied and the corresponding investment and
necessary costs for the protection of the environment are evaluated.
3. The third step refers to the conceptual analysis of the alternative financing
modes for supporting the investment required for anti-pollution policies as
well as to the quantitative assessment of the macroeconomic implications
of such investment. The modelling device utilized in this step is the
HERMES model.
The analysis has been carried out in the context of the CEC Task Force on
the completion of the European Internal Market and the Environment and
182
concerned five European countries: Belgium, France, the former Federal

Republic of Germany, Greece and the United Kingdom.
The analysis for energy demand and air pollution has shown that if the past
trends of the energy and environmental system are maintained in the future,
energy demand, supply and the emission of S02 and NO x will increase with
the completion of the Internal Market. The evaluation of the macroeconomic
impact of the increase in emission-reducing investment and the possibilities
given by the completion of the Internal Market (more growth, gain in benefits)
to finance such investment to prevent or abate pollution, were studied by
assuming different policy options. These concern the effects of an increase in
pollution abatement investment with different forms of financing.
The analysis shows that the financing of environmental protection investments will have minor impacts on the main economic variables. The increase
of emission-reducing investment by 1% of GDP, financed by price increase,
subsidies and a reduction of productive investment, will have a neutral impact
on GDP in the medium term. Environmental concern can give the incentive
for the creation of an entire sector dealing with environmental protection. The
above results will have positive impacts on overall economic activity and
employment.
The conclusion stresses the fact that the surpluses, which will be created by
the completion of the Internal Market and spent for the protection of the
environment, will constitute an 'investment' with positive long-term effects
both for the economy and the quality of life, which will overcome short-term
spending. 6
REFERENCES
d'A1cantara, G. and Italianer, A. (1982) 'European Project for a Multinational Macrosectoral Model'. Commission of the European Communities, MSll, XII/759/82.
Capros P., Karadeloglou P. and Mentzas G. (1990a) Carbon Tax Policy and its Impacts
on CO 2 Emission. Working Paper, National Technical University of Athens, April.
Capros P., Karade1oglou P. and Mentzas G. (1990b). 'New Developments for the
MIDAS Medium-term Energy Modelling Project of the EEC: the Energy Supply
Model and the Supply-Demand-Pricing Linkage'. Paper presented at the 12th
Triennial Congress on Operations Research IFORS 90, Athens 25-29 June 1990.
Chateau, B. and Lapillone, B. (1978) Long-term Energy Demand Forecasting: A New
Approach. Energy Policy, pp. 140-57.
Commission of the European Communities (1990) 1992: The Environmental Dimension.
Economica Verlag, Bonn.
Detemmerman, M. V., Guillaume, Y. and Ledoux, M. (1988) 'MIDAS Demand'. Report
to the Commission of the European Communities, DGjXVII.
ECOSIM Sarl (1986) 'The MIDAS Energy Model', Report to the Commission of the
European Communities, DGjXVII.
6For a detailed presentation see Commission of the European Communities (1990).
Appendix
183
Finon, D. (1979) Scope and Limitations of Formalized Optimization of a National

Energy System: The EFOM model. Energy Policy, Special Issue (ed. A. Strub), pp.
58-71.
Italianer, A. (1986) The HERMES Model: Complete Specification and First Estimation
Results'. Commission of the European Communities, EUR 10669 EN.
Valette, P. and Zagame, P. (eds) (1991) The HERMES Model. Commission of the
European Communities, DG/Xn, DG-Xn (forthcoming).
Van der Voort, E. (1982) The EFOM 12C Energy Supply Model within the EC
ModelliI)g System. Omega, 10, 507-24.
---11--A comparison of energy-economy

models: the French experience
Ghislaine Destais
11.1 INTRODUCTION
From the mid-1970s on, there has been a clearly felt need to understand more
precisely the links between energy trends and the economy as a whole.
Particularly in France, with its strong tradition of central economic planning,
this led to a strong interest in the modelling of energy-economy interactions.
This chapter compares and analyses several energy-economy interaction
models. The first section situates energy-economy models in relation to partial
equilibrium approaches, and suggests a typology of these models. The chapter
then focuses on the French experience in the field of energy-economy
modelling.
11.2 MODELLING ENERGY-ECONOMY INTERACTIONS:
SOME ELEMENTS OF ANALYSIS
11.2.1 The limitations of partial equilibrium approaches
Models have always been used in the energy field. On the supply side, the most
frequently used technique is optimization, essentially for investment programming. Initially used at the level of the firm, this approach was then applied to
the energy sector as a whole, as in the EFOM model (Fin on, 1976). Supply
models, however, only result in partial equilibria because energy requirements
are taken to be exogenous and the interdependence between the energy sector
and the rest of economy is neglected.
Demand forecasting models, on the other hand, do establish relations
between the evolution of the economic system and energy consumption trends.
Be they econometric, like SIBILIN (Criqui, 1985), or technico-economic, like
MEDEE (Chateau and Lapillonne, 1982), they also remain partial models for
186
A comparison of energy-economy models
two reasons: first, because they take economic variables to be exogenous, and
secondly, because they neglect the possible influence of energy supply on
demand.
Supply-demand equilibrium models avoid some of the limitations that we
have just mentioned by calculating the equilibrium levels of supply and
demand interactively. The IFFS model of the US Department of Energy
(Murphy et ai., 1984) is an example of such a model which covers the whole of
the energy sector. However, these models also remain partial in that they give
a representation of the operation of energy markets without taking into
account the cross-effects with and on the rest of the economy.
Input-output models constitute an advance in this respect in that they situate
energy flows within the framework of exchanges between the various sectors of
the economy - the breakdown may include anything up to several hundred
branches of activity. They can be used to analyse the links between structure
of production and energy demand (see Hoch and Carson, 1984). Other
applications include the repercussion of energy prices on other prices or the
impact of energy scenarios on the rest of the production system (see CERENCERNA, 1983). If the model is static, it accounts for the indirect effects induced
by intermediate consumptions, whereas with the dynamic Leontieff model,
acceleration effects on investment can also be integrated. However, in neither
case are the effects on macroeconomic evolutions taken into account. In reality,
the simple input-output model has two limitations when used in forecasting:
it only allows for scenarios which do not substantially modify macroeconomic
equilibria, and its assumption of fixed input-output coefficients limits its
application to the short term.
11.2.2 Modelling interactions between energy and the economy
Since the energy sector, far from being marginal, is central to the economy,
partial equilibrium analyses will not suffice. Instead energy-economy models
are intended to include the feedback effects between energy and economic
trends. By making economic variables endogenous, these models render studies
of energy supply and demand more coherent, while at the same time opening
up the scope of the analysis to new questions. Will energy availability limit the
possibilities of economic growth? What has been the macroeconomic impact
of the oil shocks? Or, what economic consequences can be expected from a
given energy policy, for example a moratorium on nuclear power?
Most of these models analyse energy-economy interactions within a national framework. They are very heterogeneous, ranging from the small aggregated
model, constructed by an individual researcher for the purposes of an academic
paper, to enormous model systems entailing pluri-disciplinary work by large
research teams over several years. For this reason, it is not easy to set up a
general analytical framework which would include all possible structures.
Attempts to do so have all been flawed (see, for example, Coates, et ai., 1979;
The French experience
187
Samouilidis and Mitropoulos, 1982; Devezeaux de Lavergne and Ladoux,

1989).
11.2.3 A typology of energy-economy models
Our classification of these models is based on the following criteria: the way
in which energy supply, energy demand and the overall operation of the
economy are represented. To be complete, we will add a fourth criterion, which
is the level of breakdown chosen, notably for exchanges between sectors and
the forms of energy involved.
Using these criteria we can readily obtain some idea of the structure of a
model. Without going into any detail, we simply recall here the main ways in
which each of these components may be represented. With supply and demand
modules there is a choice between a technico-economic and an econometric
approach; the latter is generally based on production and consumption
functions. These modules use either optimization or simulation techniques. The
representation of the overall operation of the economy can be based on one or
other of the various currents of economic theory. In practice, however, the
structure of the models is generally either of a neo-Keynesian or neo-classical
general equilibrium type.
The above classification makes it possible, for example, roughly to characterize the basic models in this field. The model used by Hudson and Jorgenson
(1974) is a general equilibrium simulation model, made up of nine branches of
activity (including five energy branches). Energy demand is defined econometrically and energy supply is seen as an exogenous variable. The model
developed by Manne (1977) is also of the general equilibrium type, but it is
resolved by optimization and includes only one non-energy branch of activity.
Energy demand, which is broken down into two products, is derived from a
macroeconomic production function whose parameters are seen as exogenous.
Energy supply, on the other hand, is based on a detailed technological
representation.
11.3 THE FRENCH EXPERIENCE
It was not until the 1980s that energy-economy interaction models began to
be developed in France. Unlike experience in other countries, in France these

models were mainly elaborated by public bodies. The initial research work was
undertaken by the administration for economic planning purposes. Five
models in all have been developed.
The first of these, CGP (Commissariat General du Plan), is a small one and
was built and used solely in order to prepare the VIIIth Plan. According to its
initiators, Levy-Garboua and Sterdyniak (1980), it was intended to stimulate
discussion on four issues: the role of energy prices in economic growth, energy
188
price and taxation policy, the behaviour of energy demand, and, lastly, the
penetration and financing of electric power.
The construction of the Mini-OMS-Energie (MOE) model took place
between 1979 and 1982, and involved several organizations: the INSEE
(Institut National de la Statistique et des Etudes Economiques), the Centre
d'Etudes et de Recherches Economiques sur l'Energie, the Forecasting Directorate of the French Ministry of the Economy, Electricite de France and the
Institut Fran9ais du Petrole. The declared objective was to produce forecasts
containing a description of macroeconomic trends which would fit into the
framework of the French national accounting system, and a representation of
the energy sector by means of complete and detailed balances (Brillet et al.,
1982; Insee, 1988). The model was put to substantial use in the preparation of
the IXth Plan (Commissariat General du Plan, 1983), and a group of users
made up of energy enterprises and public bodies continued to make use of it
(Caussat and Plateau, 1986). More recently, the new orientation in the
modelling policy of the INSEE has led it to hand over the management of
MOE to the Laboratoire d'Economie de l'Ecole Centrale.
This was the same laboratory that designed HERMES-FRANCE, i.e. the
French part of the HERMES model of the European Commission, between
1981 and 1984 (see Chapter 10 and Faubry et al., 1984; Moncomble and
Zagame, 1986). The idea behind this project was to provide a medium-term,
multinational and multisectoral model aimed at informing European energy
choices.
The last body to be concerned with macroenergy modelling in France is the
Commissariat a l'Energie Atomique (CEA). An initial project, which was
carried out between 1980 and 1983, was reformulated between 1983 and 1986
and gave rise to the MELOOIE model, designed to assist in defining the
long-term strategy of the CEA (see Berthelemy and Oevezeaux de Lavergne,
1987). The CEA also uses a simplified version of this model, micro-MELODIE
(see Ollevier, 1987).
11.3.1 A first difference between models: the level of aggregation
Table 11.1 shows that the French models do not all analyse energy-economy
interactions in the same detail. There are two small, highly aggregate models,
which do not take into account the evolution of intersect oral exchanges and
only distinguish between two energy carriers (domestic and imported energy in
the CGP model; electric power and fossil energy in micro-MELOOIE). MOE
and MELOOIE are medium-sized models. They include two non-energy
sectors and distinguish between five energy forms (coal, oil products, gas,
electricity and others). Lastly, HERMES allows for a far greater degree of
breakdown, since it accounts for seven branches of activity (outside the energy
sector), including three industries and eight energy carriers.
189

Table 11.1 The degree of breakdown of French energy-economy models
No. of equations
Model
Total
Behaviour
CGP
Micro-MELODIE
MDE
MELODIE
HERMES- FRANCE
40
90
700
>600
1500
20
50
175
n.d.
250
No. of
exogenous
variables
No. of
sectors
No. of
energy
carriers
n.d.
1
1
2
2
7
2
2
5
5
110
450
n.d.
400
'Outside the energy sector.

n.d., Not determined.
These characteristics of the models are important in determining where they

are to be applied. They also reveal the assumptions of the analysis concerning
the possible feedbacks between energy trends and economic structures. Generally, such effects can scarcely be taken into account when there is little
breakdown between sectors of activity.
11.3.2 Household energy demand: technical versus economic approaches

When looking at household energy demand, we should stress the contrast
between the MDE model, which adopts a technico-economic, 'bottom-up'
approach, and the other four models, which are purely econometric.
In MDE (Figure 11.1), energy use is divided up into four demand sectors:
transport, electrical appliances, domestic heating with hot water consumption
and cooking, which is exogenous. In each individual sector, energy demand
depends on three determining factors: the stock of equipment, its specific
consumption and the utilization rate. The last two variables are usually
exogenous. On the other hand, the evolution of the stock is specified endogenously by econometric relations, which employ various explanatory variables: temporal trends, number of households, total household consumption
and price of equipment (housing units or personal vehicles). However, the
allocation of energy forms used by new housing stock remains exogenous. The
only energy price used in this part of MDE is the fuel price, which has an
impact on the rate of utilization of personal vehicles.
The four other models adopt a 'top-down' approach, abolishing all reference
to the stock of equipment in order to establish a direct econometric relation
between the economic variables and energy demand. Moreover, they determine
inter-energy substitutions endogeneously according to the relative prices of the
various energy forms. Consumption for each type of energy is deduced from
total household consumption and prices, in one single stage in the aggregate
models (CGP and micro-MELODIE), and after two successive breakdowns in
HERMES and MELODIE - the intermediate level providing total energy
190
Equipment costs
Consumption or
number of households
Distribution of new
buildings by type
of energy
Find trend for household

electrical equipment
Utilization
rate
Equipment stock
-automobiles
-household electrical equipment
-buildings by type of heating
-gasoline consumption
--electricity consumption for specific uses
--energy consumption for heating and water
by energy form (electricity, gas, coal,
fuel oil, renewables)
Fig. 11.1
Specific consumption
(by energy form)
for heating and water
Energy consumption
for cooking needs
(by energy form)
Household energy consumption in MDE. (The variables in boxes are

endogenous, the rest is exogenous.)
consumption. The most frequently used allocation function is of a dynamic

translog type.
In comparison with the other models, the first advantage of MDE is that it
reveals the actual physical details of energy demand formation, whereas
breakdowns obtained from total household consumption do not correspond to
any technological reality. Moreover, the use of a large number of exogenous
variables in MDE makes it possible to test the effect of various energy
conservation policies. With the other models only price policies can be dealt
with.
On the other hand, the multiplicity of exogenous variables in the MDE
model increases the danger of inconsistency. Moreover, the user should be
aware that many of the findings do not result from the operation of the model
as such but are merely the transcription of the scenario hypotheses. The
econometric models provide greater consistency for the analysis as a whole,
and are particularly well adapted to coupling with a macroeconomic module.
191

11.3.3 Energy demand in production
Here again MDE is quite different from the other models, but the differences
are expressed at another level. Given the difficulty of taking into account the
great diversity of industrial processes, no model has explicitly introduced the
stock of energy-consuming equipment. They all establish a direct relation
between economic indicators and the energy demand of the various economic
sectors.
The principal particularity of MDE is that it bases its analysis on the notion
of the energy content of value-added; the evolution of energy contents being
defined econometrically by a temporal trend (Figure 11.2). In the aggregated
industrial sector, the model distinguishes between a specific electricity content
and a content in terms of substitutable energies for heating purposes. The use
Time
trend
Time
trend
Electricity
intensity
Fuel
intensity
Value-added
of sector
Fuel consumption
Exogenous distribution
Electricity
consumption
Consumption of
petroleum products
Gas consumption
Coal consumption
Fig. 11.2 Producers' energy consumption in MDE.
192
of energy as a raw material is exogenous. The rest of the economy - too

heterogeneous for any useful analysis - is broken down into three sectors
(services, transport, the farming and food industries along with building and
public works), and fuel and electricity demand are determined for each of these.
As in the case of household consumption, inter-energy substitutions are
exogeneous. Another particularity of the model is that it only uses energy
prices marginally (only in explaining demand in the tertiary sector).
The other models (Figure 11.3) all use production functions to estimate
energy demand along with the demand for other factors (capital, labour,
Final demand
addressed
to the sector
KLEM
production function
\~
Energy
Labour
/
Distribution function
Relative price
of factors
Materials
Relative energy
prices
Consumption by energy form

(according to the level of
breakdown within the model)
Fig. 11.3 Producers' energy consumption in CGP, MELODIE, micro-MELODIE

and HERMES-FRANCE.
193
material) in each of the sectors which have been identified. The determining
factors of energy demand are the volume of production, prices, technical
progress, scale and the rate of utilization. As in the case of households, energy
is directly broken down into two products in the CGP and micro-MELODIE
models. In HERMES and MELODIE energy demand per product is obtained
by breaking down total energy demand by means of a translog allocation
function. In the CEA models the production function is also translog; it is
dynamic in MELODIE and static in micro-MELODIE. HERMES and CGP
both use a model based on capital vintages. Technical choices are made on new
investments and then remain frozen for the life-span of the equipment. The
marginal production functions are of the Cobb-Douglas type in the CGP
model and of a two-level CES/Cobb-Douglas type in HERMES.
Finally, the approach adopted by the MDE model appears to give too little
importance to the possible impact of energy prices. On the other hand, the
macroeconomic framework of the other models appears to give too much
importance to prices and to neglect other factors. In addition, all of these
models seem to be inappropriate if we are to take into account certain energy
policy measures such as, for example, financial aid for investments involving
energy savings or substitution. Finally, disaggregation by sector is, on the
whole, insufficient if we are to take account of structural effects and contrasting
changes in energy content from one sector to another.
11.3.4 Energy supply: frequently exogenous
The following information can be ascertained (wholly or partly) from the
supply side of the various models: investment and capacity in the energy sector;
final energy production and primary energy demand of the energy sector,
energy import and export; and finally energy prices and financial accounts
within the sector. The methods used vary between the five French macroeconomic models but what is striking overall is the number of exogenous
factors used. This feature can be explained not only by the fact that these
models are mainly intended to test energy supply policies but also by the
difficulties involved in integrating a complex energy supply model within a
macroeconomic framework.
As a result, investment is only endogeneous in two cases: aggregate investment for the entire sector within the CGP model, and refining investment in
MDE. In both cases investment is dictated by changes in demand without the
intervention of any profit-related terms.
Supply and demand are interconnected in all models by adapting the level
of supply to that of demand. Supply is determined for each energy product by
the following balance between use and resources:
production + imports = domestic demand + exports
in which domestic demand is calculated by the demand modules as explained
194
previously and exports are exogenous (except in the case of HERMES). Thus,
only in HERMES is the breakdown of supply between production and imports
endogenous. In the other models either production or imports is exogenous,
the other being deduced residually.
Energy import prices are always exogeneous. On the other hand, the
production price of energy is sometimes considered endogenous, but the way
in which it is determined varies from model to model. In other cases (mainly
in CGP and MDE), the average user price is entirely exogeneous, i.e. considered as a command variable of energy policy.
In HERMES and in the non-electric sectors of the CEA models, the move
from final energy demand to primary energy demand also involves exogenous
input-output coefficients. In the other cases, the way in which primary energy
demand is calculated is the result of a detailed representation of the behaviour
of the energy sector, but a different modelling technique is used in each case.
CGP uses an aggregate production function for the whole energy sector,
based on capital vintages, as for the rest of the economy. In this way it can
incorporate demand for production factors in the energy sector (capital and
imported energy) as a function of their relative prices and technical progress
(see Chapter 3). The main problem raised by this approach is, in our view,
related to the fact that the aggregate figure includes such diverse units as
electric power stations and refineries. In contrast, the following two types of
representation are interesting because they are based on the technological
reality of the sector.
The refining model used by MDE is made up of econometric relations which
use pseudo-data. These fictitious data are obtained from the results of sixtyodd projections produced by the French Petroleum Institute's large optimization models which show how to adapt refining facilities in order to satisfy
demand at least cost. It is then possible with the resulting model to make
investment endogeneous, to determine oil demand and to calculate marginal
production costs using fewer equations while still taking the numerous complex technological constraints into account. However, though this method is
simpler, it is less transparent. In addition, its use is necessarily limited to the
study of scenarios which are more or less similar to those used in constituting
the reference sample.
In the MDE, MELODIE and micro-MELODIE models, the electricity
sector is represented by a simulation model for the management of power
station capacity. These modules can be used to determine the primary energy
used, given the level and structure of electricity demand and the exogenous
production capacity. The advantage of such a technological representation is
that it takes account of variations in the input-output coefficients as a function
of production level. But it necessarily remains a rough approximation of the
way in which power stations are managed, since in reality the latter takes place
in real time.
195
11.3.5 Similar macroeconomic structures

In line with the French macroeconomic modelling tradition, all the models
under consideration have a post-Keynesian structure.
The way they work can be described as follows (Figure 11.4). They first
determine final demand, which is made up of consumption, investment, public
Final demand (except imports)
Input-<lutput
coefficients
matrix
Production
Labour
demand
Labour
supply
Capital
demand
/
Wages
Uti lization rate

of productive
capacities
--
----
--1
----
Incomes
Capital
stock
Unemployment
Production
function
Prices
Fig. 11.4 Simplified macroeconomic structure of the models.
196
expenditure and exports. Having deducted imports, a matrix of input-output

coefficients is used to calculate the required production, intermediate consumption and value-added in volume terms. The model then deduces factor demand
and the tensions between this demand and the corresponding supply determine wages and prices. The distributed incomes then determine a new value
for final demand. Static equilibrium is reached when this indirect final demand
is equal to the initial demand. Thus there are two typically Keynesian factors
at the heart of these models' resolution processes; the driving role attributed to
demand (the hypothesis being that there is a supply surplus) and the importance given to imbalances represented by the tension variables (unemployment
and capacity utilization).
The short-term dynamics of this kind of model are influenced above all by
the effect of income on consumption, of activity on foreign trade and by the
effect of limited outlets on investment. In the medium term, this type of multiplieraccelerator movement is rectified by the growing influence of the wage-price
loop. In the longer term, when prices and tension-inducing variables tend
towards their equilibrium value, supply conditions become dominant.
Given this theoretical framework, these five models have several common
features in terms of the way in which they look at interactions between energy
and the economy. These mechanisms can be easily demonstrated by analysing
the way in which the different variables follow on from each other in an oil
shock scenario. The shock has first of all an inflationary effect which acts both
through the direct effect on price levels and the wage-price loop.
The rise in price levels then decreases the commercial balance as competitiveness is affected - unless supplementary exogeneous hypotheses are
made concerning foreign price levels. In the short term the oil shock also has
a negative effect on real household incomes since wage rises do not entirely
make up for price rises. The result is a fall in consumption. Given the reduced
possibilities offered by outlets, the level of investment also falls. The shock thus
brings about a slowdown in growth, a reduction in employment and a fall in
the rate of utilization of productive capacities.
These recessive effects do, however, have beneficial effects on foreign trade
which tend to counteract the initial effect. They first of all reduce imports
because of the fall in domestic demand and the fall in the rate of utilization of
productive capacity. At the same time, firms start to look for ways in which to
offset this fall by increasing exports.
In addition, recessive effects have complex consequences for inflation and
real household income. The increase in unemployment holds back wage
increases and leads to a decrease in the real wage bill, both of which amplify
the fall in nominal income. Then the wages-prices loop, fed by the increase in
unemployment and the fall in capacity utilization, leads to a slowdown in price
rises which helps stabilize real income.
Finally, it should be noted that though all the models determine the impact
197
of the shock on the external balance, none of them takes account of its
deterioration as a constraint on growth.
11.3.6 A comparison of results
The fact that there is a common economic logic should not eclipse the
differences between the models. We have already mentioned the specifics of
each model as regards the level of aggregation and the way energy supply and
demand are represented. Other differences stem from the choice of explanatory
variables used in the macroeconomic equations and of specifications (the form
of the equations, presence of dynamic elements and the value of the coefficients). Since it is beyond the scope of the present article to describe the models
in detail (see Destais, 1989, ch. VIII), we shall here only mention their most
salient characteristics.
One essential difference has to do with the production function used to
represent the behaviour of producers. This function is of the putty-clay type in
CGP as well as in the industrial sectors of HERMES. It is fixed coefficients in
MDE and the CEA models use a translog form. While substitution between
production factors as a result of changing relative costs is impossible in MDE,
it plays an important role in the other models. To be more exact, the CEA and
HERMES models assume complementarity between capital and energy, the
aggregate of these two factors being substitutable for labour. CGP, on the
other hand, works on the hypothesis that all factors can be substitutes.
Looking at the effects of an oil price shock, these mechanisms modify the
analysis given above which leads to a reduction in investment and employment. In CGP the levelling off of investment is attenuated by the increasing
cost of other factors which encourages very capital-intensive processes (though
it is reinforced by the increase in real interest rates). In the industrial sectors of
the HERMES model and in the CEA models, the fall in investment is accentuated by the complementarity of energy and capital. In contrast, the price of
energy has no direct effect on investment in the MDE model and in the nonindustrial sectors of HERMES. However, the formation of capital is also
slowed down because of the increase in production costs which induces a fall
in the profit rate. In addition, the unemployment caused by the oil shock is
attentuated by changes in relative prices which lead to a redeployment of production factors in favour of labour in all the models (with the exception of MDE).
Price equations also differ quite clearly from one model to another while
consumption, foreign trade and wage rate equations are relatively similar.
However the other basic difference between models has to do with their
dynamics. These dynamics are the result of the existence of variables which are
out of phase by one or more periods (years) in order to express anticipatory
behaviour or the existence of lags in adjustment as a result of inertia. From
this point of view, the quasi-static nature of micro-MELODIE and CGP
198
(which are supposed to express long-term adjustments) contrast with the

systematic introduction of phase differences in HERMES (intended to improve
the handling of short-term changes). The other models lie somewhere in
between, MELODIE having the specific characteristic of adopting a single
delay structure for all the production factors.
All in all, the particularities of each model have a certain effect on the results
it provides. To illustrate this, we shall describe a scenario which has been tested
using three of the models: MDE, HERMES and micro-MELODIE (Kalaydjian and Maillet, 1989). The variation looked at corresponded to a 50%
increase in the price of crude oil maintained over 7 years. All the other
exogenous variables were left unchanged. The results (see Table 11.2) are given
as percentage differences as compared with a reference scenario for each model.
As our analysis of the models led us to expect, some common elements can
be found, such as the overall recessive and inflationary effect of the shock, but
at the same time important differences can be seen in the numerical results for
certain variables.
1. The quasi-static nature of micro-MELODIE leads to a larger fall in growth
in the short term. However, this trend is then modified, in particular by
substitution among the factors which make competitive gains possible and
act favourably on foreign trade.
2. In MDE and micro-MELODIE inflationary effects are felt immediately,
whereas they spread slowly in HERMES. They are later attenuated in MDE
as the result of a group of factors (slowdown in activity, increase in
Table 11.2 Impact of an oil shock in MDE, HERMES and micro-MELODIE
Volume GDP
(%)
Consumer
price (%)
Nominal wage
rate (%)
Employment
(thousands)
Primary
energy consumption (Mtoe)
Final electricity
consumption
(Mtoe)
Year
MDE
HERMES
Micro-MELODIE
0
7
0
7
0
7
0
7
0
7
-1.6
-3.9
+7.1
+6.3
+6.4
+3.2
-125
-566
-7.0
-2.5
-0.7
-2.2
+1.4
+8.9
+0.6
+5.5
- 24
-215
n.a.
n.a.
-2.6
-1.6
+5.4
+2.3
+4.8
+0.8
-241
+ 79
-12.9
-10.7
0
7
-0.2
-2.1
n.a.
n.a.
+ 0.6
+ 3.8
n.a., Not available.

Mtoe=Million tonnes oil equivalent.
Conclusion
3.
4.
5.
6.
199
unemployment, fall in profit rates}. In micro-MELOOIE inflation is substantially reduced by the end of the period because of substitutions.
MOE and micro-MELOOIE both employ an immediate indexation of
wages to prices, while HERMES uses a time delay.
The large fall in employment in micro-MELOOIE in the first year is linked
to the fact that there is no adjustment delay and the final improvement in
employment is due to substitution between factors. In HERMES the
worsening of the labour market as a result of the recession cannot be
counterbalanced by factor substitution within the 7-year period because of
adjustment delays. The results provided by MOE illustrate the postulated
rigidity between the level of activity and employment.
The fall in energy consumption resulting from income and activity effects is
accentuated by the price effect in micro-MELOOIE but attenuated in MOE
because of rigidities in production technology.
Consumption of electricity falls in MOE, while it increases in microMELOOIE because of substitution between energy forms.
These examples show how results reflect the structure of the models and how
important it is to be fully aware of this structure in order to appreciate their
relevance.
11.4 CONCLUSION: CORRECT USE OF THE MODELS

The main lesson which can be learned from the preceding analysis is that
although the models described are useful tools in the decision-making process,
they should be used with caution.
It should be said at the outset that because they deal simultaneously with
mechanisms of energy supply and demand formation, exchanges between
sectors and macro-economic evolution, French energy-economy models definitely allow us to widen the field of analysis of energy-economy interactions
as compared with partial equilibrium models. However, this possibility has
been obtained at the cost of a certain number of restrictions: the analysis
remains highly aggregated; information concerning the technical mechanisms
which govern the formation of energy demand is rarely exploited; and energy
supply is essentially exogeneous.
More seriously, there is a certain lack of consensus concerning the dynamics
of energy-economy interactions. In this respect, the challenges are, first, to
bring out the dynamics which affect economic and energy phenomena (time
required to implement investments, lagged or anticipatory behaviour, etc.), and
secondly, to integrate short- and long-term phenomena. We have shown that
the formulations used by the various models are disparate and lead to
constrasting overall dynamics, which is obviously not satisfactory. More
200
generally, one can raise the question of the reliability of the quantitative
analyses provided by these models. Studying a common scenario over several
models has indeed shown that they can at times produce very different
quantitative results.
Should we conclude then, as Hoogdwin (1987), that energy-economy models
are above all heuristic and of limited use? In our view their importance in
helping us structure our analysis of the questions involved should not be
underestimated, for they have two major advantages. First they render the
hypotheses underlying the conclusions explicit. Secondly, they guarantee the
internal coherence of the analysis and can be used to evaluate the results of
various contradictory effects.
But it should be clearly understood that the models are not a neutral
simplification of reality. For example, one must know that some questions
(such as the effect of the external constraint on growth) cannot be dealt with
at all using any of the French energy-economy models, since certain phenomena are not represented (for example the financing of the external deficit). These
models also have an appropriate field of utilization linked to some of their
hypotheses. Thus we have seen that their conclusions were only valid in a
Keynesian situation of surplus supply - a heavy constraint. In addition, it is
our view that their time horizon should be limited to the short and medium
term, even if some modellers claim they can be used for long-term studies.
Consistency within the equations and stability over the long term do not
protect a model from its own econometric and aggregate nature, features which
make it impossible to incorporate long-term structural changes taking place in
the economy and in the energy sector.
Moreover, potential users should ask themselves whether or not the representation which has been adopted by the model in question actually corresponds to their own analysis of the situation. Thus, as Hourcade and Kalaydjian (1987) have emphasized, 'within the representation of structural changes,
models using KLEM production functions give more weight to substitution
between production factors under pressure of prices - and this is only one,
controversial, way of looking at such changes'.
Finally, our analysis shows that the appropriate use of the models has to be
based on an extensive knowledge of the ways in which they work. In this
respect, the comparative study of several models is, in our view, an excellent
means of evaluating the explanatory potential and the limits of each model.
However, this approach is not always rendered easy by those responsible for
the design of the models. If they really want their models to be used in the most
appropriate situations, we would suggest, in conclusion, that they adopt a
homogeneous framework for the presentation of the models and terminology
in order to facilitate access to their models. If, in addition, the modellers were
to make an effort to situate their model in relation to the others available, then
potential users would be more than delighted.
References
201
REFERENCES
Berthelemy,1. C. and Devezeaux de Lavergne, 1. G. (1987) Le modele MELODIE: un
modele energetique de long terme pour l'economie fran9aise. Revue d'Economie
Politique, 97, 649- 72.
Brillet, J. L. et al. (1982) Energie et economie: Ie modele Mini-DMS-Energie, Economie
et Statistiques, July/August, pp. 73-85.
Caussat, L. and Plateau C. (1986) 'L'analyse comparee des effets economiques de
diverses mesures de politique energetique Ii l'aide du modele Mini-DMS-Energie.
World Energy Conference, 2.3.3., 5-11 October, Cannes.
CEREN-CERNA (1983) Impacts macroeconomiques de la production et de la consommation de charbon-vapeur, 2 vol. Paris, p. 99 and appendices.
CGP (1983) Rapport du Groupe Long Terme Energie, Preparation du IXeme Plan
1984-1988. La Documentation Fran9aise, Paris, vol. 1, p. 435, vol. 2, p. 236.
Chateau, B., Lapillonne B. (1982) Manuel de description du modele MEDEE-3. IEJE,
Grenoble, p. 53.
Coates, R., Hanson, D. Juenger,s. et al. (1979) Survey of the Research into EnergyEconomy Interactions, vol. 1. US Department of Energy, Washington, DC.
Criqui, P. (1985) Le role des importations d'energie dans Ie jeu des contraintes
internationales, construction du modele SIBILIN. CGP, Paris, p. 191.
Destais, G. (1989) 'La modelisation des interactions energie-economie, une analyse
centree sur Ie cas fran9ais'. PhD Thesis, Social Sciences University of Grenoble, IEPE,
p.670.
Devezeaux de Lavergne, J. G. and Ladoux, N. (1989) Pourquoi des modeles macroenergetiques? Revue de l'Energie, no. 411, May, pp. 423-34.
Faubry, E., Moncomble, 1. E., Vidal de la Blache, O. et al. (1985) Le modele
HERMES-FRANCE. Economie et Prevision, no. 66, p. 3-29.
Finon, D. (1976) Un modele energetique pour la France. CNRS, Energie et Societe,
Paris.
Hoch I. and Carson, R. T. (1984) An Energy-oriented input-output Model. Electric
Power Research Institute, Palo Alto, Ca, p. 314.
Hoogdwin, L. (1987) 'On the character of macroeconomics, macroeconomic policy and
econometrics: the need for another macroeconomic policy conception.' Papers
presented at the Economic Modelling Conference, 21-22 October, De Nederlandsche
Bank, Amsterdam, p. 24.
Hourcade, J. C. and Kalaydjian, R. (1987) Macroeconometrie et choix technologiques
structurants: bilans et questions Ii partir du domaine de l'energie, Papers presented
at 19th Conference on Economic and Econometric Structures, ARAE, SophiaAntipolis, 21-22 May, p. 37.
Hudson, E. A. and Jorgensen D. W. (1974) US Energy Policy and Economic Growth,
1975-2000 Bell Journal of Economics and Management Science, 5, 461-514.
INSEE (1988) 'La demande et l'offre d'energie dans Mini-DMS-Energie. Working
Paper Note 320/97, Service des Programmes, p. 70.
Kalaydjian, R. and Maillet, P. (1989) Les modeles au service de la decision: analyse
comparative de trois modeles macroenergetiques fran9ais. Revue de I'Energie, no. 409,
February, pp. 67-85.
Levy-Garboua, V. and Sterdyniak, H. (1980) Coherence macroeconomique, dossier no.
6, in CGP, Commission de I'Energie et des Matieres Premieres. La coherence
macroeconomique de la politique de l'energie, Report of Working Group no. 1, Paris,
pp. 187-241.
Manne, A. S. (1977) ETA-MACRO, a model of energy-economy interactions, in
202
Modelling Energy-economy interactions: Five Approaches (ed. C. 1. Hitch), Resources for the Future, Washington, DC.
Moncomble, 1. E. and Zagame, P. (1986) Les effets d'une baisse du prix du petrole:
l'analyse du modele Hermes-France. Petrole et Techniques, no. 326, June/July, pp.
15-18.
Murphy, F. H., Conti, 1., Shaw, S. et al. (1984) An Introduction to the Intermediate
Future Forecasting System, in Analytic Techniques for Energy Planning (eds. B. Lev
et al.), North Holland, Amsterdam pp. 255-64.
Ollevier, M. (1987) 'Modelisation macroeconomique, energie electrique et energies
fossiles dans l'economie fram;aise: un essai d'approche formalisee' Memoire de DEA,
CEA, Departement des Programmes p. 78 and appendices.
Samouilidis, J. E. and Mitropoulos, C. S. (1982) Energy-economy models: a survey.
European Journal of Operational Research, 11, November, pp. 222-32.
---12--Models and projections of energy

use in the Soviet Union
Yuri Sinyak
12.1 ENERGY PLANNING AND MODELLING

IN THE SOVIET UNION 1
The use of mathematical computer models in the Soviet energy system goes
back to the late 1950s. The energy systems were one of the first sectors of the
national economy in which these new tools were applied. There was, at that
time, considerable enthusiasm over the supposed ability of such models to
solve economic planning problems but this soon turned out to be an illusion.
It was a common error in the centrally planned economies to believe that social
reality could be mechanically determined and described by mathematical tools.
But reality is infinitely more complicated than even the most sophisticated
computer model, and judgement by the human intellect can only be replaced
in the very simplest of mechanical applications. Another problem in this area
is that the performance of these models can never be better than the quality of
the data input (which in general is quite poor).
Today it is generally realized that effective planning and management of
man-machine systems is possible only through a combination of formalized
and non-formalized models. Man must retain an active role at all planning
levels, both in the formulation of goals and restrictions and in the assessment
of final results.
A specific character of Soviet economic planning has until recently been the
emphasis on material balances. The purpose of modelling in this framework
has been mainly optimization and relatively little attention has been paid to
the role of pricing, consumers' preferences and substitution between products
and factors of production.
1 The author describes the main features of the centralized planning system existing at the time of
writing. It is difficult to say how it will develop with a decentralized market-oriented economic
system among the national republics. However, the author is convinced that the scientific
background for long-term energy planning will be the same in the future.
204
Models and projections of energy use in the Soviet Union
This 'optimization' has been carried out by a multi-stage process in which

plans at the national level are linked to those at the regional and local levels
while long-term plans for technology choice (over 10-15 years) are linked with
5-year perspective plans and with yearly plans as well as the models of current
operative management,
This grandiose optimization scheme meets with a number of severe obstacles
such as the great number of elements and linkages in the energy system that
are non-linear or discrete. Furthermore, energy demand is decided by a great
number of independent decision-makers and this decentralization requires, in
a programming context, the conveyance of correct price information from the
upper levels in the planning hierarchy to its lower levels. This has earlier been
a major obstacle but the principle of marginal pricing is now accepted and
correct fuel, electricity and heat prices by region are now available but only
used for the design of new energy facilities and projects, not for everyday
operation which results in the collision of current and long-term goals. Such a
dual situation reduces the effectiveness of the pricing system and helps
maintain an inefficient administrative command system.
Recently the use of price-type information together with simulation concepts
is gaining support among planners at the expense of relying solely on
traditional optimization methods.
12.1.1 National energy planning

The application of mathematical modelling to the Soviet energy system started
with long-term optimization (Figure 12.1). The initial phase of energy forecasting consists in elaboration of several options for growth in GDP. The results
of this phase are used for the assessment of energy demand and feasible
alternatives for energy production. This information together with technology
projections is the basis for the optimization of state and regional energy
systems and their main sectors - electricity, heat, nuclear, oil, natural gas and
coal production and distribution. The final phase deals with the development
of state technological programmes, which are essential for achievement of the
specified goals.
The main objectives of energy planning and forecasting are:
1. to clarify major ways of improving energy efficiency;

2. to investigate the most promising areas of electricity penetration as a
driving force of social and technological progress and the optimal structure
of electricity generating mix;
3. to study the detailed possibilities for energy production, transportation
(pipelines, electricity grids, railway, etc.) and fossil fuel processing and
conversion - taking into account economic and ecological aspects;
4. to assess the material, labour and capital requirements of the energy sector;
5. to elaborate R&D programmes.
Energy planning and modelling in the Soviet Union
205
GOP projections
Fossil fuel
reserves and
resources
Economics of
new and
conventional
technologies
Energy
demand
options
Energy
production
options
State energy balance
Plans for
energy supply
sectors
~
>-
t::
()
a:
t3UJ
(!)
...J
cr:
...J
-
UJ
cr:
...J
C3
()
ttTT
...J
Fig. 12.1
::)
I-
()
Regional energy
balances
R&D
programmes
Energy
balances of
industrial
enterprises and
complexes
Approximate scheme of the Soviet energy optimization concept.
The procedure for the energy system optimization consists of four main
stages. At the first stage a simplified optimization of the state energy system is
used (Melentiev, 1979). The expected range for aggregate fossil fuel production
is based on the available fossil fuel reserves and the planned allocation of
capital to develop these reserves which in its turn depends on economic growth
and energy demand.
The second stage requires the application of multi variant optimization
models (see below) developed for the selection of the best solutions for each
energy supply sector to satisfy, at minimal cost, the supply options defined in
the first stage. At this stage the decision parameters are: fossil fuel production
by basin, inter-regional fuel and electricity flows, location of new fuel processing facilities and electricity generation, and levels of regional and sectoral
energy consumption. Usually at this stage more dis aggregated energy models
are used compared to the first stage.
206
The third stage consists of the investigation of uncertainties associated with

the probabilistic or uncertain character of parameters used at the previous
stages (Beliaev, 1978). Different primary energy sources are compared with
respect to cost, export value, production scale, input requirement, etc. As a
result two or three reasonable strategies for the development of each energy
resource are selected. These strategies are then evaluated under various
scenarios including the most unfavorable conditions. Preference is then given
to the strategies with least total costs under a mixture of normal and
unfavourable conditions.
Finally, at the fourth stage the selected strategies are evaluated from the
point of view of their demands on other sectors of the economy. Simplified
dynamic energy--economy models (Kononov, 1981), are used to compare their
capital and labour requirements. Lead times of construction are also taken into
account at this stage, providing the adjusted programmes of material and
equipment supply for investment in energy facilities.
The peak of modelling applications in energy-related tasks occurred in the
1970s (Aganbegian and Fedorenko, 1978; Melentiev, 1979). Since then interest
in energy modelling seems to have declined, mainly because of the lack of
feasible alternative solutions due to increasing strains on the national economy
in the early 1980s and the neglect of the economic aspects in compiling
perspective plans of industrial development. Until now energy modelling has
been considered an optional supplement to the main planning procedure based
rather on simulation than on optimization.
12.1.2 State and regional energy balance models 2
The aim of state or regional energy balance models is to investigate the main
proportions within the national energy system and major inter-regional energy
flows (USSR Academy of Sciences, 1975, 1977). A simplified linear programming approach is usually applied. The energy system is treated as static and
the models usually describe the state of the system at the end of the planning
periods (5-, 10- or 15-years). The models cover large regions (e.g., European
Soviet Union, Ukraine, Volga Region, Caucasus, Middle Asia, West and East
Siberia, Far East). Each region is characterized by the same energy consumers:
industrial furnaces, cement kilns, power plants, industrial boilers, space heating
and hot water supply. These are divided into three categories - old, under
reconstruction or new - each with different input coefficients. Sometimes the
2The practical applications of state energy balance models are carried out by the Computer Centre
of the USSR State Planning Commission and its former Institute on Complex Fuel and Energy
Problems, and the research work on the elaboration and improvement of such models is done
mainly by the Siberian Energy Institute of the USSR Academy of Sciences. Regional energy
balances, which resemble the national energy balance, are studied practically at all energy-related
institutions in national republics.
Energy planning and modelling in the Soviet Union
207
scale of energy consumers and load curves are taken into account. One of the
common points of dispute in the application of these models is the method of
demand description: in terms of useful energy or cost.
One of the most interesting attempts to overcome the limitations of aggregated models is the elaboration of a block-wise dis aggregated linear model,
in which each block has to be treated separately by different groups of
specialists and then linked up by a central coordinating module carrying out
the function of the centralized control unit (Makarov, 1989; Nekrasov, 1981).
However, practical implementation has not been very successful because the
large information flows between modules can hardly be handled with the
existing computers and because of low productivity of the teams working with
the different modules in the preparation of new versions in response to
changing input parameters specified by the central module. Therefore, many
institutions have ceased efforts in this area.
12.1.3 Electric system models 3
The first application of electric system models dates back to the late 1950s and
earlier 1960s with the elaboration of the linear planning model (Markovich
et ai., 1962). More sophisticated static and dynamic models were later developed in the Siberian Energy Institute of the USSR Academy of Sciences and
were applied for expansion planning of electricity grids in the European part
of the Soviet Union (static model) and in Central Siberia and the north-west
of the European part of the Soviet Union (dynamic models) (Syrov et ai., 1966);
Melentiev, 1971). The first models were used for studying the impact of various
factors on the optimal structure of the generating capacities, the specific
features of hydroelectric stations or load curves for different seasons, optimal
rates of nuclear energy development, etc. Later new electric system models were
elaborated in the Central Economic and Mathematics Institute of the USSR
Academy of Sciences for planning the location of base-type thermal power
plants and for electricity sector development at the level of a Soviet republic
(Nekrasov et ai., 1973; USSR Academy of Sciences, 1977). A special class of
electric system models comprises the short-term planning models dealing with
minute load-meeting or fuel supply within short time periods (e.g., a day or
week) (Sovalov, 1983).
Considerable efforts have been undertaken to elaborate more advanced
models for this sector based on the concepts of dynamic programming or with
the use of gradient optimization procedures. But these approaches are still at
an experimental stage.
3The most well-known organizations engaged in electric system modelling are the Research Institute
of Electric Grid Design and the Research Institute of Electric Energy, both of the Ministry of
Energy and Electrification, and the Siberian Energy Institute of the USSR Academy of Sciences.
208
12.1.4 Fuel supply models

Starting from the 1960s, several models for natural gas development have been
created (Smirnov and Garlyauskas, 1969; Druzhinin and Kuznetsov, 1970;
Feigin et ai., 1979}.4 They have been used for the elaboration of 5-year plans
with the main goal of investigating optimal reserve/production ratios for
different gas fields as well as the allocation between Eastern (West Siberia) and
Western (European) gas production. Recent interest has been expressed in
optimization models which require a multilevel description of the natural gas
production and distribution system (Stavrovsky and Sukcharev, 1978; and
Maksimov, 1979).
Similarly, oil models have been used to compile middle- and long-term plans
of this industrial sector (Makarov, 1989). Some of these models have special
modules describing the petrochemical industry. As a result, it has been shown
that petrochemistry has a large impact on the production rates of crude oil in
different regions and the location of the oil refineries as well as on the
distribution of light and heavy oil products. There are also models for oil
refinery optimization depending on the quality of the oil input and the local
requirements on the refinery products, including petrochemistry (Starovoitov
et ai., 1974}.5
Modelling in the coal industry 6 has gone mainly in the direction of gradual
disaggregation of the input information on coal producing facilities (Astakhov
et ai., 1967; Tsvetkov, 1967). The latest versions of the models usually consider
the whole set of the coal mining and coal processing (mainly benefaction)
enterprises with several alternatives for their development. As a result of such
a detailed approach to the description of the coal production system the static
models contain 800-900 constraints and above 6000 variables. The dynamic
versions expand the model dimension two to three times and require the use
of more complex software.
12.1.5 Recent trends in Soviet energy planning
In addition to the sectoral and regional energy models mentioned, the

development of large industrial enterprises and complexes has been studied
with specially designed models (Kluev, 1970; Nekrasov and Sinyak, 1979).
Recently energy planning based on simulation concepts, especially for
long-term applications, is getting increased attention, particularly in research
4Natural gas modelling is concentrated in the research institutions of the former Ministry of
Natural Gas Production.
SOil sector modelling was carried out in the Central Economics and Mathematics Institute of the
USSR Academy of Sciences, in the Institute of Economics and Industrial Management of the
Siberian Branch of the USSR Academy of Sciences and now is concentrated mainly in several
research institutions of the Ministry for Oil Production and Processing.
6The best known modelling results have been achieved in the Central Institute of Coal Industry
Economics of the Coal Ministry in cooperation with academic institutions.
Soviet energy demand and supply prospects
209
institutions. The experience of the last 20-30 years has known many problems
but also some important achievements:
The treatment of the energy system as a multilevel hierarchical structure
which requires the use of complex optimizing procedures and good information exchange networks.
Decision-making in the energy system on the basis of optimization procedures with incomplete or uncertain information which increases the
robustness of the final solutions.
Top-down accounting of constraints and time dynamics in planning practice
at all levels of hierarchy which increases the internal consistency of the plans.
12.2 SOVIET ENERGY DEMAND AND SUPPLY PROSPECTS

(1990-2000)
12.2.1 Energy programmes and policies
The prospects for Soviet energy systems development to the year 2000 were
defined in the Energy Programme (1984), first adopted in 1982. The Programme gave great importance to the production of natural gas, coal and
nuclear energy and energy conservation. Since then, the Programme has been
revised and its time frame extended to 2010. Since the plans for nuclear energy
development between 1980 and 1985 were not fulfilled and installation of
nuclear power plants slowed down after the Chernobyl accident (1986-90),
provisions have been made to compensate for the drop in nuclear electricity
production primarily through the accelerated growth of the national gas
industry and increased use of coal in electric power plants. Although practically
finished, the latest Energy Programme has not been approved by government
because of heavy criticism from the public and some scientists. Now the Soviet
Union has main directions for the energy policy only for the next 10-15 years
(Makarov, 1989; Makarov and Bashmakov, 1989; Volfberg et al., 1989).
Central to the national energy policy is energy conservation. The serious
drawbacks for the country's economic structure are the ineffective high-energy
intensive economic structure resulting from the long policy of self-isolation, the
high level of material intensity compared to industrially developed countries
and the low efficiency of obsolete and outdated equipment. Electrification is
considered as a driving force of social and economic progress and will proceed
rapidly from 30% today to 33-35% of primary energy demand in 2000. Special
attention is devoted to ecological aspects. 7 First, as concerns the use of
low-grade fossil fuels (in particular coal), large investments are to be under7However, up to now the problem of global warming and possible abatement measures have not
been seriously studied in the Soviet Union. The importance of these issues is presently under
investigation and most likely the Energy Programme will be revised with a view to these subjects.
210
taken to reduce the negative impacts on environment and health. The new
energy policy also stresses the importance of implementing economic mechanisms in the energy sector with a view to promoting self-supporting energy
supply in the future.
The policy suggests that the improvements should be carried out in two
phases. The first phase (up to the year 2000) will be characterized by a
continuous growth in hydrocarbon production and particularly natural gas
(mainly Tyumen gas transported to the European Soviet Union). During this
period, efforts are to be focused on radical improvements in safety and
reducing costs of nuclear energy as well as stable coal production.
In the second stage (first decade of the twenty-first century and perhaps
beyond), hydrocarbon production is to be stabilized with a further growth of
natural gas production compensating for possible reductions in liquid fuel
production. It is not excluded that energy conservation will stabilize nuclear
energy growth. Options for economic development without growth in energy
consumption are now being studied. But such a transition is likely to occur
only beyond 2005~10.
However, the last few years showed that the national energy policy has little
chance of success if the new situation in the USSR is not seriously taken into
account.
12.2.2 Energy saving
These goals are quite impossible to achieve with existing trends and tendencies,
which implies the need for new approaches for solving national energy
problems. Instead it seems more promising to follow the lines of enhanced
energy conservation. According to some evaluations, structural changes in the
productive system could provide a 50-60% reduction in the expected energy/NMP ratio equivalent to savings of 0.6 billion tonnes coal equivalent (tce)
in 2000 and 1.8~2.1 billion tce in 2010 (as compared to1985). The rest is to be
achieved by improvements in energy efficiency (technical limit of energy
conservation is equal to more than 1000 million tce/year in case of full
utilization of all known technologies, i.e., not less than one-third of today's
total energy consumption). 8 When the energy policy was compiled, more than
5000 energy-saving measures were analysed and chosen, of which 70 could
achieve energy savings of 400 million tce in 2000. For example, 50 million
tce/year could be saved by improving industrial furnaces. The same level of
energy savings can be achieved by new small steam and water heating boilers
with automatic control and automatically controlled heat supply systems.
Installation of new lighting devices, controlled electric drives, electric compensating equipment and improved transformers could result in further savings of
8 According to the author's assessment, this saving potential is approximately equal to half of the
total primary energy consumption in the Soviet Union. These figures may be compared to those
in Chapter 4.
211
50 million tce. Waste energy resources, if properly utilized, could save over 20
million tce/year. Therefore, just the above mentioned measures will result in
savings equivalent to the total coal production of the Donetsck region (the
largest coal-producing region in the Soviet Union, with about 200 million
tonnes). Further energy savings are expected with improvements in total
material production. For example, higher quality steel products and changes
in the structure of steel production technologies could save about 70 million
tce/year. Another 40 million tce could be saved in the transportation sector.
Investments will also be needed to implement energy-saving measures.
According to the Institute of Energy Research of the USSR State Committee
on Technology and Science and the Academy of Sciences (Makarov, 1989),
about 200 million tce/year could be saved without additional investments. The
saving potential for a 2- or 5-year payback period is equal to 450-650 million
tce. This will demand new capital investments of 28-30 billion roubles and the
production of new energy-saving equipment should reach 270-300 billion
roubles. Moreover, for savings of up to 600 million tce the specific investments
in energy saving remain cheaper compared to the corresponding investments
in primary energy supply systems. Figure 12.2 shows the dependence of the
900
~
Q)
r---------------------,
700
5 yr
----------:;::::::
.l:l
.sc:
o
:;
3 yr
~500
Q)
til
c:
oo
>-
...
Cl
Q)
c:
w 300
100
100
200
300
Investments in energy improvements (rbi/tee)
Fig. 12.2
Energy saving potential in the Soviet Union. (Source: Makarov and

Bashmakov, 1989.)
Solid
fuels
Liquid
fuels
Primary energy
production
1985
480
845
495-500
1990
900
600--620
950--920
2000
-20
-225
1985
Balance of trade
and changes in
1990 -(40--30) -(260--255)
stocks
2000 -(75--45) -(205-225)
460
620
Total primary
1985
energy consump- 1990
455--470
640--645
tion
2000
525-575
645-665
Electricityb
-130
-80
1985
1990 -(155-165) -(60--65)
-25
2000 -(225-245)
Other forms of
-120
-135
1985
-90
-105
energy conver1990
sion, own use,
2000 -(100--120)
-125
transportation
and distribution
losses
Year
Gaseous
fuels
-45
-60
-(80--85)
60
75
105-120
60
75
105-120
Nuclear
Steam and Other energy
energy Hydropower Electricity hot water
forms
The energy balance of the Soviet Union"
50
70
760
115-85
80
970--1000
190--220
110--115
1110--1255
-90
-10
-15
-(175-145)
-(130--255)
-20
70
40
670
795-855
100--70
80
170--200
110--115
980--1035
-125
-40
-70
+140
-(135-150) -(100--70)
-80 +(160--155)
-(120--150) -(150--180) -(100--155) +(205-235)
-305
+350
-(390--420)
+ (390--400)
-(510--520)
-20
+(495-520)
Table 12.1
2265
2635-2640
2965-3250
-345
-(490-445)
-(430--540)
1920
2145-2190
2535-2710
-305
-(370--37)
-(425--48)
-255
-(255-27)
-(340--35)
Total
1985
1990
2000
Energy delivered
1985
to consumers
1990
2000
Industry
1985
1990
2000
Transport
1985
1990
2000
Agriculture
1985
1990
2000
Residential and
1985
commercial sector 1990
2000
-5
-80
-5
-105
-5 -(115-125)
205
325
205-210
370
195-205
380-390
107
80
109-114
96
95-105
80-85
156
171-174
200-205
64
13
16
73-75
25-20
70
25
85
30-25
80
75-80
30
9-10
15
14
16-20
30-25
60
77-81
90-100
-45
-60
-100
195
210-225
250-265
121
108-114
115-125
140
160-155
205-235
88
99-94
120-140
10
10
15-20
14
17
25
28
34
45-50
5
6
8
10
88
90-95
130
350
390-400
495-520
256
292-297
350-375
bElectricity consumed, i.e. without losses in electricity generation.
a1985 factual, 1990 expected plan targets, 2000 forecast, million tonnes coal equivalent. Data from Volfberg et al. (1989).
Non-energy uses
5-10
5-10
15
15
25-35
15
15
15
-130
-170
-(220-23)
1230
1350-13
1550-16
667
720-73
775-84
166
190-19
235-24
11
130-13
165-16
28
310-31
375-40
214
energy-saving potential on investments. This huge energy-saving potential

could not be realized up to now because of the strong influence of energyproducing ministries. With economic reform, it is hoped that it will be possible
to utilize this potential to improve energy and economic efficiency.
The growing emphasis on energy saving could result in energy demand being
50% lower in 2000 (and 80% lower in 201O) compared with the level which
would have been achieved without energy savings.
12.2.3 Energy supply
Table 12.1 gives a description of the Soviet Union's energy balance over the
period 1985-2000 (Volfberg et al., 1989). The natural gas industry will remain
the most dynamic industry in the fuel and energy complex, its production rate
increasing by a factor of 1.4-1.7 (the gas share in the production of all forms
of energy is expected to rise from 33% in 1985 to 37% in 2000). Natural gas
production may reach 1.2-1.3 trillion cubic metres per year in 2010 (without
accounting for deep gas or gas hydrates). Until 2000-10, natural gas can
successfully compete with nuclear energy in the European part of the Soviet
Union, but beyond that the situation remains quite uncertain. Previously
planned levels of natural gas production for the year 2000 are now considered
as a goal for 1995. The increase in natural gas use will allow an extra 10-15
(maybe even 20) years to search for solutions to the ecological problems arising
from coal and nuclear energy. It seems that the problems connected with
natural gas can be solved more easily.
In contrast, oil production is only expected to stabilize, reducing its share in
supply from 37% to 28-32%, by 2000. It is clear now that 'active' reserves,
which can be extracted with existing technologies, can meet only about 35%
of the country's oil demand in 2010. Passive (hardly recoverable with current
technologies) reserves including those occurring in poorly permeable reservoirs,
in gas condensate fields with oil accumulation, heavy oil reserves etc., will
provide more than half of the production possibilities beyond 2000. This type
of reserves includes virtually half the potential resources in western Siberia, the
PreCaspian Lowlands, eastern Siberia and the Arctic offshore. Their exploitation depends on the development of new technologies in all spheres of the oil
industry. Meanwhile, according to A. A. Trofimuk, a well-known Soviet oil and
gas geologist, the Soviet Union possesses vast hydrocarbon resources permitting long-term forecasts of oil production. The country's undiscovered oil and
gas reserves are believed to be equal to all reserves extracted and discovered
by now. Furthermore, over 60% of the oil is left in the fields formerly
developed or currently being developed. Enhanced recovery technologies will
allow partial exploitation of this share as well. Virtually no exploitation has
begun of offshore reserves, which exceed extracted, discovered and potential
resources on the continent. No use has been made of heavy oil and solid
bitumens deposits, the reserves of which amount to tens of billions of tonnes.
215
There are thus many alternatives but they inevitably result in a considerable
increase in oil production expenditures and therefore stabilization of oil and
gas condensate production (at a level of some 650 million tonnes annually by
2010) is technically feasible, although at a high price. Instead, it appears
expedient to have an absolute reduction in oil production levels beyond
1995-2000 by a few tens of million tonnes per year. This means the development of fields with very low productivity can be avoided, as can the wide-scale
use (at least until 2010) of the most expensive technologies. But even in this
case, investments would rise considerably from 210 roubles/tonne in 2000 to
290-300 roubles/tonne by 2010. Under these conditions the expediency of
maintaining higher production levels will depend on world oil prices. The funds
saved in oil industry development can be better directed to development of
alternative sources for motor fuels and chemical feedstocks, notably those
based on natural gas.
In the early 1980s, motor fuel and fuel oil yields in oil refining amounted to
around 40% each. In the future, motor fuel yields will be up to 60-65%, fuel
oil will decline to 15-17%, and there will be a substantial increase in the share
of feedstocks for petrochemistry and non-fuel products.
This is due to expected changes in the pattern of petroleum product
consumption. Today vehicles account for only 40% of consumption; in the
future their share may rise to 65%. At the same time fuel use for electricity
generation and heat supply is expected to decline from 35 to 12%. Petroleum
product consumption will also be strongly influenced by a shift to more
economic vehicles, electrically powered railway and urban public transport,
and by the use of compressed methane for part of the intra-city freight traffic.
The refining industry will make greater use of thermo catalytic refining
processes and fuel oil hydrogenization by methane-based hydrogen. Great
importance is attached to the production of lead-free gasoline and low sulphur
diesel to reduce the ecological problems of the large cities.
Coal production is to reach 0.7-D.8 billion tce in 2010 (especially owing to
the development of coal deposits located in the eastern part of the Soviet
Union - Kuznetsk, Kansk-Achinsk, Ekibastuz basins). Coal production in the
European part will reach a stable level before 2000 and will start to decline
afterwards. The large-scale users in eastern Siberia and Kazakhstan (power
plants, iron and steel factories, cement works, etc.) consume local solid fuel
which is cheaper than nuclear energy, but ecological constraints limit the use
of coal in those regions with access to natural gas. In western Siberia and
central Asia (in view of seismology and as a result of higher capital costs of
nuclear power plants), Siberian coal will keep its competitiveness with nuclear
energy but not with natural gas (at any rate until 2010-2020). Coal production
will grow very slowly during the next decades because of ecological and social
constraints.
Because of the serious difficulties encountered recently more moderate
nuclear energy growth rates are expected during the next decade as compared
216
to earlier forecasts. The nuclear share in primary energy is now expected to

increase from 2.2% in 1985 to 6.4--6.7 by 2000, which is significantly lower than
was previously predicted. Quality and security improvements in nuclear energy
are to be given priority instead of quantitative growth. However, electricity
generation at nuclear power plants is still expected to almost triple by 2000
and increase steadily thereafter. Nuclear energy is to provide the major part of
electricity demand growth in the European part of the Soviet Union. Even this
completely revised nuclear future now seems too optimistic. The procedure for
choosing construction sites for new nuclear energy facilities must also be
improved (especially by wider public involvement and open discussion of the
problems).
The hydroelectric potential of the European part of the Soviet Union will
practically be fully utilized by 2010 and that of the Eastern region partly so by
2040. Unconventional renewable energy sources are likely to playa significant
role only after 2010. In total, the nuclear, hydro and renew abies share will
amount to 0.4-D.5 billion tce annually in 2010, although their total share in
primary energy production will not exceed 25%.
12.2.4 Summary
Primary energy consumption in the Soviet Union is thus expected to increase
only by a factor of 1.4--1.5 by 2000 (compared to 1985) and 1.5-1.6 in the year
2010 given a growth in NMP of 1.9-2.0 and 3.0-3.5, respectively.9 This means
that the energy/NMP ratio is to decline by 26-27% until 2000 and by 50-55%
by 2010 (see Tables 4.1 and 4.2). Such a policy will allow an approximately
equal primary energy production growth during subsequent 5-year periods but
the structure of this growth will change: if over the period 1985-90 more than
85% of the energy demand growth is met by fossil fuels, then in 2005-10
the share of fossil fuels will be not more than 15-20% and the share of
non-fossil fuels 80-85%.
Final energy consumption is expected to increase by 26-34% by 2000
(industrial consumption growth 16-27%, transportation 42--48%, agriculture
44--48%, and the residential and commercial sector 31--42%). Thus the share
of industry will be reduced from 54% to 51 % but it is still high compared to
the United States of America (less than 33%) and Western Europe (less than
40%).
The policy foresees possibilities for further energy export growth (notably
natural gas and coal with a virtually stabilized export of oil and oil products).
Over the past decades, energy resource export has played an important role in
balancing foreign trade deficit. Over the past 25 years, the export of hydrocarbon fuels has increased considerably: in 1988 oil and oil product exports
9These goals are established much lower now because of economic crises at the beginning of the
1990s and uncertainties about the success of economic reforms in the Soviet Union. However, the
general tendency in the national energy policy is expected to remain.
References
217
amounted to 33% of the oil produced compared to 12.1 % in 1960; for gas,
shares were 11.4 and 0.5%, respectively. In 1988, oil and oil products exports
to capitalist and socialist countries totalled 144 million tonnes, oil products 61
million tonnes and gas 88 billion cubic metres, accounting for about two-thirds
of total hard currency earnings.
Recently, the policy of exporting hydrocarbons, particularly to market
economies, was strongly criticized by Soviet economists and politicians. But
this policy will certainly have to continue over the next decades until the
national economy is restructured and the quality of products raised to world
market levels.
Here it should be noted that interpreting raw material exports as a sign of
backwardness of a country is hardly justified. The modern world is characterized by the growing division of labour. Autarchy is no alternative. Developed
nations are striving for mutually advantageous cooperation in all fields, not
excluding trade in raw materials. Efficient production of energy resources
nowadays uses just as much high technology as for example, electronics.
Rough estimates show that average labour productivity will increase by 30%
over the next 20-25 years. This will result in NMP growth of 1 trillion roubles
over the period 1986-2000. The realization of the new energy policy will
demand about the same level of capital investments (including 800 billion
roubles in the development of the fuel and energy complex, 50-80 billion
roubles in energy conservation and the use of unconventional sources of
energy, 30 billion roubles in the development of appropriate machinery
construction industries, and some 90-100 billion roubles to meet the social
needs of operational 'staff). However, the new energy policy with an orientation
towards energy conservation and natural gas results in higher economic
efficiency. Due to structural changes and improvements in the energy systems,
more than one million people now employed in the energy supply sectors will
be shifted by 2000 to more productive and efficient employment. At the same
time the policy sets the task of improving the ecological situation in big
industrial cities of the country by reducing the amount of hazardous pollutants
emitted by power-generating units by a factor of 1.5 by 2000 and more than
twofold over the following decade.
Needless to say, the elaboration of national energy programmes and policies
has required extensive use of modelling efforts on all the levels of the Soviet
hierarchical energy system: from industrial and agricultural enterprises and
households to the top-level planning authorities.
REFERENCES 10
Aganbegian, A. G. and Fedorenko, N. P. (eds) (1978) General Recommendations for the
Optimization of Industrial Development, Nauka, Moscow.
10 All
publications cited are in Russian.
218
Astakhov, A. S., Gitin, E. M. and Saratovsky, E. G. (1967) Long-Term Optimization of

Coal Industry with the Use of Linear Models, in Proceedings of the All-Union
Conference on the Application of Computers and Mathematical Methods in Industrial
Planning. Tallinn.
Beliaev, L. S. (1978) Solution of Complex Optimization Problems under Uncertainty.
Nauka, Novosibirsk.
Druzhinin, E. P. and Kuznetsov, Yu. A. (1970) The Optimization of the State Natural
Gas Supply System. Siberian Energy Institute, Irkutsk.
Energy Programme (1984) Main Principles of the USSR Energy Programme for the
Long-Run. Politizdat, Moscow.
Feigin, V. I., Frolova, E. P. and Pevzner, E. V. (1979) Modelfor Long-Term Planning of
Natural Gas Supply Systems with the Optimization of the Seasonal Storage Capacity.
Research Institute of Economics and Management in the Natural Gas Industry,
Moscow.
Kluev, Yu. B. (1970) Long- Term Energy Development of an Industrial Enterprise. Ural
Filia of the USSR Academy of Sciences, Swerdlowsk.
Kononov Yu. D. (1981) Energy and Economic System. Transition to New Energy Sources.
Nauka, Moscow.
Makarov A. A. (1989) New Concepts of Energy Development in the USSR, Energia, 4,
14-17.
Makarov, A. A. and Bashmakov, I. A. (1989) The Soviet Union: A Strategy for Energy
Development with Minimum Emission of Greenhouse Gases. Institute of Energy
Research, USSR Academy of Sciences and State Committee for Science and Technology, Moscow.
Maksimov, Yu. I. (1979) Network Models in Long-Term Industrial Planning. Nauka,
Novosibirsk.
Markovich, I. M., Brailov, V. P. and Denisov, V. I. (1962) Applications of Mathematical
Programming to Long-Term Electricity System Development (in Russia). Izvestia AN
SSSR. Energetika i Transport, 6, 5-13.
Melentiev, L. A. (ed.) (1971) Mathematical Models for the Optimization of the Electricity
Generation Development. Siberian Energy Institute, Irkutsk.
Melentiev, L. A. (1979) Energy Systems Analysis. Nauka, Moscow.
Nekrasov, A. S. (ed.) (1981) Optimization of the Fuel-and-Energy Complex. Energoizdat,
Moscow.
Nekrasov, A. S. and Sin yak, Yu. V. (1979) Management of Plant Energy Systems.
Energia, Moscow.
Nekrasov, A. S., Kretinina, Yu. S., Ershevich V. V. (1973) Base Load Thermal Power
Plan Siting (in Russia), Electric Stations, 8, 33-6.
Smirnov, V. A. and Garlyauskas, A. V. (1969) General Approach to the Optimization and
Control of the Natural Gas Supply System. Research Institute of Economics and
Management in the Natural Gas Industry, Moscow.
Sovalov, S. A. (1983) Regimes of the Unified Electricity Supply System. Energoatomizdat,
Moscow.
Starovoitov, S. N., Andreeva, L. A. and Barmina, S. N. (1974) Joint Optimization of
Crude Oil Production, Refining and Processing, in Optimization of Industrial Development, N auka, Novosibirsk.
Stavrovsky, E. R. and Sukcharev, M. G. (1978) Hierarchical Modelling System for
Natural Gas Planning, in Hierarchy of Large Energy Systems, Siberian Energy
Institute, Irkutsk.
Syrov, Yu. P., Makarov, A. A. and Zeiliger A. N. (1966) Linear Models for the
Optimization of the Electricity Generation System (in Russian) Teploenergetika, 10,
24-26.
References
219
USSR Academy of Sciences (1975) Methodology for the Optimization of the Fuel and
Energy Complex. Nauka, Moscow.
USSR Academy of Sciences (1977) Recommendations for the Optimization of Regional
Energy Balances. Moscow.
Volfberg, D. B., Demirchan, K. S., Klokova, T. I. et aI., (1989) USSR Energy Balance,
Izvestia AN SSSR. Energetika i Transport, 1, 3-7.
---13
A detailed simulation approach
to world energy modelling:
the SIBILIN and POLES
experIences
Patrick Criqui
The 1970s and 1980s have been marked by drastic changes in the international
energy markets. The early 1990s have seen the return of the oil dependency
problem, but new issues are arising, in particular the concern for the planet's
global environment. To understand these changes and to identify appropriate
strategies for the new challenges, structured models, relying on good quality
information and retrospective analyses, are highly valuable tools. In this paper,
we describe the effort that the IEPE (Institut d'Economie et de Politique de
l'Energie in Grenoble) has undertaken to attain these goals. In the first part,
we analyse the specific approach developed, that of dis aggregated models for
the simulation of the world energy system. We then describe the structure and
results of the SIBILIN (SImulation des BILans energetiques INternationaux)
model, which had been developed to carry out medium-term oil market
scenarios. The third and final part presents the POLES (Prospective Outlook
on Long-term Energy Systems) model, presently being built to analyse longterm energy scenarios and their potential consequences on the global environment.
13.1 A DISAGGREGATED SIMULATION APPROACH TO
INTERNA TIONAL ENERGY MODELLING
The margins for improving our understanding of the state and dynamics of the
world energy system are wide. Energy models have been developed at the IEPE
precisely because modelling is a learning process. This explains why disaggregated models of the recursive simulation type have been chosen. It also
accounts for the use of the scenario approach instead of predictive forecasts
222
The SIBILIN and POLES experiences
and, last but not least, for the fact that such a modelling effort relies on detailed
data bases and retrospective analyses. The basic structure and logic of the
models correspond to the concept of hierarchical 'nearly decomposable'
systems, as identified by Simon (1962). World energy markets (level 1) are fed
by imports and exports from a set of national and regional energy models (level
2) whose subsystems correspond to the main elements of the energy balance,
i.e. final consumption by sector, energy transformation sector, primary energy
production (level 3). Finally these subsystems rely on exogenous information
stemming from more detailed sectoral or 'bottom-up' studies.
13.1.1 Geographical disaggregation: the national simulation approach
Most world energy models are highly aggregated. The Energy Modelling
Forum Study 'World Oil' (EMF, 1982) shows that the most detailed model
includes 16 production and consumption regions. Most other models include
five regions, the least detailed have only two regions. Among more recent
models, dealing with climate change issues, the IEA-ORAU model (Edmonds
and Reilly, 1985) incorporates ten regions, Global 2100 (Manne, 1990) five
regions and Nordhaus and Yohe (1983) study long-term energy scenarios for
the world as a whole (see also Table 14.1). The problem is, in fact, to choose
an optimum disaggregation level while taking into account:
1. the cost of gathering and managing detailed information and the benefits of
being able to identify the particularities of the different elements of the
model;
2. the fact that the consistency of the model might decline when the number
of exogenous hypotheses and causal relationships increases.
National energy balances are the main tool for a disaggregated approach to
world energy modelling, since they enable us to take country-specific factors
and constraints into account. Some 94% of total world energy is consumed by
only 40 countries and energy balances for the last 20 years have been
constructed for most of these countries by national and international institutions (lEA, UN-EeE, OLADE, Asian Development Bank), as well as by
research groups (International Energy Research Group - Lawrence Berkeley
Laboratory or ENERDAT A-IEPE).
This is why the IEPE's world models are based on the simulation of national
and regional energy balances. These balances are connected to the upper level
- international energy markets - by energy exchange flows and prices. Their
basic dynamics derive, however, from the lower level - national energy
subsystems (production, transformation and final consumption) in which
country-specific technico-economic variables and energy policies are taken into
account. In spite of its requirements in terms of data collection, this approach
leads to greater consistency than more aggregated world energy models. This
A disaggregated simulation approach
223
is especially true for medium- and long-term energy issues, where national
energy policies playa particularly important role.
13.1.2 Modelling the markets: simulation rather than optimization

As regards international energy markets, the pros and cons of simulation
versus optimization/wealth-maximization approaches have been thoroughly
discussed by Gately (1984), in his survey of world oil models. He addresses in
particular the controversy between theoretical models based on the Hotelling
tradition, and empirical models which use recursive simuiation processes to
determine world oil prices.
According to Gately, the intertemporal wealth-maximization approach,
although internally consistent, involves highly disputable hypotheses and
simplifications, such as perfect information about future price and income
elasticities, or the neglect of lags in the adjustment of demand and supply. On
the contrary, despite their apparent simplicity, recursive simulation models of
the oil market make it possible to take into account the unavoidable uncertainty which OPEC is currently facing, as well as the 'rule of thumb' reaction
functions (for instance, a rise in price when the capacity utilization rate is
rising) which it is forced to adopt in a highly complex situation.
Similar reasoning was used by Criqui and Kousnetzoff (1987). Their main
findings for the 1970-87 period were as follows:
1. Oscillations in oil price basically depend on the capacity utilization of the
world's swing-producers (OPEC) and more particularly of the large reserve
countries (OPEC Core, i.e. the Gulf).
2. The mere direction in the movement of oil prices is given by market
fundamentals, but the magnitude of the variations and the levels obtained
very much depend on the current geopolitical context, as well as on the
organizational structure of the oil industry and international market.
3. The price-production strategy of the swing-producers in the medium term
cannot be analysed simply within the framework of wealth-maximization,
but must also include such factors as absorptive capacity and political
rivalry.
This is why the combination of 'behavioural' simulation models, driven by
the market fundamentals, with scenario hypotheses concerning the tactics and
strategy of the core swing-producers seems to be the most adequate solution.
Thus, it is necessary to avoid a fully integrated model structure, in order to
leave room for judgement in the building of scenarios.
13.1.3 Modelling demand: bottom-up or top-down?

As regards the dynamics of energy demand modelling, there is the well-known
'top-down versus bottom-up' controversy (see Chapter 2). The first approach
224
emphasizes the role of strictly economic variables (mainly prices and income)
and strictly economic behaviour (cost minimization and expenditure optimization). It is strongly rooted in the neoclassical economic paradigm. Final
energy demand forecasts derive from the use of price and income elasticities.
The second approach is often referred to as the engineering approach to energy
demand. In this case, demand forecasts are mostly based on assumptions about
energy using equipment (boilers, cars, domestic appliances). This bottom-up or
end-use approach implies a finer degree of disaggregation, in order to take into
account the basic needs which energy has to satisfy as well as the corresponding economic structure and the technology applied.
Clearly, strictly economic and technological variables are both important in
analysing and forecasting energy demand. This is why any energy demand
model should to some extent incorporate and link the two sets of variables.
One possibility may be to use strictly economic variables for short- to
medium-term studies, while structural and technological variables have to be
explicitly taken into account for long-term outlooks: empirical evidence is
given by long period historical analyses of energy intensity of GDP, which
show short periods of stability within a long-run increasing or decreasing
trend, as identified by Martin (1988).
Thus intermediate models, placing greater or lesser emphasis on economic
and technological variables according to the time horizon considered, would
appear to be best suited. One can note that the inclusion of non-economic
mechanisms or variables implies exogenous hypotheses on changes in the
technological or consumption patterns and hence, the need to use simulation
processes.
13.1.4 Energy simulation models and Strategic Planning methods
For strategic planners such as Dumoulin (1988), 'the future cannot be foretold
but it can be structured'. In fact Single-Line Forecasting is not adequate in
times of shocks and strong perturbations of the economic system. On the
contrary, the basic methods of Strategic Planning, which consists of checking
different strategies against different consistent states of the world (Le. scenarios), proves more relevant in a situation of high uncertainty. This is of
paramount importance in corporate planning, an area where neglecting the
'phantom scenario' (the worst case) can be fatal to the organization. But it also
holds true for policy-making.
'Structuring the future' means first of all taking into account the predetermined elements. Even for long-term studies, they are often more important than first thought: demography is one example, but capital stocks,
equipment under construction and technologies under development also illustrate this fact. Thereafter, structuring the future means concentrating on major
uncertainties and combining them, so as to identify those states of the world
which seem internally consistent. All this can be done without a completely
A study of the medium-term world oil market
225
formalized and quantified model. However, models - particularly detailed

simulation models - can be extremely powerful tools to address these tasks.
Models can be viewed in some respects as 'future-oriented data bases'.l This
is the case when they incorporate the relevant predetermined elements within
their exogenous hypotheses. What really matters is the quality of the information concerning future changes and therefore this information has to be
validated by independent experts on a sectoral and geographical basis. When
modelling is considered as a learning process, dis aggregated simulation models
do in fact provide a means of incorporating all relevant existing information
when building scenarios and testing strategies.
13.2 A STUDY OF THE MEDIUM-TERM WORLD OIL MARKET
USING THE SIBILIN MODEL
In the early 1980s the CEPII (Centre d'Etudes Prospectives et d'Informations
Internationales) and the IEPE decided to undertake a joint research programme in order to study the interactions between economics and energy, in
an international perspective. As part of this programme, the IEPE constructed
a simulation model for international energy balances (SIBILIN) and the CEPII
elaborated a set of procedures which made it possible to provide an image of
world energy trade in monetary terms, while working from energy balances in
physical units.
13.2.1 The SIBILIN model: structure and data bases
The major feature of this joint research is that, rather than using aggregated
models, it is based on detailed scenarios in both the economic and the energy
fields. They provide information on each main region of the world. The second
characteristic is that the modelling and data bases on specific issues are linked
to analyses of past trends and events. These analyses make it possible to take
into account the factors or events that cannot be directly modelized (see Figure
13.1).
The SIBILIN model is not therefore an integrated energy-economy model,
comparable to other models developed for France, on a national level (see
Chapter II by Destais). This arises from the difficulty of formally linking a
detailed international energy model with a detailed economic model. However,
it also reflects two preoccupations: that of using models as tools in order to
incorporate information in a form which is both coherent and structured; and
that of keeping open the possibility of simulating contrasted scenarios, particularly with regard to the hypotheses concerning behaviour patterns and strat'Future-oriented' should here be understood in the sense of the word 'prospective' in French, i.e.
the contrary of retrospective.
226
Exchange Rates
Inflation
OIL PRICE
ENERDAT A data base
Fig. 13.1
SIBILlN, linking international energy and economic scenarios.
tegies. As regards the energy markets representation, SIBILIN places itself

deliberately within the family of recursive simulation models and not among
the intertemporal optimization models.
The model is based on the energy balance accounting logic. It first of all
works back from the simulation of final consumption by sector and source to
primary consumption, having integrated transformation losses in the energy
sector on the one hand, and the structure of the electricity production system
on the other. It then calculates the import demand or the potential export
supply of the zone under consideration as the difference between its consumption and its production capacity. The balance sheets are simulated up to 1995,
and individual results are given for 24 OECD countries, 20 developing
countries and eight Eastern European countries, including the Soviet Union.
Along with these 52 national balance sheets, simplified primary energy balances are also provided for 16 geographical areas. These zones make it possible
to cover the whole world in a way which is compatible with the CEPIICHELEM economic data base.
227
The demand submodels have an intermediate structure between the classical

econometric models (top-down approach) and the technico-economic models
of the MEDEE-type bottom-up approach (see appendix to Chapter 10).
Indeed, the explanatory variables (prices, value-added and income) and functional relations are those of econometric models, but they are applied within
the framework of a disaggregation into homogeneous consumption modules,
as is the case in technico-economic models.
In fact, the logic is very similar to that used in the structural analyses of past
demand. 2 In the model applied to the OECD countries, the demand equations
are produced by dis aggregating final consumption into 20 modules (energy
sectors and/or groups of products) and then by breaking down the consumption of each module into three indicators (energy intensity, structure and
activity). Thus:
FE= 'iJEi= 'i,(FEi/VAi) x (VAi/GDP) x GDP,
i
where FE stands for total final energy consumption, FEi and VAi for the energy
consumption and the value-added of sector i, and:
F Ei/VAi = f(VAi, EPi, T),
where EPi stands for the average price of energy in sector i, and T for a time
trend.
Moving from final consumption up to source-by-source primary consumption involves taking into account the efficiencies of the various energy chains
as well as the structure of the thermal power plant system. Efficiencies are
extrapolated from past trends on the basis of national energy balance sheets.
Market shares of the various categories of power stations (coal, oil, gas) are
drawn, whenever possible, from national energy programmes or plans.
The hypotheses concerning the production capacities of the various energies
are not derived from a proper supply model. Given the time horizon used in
the forecast (5-10 years) and the lead times for energy production facilities, it
was more efficient to use exogenous hypotheses. A first 'production capacity
data base' was therefore set up for all the major energy producing countries,
using national energy programmes or international studies. This base was later
completed, modified and updated after discussion with experts from major
French energy companies.
13.2.2 A look back at key uncertainties and oil scenarios of 1987
The main assumption made on the basis of retrospective analyses and used to
build the scenarios was that oil price is a function of the ratio of OPEC
production to OPEC capacity on the one hand, and of the strategy adopted
2See for instance the analyses of Chateau et al. for Europe, Schipper et aI, for the United States
and Matsui for Japan, in Energie Internationale 1990-1991, Economica, Paris, 1990, pp. 87-98,
99-110 and 111-23.
228
by the large reserve core-countries on the other. When the capacity utilization
ratio is high (over 80%, or 28 million barrels per day (Mbd) production on the
basis of a 34 Mbd OPEC capacity), there is a high probability of a price hike.
When it is low (under 60%, 20 Mbd production), the price is very likely to
drop.
As regards the future as it appeared in 1987, the first uncertainty concerned
economic growth. At that time, analyses indicated that world medium-term
growth would be moderate or low, depending on whether the major economies
of the OECD managed to elaborate a concerted policy for re-absorbing current
trade imbalances. The second uncertainty concerned the future strategy of
OPEC: would the Organization manage to restore sufficient internal cohesion
in order to defend oil prices by limiting output? The lower the level of
economic growth, and therefore of oil consumption, the more acute this
question would become. If internal discipline could not be upheld, and if oil
prices fell once again in the short term, then a final and major uncertainty
appeared: what would be the impact on world supply and demand of a very
low oil price over several years? Taking these uncertainties into account
resulted in the tree-matrix of the scenarios (Figure 13.2).
What strategy for OPEC: price defence or market share defence?
The lesson of the 1986 counter-shock was that the weaker the demand for
OPEC oil, the greater the sacrifices required of swing-producers to defend the
price and the greater is the risk of declared or latent price wars aimed at
1. World economic growth
2. OPEC discipline
(oil price)
3. Demand and supply

price response
1\
A \
LOW
HIGH
MODERATE
LOW
HIGH
~ SHO~
o
RUN
(Reference case)
LONG
R~
Fig. 13.2 Three levels of uncertainty, four scenarios.
229
defending a new target: production levels or market shares. The simulations

carried out in 1987 with the moderate growth scenario were taken to be the
reference case. They showed a relatively low but regular growth in demand for
OPEC oil which could be a sufficient condition for a return to price regulation
by output control, since OPEC production levels would rise progressively to
26 Mbd, with a price of $24 per barrel (1986 dollars) in 1995. But in the case
of low economic growth with internal discipline, the results did not show any
short-term improvement in OPEC's oil revenues and production levels (18
Mbd in 1990,24 Mbd in 1995). This is why the exogenous hypothesis of a new
break in discipline and oil prices was introduced (10 $/bl between 1987 and
1990), followed by a rise to 25 $/bl in 1995. Two scenarios were then built in
order to analyse the consequences of this new price war situation.
What price-response for oil supply and demand: short-term or long-term?
The impact of a low price on energy consumption was computed directly by
the model using the incorporated price elasticities. However, an additional
hypothesis was introduced for those sectors in which inter-fuel substitutions
are possible: the substitution of other fuels for oil was barred as long as the
price remained under the 15 $86/bl threshold. The overall result was an
additional 2 Mbd consumption in 1990.
As regards oil supply, two hypotheses were considered
1. The first involved a rapid reduction in non-OPEC production ( - 2 Mbd in

1990), but also a quick revival after the real price upturn; this short-run
response scheme resulted, paradoxically, in a fairly balanced situation for
1995, with OPEC production at 24 Mbd.
2. In contrast, the second hypothesis was a delayed response situation, with
no short-run impact of low prices, but, because of lack of investment, a
4 Mbd reduction of non-OPEC production ( - 9% on the initial hypothesis)
in 1995; the result was clearly a return to a tight market with OPEC
production at 28 Mbd in 1995.
The four scenarios briefly described here were an attempt to depict consistent dynamics on the oil market while combining clear-cut hypotheses
concerning the main uncertainties. They can now be compared with the actual
changes since 1987, in order to identify mis-specified hypotheses, processes or
behaviour patterns and to enhance our understanding of the world oil market.
13.2.3 The SIBILIN scenarios as compared with actual events
From 1987 to the first half of 1990, world economic growth has been higher
than assumed in the SIBILIN scenarios. This can be explained by the stronger
than forecasted delayed macroeconomic effects of the 1986 counter-shock on
230
the one hand and by the fact that the structural adjustments on the US trade
and budget deficits have still been postponed on the other.
During this period, and in spite of a relatively high-growth environment,
OPEC did not succeed, or maybe did not even intend, to establish a firm
price-defence strategy. With some oversimplification one might say that 1987,
1988 and 1989 have been successively years of high, low and again high
discipline inside OPEC. Is this just the way cartels function (agreementcheating-agreement)? Or is it in some way intentional, aimed at obtaining an
intermediate price-path between price defence and price war?
This cyclical evolution might in fact have provided the core-countries with
a means of obtaining the same results as would have been obtained by a
technically difficult and politically disputed 'fine-tuning' of the oil market (see
AI-Chalabi, 1988). In any case, this price-path, with some very low points (less
than 12 $86/bl) in 1988 and again in the first half of 1990), clearly had an
impact on world oil supply and demand, accelerating the upturn in the call for
OPEC oil. This explains the fact that actual OPEC production levels have been
higher than proposed in the scenarios: 22.6 Mbd in 1989, against at most
22 Mbd in the low-growth, price war and short-run price response scenario.
The outlook has changed as a result of the Gulf crisis. A political event has
once again upset the oil market, demonstrating that oil importing countries
remain structurally vulnerable. But we have learned from the second oil shock
that oil prices can also go down. This has happened again since the crisis was
over, simply because an OPEC production level of 23 Mbd is not at all a
critical one in a medium-term perspective. During each period on the oil
market, a 'consensus price' appears among experts, forecasters and companies.
This was of 18 $/bl between 1986 and 1989. It is currently of 25 $/bl. This is
not very far from the 24 $/bl (20 1986 dollars) which was indicated for 1990 in
the SIBILIN scenarios reference case.
13.3 POLES, A TOOL FOR LONG-TERM
ENERGY-ENVIRONMENT STUDIES
Although the hypothesis of a possible climate change linked to fossil energy
consumption was first proposed in 1896 by S. Arrhenius (Grinevald, 1990), it
was given relatively little attention until the 1980s. At that time important
advances by climatologists began to show that something with important
consequences for the planet might be taking place. Energy forecasting and
modelling efforts also began specifically to address the issue of global climate
change.
A number of global energy models and forecasts already study this problem
and the strategies that might be elaborated to manage it. But their results in
terms of long-run energy consumption vary greatly: for example, the World
Energy Conference 1989 report points to a 14 million tonnes of oil equivalent
(Mtoe) consumption in 2020 (Frisch, 1989), while the Energy for a Sustainable
POLES
231
World study indicates only 8 Mtoe (Goldemberg, 1988). In order to understand

these differences and to reduce the level of uncertainty, the IEPE decided to
build a new model, POLES (Prospective Outlook on Long-term Energy
Systems). The main aim of the model is to produce detailed and consistent
long-term energy scenarios whose environmental consequences will then be
assessed by complementary studies. For a general comparison of POLES and
SIBILIN, see Table 13.1.
Table 13.1
Main features of the two models
SIBILIN
Character
Time
Periodicity
Geographical
breakdown
Use
Economic agents
Sectors of
economic activity
Data base
Endogenous
Exogenous
Size (number of
equations)
Software
Econometric and technical

relations
Medium term (5-10 years)
POLES
Technical and econometric

relations
Medium/long term (2010)
and very long term (2030)
Yearly
Yearly (1990--2010) and
5 years (2010--30)
32 world regions
14 world regions and
and 52 countries
14 or 43 countries
Projections
Projections
Scenario analysis
Scenario analysis
Backcasting
Energy consumers
Energy consumers
and producers
Three industries
Four industries
Three transport modes
Four transport modes
Residen tial
Residential
Tertiary
Tertiary
Agriculture
Agriculture
Supply sector
IEA-OECD
ENERDATA
UN-ECE
(IEA-OECD,
CHELEM-CEPII
CHELEM-CEPII
national data)
Energy demand
Energy demand
Energy exchanges
Energy supply
Energy exchanges
International energy prices
International macroeconomic International macroeconomic
variables
variables
Technical coefficients
Technical parameters
Capacity expansion plans
Energy--environment
Oil price
policies
42/0ECD country
Not yet available
Not yet available
15/Developing country
5/Centrally planned economy Not yet available
MODULECO + specific
TROLL
232
The SIBIL IN and POLES experiences
13.3.1 The main goals: to reduce uncertainties and identify the margins for
action
The model will produce world energy scenarios for the long term (2010) and
the very long term (2030). This corresponds to short time periods for climatologists and to very long ones for economists and decision-makers. However, it
is extremely important to identify what are the predetermined elements for
these next decades, in order to concentrate, in a second stage, on those
variables which could provide scope for freedom in policy-making.
The long lead-time and life-time of energy production and conversion
facilities is a well-known characteristic of the energy sector. For instance,
power plants take 5-10 years to build and may remain in operation for 30
years or more. This means that many electricity generation plants whose
construction will be decided on over the next 10 years, will still be in operation
in 2030. In the same way, given the lead-time for the research, development
and diffusion of new energy technologies, it is probable that the major elements
of the 2030 technological systems have already been identified, either at the
laboratory or at the pilot-plant stage.
Inertia also exists, but might be less important for consumption devices than
for production and conversion equipments (Chapter 3). A simplified end-use
approach to energy consumption, based on the analysis of industrial waves
(Piatier, 1989) and of corresponding consumption patterns (from the building
of infrastructures, to household and individual transport equipment and
information) will be used as a framework. Hence the simulation of final energy
demand will result from country-specific energy paths but also from international comparisons aimed at identifying trends or possible saturation levels for
each main sector: industry, transport and residentia1. This should of course be
linked to an analysis of the technological systems that might contribute to the
satisfaction of these needs.
As for possible areas of action, it is clear that the limitation of energy
consumption by way of constraints or scarcity does not provide a solution. The
low level of satisfaction currently observed in developing countries today
derives not only from the lack of consumer purchasing power, but also from
the mere unavailability of energy and infrastructures. Because of debt and
financial constraints, these situations might worsen and extend into the future.
These facts should of course be taken into account in the mode1. However, the
diffusion of energy-efficient technologies, not only at the level of consumption
but also at the production and conversion levels, clearly appears to be the
correct solution for energy policies addressing environmental issues. Many new
technologies are already known: their technical feasibility has been proved, but
the role of R&D policies remains crucial in making them cost-effective.
Different hypotheses on the diffusion rates of new energy technologies will be
at the core of the normative scenarios simulated using the POLES mode1.
POLES
233
13.3.2 The need for a global cost-effectiveness analysis

As concerns global climate change issues, strictly national greenhouse gas
(GHG) abatement policies make little sense unless they are integrated within
an international perspective. For a country willing to spend several billions of
dollars on reducing GHG emissions it should not matter if the investment is
made inside its own frontiers or in a distant foreign country (unless indirect
advantages are anticipated). A policy of cost-effective investment should imply
a comparison of such different options as overinsulation of buildings in
countries of the North versus improved cooking-stoves or hydro-electric
development in countries of the South. For a global problem, a global
assessment scheme is required and the concept of national energy policy has
to be revised, or at least be placed in a wider perspective.
The global cost-effectiveness approach might become a key aspect of
international global climate change policy issues. In fact other mechanisms
such as a reduction in proportions to past emission levels or stabilization of
emissions on a per capita basis clearly appear to be unfair. They would prove
highly disputable in the eyes of the countries of the South, since per capita
fossil fuel consumption is ten to 20 times higher in industrialized countries. 3 In
contrast, the financing of low-GHG energy technologies, wherever they present
a low investment per unit of avoided GHG, may prove logical for countries
both in the North and in the South. It might also ease the financial constraint
which in many cases prevents developing countries from choosing environmentally sound solutions. By enabling the study of the consequences of
technology diffusion scenarios, the POLES model aims to provide insights into
the analysis of these issues.
13.3.3 The structure of the model 4
The structure of the model is again based on the hypothesis that the world
energy system can be considered as a 'nearly decomposable' hierarchical
system. At the first level, energy markets are simulated by taking into account
the fundamentals for each internationally traded energy, as well as the
strategies of the main protagonists. At the second level, national energy
balances make it possible to put together primary energy consumption and
production in order to compute the exchanges. At a third level, submodels
simulate energy consumption, transformation and production in each country
as a function of technical and economic variables and hypotheses.
3See the analyses in 'World Status: Environmental Taxation?', in Financial Times Energy
Economist, no. 100, February 1990, pp. 14--22.
4For further detail see POLES (Prospective Outlook on Long-term Energy Systems). Maquette,
IEPE, Grenoble, March 1990, p. 112.
234
Level 1: international energy markets

Because it generates the basic hypotheses concerning energy prices, the
simulation of energy markets simply cannot be bypassed in any long-term
energy study. For the mediumjIong term, it is considered that the Gulf region
will remain the world swing-producer for oil. The links between the oil price
and the price of the other internationally traded fuels - coal and gas - are
complex. However, we assume that, as long as strong inter-fuel competition
exists, the price of oil will remain the leading price for energy, while also taking
into account the fact that its variations are bounded by limits imposed by theother fuels, especially by coal. This can be summed up as follows:
The price of oil is a function of the production level or capacity utilization
rate of the Gulf countries (60% of total world reserves, at very low cost) and
of the trade-offs made by those countries between two contradictory goals price defence and market share defence (See Figure 13.3).
The price of coal is a linear function of the price of oil, though it remains
somewhere between a floor and a ceiling price corresponding to the respective production costs of the low-cost coal exporter (currently Australia) and
of the high-cost exporter (USA-East Coast); the conditions affecting the
values of the oil/coal relative price thus impose boundaries on oil price
levels.
The price of gas is a linear function of the price of oil; its floor price
corresponds to the export cost for each large regional market (North
America, Europe, Asia).
Oil/coal price ceiling
Oil price ($86/bl)
:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:;:;:;:;:;:.:.:;:.:;:;:::;:;:.:.:;:;:.:;:.:;:.:.:.:.:.:.:
40
85
LEe
30
~:~
20
capacity
10
66
04-------------r------------r------------,
12
20
28
36
OPEC production (Mbd, 2 years moving average)
Fig. 13.3 The 'Snail', past evolutions and future scenarios for the oil price. (Sources
for 1966-89: OPEC Statistical Bulletin; OECD Economic Perspectives; PIW.)
POLES
235
The logic of price determination is different in the very long term. We

suppose less overcapacity in oil production and less inter-fuel competition as
oil consumption becomes concentrated on captive uses (transport fuels and
chemical feedstocks). Price levels should thus be explained in terms of fuelspecific conditions, i.e. the marginal cost of developing new resources for each
internationally traded fuel. In this case, the price model simulates the progressive mobilization of more and more expensive resources, using long-term supply
curves and taking into account the impact of technological progress on costs
(Bourrelier, 1990).
Level 2: national energy balances
This level provides the accounting framework within which information is

organized in the most relevant way. Within the POLES model it will also make
it possible to analyse the impact of the development of energy systems: degree
of energy dependency, macroeconomic cost of energy imports, investment for
energy production and transformation systems etc. At this stage it will be
possible to supersede the mere extrapolation of existing trends by simulating
different national or regional energy policies.
Level 3: energy demand, transformation and production subsystems
The energy demand modules first take account of such nearly predetermined
elements as demography, for which international organizations (UN, World
Bank) provide structured and detailed long-term forecasts. Final energy
demand is broken down into a dozen end-uses or sectors. For each module
two standardized indicators - a specific consumption indicator and an activity/income indicator - have been developed. They enable us to study the
country-specific dynamics of energy demand variables as a function of price
level. They also make it possible to take advantage of international comparisons in order to construct hypotheses concerning long-term trends or saturation levels. Last but not least, these modules will simulate the consequences of
different rates of adoption of new energy-efficient technologies. This can be
done either explicitly, through hypotheses on the specific consumption of
vehicles for instance, or else implicitly, through evolutions of the specific
consumption indicators based on more detailed bottom-up studies.
Electricity generation is at the heart of the transformation subsystems, at
least within the medium- to long-term horizon, while in the longer term other
conversion options (synfuels, hydrogen, fuels from renew abies) might acquire a
more significant weight. Two different methods are used to simulate the
evolution of electricity generation plants. In some countries, the intertemporal
cost-minimization of electricity supply, based on fuel prices and demand
expectations, is still the actual framework for investment planning. In this case,
a simplified linear optimization model is used to simulate utilities' investment
236
behaviour. In many countries, however, particularly developing countries,

constraints on expansion plans are such that decisions clearly do not derive
from a real optimization process. These constraints range from finance availability, to national resources implementation, environmental limits and social
acceptability. In these cases, a more simple approach is used which takes into
account existing programmes, natural resources (hydro) and economic or
environmental constraints.
In the mediumflong-term perspective, fossil fuel supply is investment-driven.
The combination of simple cash flow models (from fuel prices to investments)
and discovery models (from investments to reserve additions and new capacities, see also Chapter 8 by Rodriguez Padilla) provides the basis for endogenizing oil and gas supply, while coal supply is the result of investment in
production and transport. In the longer term, the approach is simplified,
inasmuch as supply is demand-driven, and the price is deduced from the
marginal cost. Thus, instead of being a driving variable in the model, energy
supply investment becomes an ex post computed variable.
13.4 CONCLUSION
The approach to world energy modelling developed within the SIBILIN and
POLES models presents some particular features. Both models are relatively
simple but dis aggregated, while most other international energy models are
more complex although more aggregated. The building of a detailed model
implies large investments in data bases and this effort has been continuously
sustained by the IEPE in the past years. Its counterpart is the possibility of
organizing the model on the basis of a hierarchical structure of interdependant
subsystems and of incorporating that information which is relevant to the
evolution of these subsystems. Last but not least, the approach is well-suited to
the building of contrasted scenarios, in order to structure the images of the
future which are needed for policy-making.
REFERENCES
AI-Chalabi, F. 1. (1988) OPEC and the Present Structural Limitations on Its Oil Price
Control. OPEC Review, 12, 115-21.
Bourrelier, P. H., Boy de la Tour, X. and Lacour, 1. 1. (1990) L'Energie a long terme:
mobilisation ou laissez-faire? Revue de I'Energie, no. 418, February, pp. 81-117.
Criqui, P. and Kousnetzoff, N. (1987) Energie 1995, Apres les chocs. Economica, Paris,
p.253.
Dumoulin, H. (1988) Petroleum Resources in Non-OPEC Developing Countries, Analysis
of Profitability and Risk. Petroleum Economics Limited-Pacific, London, June, p. 27.
Edmonds, 1. and Reilly, J. M. (1985) Global Energy: Assessing the Future. Oxford
University Press, Oxford, p. 317.
References
237
EMF (Energy Modelling Forum) (1982) World Oil, EMF Report 6, Standford University Energy Modelling Forum, Stanford, Ca, p. 111.
Frisch, J. R. (1989) Horizons energetiques mondiaux 2000-2020. Technip-Conference
Mondiale de I'Energie, Paris, p. 378.
Gately, D. (1984) A Ten-Year Retrospective: OPEC and the World Oil Market. Journal
of Economic Literature, 22, 11 00-14.
Goldemberg, J., Johansson, T. B., Reddy, A. K. N. et al. (1988) Energy for a Sustainable
World. Wiley Eastern, New Delhi, p. 517.
Grinevald, J. (1990) L'effet de serre de la Biosphere. De la revolution thermo-industrielle
Ii l'ecologie globale. Strategies energetiques, biosphere et societe, no. 1, pp. 9-34.
Manne, A. S. (1990) Global 2100: An Almost Consistent Model of CO 2 Emissions
Limits, Stanford University, Stanford, Ca, February, p. 20.
Martin, J. M. (1988) L'intensite energHique de l'activite economique dans les pays
industrialises: les evolutions de tres longue periode ... Economies et Societes, series E,
no. 4, April, pp. 9-27.
Nordhaus, W. D. and Yohe, G. W. (1983) Future Paths of Energy and Carbon Dioxide
Emissions, in Changing Climate, National Academy Press, Washington, DC, pp.
87-153.
Piatier, A. (1989) 1980-1990: dix ans de surf. Economie et Societes, series F, no. 31,
March, pp. 5-41.
Simon, H. A. (1962) The Architecture of Complexity, Proceedings of the American
Philosophical Society, 106, 467-82.
---14--Inferred demand and supply

elasticities from a comparison
of world oil models
14.1 INTRODUCTION
Analyses of oil markets frequently depend upon a relatively small set of
important parameters governing the response of oil supplies and demands to
prices and economic growth. Analysts must assign these parameter values from
a limited historical experience that includes several sharp shifts in oil market
and economic trends. As a result, one finds a range of plausible parameter
values being used by oil policy analysts that can often lead to quite different
perspectives on oil market trends and the effectiveness of various policies to
reduce dependence upon insecure oil supplies.
This paper summarizes the responses of oil supply and demand to prices and
income in 11 world oil models that were compared in a recent Energy
Modeling Forum (EMF) study. In May 1989, the EMF commenced a study of
international oil supplies and demands (hereafter, EMF-11) to compare alternative perspectives on supply and demand issues and how these developments
influence the level and direction of world oil prices. In analysing these issues,
the EMF-11 working group relied partly upon results from 11 world oil
models, using standardized assumptions about oil prices and gross domestic
product (GDP). During the study, inferred price elasticities of supply and
demand were derived from a comparison of results across different oil price
scenarios with the same GDP growth path. Inferred income elasticities of
demand were derived from a comparison of results across different economic
growth scenarios with the same oil price-path. Together, these estimates
summarize several important relationships for understanding oil markets.
The next section provides some background on the EMF study and on
general trends in the scenarios of interest that help to understand the results.
The following sections explain the derivation and qualifications of the inferred
estimates, report the results and summarize the key conclusions.
240
Inferred demand and supply elasticities

14.2 THE EMF STUDY
14.2.1 Purpose and approach
The eleventh Energy Modeling Forum study (EMF-ll) analysed the factors
determining the long-run trends in the international oil market over the next
two decades. Such issues included the growth in world oil demand, the
prospects for supplies outside OPEC and the long-run implications of these
demand and supply trends for the world's dependence on oil from OPEC
member countries and particularly from the Persian Gulf. From its inception,
the study was not designed to focus on the short-run impacts of disruptions on
oil markets. Nor did the study attempt to provide just a single view of the likely
future path for oil prices. As in previous EMF studies, the research was
conducted by an ad hoc working group of more than 40 leading analysts and
decision-makers from government, industry, academia and other research
organizations. In the EMF process, the working group pursues the twin goals
of improving the understanding of the capabilities and limitations of existing
energy models, and using these models to develop and communicate useful
information for energy planning and policy.1 The group is guided in the pursuit
of these goals by a set of design principles:
1. A model user orientation maintained by active user involvement in the

development of the study.
2. A comparison of the capabilities and limitations of many models rather
than a detailed evaluation of a single model.
3. An issue focus that directs and guides the study by applying the models to
an important energy problem.
4. Broad participation by a number of people in selecting the topic, forming
the working group, analysing the results, and disseminating key findings.
5. Decentralized analysis of scenarios by proprietors familiar with the individual models.
The group met four times over the 1989-90 period - prior to the Iraqi
invasion of Kuwait - to develop a study plan with a set of carefully selected
scenarios, analyse model results and supporting analysis and develop key
conclusions and insights. Eleven economic models of the world oil market were
run by their proprietors at their home institutions using standardized assumptions for 12 different scenarios. These results were reported to the EMF staff
and formed the basis of the group's indepth comparison of alternative
perspectives on the world oil market.
IThe EMF process and key findings from previous studies have been discussed extensively in
several papers, e.g. Huntington et al. (1982).
The EMF study
241
14.2.2 Scenarios and models

Only six of the 12 scenarios are analysed in this paper. They were developed
to analyse differences in oil demand and supply projections based upon
standardized assumptions for the oil price and economic growth. Specific
model assumptions about OPEC's behaviour or responses to market conditions are excluded from these scenarios. These results help to interpret the
results from scenarios where supply and demand conditions, including OPEC
production decisions, are allowed to determine oil prices endogenously in each
model. Three cases assume a flat oil price-path with different GDP growth
assumption - low, base and high. Another three cases use the same three
economic growth assumptions with a rising oil price-path. 2
The models in the study were developed to prepare long-run projections of
oil prices, oil production and oil consumption and to study changes in these
variables under alternative scenarios. They incorporate the behaviour of three
distinct agents: oil consumers, oil producers outside the cartel, and oil producers within the cartel. Oil consumers respond to gross domestic product
(GDP)/ energy-saving trends in technology or economic structure (if present),
and oil prices. The response of oil producers outside the cartel is governed by
assumptions about trends in resource depletion and technology in addition to
oil prices. In most models, the cartel's productive capacity is exogenous, based
upon modeller judgement of a combination of economic and political constraints.4 The cartel sets a price based upon last period's price and rate of
utilization of its capacity. In this way, oil prices, production and consumption
are determined recursively; market conditions in one year influence those in
the succeeding year.
The main model features of interest to the elasticity estimates in this paper
are summarized in Table 14.1.5 Most are simulation models that determine oil
prices recursively in the manner described above. ETA-MACRO and DFICEC are optimization models that endow oil producers and/or consumers with
perfect foresight. The first assumes that both oil producers and consumers
maximize the discounted utility of total consumption of all goods and services;
the latter assumes that producers maximize total discounted oil profits. These
models require explicit assumptions about resource cost curves - the amount
of recoverable resources ultimately available at different prices.
2The other six scenarios included three based upon an exogenous oil price-path and three in which
market-clearing prices were determined by each model. See Huntington et al. (1989) for more
information on the assumptions in all 12 scenarios.
3Shifts in the economies' structures are seldom incorporated explicitly, although a macroeconomic
model linked to the Penn-BU model contains such detail.
4Capacity is endogenous in Penn-BU, IPE and ETA-MACRO. DFI-CEC uses an OPEC resource
curve directly without any capacity constraint.
5This table is based upon a comparison of model structures reported by Kress et al. (1990). Beider
(1982) also provides a very useful comparison of similar modelling approaches used in a previous
EMF study, including the distinction between recursive simulation and intertemporal optimization
approaches.
242
Table 14.1

Key features of models
OMS
[PE
ETAMACRO"
Penn-BU
WOMS
Model
type
Recursive
simulation
Recursive
simulation
Intertemporal
optimization
Recursive
simulation
Recursive
simulation
Perfect
foresight
No
No
Producers and
No
No
consumers
Periodicity
Annual
Annual
10 years
Annual
Annual
Horizon
2010
2000
2100
2010
2010
Regions
Supply
Demand
7
6
10
Parameters
Supply
Demand
Judgement
Judgement
Judgement
Judgement
Participating
Modeller
Mark Rodekohr
Nazli Choucri Alan Manne
Robert Kaufmann Nicholas

and
Baldwin and
Peter Pauly
Richard
Prosser
Organizationb
US Energy
Information
Administration
MIT
Univ. of Penn.
and Boston
University
5
Judgement
Judgement
Stanford
University
Econometric
Econometric
Econometric
Econometric
PowerGen
"This version of ETA-MACRO is also called Global 2100.

bOrganization of individual (s) who developed and exercised the model during the study. Listed for identification
purpose; model and results do not necessarily reflect organization's views.
Table 14.1 also compares the models in terms of periodicity, horizon (last
year in the projection), number of supply and demand regions, and whether the
supply and demand parameters are direct econometric estimates or are
determined judgementally based upon a reading of the available literature on
energy demand responses. The institutional affiliation listed in Table 14.1 is
provided to identify the model and not to indicate an official modelling
framework of a particular organization. This caveat applies particularly to
BP-America, WOMS and the Federal Reserve Bank of Dallas (FRB-Dallas),
as well as the various university models.
Most models report prices and supply-demand balances annually and focus
exclusively on world oil markets. Alternative fuel prices and interfuel substitution are not explicitly represented but instead are implicitly incorporated
through the own-price elasticity for oi1. 6 (This assumes that both the relationship between oil and other fuel prices and the potential for interfuel substitution will remain the same as in the past.) An exception to this general
paradigm, ETA-MACRO focuses on the interactions between electricity, fossil
fuels and the economy in the very long run, embodying specific parameters for
6BP-America and Penn-BU are exceptions. For the latter, interfue1 substitution is incorporated in
a detailed macroeconomic model linked to the world oil model.
243
The EMF study
CERI
HOMS
PRBDallas
DFl-CEC
BPAmerica
Gately
Recursive
simulation
Recursive
simulation
Recursive
simulation
Intertemporal
optimization
Recursive
simulation
Recursive
simulation
No
No
No
Producers
No
No
Annual
Annual
Annual
5 years
Annual
Annual
2010
2010
2010
2032
2010
2010
17
7
7
5
Econometric
Judgement
Econometric
Econometric
Econometric
Econometric
Judgement
Judgement
Econometric
Econometric
Judgement
Econometric
Anthony
Reinsch
William Hogan
and
Paul Leiby
Stephen
P. A. Brown
Dale Nesbitt
Lakis
Vouyoukis
Dermot Gately
Canadian
Energy
Research
Institute
Harvard
Federal
Reserve
Bank of
Dallas
Decision
Focus, Inc.
BP America
New York
University
substitution between energy and non-energy inputs as well as for substitution

within energy between electricity and fossil fuels.
14.2.3 Demands and supplies with the flat price-path
The EMF-II working group considered several very different sustained,

long-run paths for the world oil price. Current oil prices (December 1990) have
been driven far above these assumed paths by the Iraqi invasion of Kuwait and
could become quite volatile with military conflict. Eventually, however, many
analysts expect that once the situation is resolved, market forces will return oil
prices to substantially lower prices. Thus, these price trajectories should be
viewed as establishing a reasonable range for the long-run, sustained path over
the next several decades.
A flat oil price case assumes that the US refiner acquisition cost for imported
oil rises from $14.70 in 1988 to $18 per barrel in 1989 (all prices are in 1988 $)
and remains at that level until 2010. A rising oil price case assumes that this
oil price rises gradually to $36 per barrel by 2000 and remains at that level
until 2010. In both scenarios, GDP for the market economies is assumed to
grow by 2.9% per annum through the period, with higher economic growth
(4.1 % p.a.) outside the OECD countries. In both the flat and rising price
244
scenarios, OPEC is considered to be a residual supplier of oil, meeting all the

oil demand that remains unsatisfied by non-OPEC production.
It should be emphasized that modellers were requested not to impose any
shifts in government policies in running these cases. Many working group
members thought that oil importing countries would impose taxes and other
conservation policies to limit their oil demands. Thus, the EMF scenarios
should be considered as revealing the pressures that would emerge under
alternative oil price and GDP paths if no such policies were implemented.
Table 14.2 summarizes the trends for OECD demand, market economies
demand, non-OPEC production and the residual demand for OPEC oil in the
Table 14.2 Consumption, production, and call on OPEC (MBD) with flat
oil price-path
Flat oil price

Consumption
OECD
Mkt econ.
Production
Non-OPEC
CPE exports
Call on OPEC
OPEC share
Flat price with high
Consumption
OECD
Mkt econ.
Production
Non-OPEC
CPE exports
Call on OPEC
OPEC share
1990
Model averages
2000
38.3
52.6
46.9
66.3
58.1
85.6
28.5
1.9
22.2
42.3%
25.5
1.1
39.7
59.9%
20.7
0.6
64.3
75.1%
2010
% change (p.a.)
1990-2000 1990-2010
2.0%
2.3%
2.1 %
2.5%
-1.1%
-5.4%
6.0%
-1.6%
-5.3%
5.5%
2.8%
3.2%
2.9%
3.3%
-1.1%
-6.2%
7.3%
-1.6%
-7.1%
6.6%
1.3%
1.5%
1.4%
1.6%
-1.1%
-4.6%
4.6%
-1.6%
-4.0%
4.3%
GDP
68.1
102.5
38.8
53.2
51.0
72.6
28.5
1.9
22.8
42.9%
25.6
1.0
46.1
63.4%
20.7
0.4
81.3
79.4%
43.0
60.5
49.7
71.8
25.5
1.2
33.8
55.9%
20.6
0.8
50.4
70.1%
Flat price with low GDP

Consumption
OECD
37.9
Mkt econ.
51.9
Production
Non-OPEC
28.5
CPE exports
1.9
Call on OPEC
21.6
OPEC share
41.5%
The results are averages for all models that report all components in the table. ETA-MACRO is
excluded from the averages in this table because it did not report market economies consumption
and Non-OPEC production in the study. Penn-BU is excluded because it did not report OECD
consumption. IPE results are included for 1990 and 2000 but are unavailable for 2010.
The EMF study
245
three scenarios based upon the flat oil price-path. 7 The alternative scenarios
represent a high GDP case (GDP for market economies grows by about 1
percentage point higher) and a low GDP case (GDP grows by about 1
percentage point lower). Although Table 14.2 reports model averages only,
there exists a wide variation in results across models in these scenarios.
The projected supply and demand levels for the flat oil price-paths reveal the
strong pressure for OPEC members either to expand production rapidly or
increase prices. All scenarios imply substantially higher oil demands, modestly
declining non-OPEC supplies and rapidly growing dependence upon OPEC
sources. With the baseline GDP assumptions shown in the upper rows, OECD
oil consumption would grow from 38 MBD in 1990 to 47 MBD by 2000 and
to 58 MBD by 2010. Consumption by the market economies, which includes
the less developed countries (LDCs), would grow even more rapidly, reaching
86 MBD by 2010. Non-OPEC production would decline modestly through
2000 (to 25 MBD), falling more precipitously during the initial decade of the
next century. The call on OPEC production resulting from these above trends
would climb rapidly to 40 MBD by 2000 and to 64 MBD by 2010. Demand for
OPEC production with the flat price would increase by 6.0% p.a. between 1990
and 2000. If OPEC were simply to meet this demand at the $18 price,
dependence upon OPEC sources would quickly increase to 70% or more by
2010.
In the projections immediately below these results in Table 14.2, the higher
GDP path would accentuate these trends by raising world oil demand,
increasing the call on OPEC to 46 MBD in 2000 and to 81 MBD in 2010. The
lower GDP path would reduce significantly the level of OPEC production to
34 and 50 MBD, respectively, for these two years. This second scenario would
still require a 4.6% p.a. growth in OPEC production over the next decade.
Although not shown in this table, differences in demand projections among
models dominate differences in production outside OPEC. In 2000, demand in
the market economies varies by more than 30 MBD across models, while
non-OPEC supply varies by about 7 MBD. Thus, variations in demand have
a critical effect on the different calls on OPEC observed in the various models.
The range of demand projections is emphasized quite dramatically in Figure
14.1, which shows the oil-GDP ratio for the OECD countries continuing its
historical decline of the last two decades in six of the nine models under the
flat price scenario. By 2010, the oil intensity falls by 20--25%, or by 1.0% to
1.3% p.a. Three models - HOMS, ETA-MACRO and FRB-Dallas - show the
oil intensity as initially rising before leveling out with the flat oil price-path.
All three models assume that oil demand grows 1% for each 1% increase in
economic output, holding energy prices constant. The other models assume
further declines in oil intensity with future economic growth. Both HOMS and
7Reftecting traditional data collection procedures, the models (except ETA-MACRO) exclude oil
supplies and demands in the Soviet Union, Eastern Europe and China. Net exports from these
regions are an assumption.
246
1.3
-r---------------------,
-+- FRB-Dallas
1.2
-B- HOMS
1.1
-{}- ETA-MACRO
.E].
1.0
Gately
.... WOMS
0.9
--- OMS
....... CERI
0.8
0.7
BP-America
... DFI-CEC
IPE
0.6
0.54---~---,_--_.---r_--_r--~
1980
1985
1990
1995
2000
2005
2010
Fig. 14.1 OEeD oil-GDP ratio with flat price (1988 = 1).
FRB-Dallas are based upon explicit econometric modelling of the demand

response to oil prices and GDP. Neither represents any oil efficiency improvements over time because neither modeller found evidence for oil-saving
technological progress. ETA-MACRO uses demand responses that are based
upon a reading of the available econometric evidence, but allows for oil
efficiency improvements of 0.5% p.a. 8
14.2.4 Demands and supplies with the rising price-path
Table 14.3 shows the average projection for consumption, non-OPEC production and the call on OPEC when oil prices rise gradually from $18 to $36 up
to 2000 and remain at that higher level after 2000. For comparison with the
previous results, Table 14.3 also reports supply and demand levels for this
alternative price-path with high and low GDP assumptions.
Lower calls on OPEC result in the three scenarios based upon the rising oil
price-path than in those based upon a flat oil price-path. The average net
demand with the baseline GDP reaches only 25 MBD in the rising price case
by 2000, compared to 40 MBD in the flat price case. As a result, OPEC
production increases by a relatively modest 1.5% p.a. up to 2000 in the rising
price scenario.
8 A more comprehensive decomposition of these differences is reported by the Energy Modeling
Forum (EMF, 1990). This decomposition separates the OEeD oil demand projections for each
model in the flat price case into several components: the response to GDP changes; the momentum
caused by past price changes which continue to influence demand decisions through a lagged
adjustment process; and autonomous energy efficiency improvements (AEEI) that accrue over time
and are unrelated to either future or past price changes. The latter includes shifts in economic
structure away from energy-intensive sectors as well as the emergence of new technologies and
processes introduced for reasons other than price.
247
The EMF study
Table 14.3 Consumption, production, and call on OPEC (MBD) with rising oil price
path
1990
Rising oil price
Consumption
OECD
Mkt econ.
Production
Non-OPEC
CPE exports
Call on OPEC
OPEC share
Model averages
2000
2010
37.7
51.7
38.8
55.9
44.3
67.0
28.2
1.9
21.6
41.8%
29.8
1.1
25.0
44.8%
28.2
0.6
38.2
57.0%
42.4
61.4
51.9
80.1
29.8
1.0
30.6
49.8%
Rising price with high GDP

Consumption
38.1
OECD
Mkt econ.
52.3
Production
Non-OPEC
28.2
CPE exports
1.9
Call on OPEC
22.3
OPEC Share
42.5%
Rising price with low GDP
Consumption
OECD
37.2
Mkt econ.
51.0
Production
Non-OPEC
28.2
CPE exports
1.9
Call on OPEC
20.9
OPEC share
40.9%
% change (p.a.)
1990-2000 1990-2010
0.3%
0.8%
0.8%
1.3%
0.5%
-5.3%
1.5%
-0.0%
-5.2%
2.9%
1.1%
1.6%
1.5%
2.2%
28.2
0.4
51.5
64.3%
0.6%
-6.1%
3.2%
-0.0%
-7.2%
4.3%
35.3
50.7
37.9
56.2
-0.5%
-0.1%
0.1%
0.5%
29.7
1.3
19.7
38.9%
28.1
0.9
27.1
48.2%
0.5%
-4.1%
-0.6%
-0.0%
-3.4%
1.3%
The results are averages for all models that report all components in the table. ETA-MACRO
is excluded from the averages in this table because it did not report market economies
consumption and Non-OPEC production in the study. Penn-BU is excluded because it did not
report OECD consumption. IPE results are included for 1990 and 2000 but are unavailable for
2010.
As expected, the rising price-path encourages more non-OPEC production

than in the flat price scenarios. The mean estimate calls for a relatively stable
production path of 28-29 MBD through the period. Meanwhile, oil demand in
the market economies grows noticeably slower than with flat prices. Flat
OECD consumption and moderately increasing demands for the market
economies and for OPEC production result when high oil prices are combined
with low economic growth.
Figure 14.2 shows that the oil-GDP ratio declines under the rising price
assumptions for all but two models. By 2010, the oil intensity (indexed to 1 in
1988) declines by 30-40%, or by 1.6% to 2.3% p.a. The exceptions are HOMS
248
1.3
1.2
1.1
.fI-
HOMS
-+-
FRB-Dallas
Gately
-J- ETA-MACRO
1.0
OMS
0.9
WOMS
IPE
0.8
0.6
0.5
1980
BP-America
.....
DFI-CEC
...,..
0.7
1985
1990
1995
2000
2005
CERI
2010
Fig. 14.2 OECD oil-GDP ratio with rising price (1988 = 1).
and FRB-Dallas, both of which reveal oil intensities by 2010 that are not much
lower than those in 1988. During the early 1990s in these models, oil intensity
increases in response to the price declines of the 1980s. Later in the period, oil
intensities begin to fall as future oil prices move higher.
ETA-MACRO's oil intensity trend is substantially different with rising than
with flat oil prices. With higher prices, it follows the pattern set by most models
and declines throughout the period. This trend contrasts sharply with the oil
intensity trends for the rising oil price case (Figure 14.1), where ETA-MACRO
joined HOMS and FRB-Dallas in showing rising or flat oil~GDP ratios up to
2010. The sharp swing in this model from a falling intensity in the rising price
case to a rising intensity in the flat price case reflects a strong demand response
to price, as will be discussed in the next section on inferred estimates of price
elasticities.
14.3 ELASTICITY ESTIMATES
The general oil supply and demand trends associated with the several rising
and flat price scenarios were discussed above. In this section we report some
elasticity estimates that summarize the responses of oil supplies and demands
to changes in price and income based upon these scenarios. Price elasticities of
oil supply and demand for each model are derived implicitly from a comparison of the quantity and price results from the rising and flat price scenarios.
Inferred elasticities are computed as the ratio of the percentage difference in
the quantity demanded or supplied between the two scenarios and the
percentage difference in the crude oil price in the same year. GDP levels are
Elasticity estimates
249
held constant across these two scenarios. Income elasticities of oil demand for
each model are derived implicitly from a comparison of the quantity, and GDP
results from the flat and the high GDP (with flat price) cases. They are
computed as the percentage difference in oil quantity between the two
scenarios, divided by the percentage difference in GDP levels in the same year.
Oil prices are held constant across these two scenarios.
14.3.1 Uses and qualifications

These estimates are quite useful for understanding the pressure on long-run
oil prices to change in response to shifts in supply and demand conditions. For
example, the mean results in Table 14.2 indicated rising oil demand with
limited expansion in oil supplies outside OPEC with a flat oil price-path. If
OPEC producers also limit oil supplies, market pressure would push oil prices
upward over the long run - a result that is evident in the endogenous oil price
scenarios that the participating modellers ran during the study. How much
prices would increase depends partly upon the size of the supply and demand
shifts and partly upon the response of supply and demand to price. Limited
price sensitivity requires larger increases in oil prices to re-establish an oil
market equilibrium after the supply and demand shifts. Thus, price elasticities
of supply and demand play an important role in shaping long-run oil price
projections from any particular model.
These estimates also help elucidate how different production strategies
influence the income of cartel producers exercising monopoly control. While
cartel producers may have other objectives, income is likely to remain an
important criterion in their decision-making. In any particular year, reduced
cartel production will generate additional revenue if prices rise proportionately
more than the cartel's output declines. Again, prices will tend to increase more
with a given reduction in cartel output when world consumers and other
producers outside the cartel are less sensitive to price and when the cartel's
market share is greater. Ignoring extraction costs that are likely to be minimal,
income for the period would be maximized when the net demand for cartel
output possesses a unitary price elasticity. Income over the planning horizon,
of course, would also depend upon the timing of revenue receipts and the
cartel's discount rate.
Elasticity estimates are also useful for evaluating the effects on oil markets
of various policies introduced in major oil-consuming countries to reduce
imports and prices. Taxes on petroleum consumption will have a smaller
impact on domestic national wealth when price-induced substitution away
from oil is more extensive. Moreover, taxes will have a greater depressing effect
on world oil prices, and hence a smaller impact on domestic prices (including
the tax), when domestic demands are more price-sensitive and supplies and
foreign demands are less price-sensitive.
While the inferred elasticities are quite useful summaries of the responses for
250
each model, they must be interpreted carefully. Oil demand adjusts slowly to
price as the capital stock turns over so that the complete adjustment to price
(i.e., the long-run response) is not observed for many years. This problem is
compounded by the fact that higher oil prices are phased in gradually over 12
years in the EMF rising price scenario. In most models, consumers are
assumed to consider current (and past) prices, but not to look ahead at future
prices. Thus, for over half the period, demand decisions are being made on the
basis of prices below $36, the price level used in estimating the inferred
elasticity for 2000 and beyond. By overstating the price change upon which
decisions are made, the inferred elasticities will be understating the true
elasticity. Finally, we should note that the elasticities need not be constant in
all relevant price ranges, but may in fact depend upon the price level.
It should be emphasized that the EMF-ll estimates are for crude oil and
not for petroleum product price elasticities. When refinery margins and taxes
remain relatively stable in dollars per barrel, delivered product prices will
change proportionately less than crude oil prices. Under these conditions, the
crude price elasticity will be smaller, being approximately equal to the product
price elasticity times the ratio of the crude to product prices. Such conditions
appear to apply to US oil markets. Given current prices within the United
States crude elasticities are approximately one-half product elasticities.
14.3.2 Price elasticities of demand
Table 14.4 reports the average price elasticities of demand inferred from the
EMF scenarios for the United States, OECD, non-OECD countries and all
Table 14.4 Inferred crude oil price elasticities of demand

l-yr
Average model responses

lO-yr
Long run/20-yr
US
OECD
Non-OPEC LDCs
Market economies
-0.10
-0.12
-0.11
-0.10
-0.33
-0.34
-0.21
-0.26
-0.44
-0.47
-0.30
-0.38
Estimates for other studies

Crude oil
Gately-Rappoport (1988)
Brown-Phillips (1989)a
Gasoline
Sterner and Dahl (1990)b
-0.07
-0.11
-0.38
-0.56
-0.24
-0.80
Elasticities are derived from the EMF rising and flat oil price scenarios. See text for derivation and
qualifications.
'one-quarter elasticity equals -0.08.
bSee Table 5.1 (lagged endogenous, survey average).
251
market economies. The table contains estimates for the demand response after
the first, tenth and twentieth years.9
The results reveal several conclusions. First, the responses for the United
States appear quite similar to those for all OECD countries. Price elasticities
are approximately -0.1 after the first year, rising to -0.4 or -0.5 after 20
years of adjustment in the capital stock. Secondly, these estimates appear
comparable to several recent econometric studies that have estimated the
demand response to crude oil price changes in the United States (shown at the
bottom of Table 14.4). It is not surprising that the Brown and Phillips (1989)
estimates are similar because those estimates are precursors to the FRB-Dallas
model being used in EMF-II. Also, shown in Table 14.4 are the means
reported by Sterner and Dahl in their survey of gasoline demand studies
(Chapter 5). The first-year and long-run responses of -0.24 and -0.8,
respectively, correspond roughly to crude oil price elasticities of - 0.15 and
- 0.50, given recent crude oil and US refined product prices. And thirdly, the
estimated price elasticities are lower outside than within the OECD. It should
be emphasized, however, that the modelling of oil demand in the developing
countries is very rudimentary given the existing data for these regions. Since
much less effort has been expended to estimate oil demand parameters for these
countries, one must be cautious in drawing conclusions from these estimates.
Estimated price elasticities are reported for each model in the appendix
(Table A.l) For the most part, long-run elasticities cluster in the -0.3 to -0.5
range for US and OECD demand. ETA-MACRO and an alternative version
ofHOMS (HOMS-I) have substantially higher long-run price elasticities in the
-0.8 range, while Gately and IPE reveal considerably smaller than average
responses.
The higher response in HOMS-I directly reflects the alternative assumptions
used to estimate oil demand from historical data. This version assumes that all
declines in oil intensities over the last two decades can be attributed to higher
oil prices operating with a considerable lag as the capital stock is replaced. The
version reported as HOMS 1o in the EMF study assumes that the structure of
oil demand was permanently altered in 1980, resulting in a one-time improvement in oil use efficiency independent of the oil price. Thus, part of the price
effect in HOMS-l is attributed to other causes in HOMS.
The higher price response in ETA-MACRO may depend upon its focus on
all energy rather than oil alone, as in the other models. This model explicitly
incorporates the interfactor substitution between energy and nonenergy inputs
as well as interfuel substitution between oil and other energy forms. In
addition, the model's substitution response to various prices is not estimated
9The choice of initial year (1989 or 1990) depends upon how the price change was implemented in
each model. The lOth- and 20th-year estimates were calculated from results for 2000 and 2010,
respectively.
lOThe HOMS modellers do not prefer one specification over the other. The choice of which version
to use as the main HOMS entry in the EMF study was arbitrary.
252
directly from historical data but instead is set judgementally based upon a
reading of estimates from other studies.
The lower price response in the Gately model results from assumed asymmetries in the demand response to price changes. Due to large capital costs,
investment in energy-conservation measures is not undone when prices fall
from previously high levels, so that demand would not change very much. Nor
does such investment need to be added back when prices begin to recover and
rise again, resulting in very little decline in demand. However, if prices were to
exceed their historical maximum (which is not reached in the EMF scenarios),
the price response would increase as new opportunities for investment in
conservation would emerge.
14.3.3 Income elasticities of demand
Table 14.5 reports the average inferred income elasticities of demand for the
United States, OECD, non-OECD developing countries and all market economies. It contains estimates for the demand response after the first, tenth and
twentieth years.
The mean long-run elasticities for these models lie in the 0.8-0.9 range for
all regions. This result suggests some improvements in oil efficiency in these
economies over time even without higher oil prices, because oil consumption
grows more slowly than economic output. As reported in the appendix,
however, the inferred income elasticities differ widely across models. Income
elasticities in the range of unity are found for both versions of HOMS,
Table 14.5 Inferred income elasticities of demand
l-yr
US
OECD
Non-OPEC LDCs
Market economies
0.87
0.88
0.78
0.72
Estimates from other studies

Crude oil
Gately-Rappoport (1988)"
Brown-Phillips (1989)b
Gasoline
Sterner and Dahl (1990)"
0.60
1.13
0.45

lO-yr
Long run/20-yr
0.85
0.86
0.88
0.81
0.86
0.88
0.92
0.85
1.31
Elasticities are derived from the EMF high GDP (with flat prices) and flat price scenarios. See text
for derivation and qualification.
"Estimated from annual data, 1949-85. Long-run income elasticity equals the first-year elasticity.
bEstimated from quarterly data, 1972:1-1988:1. Long-run income elasticity equals the first-year
elasticity.
CSurvey of other studies, lagged endogenous model: see Table 5.1.
253
FRB-Dallas, WOMS, BP-America, and ETA-MACROY The 20-year income

elasticities for the remaining models average 0.6 for both the OECD countries
and market economies.
Most models with a unitary income elasticity also revealed a trend effect
towards declining oil intensity that is unrelated to changes in either past or
future oil prices or income. 12 These autonomous improvements in oil efficiency
incorporate the adoption of newer more energy-efficient technologies or
processes for reasons other than oil prices. In addition, the trend includes shifts
in the economic structure away from energy-intensive industries. As a result,
these models joined the group of models with income elasticities below one in
projecting oil demand to grow less rapidly than economic growth, even with
constant oil prices. Only FRB-Dallas and HOMS (both versions) incorporate
a unitary income elasticity without any autonomous improvements in oil
efficiency. These two models also indicate the highest demand projections in
the EMF scenarios specifying exogenous oil price-paths.
Finally, the first-year elasticities are virtually the same as the long-run or
20-year responses for all regions. The appendix reveals that most models follow
this trend of relatively constant income elasticities over time. Exceptions are
CERI and Penn-BU, in which both price and income responses become
stronger over timeY The SternerjDahl survey of gasoline demand (Chapter 5)
provides some evidence that income elasticities are larger in the long run than
the short run, although the 1.31 long-term estimate can be consistent with the
EMF-ll estimates only if other petroleum products are quite income-inelastic. 14 The other two studies included in the bottom of Table 14.5 incorporate
an instantaneous adjustment to the long-run income elasticity. They reflect the
two alternative views depicted in the EMF estimates. The Gately-Rappoport
study (1988) reports income elasticities of about 0.6 while the Brown-Phillips
study (1989) estimates an elasticity that is not significantly different from unity.
14.3.4 Price elasticities of supply
Price elasticities of supply for the non-OPEC regions were calculated from the
rising and flat price scenarios in a procedure analogous to the one used for the
price elasticities of demand discussed previously. The percentage difference in
llETA-MACRO assumes an income elasticity of unity but did not run the scenarios that would
reveal an inferred income elasticity. Its responses are not reported in the appendix tables.
12This information was ascertained by comparing two separate scenarios based upon the flat oil
price-path that (a) eliminated any economic growth, and (b) eliminated both any economic growth
and any time trend towards improved oil efficiency independent of oil prices.
13The Penn-BU results cause the average income elasticities for the market economies in Table
14.5 to rise slightly. This model did not report consumption for other regions. Table 14.5 is based
upon averages that exclude IPE and DFI-CEC in order to emphasize the time pattern of the
response. DFI-CEC did not report short-run results, and IPE's horizon extends only up to 2000.
14The surveyed studies generally did not control for the number of drivers. Gately (1990) argues
that incorporating this effect would lower the income elasticity by nearly one-half.
254

Table 14.6 Inferred crude oil price elasticities of supplies
l-yr
US
OECD
Non-OPEC
Total
Excluding US

lO-yr
Long run/20-yr
0.05
0.05
0.24
0.25
0.40
0.43
0.03
0.Q2
0.21
0.20
0.40
0.38
Elasticities are derived from the EMF rising and flat oil price scenarios. Mean response excludes
DFI-CEC, an intertemporal optimization model. See text for derivation and qualifications.
quantity produced between the two cases is divided by the percentage difference in crude oil prices. Results for 1-, 10- and 20-year responses appear in
Table 14.6.
Price elasticities of supply begin a little lower than their demand counterparts 15 (Table 14.4) but increase over time until the two elasticity estimates are
roughly comparable after 20 years. Long-run price elasticities of supply average
about 0.4 in each of several regions for which responses could be calculated.
Long-run responses for total non-OPEC production range from 0.16 (CERI)
to 0.64 (HOMS-l), as reported in the appendix. The pattern of the OFI supply
elasticity deserves special consideration. Suppliers in the model optimize
production over time to maximize discounted profits. In the rising price case,
suppliers have incentives to withhold production and extract oil in later years
when profits (after discounting) become more attractive. As a result, this model
predicts less production in most regions for the rising than for the flat price
case in the early years and substantially greater production in the later years. 16
14.4 CONCLUSIONS
The EMF scenarios were designed to analyse international oil supply and
demand trends under alternative market conditions. While they were not
specified explicitly to reveal precise estimates of the relevant elasticities, the
scenarios do offer a unique opportunity to examine the approximate responses
embodied in some of the major world oil models used for policy and planning
purposes. This information is likely to be of considerable interest to policy
analysts and to other world oil modellers.
From this comparison of scenario results, we conclude that the average price
15This would be explained by the findings of Rodriguez Podilla in Chapter 8.
16The model would view the assumed rising and flat oil price-paths as being dynamically
inconsistent because producers can earn a higher discounted profit in one time period than in
another. This factor explains the wide swings in production observed for this model in response to
the two exogenous oil price-paths.
References
255
elasticity of demand (measured at the crude oil level) in these models is about
-0.1 in the short run (after the first year), about -0.3 in the intermediate run
(after 10 years) and about -0.4 in the long run (after 20 years). Most long-run
estimates lie between - OJ and - 0.5, although several estimates fall either
above or below this range.
The evidence on income elasticities is far more diverse. For the most part,
the models incorporate the full demand adjustment to income within the first
year of a change in GDP. The average estimate of 0.8 for all models is
deceiving. Half of the models anticipate no further improvements in oil
efficiency as the economy grows, unless oil prices move higher. This result is
summarized by an inferred income elasticity of unity for these models. The
remaining models show improvements in oil efficiency resulting from future
economic growth, reflected by an inferred income elasticity of about 0.6. In
addition, several models incorporate an autonomous long-run trend towards
oil-saving goods, technologies and processes, independent of price and income
changes. The income effect and the potential for autonomous energy efficiency
improvements are particularly fruitful areas for future research on energy
demand.
Like their demand counterparts, the price elasticities of supply outside
OPEC increase over time as the full adjustment to price changes is incorporated. The average crude oil price elasticity of supply is well below 0.1 in the
short run (after the first year), about 0.2 in the intermediate run (after 10 years),
and about 0.4 in the long run (after 20 years). Most long-run estimates lie
between 0.2 and 0.5, although several estimates fall either above or below this
range.
ACKNOWLEDGEMENTS
I would like to acknowledge the significant contributions of the EMF-l1
Working Group, chaired by W. David Montgomery, for improving my
understanding of certain key issues. I am also very grateful to those researchers
who exercised their models during the study. These individuals include
Nicholas Baldwin, Stephen P. A. Brown, Nazli Choucri, Dermot Gately,
William Hogan, Robert Kaufmann, Alan Manne, Dale Nesbitt, Anthony
Reinsch, Mark Rodekohr and Lakis Vouyoukas. Interpretations and conclusions are entirely my own.
REFERENCES
Beider, P. (1982) 'Comparison of the EMF-6 Models', in World Oil, EMF Report 6,
vol. 2, Stanford University, Energy Modeling Forum. Stanford, C, pp. 349-428.
Brown, S. P. A. and Phillips, K. R. (1989) An Econometric Analysis of us Oil Demand.
Research Department, Federal Reserve Bank of Dallas, No. 8901, January.
256
EMF (Energy Modeling Forum) (1990) 'International Oil Supplies and Demands'. EMF
Report 11 (draft), Stanford University Energy Modelling Forum, Stanford, C.
Gately, D. and Rappoport, P. (1988) The Adjustment of US Oil Demand to the Price
Increases of the 1970s. Energy Journal, 9, 93-108.
Gately, D. (1990) The US Demand for Highway Travel and Motor Fuel. Energy
Journal, 11, 59-73.
Huntington, H. G., Sweeney, 1. L. and Weyant, J. P. (1982) Modeling for Insights, Not
Numbers: The Experiences of the Energy Modeling Forum. OMEGA: The International Journal of the Management Sciences, 10,449-62.
Huntington, H., Kress A. and Robinson, D. (1989) EMF-ll Revised Scenario Design.
Stanford University Energy Modeling Forum, Stanford, C.
Kress, A., Robinson, D. and Ellis, K. (1990) Comparison of the Structure of International Oil Models, (draft). Stanford University Energy Modeling Forum, Stanford, C.
APPENDIX TABLES
Table 14A.l Price elasticity of demand inferred from rising and flat price cases
United States
OMS (EIA)
Gately
IPE
ETA-MACRO
CERI
HOMS
FRB-Dallas
DFI-CEC
HOMS-I
Average
OECD
OMS (EIA)
Gately
IPE
ETA-MACRO
CERI
HOMS
FRB-Dallas
DFI-CEC
WOMS
BP-America
HOMS-I
Average
1st year
1995
2000
2005
2010
-0.071
-0.137
-0.039
-0.232
-0.146
-0.078
-0.327
-0.163
-0.327
-0.171
-0.139
-0.074
-0.088
-0.140
-0.296
-0.162
-0.323
-0.177
-0.341
-0.283
-0.154
-0.084
-0.870
-0.353
-0.224
-0.405
-0.171
-0.456
-0.419
-0.287
-0.502
-0.186
-0.630
-0.778
-0.440
-0.308
-0.537
-0.184
-0.737
-0.098
-0.219
-0.333
-0.359
-0.436
-0.130
-0.137
-0.104
-0.215
-0.151
-0.161
-0.360
-0.171
-0.396
-0.181
-0.164
-0.111
-0.101
-0.063
-0.034
-0.205
-0.311
-0.205
-0.326
-0.217
-0.179
-0.181
-0.439
-0.285
-0.160
-0.164
-0.783
-0.370
-0.269
-0.404
-0.258
-0.208
-0.317
-0.547
-0.431
-0.332
-0.498
-0.338
-0.366
-0.349
-0.713
-0.761
-0.446
-0.354
-0.531
-0.362
-0.490
-0.368
-0.804
-0.117
-0.238
-0.342
-0.395
-0.469
257
Appendix tables
Table 14A.l-contd.
Non-OPEC LDCs
OMS (EIA)
Gately
IPE
CERI
HOMS
FRB-Dallas
DFI-CEC
WOMS
BP-America
HOMS-I
Average
Market Economies
OMS (EIA)
Gately
IPE
Penn-BU
CERI
HOMS
FRB-Dallas
DFI-CEC
WOMS
BP-America
HOMS-I
Average
1st year
1995
2000
2005
2010
-0.106
-0.104
-0.144
-0.126
-0.098
-0.228
-0.122
-0.144
-0.153
-0.388
-0.232
-0.347
-0.125
-0.083
-0.255
-0.232
-0.170
-0.163
-0.199
-0.178
-0.045
-0.058
-0.102
-0.096
-0.130
-0.133
-0.306
-0.193
-0.318
-0.098
-0.071
-0.181
-0.192
-0.494
-0.292
-0.386
-0.175
-0.138
-0.329
-0.291
-0.535
-0.328
-0.400
-0.191
-0.178
-0.357
-0.326
-0.112
-0.172
-0.208
-0.271
-0.299
-0.094
-0.131
-0.104
-0.013
-0.151
-0.098
-0.139
-0.224
-0.147
-0.148
-0.158
-0.360
-0.244
-0.388
. -0.219
-0.171
-0.295
-0.450
-0.288
-0.158
-0.318
-0.165
-0.025
-0.046
-0.179
-0.167
-0.139
-0.143
-0.149
-0.298
-0.190
-0.323
-0.181
-0.149
-0.181
-0.362
-0.243
-0.431
-0.305
-0.464
-0.288
-0.299
-0.342
-0.579
-0.313
-0.450
-0.329
-0.490
-0.309
-0.396
-0.364
-0.648
-0.098
-0.208
-0.255
-0.340
-0.378
FRB-Dallas, WOMS and BP-America did not report for non-OPEC LDCs. Their estimates
have been derived as the difference in the responses for the market economies and OECD.
DFI-CEC's demand response to price was calibrated to first-round OMS results in this study.
Estimate for 1st year is for the year in which the initial demand response is observed - 1990 for
OMS, Gately, IPE and BP-America, and 1989 for all others. ETA-MACRO's demand response
begins after 1990, i.e. in 1991, but is reported for every ten years only. Non-OPEC LDC response
begins in 1990 for WOMS.
Table 14A.2 Income elasticities of demand inferred from high GDP (with flat price)
and flat price cases
United States
OMS (EIA)
Gately
IPE
CERI
HOMS
FRB-Dallas
DFI-CEC
HOMS-1
Average
ex. IPE and DFI
OECD
OMS (EIA)
Gately
IPE
CERI
HOMS
FRB-Dallas
DFI-CEC
WOMS
BP-America
HOMS-1
Average
ex. IPE and DFI
Non-OPEC LDCs
OMS (EIA)
Gately
IPE
CERI
HOMS
DFI-CEC
HOMS-1
FRB-Dallas
WOMS
BP-America
Average
ex. IPE and DFI
1st year
1995
2000
2005
2010
0.601
0.875
1.199
0.626
1.000
1.099
0.731
0.936
0.972
0.462
1.000
0.972
0.693
0.982
0.756
0.944
0.769
0.946
1.000
0.711
0.918
1.048
0.493
1.000
0.968
0.646
0.984
0.503
0.994
0.975
0.723
0.987
0.486
1.000
0.991
0.627
0.995
0.914
0.867
0.846
0.846
0.844
0.847
0.840
0.860
0.831
0.864
0.801
0.751
1.397
0.376
1.000
1.000
0.568
0.774
1.079
0.403
0.973
0.974
0.605
0.991
1.248
0.973
0.594
0.782
0.593
0.798
1.000
1.111
1.000
0.604
0.764
1.140
0.371
0.969
0.984
0.572
0.980
1.249
0.969
0.440
0.976
0.982
0.645
1.006
1.257
0.988
0.458
0.978
0.996
0.559
0.996
1.262
0.991
0.937
0.880
0.860
0.861
0.859
0.863
0.852
0.878
0.848
0.884
0.694
0.834
1.221
0.770
1.000
0.560
0.891
0.854
0.610
0.994
0.390
0.994
1.062
0.991
0.934
0.570
0.916
0.598
0.944
1.000
1.000
0.701
0.201
0.508
0.849
0.989
0.478
1.000
0.377
1.000
1.045
1.020
0.744
0.649
0.996
0.379
0.996
1.043
0.976
1.012
0.710
0.997
0.325
0.997
1.054
1.021
1.045
0.825
0.775
0.801
0.830
0.828
0.880
0.838
0.895
0.855
0.921
259
Appendix tables
Table 14A.2-contd.
1st year
Market Economies
OMS (EIA)
Gately
IPE
Penn-BU
CERI
HOMS
FRB-Dallas
DFI-CEC
WOMS
BP-America
HOMS-l
Average
ex. IPE and DFI
0.401
0.779
1.000
0.223
0.446
0.900
1.000
1995
2000
2005
2010
0.522
0.872
0.975
0.337
0.517
0.929
1.033
0.557
0.991
1.188
0.922
0.549
0.895
0.565
0.914
0.900
0.889
0.900
0.508
0.840
1.044
0.395
0.417
0.902
1.029
0.523
0.980
1.101
0.902
0.378
0.575
0.933
1.043
0.572
1.000
1.221
0.928
0.400
0.620
0.948
1.055
0.495
1.004
1.240
0.937
0.744
0.715
0.786
0.786
0.804
0.812
0.809
0.836
0.818
0.854
Responses were not reported for ETA-MACRO. Elasticity equals 1 by assumption.

Estimates are approximate due to rounding of results reported to EMF staff. Estimated
elasticities in the range of 0.95 through 1.05 are not distinguishable from unity.
DFI-CEC's demand response to income was calibrated to first-round OMS results in this study.
Estimate for 1st year is for the year in which the initial demand response is observed - 1991 for
WOMS, 1990 for Gately, CERI and Penn-BU, and 1989 for all other models. This response was
not available for DFI-CEC, which reports every five years.
FRB-Dallas, WOMS and BP-America did not report for Non-OPEC LDCs. Their estimates
have been derived as the difference in the responses for the market economies and OECD.
Response in BP-America was to unexpected income growth; its response to expected income
growth is less than unity.
Table 14A.3 Price elasticities of supply inferred from rising and flat price cases
United States
OMS (EIA)
Gately
IPE
ETA-MACRO
Penn-BU
CERI
HOMS
FRB-Dallas
DFC-CEC
HOMS-I
Average
OECD
OMS (EIA)
Gately
IPE
ETA-MACRO
Penn-BU
CERI
HOMS
DFI-CEC
HOMS-I
Average
Non-OPEC Total
OMS (EIA)
Gately
IPE
Penn-BU
CERI
HOMS
FRB-Dallas
DFI-CEC
WOMS
BP-America
HOMS-I
Average
1st year
1995
2000
2005
2010
0.117
0.045
0.000
0.149
0.186
0.024
0.306
0.453
0.340
0.577
0.000
0.137
0.012
0.013
0.089
0.119
0.291
0.159
0.137
0.028
0.280
0.230
0.294
0.032
0.134
0.165
0.339
0.281
0.239
0.180
0.400
0.188
0.313
0.440
0.382
0.377
0.563
0.215
0.162
0.195
0.522
0.475
0.500
0.662
0.052
0.168
0.235
0.378
0.394
0.070
0.052
0.000
0.119
0.186
0.015
0.233
0.453
0.256
0.577
0.000
0.075
0.086
0.076
0.117
0.200
0.295
-0.011
0.294
0.163
0.294
0.069
0.187
0.167
0.287
0.423
0.308
0.422
0.222
0.310
0.597
0.536
0.596
0.290
0.224
0.246
0.701
0.654
0.699
0.051
0.175
0.252
0.402
0.428
0.047
0.045
0.000
0.000
0.050
0.012
0.013
0.135
0.287
0.078
0.153
0.198
0.254
0.227
0.401
0.126
0.246
0.384
0.195
0.441
0.223
0.560
0.050
0.023
0.076
0.090
0.178
0.024
0.108
0.159
0.136
0.127
-0.004
0.145
0.101
0.266
0.195
0.204
0.413
0.375
0.680
0.241
0.456
0.545
0.193
0.161
0.512
0.480
0.833
0.259
0.540
0.641
0.032
0.134
0.209
0.341
0.397
261
Appendix tables
Table 14A.3-contd.
1st year
Non-OPEC ex US
OMS (EIA)
Gately
IPE
Penn-BU
CERI
HOMS
FRB-Dallas
DFI-CEC
HOMS-I
Average
1995
2000
2005
2010
0.088
0.283
0.101
0.151
0.132
0.247
0.224
0.460
0.377
0.143
0.435
0.170
0.553
0.076
0.061
0.174
0.026
0.106
0.098
0.130
0.123
-0.011
0.260
0.197
0.152
0.408
0.374
0.780
0.537
0.200
0.144
0.510
0.480
0.980
0.633
0.018
0.122
0.200
0.321
0.384
0.000
0.052
0.000
0.000
0.000
0.000
0.013
Averages exclude DFI-CEC, an intertemporal optimization model in which the rate of increase in
oil prices is critical to the observed supply response. First-year response is not reported for this
model because results are reported for 5-year periods.
Estimate for 1st year is for the year in which the initial supply response is observed - 1990 for
OMS, Gately and BP-America, and 1989 for all others. ETA-MACRO's supply response begins
after 1990, i.e. in 1991, but is reported for every 10 years only.
---15--World oil market simulation

Nick Baldwin and Richard Prosser
5.1 INTRODUCTION
Most of the models used to analyse the world oil market can be categorized
under two broad headings: (i) recursive simulation models and (ii) intertemporal optimization models. 1 Both types assume that OPEC operates as a
cartel, although the way in which OPEC behaviour is modelled varies.
In recursive simulation models the market actors make decisions based on
information about past and present but not future events. Typically these
models assume OPEC uses some form of empirically derived 'reaction function'
as a price-setting decision rule. This decision rule is based on a target level of
capacity utilization. According to Gately (1984) the use of this reaction function
implies that OPEC 'is groping toward an unknowable "optimal" price-path by
implicitly following a target capacity utilization rule of thumb: increase price
when the market is tight or tightening and let it ease off when the market is
sluggish'.
Intertemporal optimization models allow at least one sector to take account
of information about future events. This sector is treated as some form of cartel,
although the modelling of the cartel's behaviour differs between models.
Regardless of their different ways of modelling OPEC all these models are
based on Hotelling's theory of wealth maximization. Consequently it is
assumed that OPEC is attempting to choose a price-path which maximizes the
net present value of its oil revenues.
Both types of model have been subject to criticism. The price reaction
function in recursive simulation models has been criticized as ad hoc because
it is based on empirical observation rather than economic theory: and as such
it lacks a rational (objective maximizing) explanation for the chosen price rule. 2
lThis categorization was used in the study by Stanford University'S Energy Modeling Forum;
EMF-6, World Oil, (1982). The same categorization has also been used by Gately (1984) and Hogan
and Leiby (1985).
2Those who advocate this argument are sometimes prepared to concede the relevance of the
reaction function in the short term if not in the long term, (e.g. Marshalla et al. (1985)) or only
under certain conditions (e.g. Pindyck, 1972).
264
Nevertheless, the price reaction function is able to explain both the 1973/74
and 1979/80 price rises, whereas intertemporal optimization models can only
satisfactorily explain the 1973/74 price rise. (Both types of model have difficulty
with the 1986 price fall.) Criticisms have also been levelled against the assumption of perfect foresight and the neglect of uncertainty about other market
actors' behaviour within intertemporal optimization models. 3
This paper presents a recursive simulation model of the world oil market the World Oil Market Simulation Model (WOMS).4 The objective was to
construct a computationally simple model which provides a transparent view
of the workings of the oil market. In the event WOMS has a number of features
which distinguish it from other published models:
1. the effect of exchange rate movements is incorporated in the supply and
demand functions;
2. both demand and supply functions are dynamic;
3. the non-OPEC supply functions account for the geological as well as the
economic aspects of supply;
4. oil prices can be determined either by OPEC setting prices (as normally
included in this type of model) or by OPEC setting volumes and market
forces determining the price; and
5. consistency checks on consumer's and producer's behaviour are incorporated to confirm the plausibility of model projections.
This chapter commences with an outline of the model structure followed by an
examination of the choice of the appropriate data. The main sections of the
chapter discuss the estimation of the demand and non-OPEC supply functions.
Finally the modelling of OPEC's behaviour is addressed. Comparisons are
made throughout with other published work. As the model was estimated
using data covering 1960 to 1985, brief comments are also made comparing the
events of 1986 with model determined values.
15.2 MODEL STRUCTURE
The structure of the model is shown in Figure 15.1. Although seemingly

complex it can be considered simply as a game being played by three actors oil consumers, non-OPEC oil producers and OPEC - all of whom trade a
homogeneous commodity within a framework that both directs and constrains
play. The 'rules' of the game require both oil consumers and non-OPEC
producers to act as price-takers, with consumers maximizing the benefits of
their expenditure on oil and non-OPEC producers maximizing their profits.
3See Gately (1984) who provides a succinct critique of intertemporal optimization models.
4In this context world refers to the World Outside Communist Areas (WOCA).
NA
CPE
LCU
R:P ratio
ONO
CUM
ECO
RES
OXLOC
OILOC
WOCA
OECOXNA
t
I WOCA
GOP
t
t
LoC GOP
LoC
DEMAND
WOCA
"
OIL RATIO
DEMAND
Fig. 15.1
Development
North Amenca
Organization of Economic Cooperation
and Development excluding North Amenca
011 exporting less developed countries
011 Importing less developed countries
World Outside Communist Areas
Other nonOPEC
Cumulative
Economic
Reserve
Reserve: production ratio
Centrally Planned Economies
Local Currency Unit
RATES
OIL PRICE BLOCK
OECDXNA
INDEX
II
I EX~~:~GE I I
OECo
INDEX
LeU OIL
PRICES
t
t
II
I I
INDEX
NA
HISTORIC
DEMAND
OPEC SUPPLY BLOCK
OPEC ~
1rPRODUCTION
CAPACITY
SUPPLY BLOCK
US $ OIL
PRICE
Ir
II
STOCK
CHANGE
t
~
SUPPLY
[NON OPEC
J
1I PROVE~ I
t
I
r-- .....---,
1r
I~
II
CPE NET
EXPORTS
NA ECO
SUPPLY
NA GED
SUPPLY
RATIO
NA A:P
NA
RESERVES
II SUPPLY
NACUM II NACUM
RES
OPEC
DEMAND
OUTPUT
OPEC
OPEC
REVENUES
OPEC
EXPORTS
l
l [t l
_I.
--------------------------
NON-OPEC
OND R:P
RATIO
SUPPLY
ONOGCO
Y
t
OND CUM
RES
Geo] fONDRES
PROVEN)
OND
SUPPLY
r------,
ONO CUM
SUPPLY
World oil market simulation (WOMS), version 1.
Ir
I ~ CONSERVATIO~
Organization of Economic Cooperation and
OIL DEMAND BLOCK
OECD
OECo
DEMAND
GOP
10lLoC GOP.1
-----t- ----------
GOP
GOP
I OIL
OECo
I I Oil
LoC RATIO,I
RATIO
CONSERVATION
10Eco
GOP
'r-------
ECONOMIC
ACTIVITY
INA
I I OECDXNA I I OXLDC
266
OPEC set either price or volume depending on market circumstance. 5 The

behaviour of the actors is determined through three equations and an identity:
Oil demand (d)
d = fn(GDP, lag GDP, price, exchange rates, lag demand, conservation)
Non-OPEC supply (s)
s = fn(reserves, time, price, lag price, exchange rates, lag supply CPE)
Market clearing
OPEC supply = d - s + / - stock adjustment
(15.3)
OPEC price reaction function

(price-lag price)/lag price = fn(OPEC supply/capacity)
(15.4)
where 'GDP' is gross domestic product and 'CPE' is the net exports of crude
oil and petroleum products from the centrally planned economies.
In addition to the basic 'rules' there are also a number of consistency checks
that can act as constraints on the actors' freedom of movement. Consumers'
efficiency of oil usage is examined using the oil ratio (i.e. oil demand per unit
of GDP). The ability of non-OPEC producers to stay in operation is examined
using reserve to production ratios. The requirement to increase oil revenues
whilst retaining control of the market provides the basis of the choices open to
OPEC when setting either price or volume. A more detailed discussion of
OPEC's 'trade-offs' between price, volume, revenues and market share is
included in section 15.6.
15.3 DATA
The main source of oil supply and demand data used in the analysis was the
BP Statistical Review of Wodd Energy. In places BP data was supplemented
with data from the OPEC Annual Statistical Bulletin, the US Energy Information Administration's Annual Energy Outlook and Monthly Energy Review, the
Oil and Gas Journal and Petroleum Intelligence Weekly.
15.3.1 Real exchange rate adjusted crude oil prices
The dramatic appreciation of the US$ between 1980 and 1985 meant that
non-US consumers saw oil prices rise when measured in terms of their own
currencies, at the same time as US$ denominated prices were falling. This
factor has been recognized by Brown and Phillips (1984), Chevron (1985), EIA
(1985), Huntington (1986) and the International Energy Agency (1985),
amongst others.
5The assumptions about the behaviour of market actors are adapted from those given in the
documentation to the model used by the US Energy Information Administration (1985).
267
Data
To account for this phenomenon a real oil price index covering the six main
developed nations was developed. The official OPEC price of crude oil was
converted from nominal to real terms using the US GNP deflator. This was
then converted into the real local currency unit price of crude oil in France,
Italy, Japan, the UK and West Germany, using real bilateral exchange rates.
Price indices, with a base of 1980, were calculated for each of the six nations.
These were weighted together using current consumption weights to produce
an aggregate Paasche 6 index (PI) as follows:
Plt =
i= 1
i= 1
L {dit,[(Xit.P$tXiSO,P$SO)]} / L d
it
(15.5)
where X is the real exchange rate and d the oil demand of country i against
the US$ and P$ is the real oil price denominated in US$ per barrel. The suffixes
t and 80 refer to time t and 1980 respectively.
The aggregate index was named the OECD index. Two other indices were
also used depending on the area being analysed: one for the US$ price alone,
called the NA index; and the other for the five nations excluding the USA,
called the OECDXNA index. All three indices are shown in Figure 15.2
The six nations chosen for the index covered 60% of the oil consumption in
the World Outside Communist Areas (WOCA) and 80% of the consumption
140
130
Weighted average of
OECD excluding US
prices
120
110
a0
Weighted average
of OECD prices
100
90
II
80
0
<Xl
OJ
70
x 60
Q)
"0
50
40
30
20
10
0
1960
1965
Fig. 15.2
1970
1975
1980
1985
Real oil price indices.
6When using the model to make projections of future price the price index becomes a Laysperes
index using base year weights. This switch will undoubtedly introduce some bias into the result.
However, in this context it should be noted that the EMF-4 study (Energy Modeling Forum, 1980)
found that elasticity estimates were not sensitive to the choice of index.
268
in the Organization of Economic Cooperation and Development (OECD)

countries during 1985. It is particularly worth noting that Canada was
intentionally excluded from the index because the C$/US$ exchange rate
remained constant in real terms over the 1980-3 period; its inclusion would
have biased the index downwards. The exclusion of Canada was broadly
counterbalanced by the exclusion of those West European countries that were
subject to depreciating exchange rates against the US$.
15.3.2 Crude oil price vs. petroleum product price
In the world oil market producers are faced by crude oil prices, whereas
consumers are faced by petroleum product prices which include government
taxes, oil company profits, distribution costs etc. In a simple model it is easier
if only one price is used; and tests showed that crude oil prices were a
statistically acceptable proxy for product prices in the demand functions.
15.3.3 Economic data
Real national accounts measures of gross domestic product were used in the
analysis because they represent a more valid picture of the volume of economic
activity than aggregates based on present or past market exchange rates. They
also enable real prices of internationally traded commodities - some of which
like oil are subject to the 'law of one price' - to be compared between
countries. The GDP data were derived from work published by Kravis et al.
(1980) and by Summers and Heston (1984) as part of the International
Comparison Project. 7
Nominal exchange rate and GDP deflators were taken from the ,IMF
International Financial Statistics Yearbook (1986). This allowed real exchange
rates to be calculated. It is recognized that for total consistency of approach
the exchange rates should have been constructed taking account of purchasing
power parity. However, at the time the initial work was undertaken, such data
were not available; this limitation no longer applies and future work will
address this problem.
15.4 DEMAND FUNCTION ESTIMATION
Numerous models of energy and oil demand have been proposed in the
literature. EMF-4 (1980) and Bohi (1982) both provide a comprehensive
overview of the model types and the functional forms employed. The complexity of these models varies from the static, based solely on own price and
income, through to the dynamic taking account of various fuels in different
7For more details see Roy (1987).
269
Demand function estimation
sectors across many countries or regions. The results reported here are
confined to the estimation of a simple single fuel dynamic model for highly
aggregated world regions. In making this choice it was recognized that factors
such as interfuel and factor substitution, producer and consumer tax regimes,
etc were being treated implicitly rather than explicitly.
Two alternative approaches were followed when modelling the demand side
- these can be characterized as 'top-down' versus 'bottom-up'. The top-down
approach involved estimating the demand function for WOCA directly. For
the bottom-up approach separate demand functions were to be estimated for
North America, OECD excluding North America and the Less Developed
Countries using the price index appropriate for each region. The separate
demand functions could then be aggregated to provide WOCA demand.
Despite repeated attempts, however, it proved impossible to estimate any form
of satisfactory demand model for the LDCs. This difficulty, combined with
problems associated with the other disaggregated demand functions and the
superior statistical characteristics of the WOCA demand function, led to the
top-down approach being incorporated in WOMS. As a consistency check, a
hybrid top-downjbottom-up approach was followed to provide demand estimates for OECD in aggregate and the LDCs. Only the results of the
estimations for the WOCA and OECD demand functions are reported here.
A dynamic linear regression (DLR) model of oil demand can be expressed in
terms of both current and lagged income and price measures and lagged values
of oil demand:
Dt=oc+
j=O
j= 1
L (Jlj"I t - j +c5 j"PIt -J+ L (Jj"D t -
+et
(15.6)
with all variables expressed in logarithms and where I is GDP, PI is the oil
price index, D is oil demand and e is the error term. In order to preserve
sufficient degrees of freedom during estimation, the value of n was limited to 2
in this exercise.
According to Spanos (1986) a DLR model with this functional form
encompasses a large proportion of the empirical econometric models in the
literature. Drawing on the work of Hendry and Richards it is possible to
identify at least nine special cases of this DLR depending on which parameter
constraints are imposed. Included with these special cases are the finite
distributed lag model, the partial adjustment model and the autoregressive of
order one model.
Following the imposition of various parameter constraints and the application of various statistical tests, it was decided that the most appropriate
statistical model to use was:
(15.7)
BCited in Spanos (1986).
270
which can be rearranged to give:

Rt =
/30 - /33' PIt + /32' Rt- 1 +et
(15.8)
where R is the oil demand: GDP ratio (DjI) or 'oil ratio'.

This functional form can be interpreted as either a partial adjustment model
or the Koyck transformation of the finite distributed lag model. The only
difference between the two forms is the structure of the residuals. In the partial
adjustment model the residuals are serially independent, whereas in the Koyck
transformation of the distributed lag model they are serially correlated. This
distinction is important because it leads to both different estimation techniques
and a different interpretation of the results.
The models were estimated using ordinary least squares regression and
rigorously statistically tested with particular attention being paid to first and
higher order serial correlation. In models with a lagged dependent variable the
Durbin-Watson statistic is not an appropriate test for first order serial
correlation as the results will be biased. Instead, a small sample version of the
Lagrange multiplier test was used to demonstrate whether serial correlation
was present. It was found that the residuals, e, were serially independent, thus
confirming that the demand function represented a partial adjustment process.
The models were also found to be structurally stable over the period 1960 to
1985. The results of the estimation and testing process are given in Table 15.1.
Having decided to use a partial adjustment model, it was important to
understand the implications. As commonly reported in the literature partial
adjustment models are based on the idea that consumers aim at some desired
Table 15.1 Demand side estimation results for the period 1961-1985
WOCA demand model
Dt = 1.0Gt -0.960G t _ 1 +0.960D t _ 1 -0.040Pt +0.140
(36.028) (-11.007) (10.157)
iP=0.987; DW=1.974; LM=3.217-F(1, 20); SE=0.014;

SSR = 0.0042; BP=6.014-xi; ARCH=0.009-xi;
GQ=0.088-F(9, 10); C=0.782-F(3, 19); W=212-xi.
OECD demand model
Dt = 1.0Gt -0.976G t _ 1 +0.976D t _ 1 -0.046Pt +0.157
(35.284) (-10.297) (9.407)
iP=0.986; DW=1.723; LM=0.376-F(1, 20); SE=0.017;

SSR=0.0062; BP=5.793-xi; ARCH=0.066-xi;
GQ=0.089-F(9, 10); C=0.757-F(3, 19); W=105-xi
D, oil demand; G, GDP; P, real oil price.
Notation for test statistics: DW, Durbin-Watson; LM, Lagrange multiplier for first order serial
correlation; GQ, Goldfeld Quant for constant variance; C, Chow for parameter time invariance;
BP, Breusch Pagan and ARCH for heteroscadasticity; W, Wald for common factor restrictions
Demand function estimation
271
or target level of demand; but due to a lack of perfect information and

structural rigidities they can only adjust towards this target gradually. However, the existence of a perceived target level to be achieved in the long run
gives the partial adjustment model of consumer behaviour a limited degree of
foresight. Admittedly this foresight is based on past and present information
but it does imply some direction to consumers' decisions.
Taking R~ as the target or desired level of the oil ratio as seen at time t and
R t as the actual level at time t, it can be demonstrated mathematically that if:
(15.9)
where r is the adjustment factor such that 0 < r ~ 1 and if the target level is a
function of price:
(15.10)
R~=(I.+8Pt
then combining equations (15.9) and (15.10) and rearranging:

R t =r(I. +r' 8 PI +(I-r)' R t - 1 + et
(15.11)
and hence comparing equations (15.8) and (15.11) gives:
f33= -r'8;
f3o=r(I.;
f32=(1-r)
which yields a long-run equilibrium price elasticity of:
8= -/33/(1-f32)
The characteristics of the demand functions are given in Table 15.2. The shortand long-run income elasticities are 1.0 as a result of the statistical tests for the
parameter constraints. Income elasticities of 1.0 are not surprising and have
been commonly reported in the literature: e.g. Adelman (1980), EMF-6 (EMF,
1982) and Gately (1983). The price elasticities reported are short-run and
long-run equilibrium elasticities. As a way of making comparisons with other
published estimates, price elasticities were calculated after specific time intervals and these are shown in Table 15.3.
The fact that the OECD price elasticities are larger than the comparable
values for WOCA is consistent with the large reduction in OECD oil demand
Table 15.2 Demand function characteristics
Elasticities
Short-run income
Price
Long-run income
Price
Median lag
Adjustment factor
'l:
WOCA
OECD
1.0
-0.040
1.0
-1.000
17 yr
0.040
1.0
-0.045
1.0
-1.875
29 yr
0.024
272

Table 15.3 Price elasticities at discrete time intervals
Region
10 years
20 years
25 years
WOCA
OECD
-0.335
-0.404
-0.558
-0.722
-0.640
-1.022
since the peak year of 1979. Between 1979 and 1985 OECD demand fell
7.2 bbl x 106 jday whilst WOCA demand fell by 5.8 bbl x 106 jday. This indicates
that demand growth in the LDCs partially offset the fall in demand in OECD.
Three comparable estimates of price elasticities for the WOCA region found
in the literature are shown in Table 15.4. Clearly the WOMS demand function
for WOCA has a smaller short-run price elasticity than other models. However,
the events of 1986 do appear to confirm that WOCA oil demand is indeed
inelastic in the short run. The estimate of additional WOCA oil demand during
1986 calculated by WOMS is 1.4 bbl x 106 jday which corresponds closely with
the actual outturn of 1.2 bbl x 106 jday.9 The 20-25-year elasticities show that
the WOMS price elasticities are broadly similar to those of others.
The median lag reported is a measure of the time taken for half of the
adjustment to occur in response to a change in price. This is sometimes referred
to as the 'half-life' of the demand response. As a first approximation it
represents half the time it takes for the entire capitaJ stock of the economy to
be replaced; it can also be taken as half the time needed to reach equilibrium.
The half-lives cited are longer than usually reported, but it is difficult to find
rigorous defences of shorter time periods. Adelman (1980) uses 10 years by
assumption and Huntington (1987) asserts that 17 years is too long. Hogan and
Leiby (1985) rely on their 'judgement' when imposing capital stock turnover
times that imply half-lives between 2 and 13.5 years. Daly, et al. (1982) also
make an assumption based on capital stock turnover times which leads to a
half-life of 6.5 years. These results (particularly for the WOCA model) compare
Table 15.4 Comparison of price elasticities for WOCA region
Model
Short run
10 years
20 years
25 years
WOCA
WOMS (1987)
Morrison
EMF-6 (1982)
Sweeny
and Boskin (1985)
-0.040
-0.065
n.a.
-0.09
-0.335
-0.35
n.a.
n.a.
-0.558
-0.60
n.a.
n.a.
-0.640
n.a.
-0.60
-0.60
9BP gives WOCA demand in 1985 as 45.27 bbl x 106 /day and in 1986 as 46.44 bbl x 106 /day.
Non-OPEC supply function estimation
273
favourably with the view given in EMF-4 (EMF, 1980) that 25-35 years
approximates to the time needed to reach equilibrium.
The adjustment factor (r) also gives an indication of the rate of capital stock
turnover. It can be taken as an approximation to the net amount of new or
replacement energy consuming capital stock installed each year. The results
suggest that 4.0% of WOCA and 2.5% of OECD stock are either newly
installed or replaced each year. This compares with average real GDP growth
rates between 1960 and 1985 of 4.2% and 3,9% per annum in WOCA and
OECD respectively. The difference in capital stock turnover times between
WOCA and OECD can be explained by the replacement of existing stock and
the addition of new stock from the LDCs within WOCA; whereas OECD is
primarily replacing existing capital stock. Again it is difficult to find comparable estimates that are rigorously defended. The report of EMF -4 stated that
'since the fraction of capital stock newly installed each year is small - certainly
less than 0.25 (i.e. 25%) and probably less than 0.1 (i.e. 10%) - the one year
adjustment parameter can be expected to be correspondingly small' (EMF,
1980). The results confirm the 'probably less than 0.1' element of this statement.
One way chosen to assess the validity of the results was to construct
'backcasts' of the past. These backcasts used actual data for price, income and
demand for 1960, together with the actual data for price and income in
subsequent years and the fitted values for demand. The backcasts indicate that
the WOCA model has a mean absolute error (MAE) of 0.5 bbl x 106/day within
the range of a maximum error of 1.7 bbl x 106/day and a minimum of
0.0 bbl x 106/day. This gives a mean absolute percentage error (MAPE) of 1.2%
for the WOCA model when operating between 1961 and 1985. The comparable
values for the OECD model are a MAE of 0.8 bbl x 106/day within the range
0.1-2.7 bbl x 106/day and a MAPE of 2.5%.
It was recognized that there is potentially an asymmetric demand response
to falling prices compared with rising prices: put another way, although
demand has fallen in response to sharply rising prices it will not necessarily
respond in an equal and opposite way to rapidly falling prices. Accordingly, a
time trend was included to test the hypothesis that a structural change within
the world economy could account for this type of response. The trend was not
found to be statistically significant. Hence, the demand functions presented
here incorporate a symmetric response to price movements; but because of the
great uncertainty surrounding this issue, the overall model was structured to
allow exogenous assumptions to be made on structural change. This question
will be re-examined when more post-1986 data are available.
15.5 NON-OPEC SUPPLY FUNCTION ESTIMATION
Non-OPEC supplies are taken to encompass all non-OPEC oil producers in
WOCA plus the net exports from the centrally planned economies (CPE). The
274
discussion here relates to the former as the CPE net oil exports are treated
exogenously. Non-OPEC producers are categorized as North American (NA)
and Other Non-OPEC (ONO).
Unlike the demand side, there are few supply side models available in the
literature and those that do exist tend to be difficult to replicate (compare
Chapters 6 and 8). Different modellers have used different approaches without
a consensus emerging on the appropriate way of estimating aggregate oil
supply. Gately (1983) uses an aggregate non-OPEC supply model of an
unspecified derivation that is based on expected and current prices. Hogan and
Leiby (1985) use the EIA (1985) supply model, which appears to be based on
some form of partial adjustment or Koyck lag model, although neither
provides evidence on the method of treatment of discoveries, development and
depletion of reserves. Both Sweeney and Boskin (1985) and Choe (1984) treat
supply exogenously and hence not directly related to price.
Some of the intertemporal optimization models use long-run marginal cost
curves which are both difficult to derive and not readily usable in a recursive
simulation model (e.g. Marshalla et al. (1985)). The work of Odell and Rosing
(1980) makes use of the various identities that relate supply and reserves but
again only accounts for price implicitly. Adelman and Jacoby (1979) propose
two versions of a supply model which they call 'disaggregated pool analysis'
and 'aggregated country analysis'; these were considered too detailed for a
simple model like WOMS.
Weyant and Kline (1982)10 use a method, based on the work of EMF-5 on
US oil and gas supply, which at first sight appeared promising (EMF, 1982).
This approach depends on establishing an econometric relationship between
the level of the oil price and the amount of oil discovered in that year, and
knowing the ultimate level of oil reserves that will be discovered. However, it
proved difficult to obtain consistent time series data for worldwide oil discoveries. The use of extentions to proven reserves, instead of discoveries, did not
provide a suitable proxy to establish a statistically acceptable relationship
between price and oil discoveries. In addition, Weyant and Kline's use of a
constant reserve:production ratio as a way of moving from reserve levels to
supply levels seems questionable (see Chapter 6). The data does not show
constant R:P ratios for countries outside North America and even the North
American ratio has been gradually declining rather than remaining constant.
Having rejected the approaches used by others, there was no alternative but
to construct an in-house model of non-OPEC supplies which would account
for both the geological and economic determinants of supply. Accordingly,
time was postulated as a suitable proxy for the geological factors that
determine the ultimate level of the oil resource. To determine the appropriate
IODaly, Griffin and Steele (1982) also use a similar model. Incorporated in their model are
unspecified 'institutional constraints' which prevent independent validation.
275
way to employ a time trend in this analysis reference was made to the work of
Hubbert (1979)Y
Hubbert's work on US crude oil reserves postulated that cumulative oil
discoveries would follow a logistic curve. During the initial phase cumulative
discoveries rise only slowly because knowledge of the underlying geology is
poor, exploratory technology is rudimentary and demand for oil is low. As
geological understanding grows and technology improves, cumulative discoveries rise at an exponential rate with most of the large fields in the province
being found in this period. After the largest fields have been discovered,
increasingly sophisticated and expensive techniques are employed to find the
remaining fields which have often been missed in earlier exploration. In this
final period, the rate at which cumulative discoveries are made approaches the
ultimate discovery level asymptotically.
This pattern for cumulative discoveries can be used to describe the rate of
oil production. Essentially the cumulative production curve is the same as the
cumulative discovery curve, but shifted to allow for the time needed to develop
fields. The annual rate of production is the first derivative of the cumulative
production curve. This produces a bell-shaped curve which represents the
geological determinants of oil production; and it is this curve which was used
as the appropriate time trend for the analysis.
Theoretically the correct approach should have been to determine the
appropriate logistic curve for cumulative discoveries, lag this by a predetermined time interval and then take the first derivative to determine the annual
supply curve. In practice this approach presents a number of problems. First,
there was the question of what lag to use between the discovery and production
curves. Second, the data concerning discoveries was limited to annual levels of
proven reserves, and therefore to annual additions to reserves rather than
discoveries. Although this was considered an acceptable proxy another problem was encountered using this data. The published level of proven reserves
shows considerable variations with time as reserves are re-evaluated, accountancy practices change or nations try to change their international credit rating.
In addition, as noted by BP (1979) and emphasized by Odell and Rosing (1980),
true levels of proven reserves at a given time can often be determined only 20
or 30 years after the initial assessment.
To resolve these problems supply, rather than reserve, data were used to
provide the necessary logistic curve, which thereby also provided the annual
production curve directly. Using the level of the ultimate oil resource derived
from the supply data it was then possible to estimate the cumulative reserve
curve. From this, identities were used to derive the annual level of proven
reserves and reserve production ratios as consistency checks.
"The inspiration to use Hubbert's work came from a report by Chase Econometrics (1986).
276
The logistic curve for cumulative supply is given by:

CS t = CS*/[1 +A "exp( -t/(X)]
(15.12)
where CSt is the cumulative supply at time t, CS* is the ultimate recoverable
supply of oil and A and (X are both constants.
The cumulative supply logistic curve was estimated in the form:
(15.13)
Po
where = InA; P1 = -1/(X and t is the error term.

The geological annual production curve is given by the first differential of
the logistic curve:
dCS t
GSt=Tt =CSt-CSt- 1
(15.14)
where GS is the geologically determined supply.

Following the same procedure as for the demand side estimation, the
economically determined supply can be expressed in terms of a dynamic linear
regression model. This includes current and lagged values of geological supply
and price, and lagged values of economic supply as follows:
St=(X+
j=O
i=l
L (J,tj"GSt- 1 +bj"PIt- j)+ L O"j"St-j+et
(15.15)
where S is economically determined supply, GS is as above, PI is the oil price

and e the error term. Estimations were conducted using n = 3 to preserve
degrees of freedom.
The cumulative discovery logistic curve was estimated in the form
(15.16)
where CS* is determined from equation (13), CD is cumulative discoveries (as
proxied by additions to proven reserves) and the other terms are as above.
The proven reserves, R" were determined from the identity:
(15.17)
t= 1
When estimating the logistic curves for cumulative supply and cumulative
discoveries it was necessary to have estimates of cumulative supply before 1960.
Data from Jenkins (1985) and Grossling and Neilsen (1985) were used to
provide estimates of cumulative production of 64 bbl x 109 in North America
and 8 bbl x 109 in the ONO nations prior to 1960.
The logistic curves were estimated using a grid search for the value of CS*
which minimized the sum of square residuals of the regression. This approach
worked successfully for the North American data where the level of discoveries
is already past the half-way point to finding the ultimate level of reserves. The
277
ultimate level of oil reserves in North America was estimated as 256 bbl x 109
by this procedure.
However, it was less successful in the ONO region where discoveries and
supply are still on the rising part of the respective logistic curves. In this case
it was not possible to find a level of CS* which minimized the sum of squared
residuals. Accordingly two alternative approaches were adopted. The first used
the Weyant and Kline (1982) estimate of total level of non-OPEC ultimate
reserves at 500 bbl x 109 and then subtracted the North American level of
256 bbl x 109 to give a rounded estimate of 250 bbl x 109 for the ONO region.
The second approach involved constructing a logistic curve for Non-OPEC in
aggregate which was found to minimize the sum of squared residuals at
650 bbl x 109 . By differencing the Non-OPEC and North American estimates
it was possible to arrive at a second estimate of the ultimate level of reserves
in the ONO region of 400 bbl x 109 The results of the estimations of the
various logistic curves are shown in Table 15.5.
Table 15.5
Logistic curve estimation results, 1960-1985
NA cumulative supply
In[(256 CS) - 1] = - 0.0704t + 1.044
( - 366.65) (373.05)
iF =0.9998; DW=0.261; SE=0.0073; SSR=O.0013
NA cumulative reserve additions
In([256 CR]-I)= -0.0724t+0.3413
( - 366.65) (373.05)
iF = 0.998; DW = 0.62; SE = 0.057; SSR = 0.0786

ONO cumulative supply (250 bbl x 10 9 ultimate reserves)
In([250 CS] -1)= -0.0858t + 3.475
(- 57.86) (160.723)
iF =0.993; DW=0.105; SE=0.056; SSR=O.077
ONO cumulative reserve additions (250 bbl x 10 9 ultimate reserves)

In([250 CR]-I)= -0.135t+2.705
(-79.29) (108.68)
iF = 0.996; DW = 1.458; SE = 0.065; SSR = 0.102
ONO cumulative supply (400 bbl x 10 9 ultimate reserves)

In([400 CS] - 1) = - 0.0825t + 3.945
(-65.84) (215.89)
iF =0.994; DW=0.11; SE=0.048; SSR=0.055

ONO cumulative reserve additions (400 bbl x 10 9 ultimate reserves)
In([400 CR] -1)= -0.116t+ 3.115
(-66.21) (122.26)
iF =0.994; DW=1.005; SE=0.067; SSR=O.l07
For notation, see Table 15.1.
278
When considering these estimates of the ultimate level of the oil resource in
the non-OPEC region it is important to note that these are the implicit levels
being assumed by the producers of oil derived from their production behaviour; that is, oil producers are supplying oil as if they believe the ultimate level
of the North American oil resource is 256 bbl x 109 or for the ONO region is
in the range 250-400 bbl x 109 . It is not intended as an estimate of the resource
base derived from a geological analysis of the regions.
Comparable estimates of non-OPEC ultimate reserves were not readily
available from the literature, because estimates are usually provided by
geographical regions rather than by producer groupings. However, it was
possible to make comparisons with estimates for North America. These
estimates of ultimate crude oil reserves are given by Nehring (1978), Master et
al. (1987) and for the lower 48 states of America by Hubbert (1979). Remembering that the WOMS estimates include NGLs as well as crude oil, a
comparison of estimates is given in Table 15.6.
The approach used to estimate the economically determined supply function
was the same as that used for the demand functions. A generalized model was
postulated, parameter constraints imposed and tested, the error term examined
and the model tested for structural stability. The results of this estimation
process are shown in Table 15.7.
This process revealed that the supply functions for both regions were
unstable either side of the 1973 oil price rise. Between 1960 and 1973 supply
was independent of the geological production curve and negatively correlated
with price. For this period a simple partial adjustment model based on lagged
supply and current prices adequately reflected supply behaviour. This suggests
that prior to 1973 supply was demand rather than geologically or price-driven.
The opposite applies post 1973, when supply is driven by geology and price
with the model based on geology, current price and price-lagged 3 years.
The divergence in behaviour either side of 1973 is not a surprising finding.
During the 1960s oil demand was growing at a rate of some 7% per annum
whilst prices declined in real terms; and oil supplies were developed according
to lowest cost of production. From 1973 onwards OPEC set prices and from
1982 onwards also controlled production to defend those prices. Prices have
Table 15.6 North American reserve estimates
Hubbert (1979)'
Masters et at (1987)
WOMS
Nehring (1978)b
'Crude oil in lower 48 states 0[- USA.
bCrude oil in Canada. Mexico and USA.
170
242.5
256
289-380
279
Table 15.7 Non-OPEC supply estimation results for each region divided 1961-1973
and 1974-1985
NA supply (1961-73)
St=0.914S t _ 1 -0.061Pl t +0.58
(30.44) ( - 3.12)
(3.56)
iF =0.99; DW=2.62; LM=1.12~F(1.8); SE=O.013; SSR=O.0017
NA supply (1974-85)
St = 1.66GS t -1.66GS t _ 1 + St-l +0.052Pl t _ 3 -0.191
(2.026)
(5.219)
( - 5.82)
iF =0.78; DW=2.42; LM=0.392~F(1.7); SE=0.014; SSR=0.178.
ONO supply (1961-73)
St =0.924S t _ 1 -0.301Pl t + 1.199
(28.05) (- 3.52)
(3.91)
iF =0.989; DW =2.192; LM =0.118 ~ F(1.8); SE=0.046; SSR =0.021.
ONO suppy (1974-85)
(250 bbl x 109 ultimate reserves)
St = 1.345GSt +0.149Pl t +O.lOPl t _ 3 -2.695
(10.245)
(2.786)
(3.073) ( - 8.215)
iF = 0.995; DW = 1.43; SE = 0.026; SSR = 0.0058.
ONO supply (1974-85)
(400 bbl x 10 9 ultimate reserves)
St = 1.178GS t +0.147Pl t +0.117Pl t - 3 + 1.983
(7.255)
(1.961)
(2.668)
(- 5.332)
iF =0.991; DW=1.29; SE=0.037; SSR=0.012.
S, annual supply; GS, geological supply; PI, oil price index. For other notation see Table 15.1.
been at levels considerably higher than the cost of production that would have
been incurred if oil reserves had been developed rationally. Non-OPEC
producers have responded to higher prices by developing resources previously
considered supra-marginal. With OPEC prepared to limit output non-OPEC
producers have been able to sell all the oil that they could produce subject to
the limitations of geology and profitability.
Following the procedure adopted for the demand side estimation, 'backcasts'
were again constructed to examine the accuracy of the various supply models.
Because the supply models were found to be structurally unstable either side
of 1973 the backcasts were constructed using the appropriate model for each
period to give a combined back cast for the entire period 1961-85. For North
America the mean absolute error (MAE) of the backcast was 0.1 bbl x 106/day
within the range 0.0-0.3 bbl x 106 /day. This gives a mean absolute percentage
error (MAPE) of 0.9. For the ONO region based on an ultimate resource of
250 bbl x 109 % the comparable figure was a MAE of 0.1 bbl x 106/day within
the range of 0.0-0.7 bbl x 106 /day and a MAPE of 2.5%. Slightly different
results were obtained from the ONO model with an ultimate resource of
280
400 bbl X 109 Here the MAE was 0.1 bbl x 106 /day within the range of 0.00.6 bbl x 106 /day giving a MAPE of 2.5%.
It is particularly worth noting that the North American supply function does
not include a current year price term; the analysis suggested that only 3-year
lagged price effects were statistically significant. The experience of 1986
indicates that current year prices are important. According to BP, North
American production fell by some 0.3 bbl x 106 /day during 1986 whereas
calcultions with the North American supply function show production only
marginally below the 1985 level. At the same time the ONO demand function,
which does include current as well as lagged prices, shows a fall of almost
1 bbl x 106 /day in 1986, when BP shows it actually increased 0.2 bbl x 106 /day.
This issue will be re-examined as more data becomes available.
15.6 STRATEGIES FOR OPEC
Analysis of the econometrically estimated functions shows that supply and
demand can be brought into balance for a range of mutually dependent values
of oil price and OPEC output. This allows OPEC to adopt one of two
strategies: eit~er it can set price, in which case its output is constrained; or it
can set output, in which case the price is constrained. This section addresses
OPEC's behaviour in either case and proposes a framework for assessing the
appropriate strategy.
15.6.1 OPEC sets price
Prices can be set either exogenously or endogenously using an empirically

derived price reaction function. As the first method is self explanatory only the
second method will be discussed.
Whilst there is a lack of consensus about non-OPEC supply models there is
better agreement on an appropriate model to simulate OPEC's price-setting
behaviour. Most models reported in the literature are similar to that used by
the EIA (1985), which is based on the view that OPEC sets (official) prices so
as to achieve an optimum level of capacity utilization. This is modelled using
a price reaction function in the form:
(15.18)
where PS is US$ the oil price and V is the level of capacity utilization (i.e.
production/maximum sustainable capacity).
The price reaction function was estimated using the EIA data of maximum
sustainable capacity; it proved difficult to achieve a good fit with other data.
The estimation results are shown in Table 15.8. The estimated reaction
function is compared with the data in Figure 15.3. Although the statistical
characteristics of this estimation are not wholly acceptable, this function does
281
Strategies for OPEC

Table 15.8 OPEC price reaction function estimation results, 1974-1985
(PSt-PSt-d/PS t - 1 = -0.240+0.047/(1- U)
(-4:633) (6.177)
IP=0.788; DW=2.467; SE=0.088; SSR=O.071.

ps, US$ oil price; U, OPEC capacity utilization (i.e. production/maximum sustainable capacity).
For notation see Table 15.1
100
90
~
0
Q)
Cl
c:
<IS
.c
u
Q)
.~
n;
~
80
70
60
50
40
30
20
10
0
-10
-20
-30
-40
-50
1985-
--
1984
1983
50
55
60
65
1982
1976
70
75
1978
-1975
80
85
90
95
100
Capacity utilization (%)
Fig. 15.3 OPEC price reaction function.

give an acceptable backcast of the past. When the fitted values of OPEC
capacity utilization are used the function has a MAPE of 5.2% over the period
1975-85. In the absence of any alternative formulation from other workers in
the field, it was decided to accept this function. 12 Lastly, it is interesting to note
that although this or similar reaction functions are widely used no estimation
results appear to have been published.
15.6.2 OPEC sets output
Although the price reaction function provides a mechanism for explaining

OPEC's price-setting behaviour over most of the period since 1973, it has
difficulty in explaining the price fall of 1986. During 1986 OPEC increased
rather than reduced production as prices fell. For most of the year crude oil
12Hogan and Leiby (1985) have also reported difficulty in obtaining a satisfactory fit with this
model.
282
prices were being determined by fluctuations in the crude and petroleum

product spot markets. This was not the first time, since OPEC took control of
the market in 1973, that spot markets have effectively determined official prices.
Observation shows that during 1979-81 spot prices led official prices by one
or two quarters. Verleger (1982) has demonstrated a statistical relationship
between spot and official prices between 1975 and 1980.
Clearly an alternative price-setting mechanism was required to account for
price excursions. It was assumed therefore, that sudden price excursions
occurred when OPEC either constrained output or attempted to flood the
market. 13 The mechanism adopted to simulate price excursions uses an
exogenously set OPEC production level and then iteratively solves for a
market clearing price to balance overall supply and demand. In effect a price
is found which stimulates demand and constrains non-OPEC production, or
vice versa, to bring the market into balance. The inherent short-run price
inelasticity of both demand and non-OPEC supply ensure that quite small
changes in OPEC output lead to large price movements.
15.6.3 OPEC trade-offs
A framework is required to test the success of various strategies for OPEC.

Their concern to increase revenues is self-evident; but their concern about
market share has emerged in recent years and was most explicitly stated in the
so-called 'market share' policy adopted at OPEC's December 1985 meeting
which led directly to the 1986 price collapse. 14 Market share in this context
was defined in terms of OPEC output as a percentage of WOCA oil supply.
(OPEC's market share fell from 60% in 1979 to less than 40% in 1985.)
Taking revenues and market share as the two key indicators, two complementary measures of success or failure were developed. The first looks at the
trade-off OPEC makes between price and volume to achieve revenue, and the
second plots the trade-off between revenue and market share.
The price/volume trade-off was taken from Gately (1986) and is similar to a
method used by Shell (1986) for demonstrating what is described as 'OPEC's
dilemma'. Oil export revenues are the product of price and export levels - a
given level of revenues can be achieved by a variety of combinations of price
and volume. By plotting the locus of intersection with revenue isoquants
through time, it is possible to see how successful OPEC is in choosing the
13The US Department of Energy (1987) provides evidence that during the 1973/74 Arab oil
embargo some 1.5 bbl x 106 /day was withdrawn from the market over a 6-month period and that
during the Iranian Revolution in 1978/79 an average of some 2.5 bbl x 106 /day was withdrawn for
a similar period. During 1986 OPEC successfully increased output by 2.2 bbl x 106 /day. Calculations with WOMS suggest that this is some 1.5 bbl x 106 /day more than they would have managed
if prices had remained constant in normal terms at the 1985 level.
14The communique issued at the end of the OPEC meeting of 7-9 December 1985 stated 'the
conference decided to secure and defend a fair share in the world oil market consistent with
necessary income for member countries development' (emphasis added). Cited in Petroleum
Intelligence Weekly, 16 December 1985.
Acknowledgements
40
(i)
35
Q)
.~
283
300
400
Constant
revenue
isoquants
50
1981
C. 30
1980
co
(j)
25
.0
.0
~ 20
Q)
Ll
.~
15
co 10
Q)
0:
5
0
1963
8
1969
12
16
20
24
28
32
36
OPEC exports (million bbl/day)
Fig. 15.4
OPEC trade-off No. 1.
correct mix of price and volume. This trade-off can be seen in Figure 15.4.
The second trade-off plots the locus of intersection with OPEC 'utility'
isoquants through timeY The term 'utility' is taken to represent OPEC's set
of preferences at a point in time and is defined as the product of export revenue
and market share. As with the first trade-off, OPEC are faced with the need to
choose the appropriate combination of revenues and market share which
satisfies their implicit utility function. The second trade-off is shown in Figure
15.5.
Both trade-offs are consistent with Gately's view (1984), mentioned previously, that OPEC 'is groping toward an unknowable "optimal" price-path'. The
loci shown in Figures 15.4 and 15.5 seem to indicate that OPEC is learning
from its efforts to manipulate the oil market. Indeed, it seems plausible that
OPEC will be able to learn enough about the adjustment processes at work
on both the supply and demand sides to gradually move towards some
'optimal' level of revenues and market share.
ACKNOWLEDGEMENTS
The authors are grateful for the comments they received from Robert Bacon,
Dermot Gately and Aris Spanos on the model and on earlier drafts of this
15The idea for this trade-off came from Dr R. W. Bacon and the Oxford Institute for Energy
Studies.
284
c:
.12
1i
<I>If)
::J
0
co
OJ
~
~
(/)
c.
120
'0
100
80
60
40
20
'0
Q)
:J
iii
>
260
240
220
200
180
160
140
Q)
Constant 'utility'
isoquants
150
280
1963
36
38 40
42
44
46
48
19651967 1969 1971

1964 1966 1968 1970
50
52 54
56
58
60
62 64 66
OPEC market share (%)
Fig. 15.5 OPEC tradeoff No.2.

paper which has earlier appeared in Energy Economics, July 1988. They also
acknowledge the help and comments they have received from their colleagues.
The views expressed in this paper are those of the authors alone and do not
necessarily reflect those of Power-Gen and National Power.
BIBLIOGRAPHY
Adelman, M. A. and Jacoby, H. D. (1979) Alternative Methods of Oil Supply
Forecasting. Advances in the Economics of Energy and Resources, 2, 1-38.
Adelman, M. A. (1980) Energy-income Coefficients and Ratios: their Use and Abuse.
Energy Economics, 2, 2-4.
Bohi, D. R. (1982) Price Elasticities of Demand for Energy - Evaluating the Estimates,
Final report on Research Project 1220--1, EPRI EA-2612, Electric Power Reseach
Institute, Palo Alto. Ca.
British Petroleum, (1979). Oil Crisis ... Again?, brief by Policy Review Unit, British
Petroleum Company Ltd.
British Petroleum (various years) Statistical Review of World Energy.
Brown, S. P. A. and Phillips, K. R. (1984) The Effects of Oil Prices and Exchange Rates
on World Oil Consumption. Federal Reserve Bank of Dallas Economic Review, July.
Chase Econometrics (1986) The Next Oil Shock, Bala Cynwyd, PA.
Chevron Corporation (1985) World Energy Outlook. Economic Department, San
Francisco.
Choe, B. J. (1984) 'A Model of World Energy Markets and OPEC Pricing'. World Bank
staff working papers no. 633.
Daly, G., Griffin, J. M. and Steele, H. B. (1982) Recent Oil Price Escalations:
Bibliography
285
Implications for OPEC Stability, in OPEC Behaviour and World Oil Prices, (ed 1. M.
Griffin and D. 1. Treece), London, Allen and Unwin.
EIA (Energy Information Administration) (1985) Annual Energy Outlook 1984, US
Department of Energy, Washington, DC, January.
EIA (Energy Information Administration) (1985) The Oil Market Simulation Model:
Model Documentation Report, US Department of Energy, Washington, DC, May.
EIA (Energy Information Administration) (1986) International Energy Outlook 1985. US
Department of Energy, Washington, DC, April.
EIA (Energy Information Administration) (1986) Monthly Energy Review. US Department of Energy, Washington, DC, December.
EMF (Energy Modeling Forum) (1980) Aggregate Elasticity of Energy Demand, EMF,
4, Summary Report on Research Project 875, EPRI EA-1559. Electric Power
Research Institute, Palo Alto, Ca.
EMF (Energy Modeling Forum), (1982) US Oil and Gas Supply, EMF-5, Summary
Report. Stanford University Energy Modeling Forum, Stanford, Ca.
EMF (Energy Modeling Forum) (1982) World Oil, EMF-6, Summary Report. Stanford
University Energy Modeling Forum, Stanford, Ca.
Gately, D. (1983) OPEC: Retrospective and Prospects 1973-1990. European Economic
Review, 21, 313-31.
Gately, D. (1984) A Ten Year Retrospective: OPEC and the World Oil Market. Journal
of Economic Literature, 22, 1100--14.
Gately, D. (1986) Lessons from the 1986 Oil Price Collapse. Bookings Papers and
Economic Activity, 2. 237-84.
Grossling, B. F. and Nielsen, D. T. (1985) In Search of Oil. Financial Times Business
Information, London.
Hogan, W. W. and Leiby, P. N. (1985) Oil Market Risk Analysis, Report of the Harvard
Energy Security Project, discussion paper no. H-85-03. Kennedy School of Government, Harvard University, Cambridge, Mass.
Hubbert, M. K. (1979) Hubbert Estimates from 1956-1974 of US Oil and Gas, in
Methods and Models for Assessing Energy Resources (ed. N. Grenon), Pergamon
Press, Oxford.
Huntington, H. G. (1985) The US Dollar and the World Oil Market. Energy Policy,
August pp. 299-306.
Huntington, H. G. (1987) A Reply: the US dollar and the World Oil Market. Energy
Policy, April.
International Energy Agency, (1986) Energy Policies and Programmes of lEA Countries:
1984 Review. OECD lEA, Paris.
International Monetary Fund (1986) International Financial Statistics Yearbook.
Kravis, I., Heston, A. and Summers, R. (1980) International comparisons of Real
Product and its Compositions; 1950--1977. Review of Income and Wealth, March.
Jenkins, G. (1985) Oil Economists Handbook. Elsevier, London.
Marshalla, R. A., Nesbitt, D. M., Haas, S. M. et al. (1985) A Model of the World Oil
Market with an OPEC Cartel. Energy, to, 1089-103.
Marshalla, R. A. and Nesbitt, D. M. (1986) Future World Oil Prices and Production
Levels: an Economic Analysis. Enegy Journal, 7, 1-22.
Masters, C. et al. (1987) World Resources of Crude Oil, Natural Gas, Natural Bitumen
and Shale Oil. US Geological Survey, Reston, Va.
Morrison, M. B. (1987) Will Oil Demand Recover? A Challenge to the Consensus.
Petroleum Review, July.
Nehring, R. (1978) Giant Oilfields and World Oil Resources. Rand Corporation for the
CIA.
Odell, P. R. and Rosing, K. E. (1980) The Future of Oil. Kogan Page, London.
286
Oil and Gas Journal (1985) World Oil and Gas at a Glance, OGJ, 30 December.
OPEC (1983) Annual Statistical Bulletin. OPEC, Vienna.
Pindyck, R. S. (1982) OPEC Oil Pricing and the Implications for Consumers and
Producers, in OPEC Behaviour and World Oil Prices (eds 1. M. Griffin and D. 1.
Taece), Allen and Unwin, London.
Petroleum Intelligence Weekly, various.

Roy, D. 1. (1987) International Comparisons of Real Value-Added Productivity and
Energy Intensity in 1980. Economic Trends, no. 404, June.
Shell, (1986) The World Oil Scene and OPEC. Shell Briefing Service no. 2.
Spanos, A. (1986) Statistical Foundations of Econometric Modelling. Cambridge Univer-
sity Press, Cambridge.

Summers R. and Heston, A. (1984) Improved International Comparisons of Real
Product and Its Compositions: 1950-1980. Review of Income and Wealth, series 30.
no. 2 June.
Sweeney,1. L. and Boskin, M. 1. (1985) Analysing Impacts of Potential Policy Changes
on US Oil Security. Energy Journal, 6, special tax issue, 89-108.
US Department of Energy (1987) Energy Security - A Report to the President of the
United States. US DOE, Washington, DC.
Verleger, P. K. (1982) The Determinants of Official OPEC Crude Prices. Review of
Economics and Statistics, 64, 177-83.
Weyant,1. P. and Kline, D. M. (1982) OPEC and the Oil Glut: Outlook for Oil Export
Revenues during the 1980s and 1990s. OPEC Review, Winter, pp. 333-64.
---16--International Energy Workshop

projections *
Alan S. Manne and Leo Schrattenholzer
The International Energy Workshop (lEW) is a loosely organized group of

energy analysts who all produce and/or make use of energy projections (see the
OPEC Review Autumn 1986 and Spring 1988 issues for earlier presentations).
Organized jointly by Stanford University and the International Institute for
Applied Systems Analysis (IIASA), the lEW collects energy projections submitted by its members from all over the world on a standard poll form. These poll
responses include projections of the international crude oil price, gross national
product growth, electricity generation, primary energy consumption, production and trade. For reasons of comparability, the responses are grouped into
the following regions:
Soviet Union and Eastern Europe (SU/EE)
China
OECD
OPEC
Non-OPEC developing countries (NODC)
Market economies, subtotal
World, total
Individual countries and other regions.
The poll responses are collected continually, and they are published semiannually. The publications normally include only those responses that are no
more than 3 years old. Another recurrent lEW activity is the organization of
annual meetings at which the lEW poll results are selected and focal themes
discussed. This chapter reports on the June 1989 meeting at the IIASA,
attended by some 120 participants. In additon to a presentation of the poll
results, this report gives an overview of the topics presented at that meeting.
The original idea behind the lEW was to learn from the diversity of different
*An earlier version of this chapter has appeared in the OPEC Review, Winter 1989 issue.
288
projections for the same item, the diversity stemming, of course, from the
different views taken by the individual authors. Due to the continued cooperation of the lEW members, this diversity in space can now be extended by a
diversity in time, i.e. by different views taken at different times. Moreover, what
used to be a medium-term projection in one year becomes short-term in
subsequent years. Due to its neutral position as an observer, the lEW thus
provides a valuable record of forecasting history. Since it would be inappropriate to classify projections as wrong or right, we shall instead speak of the
stability of forecasts (over time) and the degree of agreement between forecasters (at a given point in time). We shall present the lEW poll results in the light
of these two criteria, permitting the reader to judge how easy it was to foresee
the individual items.
16.1 THE INTERNATIONAL PRICE OF CRUDE
Figure 16.1 shows the histogram of several recent projections of oil prices. Each
poll response is marked by an asterisk, except that the median response is
marked by an 'M' (in the case of an even number of responses, the median lies
between the two responses marked with 'M'). The medians for the years 1990
and 2000 are $19 per barrel and $26jbl, corresponding to an average annual
growth rate of almost 3.2 per cent during the 1990s. The implicit annual growth
for the following decade is more than 4%, leading to a median price projection
of just under $40/bbl for 2010 (all prices are expressed in terms of constant
dollars of 1985 purchasing power).
$/b (1985)
~r----r----------------~-----------------T------
40
1----+------------------+-----------------+M.. M -
32~--+----------------+----------------+-----
MM
24 I----+------------------+----------------~------
16 I----+-~~-------------+~--------------~------
8 ~--~----------------~----------------~-----1990
2000
Medians: 19.00
26.05
Fig. 16.1
2010
39.35
Oil price projections: 1988 responses only.
289
The international price of crude
We regard the median as the most informative single number to describe

more than one poll response. A second informative statistic is the range. For
the oil price projections in 20()(), the range stretches from $15fbl to $45/bl. A
statistical measure of this range is the standard deviation, calculated under the
assumption that the responses are distributed log-normally.! The result is
shown in Figure 16.2 which, for each point in time, gives the interval which is
two standard deviations wide and centred around the mean. If the projections
were to describe the underlying distribution correctly, then this range covers
about two-thirds (68%) of the total probability. In any case, the range describes
the degree of agreement of the responses with each other and the inherent
uncertainty. With this definition, the range for the year 2010 is smaller than for
20()() - as if the uncertainty decreased with increasing distance into the future.
This may be attributed to some peculiarity in the poll responses, but it has also
been observed in past samples of oil price projections (see Manne and
Schrattenholzer, 1988).
The record of oil price projections for 1990 is depicted in Figure 16.3, in
which the lEW poll medians in a given year are compared with the current
price. Apparently, the current price has a significant influence on the forecasts.
Although it is understandable that expectations are adapted to changing
current conditions, this at the same time indicates that these conditions have
changed in a largely unforseen way. The most recent projection of $19/b (at
1985 prices) corresponds to about $23fb at 1990 prices.
$/b (1985)
50
~ ~--~~---------+----------~----~
30
I---+-----t-----:::
20 1-------'1----10
~----I---------~----------;_----------r_---
1985
1990
2000
2010
Fig. 16.2 Ranges of oil price projections: 1988 responses only.

'This means that we assume the logarithms of the oil price to be distributed normally. This
assumption not only excludes negative prices, but also assures that the range (100%, 200%) is as
probable as (50%, 100%) all percentages expressed in terms of the average value. The untransformed normal distribution would result in the meaningless projection that 'plus and minus 100%'
are equally probable.
290

$/b (1985)
~ r---------------------------------------------~
40
20
~
10 f-
:;
~~
~~
~.:
~~4:~
;:
~
__~~~~~~__~~~~~~~~~__~~~
1983
1984
1986
1985
~ Current price
f;;;;;;:::1
1987
1988
1989
Forecast for 1990
Fig. 16.3 Current oil price and projections for 1990.
16.3 TOTAL PRIMARY ENERGY CONSUMPTION

Figure 16.4 shows, again expressed as poll medians, the projected consumption
of total commercial primary consumption for five lEW regions which comprise
virtually the entire world. The values are given for 1985 and 2000. The
10~oe
5
4
r--------------------t
Su/EE
China
OECD
OPEC
NO DC
1985
Fig. 16.4 Total primary energy consumption in 1985 and 2000: 1989 poll medians in
five regions.
Total primary energy consumption
291
individual estimates add up to about 7,200 million tons of oil equivalent (mtoe)
for 1985 and 9,950 mtoe for the year 2000. On average, this corresponds to an
annual growth rate of just over two per cent. The regional growth rates.
together with the ranges defined above, are given in Table 16.1. The ranges are
significantly narrower than those for the oil price projections shown in Figure
16.2. Together with the time series of poll medians for the year 2000 (Figure
16.5), this clearly shows that it has been much easier to project energy
consumption than to project oil prices.
16.3.1 The consumption of individual fuels

Rather than giving a large amount of detail of the poll results on individual
energy carriers, we present Figure 16.6 as a summary of the projected changes
between 1985 and 2000, expressed as differences of the poll medians. In
Table 16.1 Growth rates and 68 per cent ranges, total commercial primary energy
consumption in five regions
Average annual
growth rate
1985-2000(%)
68 per cent range, year 2000
lower bound
upper bound
(mtoe)
SUjEE
China
OECD OPEC
NODC
Total
2.1
3.8
1.2
3.8
3.9
2.1
2.160
2.480
943
993
4.310
4.960
367
518
1.310
1.660
9.090
10.600
109toe
6
~-------------------~~.--------------------------~
~-------------------I
~-------------------
SUlEE
China
1::::::::::11983
~ 1984
~ 1987
ILQjJ 1988 DJIII] 1989
.1985
OECD
bd 1986
OPEC
NODC
Fig. 16.5 Total primary energy consumption for 2000, in five regions: polls 1983-9.
292

10&toe
1200
1000
800
600
400
200
0
-200
SUlEE
_
Oil
China
1:::::::;:;1 Gas
OECD
OPEC
rzl2ZI Coal Ed.dJ Nuclear
NODC
~ 'Conservation'
Fig. 16.6 Changes in consumption rates of primary energy sources plus 'conservation',
1985 and 2000: poll 1989.
addition to the supply sources (coal, oil, gas and nuclear), the figure also shows
'conservation', The quotation marks indicate that this is a derived value,
calculated as the difference between the hypothetical growth of total primary
energy consumption at the same rate as those implied by the GDP medians of
1985 and 2000 and the actual medians of total primary energy consumption
for the year 2000. A number of points are worth notiqg:
Despite significant reductions in the growth of the rate of energy consumption in the past, there remains great potential for energy savings in the
OECD region exemplified by the overwhelming share of 'conservation' in
Figure 16.6.
The regions SU/EE and China also show large amounts of 'conservation'.
The energy-rich OPEC region is the only lEW region to which energy
consumption growth is projected to exceed GDP growth.
The region SU/EE is the only one for which a decrease in oil consumption
has been projected. This is compensated for by enormous increase in
projected natural gas consumption. Gas attains a 43% share of total primary
energy consumption in this region by the year 2000.
This last point was implicit to a presentation by Professor Peter Odell at the
June meeting of the lEW. He argued that the consumption of natural gas in
Europe falls far below its potential because the marketing and distribution
systems have not been modernized since the era of city gas (see also Chapter
7). Thus inflexibility of old structures, underestimation of available supply and
subsequent regulatory measures aiming at a restriction of natural gas demand
Carbon emissions
293
to premium uses have reduced considerably the potential growth of natural

gas. Although Professor Odell views the continuation of the status quo as the
most likely scenario for the future of natural gas in Western Europe, he
mentioned the following new elements as possible factors to bring its share up
to 30% of total primary energy consumption (Odell, 1989):
The Single European Act, promoted by the Energy Directorate of the
European Economic Community and likely to be supported by environmental interests, could lead to greater competition - and eventually to the
abolition of national monopolies in the gas industry.
Changes in the Soviet Union are likely to encourage greater gas exports.
Perestroika is paving the way for Soviet gas to be used to earn foreign
exchange for Western European consumer and capital goods, and the export
infrastructure to export up to one-and-a-half times the current volume of gas
will be in place by the early 1990s. Up to 90 billion cubic metres could be
carried via pipelines to Germany, Italy and Scandinavia by the year 2000.
Many large manufacturing groups in Western Europe, as well as utility
companies, are interested in responding to larger gas supplies from the Soviet
Union.
An enhanced gas trade between Eastern and Western Europe is in line with
current East-West political developments and with environmental concerns.
16.4 CARBON EMISSIONS

The programme of the June 1989 lEW meeting reflected the worldwide
concern that increasing global concentrations of CO 2 and other 'greenhouse'
gases (such as methane and chlorofluorocarbons) may lead to significant
changes in climate. Figure 16.7 provides the point of departure for some lively
discussion. It shows the calculated amounts of global carbon emissions. These
amounts are based on the lEW poll medians of fossil fuel consumption and on
average carbon emission coefficients reported in Manne and Richels (Chapter
18). It can clearly be seen that the carbon emission problem is of a truly global
scale - no single world region (let alone an individual country) can hope to
solve this problem independently.
Furthermore, Figure 16.7 shows that much of the burden of a global
reduction in carbon emission would rest on the shoulders of the developing
countries, in particular China. It is highly questionable - to say the least whether these countries are able and prepared to undertake the necessary
economic effort to reduce the global carbon emissions significantly Manne and
Richels give estimates of between 0.8 and 3.6 trillion (million million) US
dollars (until the end of the next century, discounted at 5% per year) as the
economic cost of stablizing US carbon emissions to 80 % of the 1990 level by
the year 2020.
294
109 tons/year
10 .-------------------------------------------~
r-.-------------------
I- - - -.r:~: : :~:~: : :~:~: :'!:~: : '!~:~: : =~:~: : =~;~: : :~:~: :J:~ --- ~~~~~~:~~ --::::::::::::=:::::::::::::::
, ,
~
~i~i~i~i~i~i~i~i~i~i~i~i~i~i~
1985
1990
2000
flmI
US
rn
China
f::::::::J Rest of the world
Other OECD
2010
~ Su/EE
Fig. 16.7 Carbon emissions implied by poll medians.
The discussants did not agree on the measures that should be taken in the
near future. The opinions ranged from the immediate introduction of drastic
constraints on the use of fossil fuels to a 'wait-and-see' policy with respect to
the adverse consequences of future climatic changes. There was general
agreement on the necessity to monitor climatic change and to further understanding of the underlying processes.
16.5 POLL MEDIANS - A '3-2-1' HYPOTHESIS ON
THE DEMAND FOR ENERGY
The conventional wisdom is sometimes described as a '3-2-1' hypothesis. That
is, for the market economies as a whole, GDP will grow at the annual rate of
3 %, total commercial primary energy consumption at 2 % and oil consumption
at 1%. If we consider the 25-year period extending from 1985 to 2010, the poll
medians are roughly consistent with these numerical values (Figure 16.8). The
poll medians also imply that oil prices will grow at the average annual rate of
1.5% over this period.
Is there any underlying rationale to the seemingly independent projections
of these four growth rates? One approach would be to undertake a detailed
'bottom-up' analysis of the end-uses of energy within individual regions (see
Chapter 13). A bottom-up analysis provides realistic details that are not
available from aggregative economic models, but it is data-intensive, timeconsuming and does not always account for new uses of energy.
295
Poll medians
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0
GOP
Energy consumption
Oil consumption
Fig. 16.8 Growth rates of poll medians in market economies, 1985-2010.
As an alternative, it is instructive to see that there is a highly aggregative

economic model that provides a logical link between the four growth rates. The
model is parsimonious. It depends upon just two parameters - one that
summarizes the effects of price-induced energy conservation and the other
that summarizes the effects of autonomous (non-price, 'structural' or policyinduced) conservation.
Let the symbol AEEI denote the annual rate of autonomous energy
efficiency improvements and let ELAS denote the elasticity of the derived
demand for crude oil with respect to its international price. Also, let the
symbols OIL-PRICE, OIL-CON, TPE and GOP refer, respectively, to the
annual rates of growth of the international oil price, oil consumption, total
primary energy and GOP.
The 3-2-1 hypothesis is consistent with the two following equations. The first
indicates that, if overall energy prices remain constant, autonomous energy
conservation is the only factor that can account for a decoupling between GOP
growth and total primary energy consumption. Autonomous energy conservation must then occur at the annual rate of 1%:
TPE=GOP-AEEI
2=3-1
(16.1)
The second equation refers to oil consumption - and to the impact of oil
prices upon the growth in demand. Implicitly, equation (16.2) refers to interfuel
substitution between oil and other forms of primary energy. If crude oil prices
grow at the average annual rate of 1.5% (in real terms) from their 1985 level,
equation (2) implies that ELAS (the price elasticity of demand for oil) must be
296
2/3:
OIL-CON = GDP - AEEI - (ELAS) OIL-PRICE
1
3
1 - (2/3)
1.5
(16.2)
These two aggregate equations are easy to understand, and they provide a
cross-check on more detailed models of energy futures. Together, they are
consistent with the view that both autonomous and price-induced conservation
are significant determinants of the growth in oil demand. They depend
critically, however, upon the estimates of AEEI and ELAS.
In the following set of controlled comparisons between models participating
in the lEW poll, we shall see that there is a great deal of variability in the
implicit values of these two parameters. Accordingly, one should not be
surprised to see that there is variability in the poll responses. We shall also see
that there is a much wider dispersion in the projections of price than in the
quantities supplied and demanded.
16.6 A CONTROLLED COMPARISON OF POLL RESPONSES
In conjunction with the Energy Modeling Forum, the lEW has undertaken a
controlled comparison of oil demand under two scenarios - an 'upper' and a
'lower' trend of international crude oil prices between 1985 and 2000. For
purposes of this comparison, the rate of GDP growth was to be held
approximately constant. As of the June 1989 lEW meeting, responses had been
received from eight modelling groups:
AMOCO
CEGB
CONCO
ECON
EIA
GRI
lEA
LTM
AMOCO Corporation
Central Electricity Generating Board, UK (now Powergen)
Conoco Corporation
ECON, Centre for Economic Analysis, Oslo
Energy Information Administration, USA
Gas Research Institute, USA
International Energy Agency
Long-Term Model, Stanford University, USA
Because several of the modelling groups reported oil demand, but not total
primary energy consumption, there is no direct way to use their results to infer
the AEEI. There is, however, a close relationship between the AEEI and the
oil consumption-GDP elasticities that are implicit in the individual models.
For example, under the 3-2-1 hypothesis - with constant oil prices - the oil
consumption-GDP elasticity would be 2/3 (recall equation (16.2) and insert
zero as the rate of oil price growth).
Figure 16.9 indicates the GDP elasticities of oil consumption at three
different levels of regional detail: the US, the OECD and the market economies
as a whole (see further Chapter 14). In most cases, these elasticities are close to
297
A controlled comparison of poll responses
1.0.-----------------------.,
I:":'~
0.8 I - - - - - - - = . . . , . = , = - - - - - - - - - - - - - - - - - - - I : { ,
::::1'"
,
,
~~~
r-
.... i'~
I-I'""-,--'L~ ---=,~-r_--I':'--~---I ---~-::::--
0.6
",
'~',
:::,,'--1-'1- I' ,::: "'
,::::,
:0:0'"
0.4
..,"" ,
..
::::...
...
': : : ' , '
0.2
::: ,
"r-
.'."
::: '
o ~~:::L'~
AMOCO
us
~r<,
1::::1':':1-'
::::~,.:.:
"
"::.::::.:: :~
:.::.:1-.
1-'
CEGB CON CO
ECON
..
:. . :.. 1-,
~} _
:::: '.'-
::::1- ~
"'1-.
::::
::::
,- I---I~~.~:.'- t.:~..:"
::.::. '
: :.: :.
:.:v,.
.... ...
::::
..
.:.: It.
::::,
t~~
..
~:.:~ ',-
::::
---I?.:.:
:.: v
:::It~
:::v~
:::v.
__~~'~~::~::~~~~~~:~::L'L_~______~:::~:__~~~~
,,
(:;:;:;J
~':".~.:
OECD
"
::: It'
EIA
GRI
::::
.:.~..:.~V
lEA
LTM
~ Market economies as a whole
Fig. 16.9 GDP elasticities of oil demand.
the value of 2/3 that is implicit in the 3-2-1 model. The principal exception is
LTM, where the authors (Manne and Rutherford, 1989) adopted unitary
long-term GDP elasticities and made no provision for the role of autonomous
energy conservation. The econometric evidence on this issue is conflicting.
Both Hogan (1988) and Brown and Phillips (1989) found no reason to reject
the hypothesis of unitary GDP elasticity.
0.7.---------------------------------------,
0.6
.. ------------~~=.-----~
r-~--------I
0.5 r - - - - - - - - - - I
0.4 r-------F~~----I
0.3 t - - - - - - -
0.2
0.1
?::::::::
--
:::v - -
~h
r~
f::
:::: , 1 - - - - - - - - - 1 :.:
::::.' ~
tt
./::-). .
-~---1r--
_- .
---
H:
rnb- t, - ; -~~
':.~ :
r- - -
~ ~.:
/., -
~: ~:. \l.J.l----l
_-
Ou-~~__~~~~~~~~~~~_____W~_L~~~~~:::L~1J
AMOCO
Dus
CEGB CONCO ECON
E:::::::::~ OECD
EIA
GRI
lEA
LTM
~ Market economies as a whole
Fig. 16.10 Price elasticities of oil demand.
SPIL
298
There is a wide variation in the oil demand price elasticities that are implicit
in the eight models (Figure 16.10). Only a few of these are high enough to be
consistent with the long-run value of 2/3 suggested by the 3-2-1 model. Various
interpretations were suggested by workshop participants. One possibility is
that there is a significant difference in demand elasticities for the IS-year period
1985-2000 and the 25-year period 1985-2010. Tax policies differ from one
region to another, and this may in turn affect demand elasticities. Another
possibility is related to the difference in the elasticity of demand for white and
black petroleum products. This would suggest that the price elasticity of the
derived demand for crude oil is very different at different price levels.
REFERENCES
Brown, S. P. A. and Phillips, K. R. (1989) An Econometric Analysis of us Oil Demand.
Federal Reserve Bank of Dallas.
Hogan, W. W. (1988) Patterns of Energy Use Revisited. Harvard University, Cambridge,
Mass.
Manne, A. S. and Rutherford, T. F. (1989) A Long-term Model of Oil Markets, Economics
Growth and Balance of Payments Constraints. Stanford University, Stanford, Ca, and
University of Western Ontario, London, Ont.
Manne, A. S. and Schrattenholzer, L. (1986/88) The International Energy Workshop: a
progress report. OPEC Review, Autumn 1986 and Spring 1988.
Odell, P. R. (1989) Der westeuropiiische Gasmarkt: Gegenwiirtige Situation und
alternative Zukunftsaussichten, Zeitschrift fur Energiewirtschaft, 12(2).
---17--Environmental regulations and

innovation: a CGE approach
for analysing short-run and
long-run effects
Gunter Stephan
17.1 INTRODUCTION
As is well known, there is a substantial gap between the rigorous and
theoretically derived definitions of the costs and benefits of environmental
regulations and their empirical estimates. Nevertheless in applied cost-benefit
analysis four criteria have proved convenient, if environmental improvements
forced by governmental interventions are to be evaluated (see Siebert, 1987)
1. economic efficiency;
2. ecological efficiency;
3. distributional effects;
4. political feasibility.
Although these criteria are important in quantifying the overall impact of

environmental regulations, they are static and will typically be used in
comparative static analyses. Environmental problems, however, have a temporal dimension. Today's emissions influence the quality of the environment in
the short- and in the long-run; an economy cannot adjust to a change of
environmental policy without short-run frictions; and environmental regulations feed back into economic growth which is closely related to invention,
innovation and technical progress on the one hand, and to resource use and
pollution on the other.
The interplay between invention, innovation, resource use and pollution is
particularly important, since technological progress in pollution abatement can
significantly expand the boundaries of environmental policy design. Some
authors (for example, see Milliman and Prince, 1989) argue that the impact of
300
Environmental regulations and innovation
regulations on the pace of innovation is over the long run the most important
criterion to judge environmental policies and the key to an effective solution
of environmental problems.
With this perspective in mind and considering the importance of environmental restrictions on the production and consumption of energy, it is the aim
of this paper to explore an intertemporal computable equilibrium framework
which allows us to analyse the intertemporal impact of environmental regulations on the allocation of resources, the distribution of income, on economic
growth and innovation. After discussing in section 17.2 how policy instruments
may differ with respect to their short-run and long-run effects on innovation
and technological change, section 17.3 is a verbal presentation of the theoretical framework. 1 Section 17.4 illustrates the functioning of the theoretical
approach by means of a numerical example.
17.2 INNOVATION INCENTIVES OF EFFLUENT
CHARGES AND STANDARDS
In economic literature it is generally argued that market conform applications
of the so-called 'polluters pay' principle are the best policy tools for controlling
environmental externalities (for example, see Baumol and Oates, 1988). Via the
invisible hand, effiuent charges and environmental taxes guarantee an efficient
and Pareto-optimal allocation of resources. They stimulate firms to lower
emissions down to the level where the marginal costs of reduction are equal to
the unit rate of the charge or tax. Direct governmental regulations like
standards on emissions or technology are less likely to promote static efficiency.
There are, however, conditions in which direct governmental interventions
by standards are preferred. This might be the case in situations where toxic
pollutants have to be reduced immediately, where high probabilities exist that
emissions exceeding certain limits lead to irreversible damages, but also for
situations in which cost functions are hard to estimate or when market
conform instruments for various reasons would be impractical or very costly
to operate. Moreover, studies on the political feasibility of policy instruments
stress that direct governmental regulations have higher chances to be accepted
than market conform applications of the polluters pay principle (see Lee, 1984).
Important reasons for this observation are that standards on emissions or
technology are easier to implement in existing legal regulations, seem to be
more reliable than an effiuent charge and can be changed more efficiently and
rapidly in urgent situations.
Despite these typically short-run advantages, environmental standards seem
to have the disadvantage that they provide almost no dynamic incentive for
1
The reader interested in a formal presentation should consult the Appendix.
Modelling innovation and adjustment in applied analyses
301
innovating less polluting technologies. It is argued in most articles that under

direct regulations polluters have to abate only those emissions that exceed the
upper limits set by government, whereas the remaining emissions can be
discharged at price zero. Consequently, firms have no economic incentive to
lower emissions below the given level by employing less polluting technologies
and/or advanced abatement technologies.
The situation is even worse if technology standards are prescribed. An
economic agent who intends to implement more sophisticated production and
purification technologies would have to change the existing technology standards and the other polluters will then be forced to adjust their techniques.
Hence, producers are motivated to avoid new technologies becoming standard
which is a disincentive for technological change, invention and innovation
(Faber and Stephan, 1987).
One should be aware, however, that technological change is performed in at
least two steps (for a reference in the industrial organization literature, see
Greer, 1984): (1) invention of the initial idea, including a crude proof that the
new technology works on an industry level; and (2) innovation which is defined
as the process of capital accumulation and structural change in order to adopt
the new and approved techniques.
With respect to innovation the traditional argument overlooks that imposing a standard on emissions and or technology has at least two effects. First,
emissions have to be reduced by employing suitable reduction strategies. The
strategies at hand are either to install an end-of-the-pipe-abatement technology, or to reduce production and consumption or to substitute inputs. The last
two options can usually be used only to a limited extent, and if so, over the
very short-term only. Secondly, implementing an end-of-the-pipe emission
abatement technology and satisfying purification standards implies additional
costs. Under perfect competition producers cannot avoid these costs. Hence,
the only way to get rid of regulation-induced 'environmental' costs is to look
for alternative and less polluting production technologies. In other words, if
the overall costs of installing and operating less polluting technologies are
lower than the costs of fulfilling effluent standards by using suitable endof-the-pipe abatement technologies, standards have a long-run effect on innovation, investment and the development of an economy which in our particular
case, is similar to the long-term impact of effluent charges. Exactly such an
outcome will be observed in section 17.4.
17.3 MODELLING INN OVA nON AND ADJUSTMENT
IN APPLIED ANALYSES
Most of the existing literature dealing with the interplay between technological
change and environmental regulations is of a partial nature and neglects the
impact of pollution control policy on the economy as a whole (for an overview,
302
see Milliman and Prince, 1989). It seems therefore necessary to establish a

general equilibrium approach which enables us to take the temporal structure
of production and investment explicitly into account, to illustrate the innovation process and its effects on an economy as a whole, and to analyse efficiency
as well as distributive consequences of a specific environmental regulation.
The instruments at hand for analysing environmental problems on the
economy level are input-output models, the optimization approach and the
computable general equilibrium (CGE) framework. The first two methodologies neither satisfy the three requirements above nor do they contain variables
which can be regarded as policy instruments in a market economy.
Input-output models are generally characterized by fixed production coefficients and demand structures. Hence, price-dependent substitution or the
innovation of new production techniques cannot be analysed and the criterion
of economic efficiency is not applicable.
An optimization approach permits a more flexible treatment of the production side of an economy, but is still rigid in the requirements placed on the
consumers' side. Generally, it starts from a centrally planned economy with a
single decision-maker who is optimizing a social welfare function. This implies
that the analysis can be concentrated on efficiency and long-run development,
but is not designed for analysing distributive and short-run effects.
The most consistent way to capture allocational as well as distributional
effects is to model all agents in a general equilibrium framework. As a
qualitative description of a competitive economy, the general equilibrium
model based on Walras's conceptional idea is already in its second century of
intellectual life. But as a quantitative tool its history is much more recent. It
was not until Scarf's pioneering work (1967) that rigorous techniques became
available to solve such models numerically. And to our knowledge no other
tool developed so far has the ability to be applicable for theoretical as well as
empirical analyses, to trace the consequences of a change of policy through the
entire economy, to provide a unified framework for analysing the trade-offs
between economic efficiency and equity and to illustrate the operation of an
economic system in which all decisions are price-guided on a micro level.
Despite these advantages computable equilibrium models are unsatisfactory
in many respects, since they rely on abstractions which can be regarded as
unrealistic and have been questioned by many economists (for a discussion, see
Hahn, 1989). For example, it is generally supposed that economic agents
behave as price-takers and maximize profits or utilities. Equally unrealistic are
the assumptions about the organization and functioning of markets: markets
are perfect and well organized, the market process depends on price signals
only and automatically assures an equilibrium.
17.3.1 Time and adjustment in computable equilibrium models
More important for our aim of providing a unified framework for analysing
innovation and adjustment, efficiency and equity is that there exists no easy
303
way to introduce time and dynamics into a computable equilibrium model.

In formulating dynamic computable equilibrium models it is state of the art
to distinguish between two approaches (Stephan, 1990). One is called 'clairvoyant' and the other is labelled 'myopic'. The first implies the concept of an
intertemporal equilibrium in the sense of Arrow and Debreu, and the second
leads to a recursive structure in which the economy evolves in a sequence of
(intertemporally uncoordinated) temporary flow equilibria.
Within an intertemporal Arrow-Debreu framework it is assumed (1) that all
commodities are traded on a complete set of well-organized future markets,
and (2) that at the initial moment economic plans are made and coordinated
for the whole economic horizon. After that markets are closed and will never
be reopened.
These assumptions significantly influence how time, innovation, adjustment
and the evolution of an economy are reflected by a clairvoyant approach.
Broadly speaking, an important role of future markets is to disseminate
information regarding future supply and demand conditions. Hence, with
future markets for all commodities, economic agents can correctly forecast the
future states of the economy and fully utilize this knowledge in decisionmaking. Furthermore, since all commodities, including all types of durable
factors of production (capital goods) and pollutants, are traded on perfect and
well-organized markets, the existing production system can be adjusted at any
time to changes in the state of the economy simply by selling used and buying
new capital goods.
Adding these two effects, a clairvoyant approach implicitly supposes (1) that
an economy can react instantaneously to any change of prices and exogenous
parameters, and (2) that capital stocks, hence technologies, can be changed
from period to period, and innovation, the diffusion of new techniques into an
economy, is almost timeless. Consequently, the clairvoyant approach may be
used to study economic development which is based on long-run equilibrium
price formations, but short-run adjustment, temporary frictions and the irreversibility of economic processes cannot be analysed within such an approach
(Stephan, 1989).
Additionally, an intertemporal equilibrium framework usually neglects that
intertemporal information structures are asymmetric. In reality, the past and
present events may be certain, but future ones are definitively not known for
sure. Some of the future events can be associated with objective or subjective
probability distributions based on past experience, but the future may also
contain novelty that is definitely unknowable, and thus cannot be associated
with probability distributions.
In contrast to a clairvoyant approach, myopic models suppose (1) that all
decisions are made period by period, and (2) that agents use only the
information available in the current period. One reason for such behaviour in
decision-making is that no future markets exist. Hence, there is no institutional
structure which enables exchange of information about the future or coordination of intertemporal plans via market processes. Economic transactions take
304
place on spot markets only and the economy evolves in a sequence of flow
equilibria.
Some authors (for example, see Faber et a!., 1990) argue that a myopic model
is a more realistic approach, since it allows for reopening of markets and
accounts for the fact that agents might be completely ignorant during decisionmaking. Such a myopic model structure captures the second feature of time
mentioned above. The advantage of a clairvoyant approach is, however, that
the economic development and decisions are handled in a logically and
intertemporally consistent way. Savings and investment result from intertemporal optimization which incorporates future development and expectations
systematically. A myopic approach may easily lead to implausible behaviour
with economic agents repeating the same mistakes from period to period.
17.3.2 Innovation and production: neoclassical and neo-Austrian approaches
Independent of whether a clairvoyant or a myopic approach is used, the
resulting general equilibrium structures share one important property: for
every period and commodity a well defined market exists: spot markets in the
case of a myopic model, future markets in an intertemporal framework.
This allows us to employ a conventional approach to intertemporal production theory, where the intertemporal course of production is vertically disaggregated into a sequence of one-period production activities which are linked
by market transactions (Stephan, 1988a). Formally, intertemporal production
is described by two elements: (1) a technology set V which represents the
technological knowledge available, and (2) a sequence of one-period production activities {(x(t),y(t)), t= 1, 2, ... }. At the beginning of period t= 1, 2, ... the
input vector x(t) is used to produce the output vector y(t) subject to the
condition that the pair (x(t),y(t)) is an element of the technology set V. x(t) and
y(t) are vectors of the same dimension and include a complete list of all goods
in process; this means, final goods such as consumption goods, primary inputs,
intermediate factors, but also all types of capital goods, pollutants etc.
Figure 17.1 illustrates how a conventional approach characterizes the
intertemporal course of production. A possible interpretation is that produc-
Markets
Markets
Markets
t+ 1 Time
t-1
____________________________________
x(t-1) - - - . , y(t-1)
VI
~ y(t)
x(t)
Fig. 17.1
Conventional approach to intertemporal production theory.
305
tion is carried out by a sequence of producers. Each producer operates his

enterprise for only one period. At the beginning of the period he buys at
existing equilibrium prices all input factors which are technically necessary to
produce the desired output starting from scratch. At the end (i.e., the beginning
of the following period) he sells his total output at current prices, including his
stocks of used, depreciated and semi-finished capital goods, emissions etc.
Seen theoretically, the vertically disaggregated characterization of intertemporal production has the advantage of being transparent and logically consistent (Burmeister, 1974). But with respect to an adequate treatment of time,
innovation or adjustment, and in view of potential empirical applications, these
assumptions become crucial for at least three reasons (Stephan, 1988b):
1. Describing production in such a way means that the model-builder must
have complete knowledge about all goods and services used during production, independent of whether these goods are marketed or not. On the
contrary, this approach does not allow us to distinguish between goods
which are actually bought or sold by firms and factors which are owned by
firms, which are never sold or for which no markets exist. Equilibrium
prices are generated for each commodity and period and any commodity is
marketed.
2. Since durable factors of production can be sold or bought at any point in
time at existing equilibrium prices, the temporal structure of constructing
and demolishing capital stocks is not explicitly modelled. Hence, such an
approach does not reflect the fact that particular goals, like innovating less
polluting technologies, require time and a particular temporal sequence of
actions: for example, if a specific pollution abatement technology is to be
operated, the necessary capital equipment has to be constructed first.
3. Every production activity embodies a certain production process and thus
a certain technological knowledge. As discussed above, the conventional
approach permits the use of different production processes from one period
to another. This means that production technologies can change abruptly.
For this reason, a conventional production theory can neither consider
short-run restrictions resulting from the immobility of capital stocks nor is
it suited to analyse the time profile of an innovation process.
To overcome some of the short-comings just mentioned, this chapter
presents an alternative formulation of a computable equilibrium model which
is built on two elements (Stephan, 1988b, 1989, Appendix).
1. In order to determine the long-run behaviour of economic agents in a
logically consistent way a conventional market equilibrium framework is
employed as far as possible. That means: all individuals are clairvoyant (i.e.,
they have identical and certain price expectations) and behave as pricetakers. Households optimize intertemporal utilities, producers maximize
306
present values of profits. Money is neutral and the system is driven by

relative prices. Environmental regulations are imposed by governmental
interventions and have to be obeyed by every producer at any time.
2. In order to allow for time lags in adjustment and the innovation of
techniques there is, however, an important point of departure from the
neoclassical paradigm. We suppose in the following that the set of commodities is divided into two disjunct subsets: complete and well-organized spot
and future markets exist for all non-durables such as consumption goods,
labour, imports and exports, but do not exist for durables like used or
semi-finished capital goods and emissions. By this assumption, markets for
durable factors of production might be tight and restrict production
possibilities in the short-run. Hence, time is required to adjust to changes
in prices or exogenous parameters.
It should be obvious that such an assumption is incompatible with a
conventional approach to intertemporal production theory. Instead of assuming that production can be vertically disaggregated into a sequence of oneperiod activities with a complete reference back to the market, we now have to
employ a vertically aggregated characterization of production where producers
cannot sell or buy the necessary capital equipment at equilibrium prices, but
have to build up their own capital stocks.
The vertical aggregation of production is the specific feature of the Austrian
capital theory (see Hicks, 1973; Faber, 1986). In the Austrian view, production
is a process in time starting from a sequence of dated primary inputs which
mature to final outputs. The delay between inputs and outputs is subject to a
variable time distribution and pure intermediate goods as well as capital goods
are subsumed in the description of the production process.
Describing intertemporal production formally, the principal difference between a conventional and the neo-Austrian approach can be summarized as
follows: in conventional production theory the characterization of production
is mainly based on the concept of production processes (activities). In this way
a technically feasible correspondence between inputs and output is defined, but
the fact is not reflected that applying a certain technology requires a suitable
structure and organization in the production system. From a neo-Austrain
viewpoint production has to be formalized by applying two concepts: the
notion of a production process which is similar to the conventional one, and
the concept of a production technique. Following Faber and Proops (1990) (see
also Stephan, 1988b), a production technique is a minimal combination of
production processes which allows us to produce the final output (consumption goods) from non-produced (primary) inputs. Such a production technique
contains all stages of production including the construction of the necessary
intermediate goods and capital equipment. Applying the concept of a production technique means therefore that in characterizing the intertemporal course
307
of production a third phenomenon 2 of time is reflected: innovation and

technological change are time-consuming, since technologies and production
structures have to be provided in a temporal structure before a specific
production process can be applied.
In this analysis we adopt the Austrian interpretation of a production
technique. But, instead of dealing with primary and final goods only, we
develop a characterization of production which is based on marketed and
non-durable commodities. To this end, we modify the concept of a production
technique (for the original definition, see Hicks, 1973).
In our view, a production technique which is implemented in period t
consists of three elements:
1. A production process (k(t),x(t),y(t)) chosen from the technology set V which
is as usual the sum of technological knowledge available over the time

horizon considered. In contrast to the conventional notion above, durable
inputs are separated from non-durable ones. Hence, k(t) denotes the services
of a capital stock which is technically necessary to operate the production
process (k(t),x(t),y(t)), and the vectors x(t) and y(t) represent the inputs and
outputs of non-durable and marketable commodities only.
2. A sequence of non-durable inputs x(t - s), s = 1, ... S, which go into a capital
gestation process (x(t-S), ... , x(t-l)) to build up the capital stock k(t). This
reflects the fact that there are no markets for new capital goods. Hence,
capital stocks have to be constructed by each producer.
3. A sequence (x(t+ 1),y(t+ 1), ... ,x(T),y(T)) of non-durable inputs x(t+s) and
outputs y(t + s), s = 1, ... , T - t, to describe the production which is feasible
from this technique in subsequent periods. Once a capital stock is established, it cannot be sold and the production process (k(t),x(t),y(t)) has to be
applied in future periods. But capital goods may deteriorate and additional
inputs are necessary to maintain the service of an existing capital stock
(Stephan, 1988b). Hence, the level of production in future periods depends
on the depreciation of the existing capital stock, and the expenditure for
maintenance.
We thus have a vintage production function similar to the one described in
Chapter 2 of this volume.
Because of the lag structure of inputs and outputs it takes time to implement
new production processes and the existing technologies cannot be abandoned
immediately (see Figure 17.2, in which the gestation of capital stocks takes one
period). Thus the short-run substitution potential, and adjustment possibilities,
are restricted by the set of production techniques which already exist. But over
2 As mentioned earlier the other two features of time are asymmetry between future and past, and
novelty (Faber and Proops, 1990).
308

Production technique (used first in period t)
Gestation period
-L
__________
Production process
~I
Depreciation
___________LI__________
t-1
(x(t-1 ). k(t))
(k(t) . x(t) . y(t))
t+1
~I__..
Time
t+2
(x(t+ 1) . y(t+ 1))
Fig. 17.2 Neo-Austrian concept of a production technique.
the long term, new production processes can be introduced and the short-run
substitution possibilities are small compared to the economy's long-run adjustment and innovation potential. This is in contrast to a conventional approach
to production theory, where the short-run and the long-run substitution
potential are identical (Figure 17.1).
17.4 NUMERICAL ILLUSTRA nON

For a simple general equilibrium model which consists of a few equations only
it might be possible to examine analytically the effects of environmental
regulations on economic growth, the development of emissions and the
innovation of new and less polluting production technologies. However, the
theoretical model discussed so far is too complex to allow for an analytical
investigation. It is therefore the purpose of the following exercise to illustrate
the operation of the theoretical framework in a simulation procedure. As such
this analysis does not aim for a very detailed and realistic simulation, and it is
not designed for policy analysis or recommendations. It simply serves as a
numerical illustration of the functioning of a model that cannot be solved
analytically.
Solving general equilibrium models numerically requires strong interactions
between model-building and the development of solution methods. On one side
the formulation of a numerically solvable model is influenced by the numerical
tools available and on the other side the existing economic structures motivate
the search for suitable computer codes. Among the various solution methods
available there are those that depend on gradient evaluations such as the
Newton ones and non-gradient methods such as the various versions of
fixed-point algorithms. The former have advantages in computational speed
while the latter are guaranteed to find a solution provided the assumptions of
Kakutani's fixed-point theorem are fulfilled (for an overview, see Manne, 1985).
We are employing the so-called MPS solution method, drawing from the work
of Mathiesen (1985) and Rutherford (1989), which combines the strengths of
both approaches.
Numerical illustration
309
17.4.1 Description of the numerical illustration

The analysis distinguishes three regions. For vividness let Region 1 be the
northern and highly industrialized part of Western Europe covering the United
Kingdom, France, Belgium, the Netherlands, the former West Germany,
Switzerland and Austria. Region 2 should correspond to the southern, less
industrialized countries, Portugal, Spain, Italy and Greece. Region 3 represents
the rest of the world, including Western Europe's main trade partners, the
United States and Japan.
The time horizon is divided into six periods, each of a 5-year.length except
for the first one which contains 7 years. Base year for calculations is 1973, the
last year reported is 1995, when the common market will be established.
It is assumed that each region produces a single, but specific commodity
which can be consumed domestically, exported and might be used as input
factor into capital gestation processes. Besides services of domestically constructed capital stocks further inputs into regional production are labour and
environmental services. It is supposed that labour cannot be transferred
between regions, and inputs of environmental services reflect the use of nature's
ability to degrade pollutants from production and power generation. On this
basis, there are 12 different types of commodities and services in each period.
Since it was supposed in section 17.3.2 that there are no markets for durable
commodities and environmental services, for each period non-durables, produced outputs and regional labour inputs are obtained which are traded on
perfect future markets (see Appendix).
As mentioned at the beginning, we are interested in answering the question
as to how governmental regulations affect the innovation of less polluting
technologies. For this purpose a counterfactual analysis is employed and
simulations are carried out under two scenarios. In Scenario 1 it is supposed
that environmental regulations remain unchanged over the time horizon
studied. Hence, Scenario 1 supposes that only those standards on emission
abatement have to be fulfilled which were implemented prior to 1973.
Economic growth means, however, that more and more raw materials and
energy have to be put into circulation, producing an increasing amount of
pollutants. At a limited extent nature has the ability to degrade pollution, but
without some kind of pollution control, nature's assimilative capacities could
be reached or even be surpassed in the long-run and could lead to an
irreversible destruction of the ecological system. Under this perspective it seems
reasonable to suppose that the national governments have to set higher
environmental standards to assure the environment's quality. In practice, the
implementation of such regulations would mean requirements on the emission
abatement technology and equipment.
Given these considerations, for the alternative case (Scenario 2) it is assumed
that in every region the standards on the emission purification technology are
310
changed during the time horizon considered. 3 In Scenario 2 from 1980

onwards each region has to run end-of-the-pipe pollution treatment plants
which reduce emissions by 95%.
Setting higher requirements on end-of-the-pipe emission treatment implies a
more then proportional increase of environmental costs for those who have to
treat their emissions in such equipment (see Appendix, where cost functions for
end-of-the-pipe abatement are discussed). It will be shown in the following that
such an increase of regulation-induced costs provides an economic incentive
for innovating less polluting technologies and technological change.
17.4.2 Results
Figures 17.3, 17.4 and 17.5 show the development of annual gross domestic
production under both scenarios. Under reference case assumptions (Scenario
1), gross domestic production is growing continuously in each region. The
highest growth rates are for Region 3 and are close to 4.5%. The lowest growth
rates are for the southern part of Western Europe (Region 2). On average they
are lower than 1.5%. Growth rates in the northern countries of Western
Europe are intermediate, below 3% per year.
As one expects, tightening standards on abatement technology has only little
impact on economic growth and production in total, but there are important
6000
GNP
5000
4000
3000
2000
1000
0
1973
1980
_
Scenario 1
1985
1990
1995
~ Scenario 2
Fig. 17.3 Economic growth in the northern part of Western Europe (Region 1).
3Note, that in a conventional counterfactual analysis changes of governmental regulations happen
typically at the beginning of the time horizon. However, assuming that governmental regulations
are changed during the time horizon considered provides in our view a better understanding of
the time-path of adjustment.
311
1800
GNP
1600
1400
1200
1000
800
600
400
200
0
1973
1980
_
Scenario 1
1985
1990
1995
~ Scenario 2
Fig. 17.4 Economic growth in the southern part of Western Europe (Region 2).
GNP
30,---------------------------------------------~
25
20
15
10
5
1973
1980
_
Scenario 1
1985
1990
1995
~ Scenario 2
Fig. 17.5 Economic growth in the rest of the world (Region 3).
differences across single regions when emission standards are changed simultaneously. Region 2 and Region 3 are worse off, if higher purification
requirements are imposed: in the southern part of Western Europe (Region 2)
economic growth is cut back almost half in the long run, whereas Region 3,
Japan and the United States, suffers only a slight reduction of economic
development. In contrast to these two regions, economic growth is increased
in the northern countries of Westen Europe (Region 1) under Scenario 2..
This last result and the differences between regions are of particular interest.
312
Industrial organization studies usually indicate that although the impact of

environmental regulations on economic growth and productivity is fairly small,
it is always negative (for example, see Barbera and McConnell, 1990). Our
calculations parallel the first part of these findings, but they contrast with the
second: compared to the reference case, the northern part of Western Europe
has extended its exports to other regions under Scenario 2, whereas tightening
effluent standards has the opposite effect on the export prospects of Regions 1
and 3. Hence, a distributional effect across regions is observed. 4
Figures 17.6, 17.7 and 17.8 show the development of annual emissions.
Under both scenarios a gap between economic growth and the development of
pollution generation is observed. But a significant reduction of emissions can
be obtained only if higher emission standards are imposed in each region.
Under reference case assumptions (that is, no change in the existing environmental regulations), emissions develop at almost half the rates of economic
production. If the new standards have to be fulfilled from 1980 onwards, the
overall picture changes significantly. Only Region 3 is still characterized by an
increase of the annual aggregate emissions. In Western Europe as a whole, the
aggregate volume of annual pollution declines over time, but with a different
timepath in each region. In the northern part (Region 1) a peak for emissions
is calculated for 1980, whereas in the south of Western Europe (Region 2)
emissions will be reduced continuously over the whole time horizon.
How can the drastic reduction of emissions be explained? What are the
200
Emissions
150
100
50
0
1973
1980
_
Scenario 1
1985
1990
1995
~ Scenario 2
Fig. 17.6 Annual emissions in the northern part of Western Europe (Region 1).
4In our view one reason for these different results is that industrial organization studies usually
employ a partial equilibrium analysis which does not allow us to capture distributional effects in
a systematic and consistent way.
313
Emissions
70
60
50
40
30
20
10
0
1973
1980
_
Scenario 1
1985
1990
1995
~ Scenario 2
Fig. 17.7 Annual emissions in the southern part of Western Europe (Region 2).
1000
Emissions
800
600
400
200
0
1980
1973
_
Scenario 1
1985
1990
1995
~ Scenario 2
Fig. 17.8 Annual emissions in the rest of the world (Region 3).
reasons for the emission peak obtained in the period in which the emission
standards are changed?
As discussed above, in principle two sets of strategies are available to reduce
pollution generation. In the short run, emissions can be reduced by cutting
back production or by input substitution; this means a higher percentage of
foreign produced commodities as inputs into production, hence an export of
environmental pollution into other regions. In the long run changing the
production technologies and implementing less polluting production processes
seems to be a suitable strategy.
314
This explicit distinction between short-run and long-run strategies serves as

an explanation for the 1980 emission peak in the northern part of Europe.
From the model formulation (see Appendix) it follows that time is required to
adjust to a change in environmental regulations for two reasons: first, construction of capital stocks, hence the innovation of new production techniques
consumes time, and secondly capital stocks and production techniques are
fixed over the short run and cannot be abandoned (see section 17.3.2).
Consequently, the time until 1980 was too short to construct a sufficient
amount of new and less polluting capital equipment: production is mainly
based on old techniques and the only emission reduction policy that could
work sucessfully was to apply short-run strategies. But with each investment
less polluting production processes can be implemented, and a switch from
short-run to long-run emission reduction strategies can be observed. This
explains the drastic reduction of emissions after the new emission standard
came into force.
To support this argument Table 17.1 reports the emission coefficients for the
sequence of production processes introduced over the time span 1973-95.
Emission coefficients measure the volume of emissions discharged per unit of
output produced by a specific production process. Hence, to allow for comparison between different vintages, production processes are normalized to
produce one unit of specific regional output. Table 17.1 shows that emission
coefficients do not change significantly over time in the reference case.
Obviously, the historically given emission standards are too low to enforce
further implementation of less polluting production techniques over the timespan studied. But, in Scenario 2, from vintage to vintage emission coefficients
decrease for each region. Hence in all regions, less polluting production
processes are implemented over time.
The reason for this outcome is quite obvious. As the input requirement
function for end-of-the-pipe abatement (see Appendix) indicates, end-of-thepipe pollution treatment costs will rise significantly if the emission standards
are tightened. This drastic increase of the end-of-the-pipe treatment costs
provides an economic incentive for implementing less polluting techniques:
Table 17.1
Region/year
Development of emission coefficients

1973
1980
Emission coefficients: scenario 1

0.036
0.032
1
2
0.042
0.038
0.040
0.038
3
Emission coefficients: scenario 2
0.036
0.021
1
2
0.042
0.031
3
0.040
0.028
1985
1990
1995
0.029
0.043
0.032
0.033
0.043
0.036
0.039
0.043
0.036
0.019
0.029
0.023
0.017
0.027
0.020
0.016
0.027
0.020
Acknowledgements
315
implementation of less polluting technologies leads to a reduction of emissions

which have to be treated in an end-of-the-pipe abatement plant and thus
implies a reduction of 'environmental' costs. However, this is not an everlasting
process. As Table 17.1 shows, the emission coefficients stay almost constant at
the end of the time horizon. This seems to indicate that tightening the emission
standards only motivates implementing less polluting production techniques in
the medium term (15 years).
The actual time-path of emissions (see Figures 17.6, 17.7, 17.8), however, not
only depends upon the implementation of less polluting production processes.
In each period a certain amount of total pollution is generated by old
production techniques continuing in existence. Hence, the quantities emitted
per period depend upon the mixture between old and new production
possibilities. The faster old production capacities evaporate, the lower the
amount of old capital goods and the faster new processes can be implemented,
thus the faster total emissions from production and power generation will
decline. As shown by Forsund and Hjalmarsson in Chapter 3, the development
of emissions depends upon the speed of substitution between new and old
production equipment. The present author has analysed how different assumptions about the speed of short-run adjustment affect the temporal structure of
pollution generation (Stephn, 1988b).
Summing up, the numerical analysis shows the following important results:
1. Tightening environmental standards has only little impact on economic

growth and productivity, but can have significant distributional effects.
International trade, imports and export prospects of single regions may
change significantly, as the counterfactual analysis indicates.
2. The finding of many industry studies that governmental interventions for
regulating environmental externalities necessarily lead to a slow-down of
economic growth and productivity cannot be supported by this analysis.
3. In contrast to the conclusions of most theoretical studies, it is shown that
not only charges but emission standards can provide a dynamic incentive
to reduce pollution generation in the long run by implementing less
polluting production techniques. The economic reasoning for this outcome
is quite obvious. We assume filter technology to be needed and thus that
complying with standards implies (shadow) prices for emissions. If standards are high enough, their shadow price can equal the optimal effluent
charge. In the absence of uncertainty, as in this chapter, imposing standards
and effluent charges may therefore lead to the same behaviour on the part
of the polluters.
ACKNOWLEDGEMENTS
Helpful comments and suggestions have been provided by T. Sterner, R. van
Nieuwkoop and T. Wiedmer. The usual disclaimer, however, applies.
316

REFERENCES
Barbera, A. and McConnell, V. (1990) The Impact of Environmenal Regulations on

Industry Productivity: Direct and Indirect Effects. Journal oj Environmental Economics and Management, 18, 50-65.
Baumol, W. and Oates, W. (1988) The Theory oj Environmental Policy. Cambridge
University Press, Cambridge.
Burmeister, E. (1974) Synthesizing the Neo-Austrian and Alternative Approaches to
Capital Theory. Journal oj Economic Literature, 12, 413-56.
Faber, M. (ed.) (1986) Studies in Modern Austrian Capital Theory, Investment and Time.
Springer-Verlag, Berlin.
Faber, M., Niemes, H. and Stephan, G. (1987) Entropy, Environment and Resources.
Faber, M. and Proops, 1. (1990) Evolution, Time, Production and the Environment.
Faber, M., Proops, 1. Ruth, M. and et al. (1990) Economy-Environment Interactions in
the Long-Run: a Neo-Austrian Approach. Ecological Economics, 2, 27-55.
Faber, M. and Stephan, G. (1987) Umwe1tschutz und Technologiewandel.I, in Technologie, Wachs tum und BeschiiJtigung (ed. R. Henn), Springer-Verlag, Berlin, pp.
933-49.
Greer, D. (1984) Industrial Organization and Public Policy. Macmillan, New York.
Hahn, F. (ed.) (1989) The Economics oj Missing Markets, InJormation and Games.
Claredon Press, Oxford.
Hicks, J. (1973) Capital and Time: a Neo-Austrian Theory. Oxford University Press,
Oxford.
Lee, D. (1984) The Economics of Enforcing Pollution Taxation. Journal oj Environmental Economics and Management, 11, 147-60.
Manne, A. (1985) Economic Equilibrium: Model Formulation and Solution.l, in
Mathematical Programming Study 23 (ed. A. S. Manne), North Holland, Amsterdam,
pp.I-22.
Mathiesen, L. (1985) Computation of economic equilibria by a sequence of linear
complementarity problems, in Mathematical Programming Study 23 (ed. A. S. Manne),
North-Holland, Amsterdam, pp. 144-62.
Milliman, S. and Prince R. (1989) Firm Incentives to Promote Technical Change in
Pollution Control. Journal oj Environmental Economics and Management, 17, 24765.
Rutherford, T. (1989) General Equilibrium Modelling with MPSjGE. Department of
Economics, University of Western Ontario, London, Canada.
Scarf, H. (1967) The Computation of Equilibrium Prices.l, in Ten Economic Studies in
the Tradition oj Irving Fisher (ed. T. C. Koopmans), Wiley, New York, pp. 68-93.
Siebert, H. (1987) Economics oj the Environment. Springer-Verlag, Berlin.
Stephan, G. (1988a) Economic impact of emission standards: A computational approach
to waste water treatment in Western Europe, in Welfare and Efficiency in Public
Economies (D. Bos, M. Rose and C. Seidl eds.), Springer-Verlag, Berlin etc., 401422.
Stephan, G. (1988b) A Neo-Austrian Approach to Computable Equilibrium Models:
Time to Build Technologies and Imperfect Markets. Schweizerische Zeitschrift Jur
VolkswirtschaJt und Statistik, 124, 48-64.
Stephan, G. (1989) Pollution Control, Economic Adjustment and Long-run Equilibrium.
Stephan, G. (1990) Comments, in Heidelberg Congress on Taxing Consumption (ed. M.
Rose), Springer-Verlag, Berlin pp. 178-180.
Appendix
317
APPENDIX: MATHEMATICAL FORMULATION OF THE MODEL

17A.l Basic assumptions
To translate ideas into a formal model it is supposed: (1) time is measured in
discrete units (periods) and evolves over a horizon of T periods. (2) Space is
split into R regions. (3) Each region r = 1, ... , R produces a specific commodity
which can be consumed, exported or used as input into domestic capital
gestation processes and into end-of-the-pipe emission abatement (see section
17.4.1). For each period t = 1, ... , T, region r's gross output, consumption,
exports to the other regions j = 1, ... ,R, inputs into the capital construction
and emission treatment are denoted by Yr(t), cr(t), mt(t), (m~(t) = 0), xr(t) and
fr(t), respectively. (4) Since an Armington rule for modelling interregional trade
is used, region r's exports enter into the other regions' production, but not into
consumption (see Stephan, 1988a). Further inputs into production are
exogenously supplied labour lr(t), services of domestically owned capital stocks
kr(t) and environmental services which are converted into emissions er(t).
(5) Given the model assumptions from section 17.3.2, there are no markets for
capital goods and environmental services. As such in each period t = 1, ... , T
there are 2R non-durables (produced output and labour) which are traded on
perfect future markets. The present value prices of produced commodities and
labour will be denoted by Pr(t) and wr(t), r = 1, ... , R, t = 1, ... , T. (6) In each
region r= 1, ... , R there is a representative consumer and a representative
producer. Both act as price-takers with households optimizing intemporal
utilities and producers maximizing present value of profits.
17A.2.1 Consumers' decision
As is usual in a microeconomic framework, the representative consumer of
region r = 1, ... , R owns the endownment of the exogenously given labour force
Lr(t), t = 1, ... , T, receives profits from production Gr and chooses an intertemporal consumption bundle {cr(t), t = 1, ... ,T} which maximizes his utility
subject to his budget constraint. Let region r's labour force grow exponentially
with the rate r and let
T
U r=
L 0~ log(cr(t))
(17 A.1)
t= 1
reflect the preference ordering of the representative consumer in region r with

the utility discount rate Or. Then his optimization problem is:
T
Max
L 0~ log(cr(t))
t= 1
318
subject to
T
L Pr(t)cr(t):::; L wr(t)Lr(t) + G"

(=
(=
(17A.2)
where Gr denotes the present value of profits in region r.
17A.3 Producers' decision

For modelling the producers' decision problem it is assumed that producers
behave as present value maximizers: in each region r= 1, ... , R the representative producer selects a sequence of technically feasible production techniques
such that the present value of production less costs for end-of-the-pipe emission
treatment is maximized.
As mentioned in section 17.3.2 a production technique consists of: (1) a
technically feasible production process, and (2) a lag structure of inputs and
outputs into the capital gestation process and into the utilization of the
production process after its installation. For applications, these elements are
specified as follows:
(1)
(kr(t), lr(t),
m~ (t),
... , mi(t), Yr(t), er(t))
denotes a technically feasible production process, if

Yr(t) = {Dr [kr(t)~ lr(t)P 1tjmj(tr]b - Erer(t)b} l/b.
(17A.3)
With this formulation it is technically feasible to combine the non-environmental inputs labour, imports and services of the capital stock as in a
Cobb-Douglas production function. 1 This index of inputs is in turn linked with
environmental services (expressed in terms of emissions) through aCES
formulation with substitution elasticities which are less than unitary as the
entropy law tells (see Faber et al., 1987).
(2) If we assume that the temporal structure of capital gestation and depreciation are identical for all production techniques, then the lag structure of
inputs and outputs can be represented by an array of matrices {A(s), B(s),
s= - T, ... , T} such that
Xr(t -s) = A(s)kr(t),
(17AA)
with
A(s) = 0, s>O,
Yr(t + s) = B(s)Yr(t),
(17 A.5)
Ir(t + s) = B(s)lr(t),
(17A.6)
'Note that region r's imports of produced goods from regionj equal regionj's exports to region r.
319
Appendix
mj(t +s) = B(s)mj(t),
(17 A.7)
er(t +s) = B(s)er(t),
(17A.8)
with
B(s)=O, s<O.
Equation 17 A.4 determines the sequence of inputs x(t - s) into the capital stock
gestation process (xr(t- T), ... , x r(t-1), kr(t))=(A( - T)kr(t), ... , A( -1)kr(t),
kr(t)); equations (17 A.S) to (17 A.8) denote the depreciation once the capital
stock kr(t) and thus production process (kr(t), lr(t), m~ (t), ... , mHt), Yr(t), er(t)) is
established.
Let {(kr(t), lr(t), m'i{t), ... , m~(t), Yr(t), er(t)), t= 1, ... , T} be a sequence of
technically feasible production techniques. The net output zr(t) produced from
this choice of techniques in period t = 1, ... , T is defined by
1
Zr(t)=
s= 1
B(t-s)Yr(S)-
s=l+ 1
A(t-s)xr(s).
(17A.9)
B(t-s)Yr(s) denotes the gross output produced by means of the production techniques which have been implemented prior to period t. LJ=I+ 1
A(t - s)xr(s) is the investment in capital construction in period t which is
expected to produce capital stocks in future periods.
Since inputs into production during period t also depend on the choice of
production techniques which have been installed in prior periods, we observe
the labour inputs (see 17 A.6).
L!=l
L B(t -
s= 1
(17A.I0)
s)lr(s),
and the inputs of produced commodities from other regions (see 17 A. 7)

1
L B(t -
s= 1
(17A.l1)
s)mj(s).
Given these conventions the firms' present value maximization problem can
now be written as
Max
Jl
Jl
Jl ttl pj(t{st
{Pr(t(tl B(t-s)Yr(S)-
Jl
A(t-S)Xr(S)]}-
Wr(t)Lt B(t-S)lr(S)]-
B(t-s)mj(S)]}-
Jl
Fr(t).
320
Fr(t) denotes the present value of end-of-the-pipe emissions abatement costs in

region r.
17A.4 End-of-the-pipe treatment and equilibrium conditions

As mentioned in section 17.4.1, it is supposed that each region r = 1, ... ,R faces
an emission's standard which requires the application of suitable pollutant
abatement technologies. Engineering analyses of such end-of-the-pipe systems
suggest that the principal determinants for purification are the volume of total
emissions Er(t), the concentration of waste and the emission standard Sr (see
Faber, et al., 1987). (1) For a fixed standard Sr emission treatment is usually
characterized by economies of scale. Hence, marginal costs of purification
decline with increasing flow size. (2) For fixed volumes of pollution Er(t),
however, marginal costs gradually increase, if the standard Sr is raised. This
reflects increasing difficulties of waste extraction.
To cope with this technical relationship in a computable general equilibrium
framework, a two-argument three-parameter function is used to express the
input requirements for an end-of-the-pipe pollution treatment system
fr(t) = Er(t)(c - b In(100- Sr )).
(17A.12)
The parameters c and b determine marginal costs of waste extraction depending on the effluent standard Sr.
Since total emissions are determined by the emissions due to the chosen
mixture of technologies, and since only domestically produced goods enter as
inputs into end-of-the-pipe pollution abatement, for each period t = 1, ... , T
the costs of running the end-of-the-pipe equipment in region rare
Fr(t) = Pr(t)
ttl
[B(t - s)er(s)](c - b In(100 - Sr))}.
(17A.13)
Summing up, we are able to formulate the equilibrium conditions:

Definition:
Let (P, W)= {Pr(t), wr(t), t= 1, ... , T, r= 1, ... , R} be a sequence of present
value price vectors. (P, W) is a (long-run) equilibrium price system, if for any
region r and period t (17 A.14-17 A.16) are fulfilled.
1
s= 1
s=l+ 1
L B(t - s) Yr(s) - L
A(t - s)x(s) ~
L
s= 1
[B(t - s)er(s)](c - b In(100 - Sr )),
(17A.l4)
Appendix
321
L B(t -
s)lr(s) ~ Lr(t),
(17 A.15)
8= 1
1=1
8=1
1=1
8=1
L wr(t) L B(t-s)lr(s)-
L Pr(t) L
(17A.16)
[B(t-s)er(s)](c-b In(100-Sr )).
Equation (17 A.14) means that in each period demand for produced nondurable commodities (i.e., consumption, exports and inputs into end-of-thepipe emission treatment) is covered by net supply. Equation (17A.15) says that
demand for labour is less than the exogenously given endowment. Equation
(17A.16) is a restricted formulation of Walras's law and implies that in an
equilibrium the existing markets for non-durables clear at positive prices. 2
2 For
a proof of existence and computability, see Stephan (1988b).
---18
CO 2 emission limits: an economic
cost analysis for the
United States of America
Alan S. Manne and Richard G. Richels
18.1 INTRODUCTION
Within the scientific community, there is a growing consensus that rising
concentrations of certain trace gases in the earth's atmosphere may lead to
significant changes in climate. The greenhouse effect has evolved from a purely
scientific issue to an important public policy debate. During the 100th US
Congress (1988-9), more attention was devoted to hearings on the climate than
to any other single environmental issue, including acid rain. The result has
been a steady flow of legislative proposals to limit emissions of the major
greenhouse gases: carbon dioxide (C0 2 ), methane (CH 4 ), nitrous oxide (N 2 0)
and chlorofluorocarbons (CFCs).
Figure 18.1 shows the estimated current contribution of the various manmade greenhouse gases to global warming. CO 2 (believed to be responsible for
approximately half the problem) is produced primarily from the burning of
fossil fuels. The energy sector therefore plays a central role in proposed
strategies to delay climate change. Over the next few decades, such strategies
typicaly call for a concerted push toward greater energy efficiency, and - to
whatever extent is possible - switching away from coal and oil toward natural
gas with its lower carbon emissions per unit of energy. For the longer term,
proposed strategies tend to emphasize greater dependence on carbon-free
alternatives such as solar (in several different forms), fission and fusion.
Although many of the legislative proposals have set physical targets for the
reduction of emissions, little attention has been paid to the costs of meeting
these targets. This presents a serious dilemma to policy-makers. Without
information on the cost of emissions abatement, it is difficult to assess the
feasibility of alternative proposals, and to determine which measures are
cost-effective. Moreover, a reduction in emissions is not the sole policy
response to potential climate change. There is a point at which further
324
CO 2 emission limits
49%
Nitrous oxide
6%
Fig. IS.1
Man-made contributions to the greenhouse effect. (Source: Rind (1989).)
reductions could become so expensive that it would be preferable to shift to

other options such as adaptation. Without careful analysis, it is difficult to
know where that point might be.
This paper describes Global 2100, an analytical framework for estimating the
costs of a carbon emissions limit. The model is designed to evaluate CO 2 energy-economy interactions. Although the analysis is being implemented on
a global scale, this paper will report only upon its initial application to the
United States. Because of the worldwide scope of the problem, it is clear that
a unilateral limit would be futile unless there were parallel abatement actions
by other nations.
Sections 2 and 3 contain a description of Global 2100, the rationale for its
design and a discussion of key macroeconomic p~rameters that affect the
demand for energy. The next two sections summarize the cost and performance
characteristics of both electric and non-electric supply technologies. We then
present a quantitative assessment, estimating the costs of a carbon limit under
five alternative scenarios. Results are also presented on: CO 2 emissions in the
absence of any limits, the size of the carbon tax that would be required to
reduce emissions to the target levels, and the benefits of various alternatives for
reducing CO 2 emissions. In the final section, we conclude with some general
observations on the costs of a carbon constraint to the United States.
17.2 MODEL STRUCTURE
The name Global 2100 has been adopted in order to emphasize both the global
nature of the carbon emissions problem and also the need for a long-term
Model structure
325
perspective. There are long time lags inherent in the build-up of CO 2 and in
the transition away from carbon-based fuels. Our model is benchmarked
against 1990 base year statistics, and the projections cover 11 ten-year time
intervals extending from 2000 to 2100. This is an intertemporal rather than a
recursive model. It is assumed that producers and consumers will be sufficiently
farsighted to anticipate the scarcities of energy and the environmental restrictions that are likely to develop during the coming decades.
In its present form, Global 2100 is based upon parallel computations for five
major geopolitical groupings: the United States, other OECD nations (Western
Europe, Canada, Japan, Australia and New Zealand), the Soviet Union and
Eastern Europe, China and ROW (rest of world). Each of these areas is
endowed with limited amounts of oil and gas resources, and each is a
contributor to global carbon emissions. Because each region is likely to pursue
its own individual interests rather than the global welfare - and because there
are differences in the relative costs of emission abatement - it would be
desirable to analyse this problem within a computable general equilibrium
framework. As an initial step in this direction, we make a series of assumptions
on the future path of international crude oil prices - and also place bounds on
the willingness of each region to import or export oil. Moreover, it is assumed
that if a carbon emissions quota is assigned to each region through international negotiations, there is no practical way to trade these quota rights. At
some point in the future, we hope to adopt a computable general equilibrium
framework. A CGE framework would allow us to deal explicitly with issues
such as trade in carbon quota rights, trade in carbon-intensive commodities
and the impact of carbon quotas upon the international division of labour.
In undertaking a global analysis, we have avoided the data-intensive
approach required for end-use models. Because of the difficulties of gathering
a consistent international data set and then arriving at a meaningful summary
of results, we have adopted a much more aggregative approach than would be
appropriate for analysing the United States by itself. The categories are
consistent with those of the International Energy Workshop, and the projections have been benchmarked against the poll's median results (see Chapter 16).
Within each region, the analysis is based upon ETA-MACRO, a model of
two-way linkage between the energy sector and the balance of the economy. 1
This is a merger between ETA (a process analysis for energy technology
assessment) together with a macroeconomic growth model providing for
substitution between capital, labour and energy inputs (See Figure 18.2).
ETA-MACRO is a tool for integrating long-term supply and demand projections. It is designed to compare the options that are realistically available to
each region as the world moves away from its present heavy dependence upon
oil and gas resources - toward a more diversified future energy economy. This
type of model may help to promote 'second order' agreement. For example,
two analysts may disagree on the costs of solar electricity generation, but might
lFor a detailed description of ETA-MACRO, see Manne (1981).
326
Exhaustible resources
(petroleum, natural gas)
Electric and
non-electric
energy
conversion
technologies
(coal, nuclear
and renewab les)
----.
Labour
consumption
electric, non-electric
energy
MACRO
ETA
energy costs
investment
capital
Fig.18.2
An overview of ETA-MACRO
agree on a logical framework within which to estimate the impact of these cost
estimates.
ETA-MACRO allows explicitly for:
Energy-economy interactions: rising energy costs and limited supplies will
prevent the economy from achieving its full potential GNP growth rate, and
this in turn will slow down future capital accumulation.
Cost-effective conservation: rising prices will induce substitution with capital
and labour, thereby reducing energy demands below the amounts projected
from historical trends.
Autonomous conservation: changes in government policy and in the structure of the economy will help to reduce the amount of energy required per
unit of GNP.
Interfuel substitution: changing relative prices will induce consumers to
replace oil and gas with electricity, e.g. heat pumps in place of fuel burners.
New supply technologies: each has its own difficulties and uncertainties on
dates and rates of introduction.
For each region that is parallel, a dynamic non-linear optimization is
employed to simulate either a market or a planned economy.2 Supplies and
demands are equilibrated within each individual time period, but there are
'look-ahead' features to allow for interactions between periods. These interactions are particularly important for the depletion of exhaustible resources and
for the accumulation of capital over time. Savings and investment decisions are
mod~lled so that consumers will receive equal benefits from an additional
2The model is formulated and solved by means of the GAMS/MINOS system, (see Brooks et al.
(1988)). In a representative example, there are approximately 200 constraints, and 400 variables.
The solution of two successive cases - with and without a carbon constraint - requires 5 minutes
on a 25 Mhz desk-top computer.
Model structure
327
dollar's worth of current consumption and a dollar's worth of investment.

In order to focus upon the long-run issues of energy-economy interactions,
resource exhaustion and the introduction of new technologies, each region is
described in highly aggregative terms. Outside the energy sector, all economic
activity is represented in terms of dollars of constant real purchasing power.
Within the energy sector, only two end products are distinguished: electricity
and non-electric energy.
Figure 18.2 provides an overview of the principal static linkages between the
sectoral and the macro submodels. Electric and non-electric energy (denoted,
by the symbols E and N respectively) is supplied by the energy sector to the
rest of the economy. Like the material balance equations of an input-output
model, aggregate economic output (Y) is allocated between interindustry
payments for energy costs (EC) and 'final demands' for current consumption
(C) and investment (I). Thus:
Y=C+I+EC.
(18.1)
Each component of equation (18.1) is measured as an annual flow (measured

in trillions of constant dollars). For an economy-wide production function,3 we
assume that gross output (Y) depends upon four inputs: K, L, E, N - capital,
labour (measured in 'efficiency' units that represent the sum of labour force
growth and productivity gains), electric and non-electric energy. To minimize
the number of parameters that require either calibration or econometric
estimation, the long-run static production function is described by a nested
non-linear form:
Y=
[a(K~Ll-~y
+ b(EP N1-Py]1/p
(18.2)
where p=(a-1)/a (for a;60,1,00).

Equation (18.2) is based upon the following assumptions:
there are constant returns to scale in terms of these four inputs;
there is a unit elasticity of substitution between one pair of inputs - capital
and labour - with ex being the optimal value share of capital within this pair;
there is a unit elasticity of substitution between the other pair of inputs electric and non-electric energy - with f3 being the optimal value share of
electricity within this pair;
there is a constant elasticity of substitution between these two pairs of inputs
- the constant being denoted by a( = ESUB);
the scaling factors a and b are determined so that energy demands in the base
year are consistent with the 'reference price' for non-electric energy; and
there are autonomous energy efficiency improvements (AEEI) that are
summarized by growth in the scaling factor b.
3For the concepts and terminology of macroeconomic production functions and neoclassical
growth theory, see Allen (1968).
328
Energy-economy interactions occur in two ways. According to (18.2), energy

is an input to the economy. According to (18.1), energy costs represent one of
the claims upon the economy's output. Tighter environmental standards
and/or an increase in energy costs will reduce the net amount of output
available for meeting current consumption and investment demands. This is
why the potential growth rates of GNP do not uniquely determine the realized
rates. Since investment determines the accumulation of capital stocks, a lower
rate of current investment will in turn reduce the amount of output available
for meeting current consumption and investment demands. This is why the
potential growth rates of GNP do not uniquely determine the realized rates.
Since investment determines the accumulation of capital stocks, a lower rate of
current investment will in turn reduce the amount of output available for future
consumption. Alternative carbon emission scenarios will therefore be measured
in terms of their impact upon present and future levels of consumption.
In addition to resource depletion, there are the following intertemporal
elements of the model. First savings and investment decisions are determined
so as to maximize the discounted utility of consumption. For simplicity, the
utility function is logarithmic in form. The utility discount rate is chosen so
that the rate of return on capital will remain constant if the realized growth
rate of the economy coincides with the potential rate determined by the growth
of the labour force. Second, the rate of depreciation of equipment and
structures governs the time lags of demand in response to changing prices. This
is sometimes termed a 'putty-clay' approach. That is, the input-output coefficients for each successive age cohort of equipment are optimally adjusted to
the future trajectory of prices, but there are no changes possible in these
coefficients subsequent to the installation of a given cohort. Third, upper
bounds are imposed upon the rates of market penetration for new supply
technologies. Moreover, there are lower bounds to ensure that older technologies will not be phased out too rapidly.
18.3 KEY DEMAND PARAMETERS
Three of the demand parameters (potential GNP growth, ESUB and AEEI)
are crucial to the debate over energy and environmental futures. There is no
easy way to estimate these coefficients econometrically. The values adopted
here have been determined so that our model will track closely with the
conventional wisdom expressed, for example, by the median poll responses of
the International Energy Workshop.
One key parameter is the rate of growth of the labour force and potential
GNP. In parallel with the slowdown of population growth during the twentyfirst century, there will be a diminishing rate of growth of GNP, and hence a
slowdown in the demand for energy. It is assumed that the United States'
potential annual GNP growth rate will be nearly 3% from 1990 to 2000, and
Assumptions for electricity generation
329
that it will slow down to 1% during the latter half of the twenty-first century.
Even with a stationary population, this would allow for a modest increase in
per capita living standards.
ESUB represents the elasticity of price-induced substitution between capitallabour and energy. (For a demonstration of the importance of this parameter,
see EMF, 1977.) Over the long run, there is a good deal of possible substitutability between the inputs of capital, labour and energy. The degree of substitutability will affect the economic losses from energy scarcities and price
increases. One example of such a trade-off would be insulation to replace
heating fuels in homes and other structures. A second example would be the
increased use of heat exchangers and of cogeneration within industry. In the
aggregate, the ease or difficulty of these trade-offs is summarized by ESUB,
here taken to be 0.40. The higher the value of ESUB, the less expensive it is to
decouple energy consumption from GNP growth during a period of rising
energy prices. When energy costs are a small fraction of total output, ESUB is
approximately equal to the absolute value of the price elasticity of demand.
Finally, there is AEEI, the rate of autonomous (non-price-induced) energy
efficiency improvements. In econometric investigations of the post-1947 historical record, there has been no evidence for autonomous time trends of this type
(see Hogan, 1988; Brown and Phillips, 1989; Jorgenson and Wilcoxen, 1989).
Technologically orientated end-use analysts, however, have suggested that
non-price efficiency improvements may be induced by changes in government
policy, e.g. a mandatory doubling or quadrupling of the average fuel efficiency
of automobiles during the course of several decades (see Goldemberg et ai.,
1987). Clearly the AEEI parameter is highly controversial. In order to represent
two distinct viewpoints, we begin with a zero value for this parameter, and then
explore the implications of a high efficiency scenario.
18.4 SUPPLY AND COST ASSUMPTIONS FOR ELECTRICITY
GENERATION
Table 18.1 identifies the alternative sources of electricity supply that are
included in Global 2100. The first five technologies represent the sources of
electricity that exist within the United States today: hydroelectric and other
renewables, gas-, oil- and coal-fired units and nuclear power plants. The second
group of technologies includes the new generation options available for the
future. These differ in terms of their projected costs, carbon emission rates and
dates of introduction. Table 18.2 contains a summary of our cost and
performance estimates for new electricity supply technologies.
Figure 18.3 shows the breakdown of electricity generation by source for
1985. In that year, natural gas-fired plants produced 12% of the electricity in
the United States, and coal-fired units produced 57%. Coal produces almost
twice as much carbon per kilowatt hour as natural gas. If sufficient natural gas
330
Table 18.1 Identification of electricity generation technologies
Technology
Earliest possible
introduction data"
Existing
Hydro
Gas-E
Oil-E
Coal-E
LWR
New
Gas-N
Coal-N
Coal-R
ADV-HC
ADV-LC
Identification
Hydroelectric
Remaining initial
Remaining initial
Remaining initial
Remaining initial
1995
1990
2005
2015
2010
2010
gas-fired
oil-fired
coal-fired
nuclear
Advanced combined cycle

New coal-fired
Coal gasification (20% CO 2 removal)
Coal gasification (93% CO 2 removal)
High-cost non-carbon-based
Low-cost non-carbon-based
aEstimated year when the technology could provide 0.1 trillion kWh (approximately 20 GW of
installed capacity at 60% capacity factor).
were available, a conversion from coal to natural gas would therefore make it
possible to achieve a substantial reduction in carbon emissions.
It is expected that new gas-fired capacity for base load electricity will be
produced by combustion turbine combined cycle plants. These units have a
high thermal efficiency and relatively low costs. If natural gas prices remain at
their 1988 levels, this technology would represent an attractive source of
electricity. In the absence of large-scale discoveries, however, domestic natural
gas resources will gradually become exhausted, and fuel prices will rise. For
example, in a baseline projection for the Gas Research Institute, Woods (1988)
projects a tripling of wellhead prices by 2010. With such an increase, gas-fired
electricity would lose its competitive advantage over coal, see also section 7.3.
Two categories of coal-fired technologies are considered - those without and
those with CO 2 emissions control. The first category includes both existing and
new pulverized coal technologies. Most of the existing pulverized coal plants
do not have flue gas desulphurization units. As a result, they have much lower
operating costs than new pulverized coal plants.
CO 2 can be separated either from the flue gas of an atmospheric boiler or
from the fuel gas produced within an integrated gasification-combined cycle
(IGCC) plant. According to the studies reviewed by Vejtasa and Schulman
(1989), the latter appears to be more cost-effective for new power plants. Table
18.2 contains the cost and performance data for coal gasification technologies
with 20% and 93% CO 2 recovery.
Separation of CO 2 does not solve the problem of permanent disposal.
Technically, this gas could be injected into the oceans or into depleted natural
gas fields. For purposes of Table 18.2, disposal costs are based upon compressing the recovered CO 2 at the generating station, transporting it by a new
10,000
10,200
1905
200
2105
2600
1750
0.000
0.000
0.022
0.187
1695
70
1765
9300
12,300
0.109
0.246
Carbon
emission
coefficient b
580
1445
Capital
cost ($/kW)
7500
9800
Heat rate
(Btu/kWh)
0.80
1.50
1.50
2.34 d
1.50
Fuel cost
($/MBtu)
8.2
18.5
14.0
17.5
14.7
Fuel
10.0
11.5
14.7
3.0
10.4
1.0
3.3
10.2
O&M
100.9
31.3
35.2
3.6
31.3
1.2
10.4
26.7
Capital
Cost components C
Cost and performance estimates for new power generation technologies'
68.4
6.6h
75.0
110.9;
51.Oi
55.7
2.2"
57.9
31.2 e
51.6 f
Total
'Cost and performance data are from Vejtasa and Shulman (1989).
bCarbon Emission Coefficient (CEC) is defined as billions of tonnes of carbon emitted to the atmosphere per trillion kWh (billion tonnes/TkWh) of
electricity generated.
CUS$m (December 1988)/MWh.
d'"Based on price of gas in 1988. Gas prices are linked to that of oil, and they rise over time.
fCosts similar to AFBC and IGCC without CO 2 control.
.,hPipeline at 25% capacity.
iAdvanced solar and biomass-based.
iAdvanced nuclear technology with passive safety features.
Coal gasif. (93% CO 2 removal)

CO 2 disposal-1oo miles pipeline
Total
ADV-HC
ADV-LC
Gas-N
Coal-N
Coal-R
Coal gasif. (20% CO 2 removal)
CO 2 disposal-100 miles pipeline
Total
Technology
name
Table 18.2
332
Coal
Gas
12%
Fig. 18.3 Electricity generation (by source, 1985)
pipeline for 100 miles, and then disposing of the gas either in the oceans or in
distant natural gas fields via the existing pipeline network. The feasibility of
these disposal options is highly speculative. In the following analysis, we
calculate the benefits of solving the problem of permanent disposal.
ADV-HC and ACV-LC respectively refer to high- and low-cost noncarbon-based electricity generating technologies. Although any of a number of
technologies could be included in these categories, the cost and performance
data contained in Table 18.2 are based upon specific designs considered in
EPRI's Technical Assessment Guide (1989). The representative high cost source
is an advanced solar technology with cost and performance characteristics
similar to those for concentrator photovoltaic cells. (Alternatively, this might
be a biomass-based generating unit or some combination of the two.) The low
cost source is an advanced nuclear design with passive safety features. In our
judgement, 2010 is the earliest availability date for 0.1 TkWh of electricity from
this technology.
18.5 NON-ELECTRIC SUPPLY TECHNOLOGIES
The non-electric energy supply technologies are listed in Table 18.3. The
individual fuels are ranked in ascending order of their cost per million BTU of
non-electric energy. The least expensive is CLDU. This category accounts for
direct uses of coal in industries such as iron and steel, cement, etc. Its growth
2.00
2.50
1.50+
1.25 additional
distribution costs
5.00 in 2000,
rising to 10 from
2030 onward
10.00
Coal - direct uses
Oil - domestic
Natural gas
Oil imports--exports
Synthetic fuels
Non-electric backstop
CLDU
Oil-D
Gas
Oil-MX
SYNF
NE-BAK
0.0203
0.408
20 quads
Growth limited to
15% p.a.
Growth limited to
15% p.a.
0.0000
0.0203
0.0145
0.0251
Growth at 20% the

rate of GNP
Constant ratio model'
Constant ratio model'
Carbon emission
coefficient b
'US$ 1988/mBtu.
bTonnes of carbon/mBtu. Source of carbon emission coefficients: Edmonds and Reilly (1985). To allow for non-energy uses of oil and
synthetic fuels, their respective carbon emission coefficients have been multiplied by a factor of 0.89.
Oil-D Gas
'Constant ratio model is based on the following estimate of reserves and resources of domestric oil and natural gas:
345
327
dProven reserves. (quads)
470
474
Undiscovered resources. (quads)
5%
5%
Resource depletion factor (RDF), annual gross additions to reserves, % of undiscovered resources.
20.00
Unit costa
Description
Upper
bound,
if any
(quads)
Non-electric energy supplies
Technology
name
Table 18.3
334
rate is taken to be only 20% that of the GNP. Next in the 'merit order' are
domestic oil and gas. These exhaustible resources are available at constant
marginal costs, but are subject to upper bounds based on a model of reserves
and resource depletion.
For specifying the upper bounds on exhaustible hydrocarbon resources, we
draw a sharp distinction between current reserves and the remaining stock of
undiscovered resources. Because cost estimation is exceedingly hazardous in
this area, we do not attempt to provide an explicit economic rationale through
rising marginal cost curves. Instead, a constant ratio model is employed to
determine an upper bound on the annual rate of oil and gas production. There
is also the possibility of delaying the exploitation of these resources.
Reserves of exhaustible resources are depleted by current production, and
are augmented by new discoveries. Production is a fixed fraction of reserves
(the 1990 production-reserve ratio), and new discoveries are a fixed fraction
(5% per year) of the remaining undiscovered resources. When the productionreserve ratio exceeds the resource depletion factor (RDF), it can be shown that
the RDF governs the ultimate rate of decline. Figure 18.4 illustrates the
comparison between a 3% versus a 5% value of the RDF. With 3% for this
parameter, production drops off even more rapidly during the years up to 2030.
Our reserve and resource estimates are taken from the 5th percentile point
along the probability distributions available from the Geological Survey work
by Masters et al. (1987). This source provides a modal (i.e., most likely) estimate
of resources along with the 5th and 95th percentile. For practical purposes, the
95th percentile point indicates a lower bound on undiscovered resources, and
the 5th percentile indicates an upper bound. That is, according to the USGS,
there is only a 5% probability that undiscovered conventional resources will
exceed the 5th percentile values. For Global 2100, we have adopted the USGS
upper bound on natural gas resources. Had our calculations been based upon
the modal or the 95th percentile, the prospects for domestic natural gas
production would be considerably more pessimistic than the case examined
here. According to Figure 18.4, there is no prospect that conventional domestic
gas resources will permit a significant expansion of consumption above its 1990
level. Since the production-reserve-resource ratios for domestic crude oil are
similar to those for natural gas, a similar conclusion also holds for this
hydrocarbon resource.
Excluding economic rents, it is assumed that oil imports net of exports
(abbreviated OIL-MX) are more expensive than domestic supplies. International oil prices are projected to rise over time, but are assumed to be
independent of the quantities imported by the United States at anyone point
in time. Typically, this option is also pushed to its upper bound - a limit of 20
quads based upon national security considerations. In the absence of this
bound, the United States would import significantly higher quantities of oil.
For modelling purposes, it is assumed that oil imports would be limited either
by tariffs or quotas, not by 'voluntary export restraints'.
335
A constrained energy supply scenario

quads
25r------------------------------------------,
20
15
10
~~
1960
__
L_~_ _~~_ _~~~~_ _~~_ _L_~_ _L_~
1980
2000
2020
2040
2060
2080
2100
Fig. 18.4 Natural gas consumption (two alternative RDF values)
According to Table 18.3, there are two high-cost backstop options - both
available in unlimited quantities: SYNF (synthetic fuels based on coal or shale
oil) and NE-BAK (e.g., biomass fuels or hydrogen by electrolysis, using a
non-carbon-based source of electricity). NE-BAK emits no carbon, but is likely
to be more expensive than synthetic fuels based upon coal or shale oil. One or
the other of these high-cost technologies will impose an upper bound upon the
cost of non-electric energy - depending on whether or not there is a carbon
constraint.
18.6 A CONSTRAINED ENERGY SUPPLY SCENARIO

In this section, we explore the implications of a carbon constraint for the
energy sector and for the economy as a whole. Carbon constraints can take a
336
variety of shapes and forms. Legislative proposals have ranged from slowing
the future growth rate to reducing CO 2 emissions to half their current levels.
Because of the wide range of options under consideration, Global 2100 has
been designed with a great deal of flexibility regarding the imposition of carbon
constraints. Here we illustrate the capabilities of the model by calculating the
economic costs associated with just one set of emission reduction targets.
Specifically, we investigate the costs of restricting carbon emissions to 1.37
billion tons (their 1990 rate) up to 2000, reducing them gradually to 80% of
this level by 2020, and stabilizing them thereafter. Although these targets are
not as stringent as those contained in some proposed legislation, they nevertheless represent a substantial reduction in future emissions when compared
with a business-as-usual view.
The impacts of a CO 2 limit will depend on the technologies and resources
available for meeting energy demands as well as the demands themselves. Table
18.4 summarizes five energy supply-demand scenarios under which the impacts
of this carbon constraint will be analysed. Scenario I represents the most
constrained case - both from the perspective of supply enhancement and
demand conservation. On the supply side, we have excluded the coal technologies with CO 2 removal capabilities (Coal-R) and the low cost non-carbonbased sources of electricity (ADV-LC). We begin with such a highly constrained supply scenario in order to establish a basis for calculating the benefits
of alternative generation options having lower CO 2 emissions. These alternatives include advanced nuclear power and also coal gasification with CO 2
removal capabilities.
On the demand side, the distinguishing characteristic of scenario I is the rate
of autonomous energy efficiency improvements. We assume a zero value for
the AEEI parameter. As in the case of electricity supplies, we start with the
most constrained case and then assess the benefits from measures which reduce
CO 2 emissions.
A carbon constraint will have both direct and indirect consequences for the
economy. Because of the absence of low cost alternatives, the economic impacts
Table 18.4 Five energy supply-demand scenarios
Scenario
Supplies
Demands
I
II
Constrained case
Constrained case
plus Coal-R
Constrained case
plus ADV-LC
Constrained case
Constrained case
plus Coal-R and ADV-LC
AEEI = 0.0% per year

III
IV
V

337
will be highest in scenario I. To understand the source of these impacts and to

gain some insight into the workings of the model, it will be useful to explore
how the energy sector might evolve over the next century under such a highly
constrained scenario.
Figures 18.5a and 5b show a snapshot of the energy sector at two points in
time, 2010 and 2030, under scenario I. In each instance, the optimal combination of supply alternatives is shown - with and without the carbon constraint.
Within the electricity sector (Figure 18.5a), note the importance of priceinduced conservation. A carbon constraint limits the options for electricity
generation, raises the price of electricity, and this in tum drives down demand.
With a carbon constraint and in the absence of an economical CO 2 removal
capability or a low cost carbon-free alternative, the options for meeting
electricity demand are severely limited. One alternative is greater reliance on
natural gas. Indeed, Figure 18.5a shows a significant rise in the use of this low
carbon fuel in the electric sector. Recall that gas-fired plants produced only
12% of total electricity in 1985. According to Global 2100, the share of gas will
rise to 27% by 2010 if we are in a carbon-constrained environment.
Increased demands for natural gas from the electric sector will eventually
place tremendous pressure on natural gas markets. Our calculations show that
by 2010 the price of natural gas will approach the point where high cost
renewables (ADV-HC) become an attractive source of electricity. Despite the
cost, we see a substantial role for this supply category if carbon emissions are
constrained, and low cost alternatives are unavailable. In such a world, high
cost renewables would by necessity become the marginal source of electricity
supply.
Without a carbon constraint, the picture is entirely different. According to
Figure 18.5a, coal would once again be the fuel of choice and would supply an
increasingly larger share of the load. Coal-fired electricity is plentiful and
relatively inexpensive. Although natural gas prices would not rise as rapidly as
with a carbon limit, the geological resource constraints and competing demands from the non-electric sector will nevertheless lead to significant price
increases. The results of Global 2100 are broadly consistent with the estimates
that appear in Woods (1988). In his baseline projection for the Gas Research
Institute, he projects a tripling of wellhead prices by 2010.
Now consider the non-electric side of the overall energy balance (Figure
18.5b). In the short term, up to 2010, a carbon constraint would be felt mainly
through its impact on the price of natural gas. High prices will in tum lead to
lower demand through price-induced conservation.
The impacts of a carbon constraint on the non-electric sector become more
pronounced over time. As indicated by Figure 18.5b, if CO 2 emissions were not
a concern, the burden of meeting non-electric demands would eventually shift
to carbon-based synthetic fuels. Such a shift is infeasible, however, if there are
stringent limitations on carbon emissions. We would then have to rely upon a
non-carbon source such as hydrogen produced by electrolysis (NE-BAK).
Electric energy-2010
E0 Gas
IZmCoal
iIi Ii Ii lili II
I
I I I I I I I
Fig. 18.5
NE-BAK
rn Oil-D
D Oil-M
Electric energy-2030
20
40
60
80
IZZZJ Coal
ADV-HC
~ ~ ~
...................
ffiW7l%0i
~
NE-BAK
~CLDU mGas ~SYNF r:::r:::JOil-D C]Oil-M
~ ~ ~ ~
No carbon limit
1/66(6((6((66'(6(66'
Non-electric energy-2030
till Gas
Carbon limit
Hydro
I 3P20Bb
Carbon limit
[0".-.-.-.-...-.......................... ,
quads
oI
No carbon limit
Electric energy projections (scenario I); b, non-electric energy projections (scenario II).
Gas ~ SYNF
~.~
[gWjpjj _JlI
.1'1'1'1'1'1"1
Carbon limit
E2ZI CLDU 0
20
40
60
No carbon limit
80 ir---------------------------~
ADV-HC
@fffpiff@ff@1
Non-electric energy-2010
Hydro [IIJ] LWR
firniTiTUFiTi"i Ti"i"Fl
quads
oI
Carbon limit
8
No carbon limit
ThWh
10 ri--------------------------------~
10 r-------------------------------~
TkWh
339
According to Table 18.3, this type of fuel is likely to be considerably more

expensive than, for example, gasoline from oil shale or from direct coal
liq uefaction.
Using Global 2100, we may add together the costs throughout the economic
system and calculate the annual losses in consumption due to the carbon
constraint. Figure 18.6 shows the losses in each time period for scenario I.
The effects of a carbon constraint do not begin to have measurable macroeconomic consequences until 2010. At that point the rise in energy prices
begins to have a significant effect upon the share of gross output available for
current consumption. By 2030, roughly 5% of total annual macroeconomic
consumption is lost as a consequence of the carbon constraint. This percentage
remains relatively constant for the remainder of the time horizon. Adding over
all the years from 1990 to 2100, the present value of these losses would be $3.6
trillion, discounting to 1990 at 5% per year. 4
It is instructive to look at the time-path of CO 2 emissions in the absence of
a carbon constraint. From Figure 18.7, we see that there could be a six-fold
increase between 1990 and 2100. This represents an average annual growth rate
$ trillion
100c-----------------------------------------------------~
10
Annual losses due
0.1
1990
__
__
2000
~~
2010
__
__
2020
____
2030
2040
___ L_ _
2050
~~
2060
_ _ ~_ _ ~_ _ _ _ ~_ _ ~
2070
2080
2090
2100
Fig. 18.6 Aggregate consumption and losses due to carbon limit - scenario I
4A
5% discount rate is consistent with the numerical assumptions that underlie the economy-wide
production function, equation (18.2). For all cases reported here, we employ 24% as capital's share
of the GNP, 2.4 as the initial capital-GNP ratio and 5% as the net annual rate of depreciation of
the capital stock.
340
Billion tonnes of carbon
10
r--------------------------------------------,
No carbon limit
Carbon limit
o~--~--~--~--~--~--~--~--~--~--~--~
1990 2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Fig. 18.7 Carbon emissions - scenario I

of about 1.7% per year. Although the increase is large, it should come as no
surprise. In the absence of a carbon constraint, carbon based fuels are the most
economical source of supply in both the electric and non-electric sectors.
There are a variety of policy instruments available for reducing emissions to
the desired levels. One frequently discussed option is to impose a uniform tax
upon those activities responsible for carbon emissions. Figure 18.8 shows the
size of such a tax for scenario I. The tax is relatively low in 2000 ($29 per ton
of carbon), then rises sharply as emission limits are tightened. In the absence
of low-carbon supply alternatives, consumers are willing to pay a high price to
burn carbon-based fuels. The tax must be sufficiently high to discourage these
demands. By the middle of the twenty-first century, sufficient additional
capacity is available to stabilize the tax at about $250 per ton of carbon. In
this scenario, the long-run equilibrium tax is determined by the cost and
emission coefficients of the synthetic fuels and non-electric backstop supply
technologies.
18.7 REDUCING THE COSTS OF A CARBON CONSTRAINT

Of our cases, scenario I is the most constrained - from the perspective of
supply enhancement and demand conservation. Experience has shown that
energy forecasting, even over a few decades, is a highly inexact art. At best, one
can ask a series of 'what if' questions' in the hope of gaining some insights into
341
Reducing the costs of a carbon constraint

Dollars per tonne of carbon
700 . - - - - - - - - - - - - - - - - - - - - - - - - - ,
600
500
400
300
200
100
O~-k--~-~-~--~-~-~-~-~~-~
2000 2010
2020
2030
2040
2050 2060
2070 2080
2090
2100
Fig. 18.8 Carbon tax - Scenario I
the relative attractiveness of various means of reducing CO 2 emissions. It is in

this spirit that we have examined several alternatives - first individually and
then in combination.
In scenario II, we calculate the benefits that might accure from the successful
development of a cost-effective means of CO 2 removal and disposal of the gas
in a manner that prevents it from reaching the atmosphere. Recall that the cost
and performance data contained in Table 18.2 are for coal gasification with
20% and 93% carbon removal.
Scenario III explores the benefits from a low-cost non-carbon-based source
of electricity. Here our costs and performance data are based on those for an
advanced nuclear technology with passive safety features (see EPRI's Technical
Assessment Guide, 1989).
Figure 18.9 shows these benefits in terms of reductions in the costs of a

carbon constraint. For developing a CO 2 removal and disposal capability with
the assumed characteristics, the discounted benefits are about $0.6 trillon (the
difference in consumption losses between scenarios I and II). The discounted
benefits from a low-cost non-carbon-based source of electricity are $1.2 trillion.
The benefits from these two technologies would not be additive. To a certain
extent, the two technologies would be substitutes for each other. However,
because of constraints on the rate that any single technology can be introduced,
there are benefits from deploying both.
On the demand side, a major issue is the rate of autonomous energy
efficiency improvements. Up to this point, we have assumed that the AEEI is
zero. That is, there are no energy efficiency improvements except those that are
342
$ trillion
6r---------------------------------------------,
5
4
Constrained
Scenario I
Scenario II
Scenario III
Scenario IV
Scenario V
Fig. 18.9. Consumption losses (1990-2100)
price-induced. For scenario IV - the high efficiency case - we assume an AEEI

of 1.0% per year. By comparison with scenario I, energy demands in 2050
would be nearly halved. Figure 18.9 shows that this huge reduction in energy
requirements would significantly lower the cost of a carbon constraint. The
total discounted consumption losses would drop to $1.8 trillion. By definition,
no consumption losses are imputed to autonomous energy efficiency improvements. It is highly controversial whether such rapid improvements are
indeed achievable. Through scenario analysis, we have sought to bound the
broad range of viewpoints on this issue.
The rightmost bar on Figure 18.9 shows the costs of a carbon limit in the
best of all worlds defined by these five scenarios. That is, both Coal-R and
ADV-LC are available, and autonomous (non-price) demand reductions occur
at the rate of 1.0% per year. In this case, the discounted costs of the carbon
limit fall to $0.8 trillion. This would be only 22% of the losses incurred under
the constraints of scenario I.
Finally, we compare the emissions for all five scenarios - both with and
without a carbon limit (Figure 18.10). When carbon emissions are unconstrained, the profiles for scenarios I and II are identical. There would be no
purpose in deploying the carbon removal and disposal technology (Coal-R) if
carbon emissions were unconstrained.
Scenario III includes a low-cost non-carbon-based source of electricity
(ADV-LC). Such a technology would be attractive for economic reasons alone.
If it were introduced in 2010, it would assume an increasing share of the electric
load thereafter. This automatically leads to a substantial reduction in the rate
of growth of carbon emissions.
343
The potential benefits from Rand D

Billion tonnes of carbon
10~---------------------------------------'
- .......-......: ..-:,. . ..::. ....:.-2
-- --
.:-; .. . oi
.. , ...;..'.:-' "AEEI=1.0 /o
AOV-LC and AEEI=1.0%

Carbon limit
O~--.-~r--.---.---'---r---r--.---.---'-~
1990 2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Fig. 18.10 Carbon emissions with and without carbon limit
Scenario IV differs from scenario I only on the demand side. It includes

energy efficiency improvements at the annual rate of 1.0%. This leads directly
to lower demands for electric and non-electric energy - and therefore to lower
carbon emissions. When these lower demands are combined with a low-cost
and non-carbon-based source of electricity, carbon emissions are reduced still
further (scenario V). Even with the technological advances implied by scenario
V, unconstrained carbon emissions would still exceed our carbon limit by a
factor of two.
18.8. THE POTENTIAL BENEFITS FROM R&D
In this chapter, we have calculated the macroeconomic impacts oflimiting CO 2
emissions under alternative scenarios. If emission controls are required, there
will be significant costs, but it is clear that the nation can reduce the size of the
ultimate bill through R&D in both the supply and demand sides of the energy
sector. According to our calculations, the combination of potential improvements could reduce the costs of a carbon constraint - perhaps by several
trillion dollars.
These savings will not occur unless there are sustained research and
development programmes on a wide variety of fronts - both in the public and
the private sector. On the supply side, consistent long-term funding is needed
to promote non-carbon energy sources such as solar, fission and fusion. R&D
on CO 2 emissions control could have a substantial payoff - provided there is
an economical solution to the problem of permanent disposal. There are also
344
gains to be anticipated from more efficient processes for the conversion of

conventional fuels into secondary energy forms such as electricity.
On the demand side, there may also be an enormous potential. Following
the oil price explosion of 1973, there has been a remarkable improvement in
the efficiency of energy utilization. It is unclear how long this trend will
continue. Some of this may be the result of an autonomous time trend, but a
good deal may be attributed directly to the price mechanism. Research is
needed to clarify the role of conservation, and to ensure that cost-effective
options are available to the greatest extent possible.
This paper has focused on the costs associated with one set of emission
reduction targets. Given the current state of knowledge, it is unclear whether
it would be justified to incur these costs. There remain a number of uncertainties in our understanding of the greenhouse effect and in the likely effectiveness
of various countermeasures. As observed by Clark (1985), 'every responsible
scientific assessment of the last several years has noted (if not always emphasized) how thoroughly uncertainties pervade the carbon dioxide question'.
These uncertainties will not be resolved in the near future.
It would be extremely costly to wait for scientific certainty on the impact of
greehouse gases upon global climate before committing to a vigorous R&D
programme. New technologies require many years for market penetration. If it
turns out that substantial reductions in CO 2 emissions are needed, it will be
important to have the means available for achieving such reductions in a timely
manner. This can only be accomplished through a sustained commitment to
R&D.
ACKNOWLEDGEMENTS
The research reported in this paper was funded by the Electric Power Research
Institute (EPRI). The views presented here are solely those of the authors, and
do not necessarily report the views of EPRI or its members. The authors are
much indebted to Diane Erdmann and to Lawrence Gallant for research
assistance. Helpful comments have been provided by George Booras, Jae
Edmonds, George Hidy, William Hogan, Dale Jorgenson, Stephen Peck, Scott
Rogers, John Rowse, Chauncey Starr, Stanley Vejtasa, Gary Vine and Robert
Williams. An earlier version of this chapter appeared as a paper in The Energy
Journal, 2(2), April 1990.
REFERENCES
Allen, R. G. D. (1968) Macroeconomic Theory. Macmillan, New York.
Brown, S. P. A. and Phillips, K. R. (1989) An Econometric Analysis oj US Oil Demand.
Federal Reserve Bank of Dallas.
Brooke, A., Kendrick, D. and Meeraus, A. (1988) GAMS: A User's Guide. Scientific
Press, Redwood City California.
References
345
Clark, W. C. (1985) On the Practical Implications of the Carbon Dioxide Question.

International Institute for Applied Systems Analysis, Laxenburg, Austria.
Edmonds, J. and Reilly, J. M. (1985) Global Energy - Assessing the Future. Oxford
University Press, New York.
EIA (Energy Information Administration) (1989) Monthly Energy Review, US Department of Energy, Washington, DC.
EMF (Energy Modeling Forum) (1977) Energy and the Economy. Report 1, Energy
Modeling Forum, Standford University, Stanford, Ca.
EPRI, (1989) Technical Assessment Guide. Electric Power Research Institute, Palo Alto,
Ca.
Goldemberg, J., Johansson, T. B., Reddy, A. K. N. et al. (1987) Energy for a Sustainable
World. World Resources Institute, Washington, DC.
Jorgenson, D. W. and Wilcoxen, P. J. (1989) Environmental Regulation and US Economic
Growth. Harvard University, Cambridge, Mass.
Hogan, W. W. (1988) Patterns of Energy Use Revisited. Harvard University, Cambridge,
Mass.
Manne, A. (1981) ETA-MACRO: A User's Guide. EA-1724, Electric Power Research
Institute, Palo Alto, Ca.
Masters, C. D., Attanasi, E. D., Dietzman, W. D. (1987) World Resources of Crude Oil,
Natural Gas, Natural Bitumen, and Shale Oil. 12th World Petroleum Congress,
Proceedings, vol. 5.
Rind, D. (1989) A Character Sketch of Greenhouse. EPA Journal, 15, no. 1.
Vejtasa, S. A. and Schulman, B. L. (1989) Technology Datafor Carbon Dioxide Emission
Model: Global 2100. SFA Pacific, Inc., Mountain View, Ca.
Woods, T. J. (1988) The Long-Term Trends in US Gas Supply and Prices: the 1988 GRI
Baseline Projection of us Energy Supply and Demand to 2010. Gas Research Institute,
Washington, DC.
Index
Abatement 314
Accommodation 157
Adjustment mechanism 143
Advanced supply technology 330
AEEI 295, 326, 342
Aggregation 11, 188
Agriculture 53
Alternative models of transport 75
Appraisal 97
Asymmetric elasticities 158
Autonomous energy efficiency, see AEEI
Average practice 31
Backstop 335
Balance sheets 227
Behavioural lag 14
Beneficiation 94
Best-practice 31
Bias 77
Budget deficit 170
Capacity 32
expansion 38
utilization 33, 228
Captive markets 235
Carbon
emissions 293
quota rights 325
tax 179,340
see also CO 2
Cartel 263
see also OPEC
CGE 145,299
CGP 187
China 287
CO 2 emission 180, 323
CO 2 removal 336
Coal 81
characteristics 86
depletion 87
seam thickness 90
supply modelling 86
transport cost 87
Cobb-Douglas 33
Co integration 75
Comparative statics 164

Composition of economic output 49
Computable general equilibrium, see
CGE
Concession 129
Conditional factor demand 41
Conservation 209, 244, 292, 326, 340
Continous mining 91
Contract form 129
Cores 98
Cost function 16
Cross-section 69
Cumulative discoveries 275
Decomposable systems 233
Deflationary spiral 169
Demand
distributed lag 71
dynamic model 2, 68-74
elasticities 8-9
growth 37
Koyck model 70
partial adjustment 69
static model 68
Demography 224
Derived demand 67
Detailed simulation 221
Developing countries, see LDC
Disaggregation 227
Discovery 97, 124
Distributed lag 270
Distibution infrastracture 113
Distributional effects 315
Divisia index 47
Drilling effort 108
Dynamic adjustment 198
Dynamic linear regression 269
Eastern European 226
EC (European Community) 180,293
ECC 141
Ecological efficiency 299
Economic planning 187
Economic structure 47
Efficiency 227
348
Index
Effluent charges 300

EFOM 1, 175, 185
Elasticity
inferred 239
oil and product 248-50
scale 35
Electricity demand 26, 53
Electricity generation 330
Electricity intensity 59
Electrification 57, 209
Embodied technical progress 31
EMF 17,61,239
Emission
abatement 323
permits 143-5
Endowment inventory 82
End-use energy service 113
Energy 191, 227
aggregate intensity 48
balance 206
efficiency 26, 204
exchange flows 222
household 189
industry intensity 49
output ratio 51
policy 157
price 37
production 191
programmes 209
saving, see Conservation
supply 193
tax 157-64
Energy Modeling Forum, see EMF
Energy-economy interaction 27, 186,326
Energy-environment 230
Engineering approach 224
ENMARK 148
Environmental cost functions, see ECC
Environmental policy 180,299, 312
Error correction 76
ETA-MACRO 242, 325
Excess supply 171
Exhaustible resources 334
Exogenous supply 193
Expected profit 123
Exploration 97, 118
Extraction cost 85
Flexible functional forms 16
Forecast 11, 60
Fuel choice 13
Fuel intensity 58
Fuel prices 19
Gas
resources 107
transportation 112
Gasoline, see Transport fuel
Geographical distribution 111
Geographical disaggregation 222
Geological and technical factors 127
Geological assurance 83
Geological risk 98
Global 2100 324
Global warming 323
Greenhouse gas 233, 323
GRI (Gas Research Inst) 296
Growth impact 315
Gulf 234
Heat 52
HERMES 159, 175, 188
Hierarchical models 209
Historical trends 58
Hydrogen 337
lEA 1, 296
lEW 287,328
IIASA 1, 141, 287
Incentives 128
Income elasticity 76, 253
Individual fuels 291
Industrial organization 312
Innovation incentives 300
Input coefficients 31
Interetemporal equilibrium 303
Interfuel competition 27
Interfuel substitution 326
Internal market, see EC
International energy workshop, see lEW
Intertemporal production theory 306
Inverted V 72
ISTUM 18,61
Judgmental attitudes 110

Kakutani's theorem 308
Koyck 270
Kuwait 243
Lag structure 71
LDC 133-7,226
Less polluting technologies 309
Level of recovery 93
Licence 129
Linked models 179
Long-run 251
Index
Macroeconomic
cost 160, 339
effects 229
modelling 158
shocks 27
structures 195
MARKALI
Market economy 57
Market power 147
Maximum usefulness factor 89
MDE 189
MEDEE 175, 185
Median lag 272
Median poll 289
MELODIE 188
Methane 111
MIDAS 175
Model
bottom up 189, 223
disaggregated 221
duality 15
dynamic 207
econometric 14, 89, 120
equilibrium 21, 186
fuel supply 208
hybrid 17, 224
input-output 186
macroeconomic 157-77
neo-Keynesian 159
optimization 15, 120, 179
probabilistic 15, 108
process and technical 13, 89, 178
top-down 13, 189, 223
typology 185
Monetary 157
Monopolization 132
Motoring, see Transport fuel
Multilevel planning 209
Myopic models 303
Natural gas
interregional trade 114
resources 105
Neo-Austrian school 304
Netback 123
NMP- 53
Non-carbon 337
Non-conventional resources 111
Non-linear optimization 326
Non-OPEC supply 273
Nuclear 292
OECD 226, 287
Oil exploration 117

Oil products 158
Oil supply 117-18
see also World oil
Oligopolistic game 137
Omitted variable 77
OPEC 227, 240, 263
dilemma 282
strategy 228
Optimal policy 34
Optimal price-path 283
Optimization 1, 263
Overburden 89
Perestroika 293
Phillips curve 177
Planning 203
POLES 221-30
Political factors 132
Political risk 128
Poll responses 287
Pollution treatment 310
Polynomial distributed lags 76
Pooled data 75
Post-Keynesian 195
Power parity 268
Price elasticity 76, 271
Price hike 228
Price of crude 288
Price of reserves 122
Price reaction function 264
Price-dependent substitution 302
Price-response 229
Price-takers 264
Primary energy 194, 290
Production
average 31
best-practice 31
putty-clay 177
sharing 130
structure 38
technique 306
Projections 203
Public sector 170
Purchasing 268
Putty-clay 197
Quasi rent 38
R&D 343
RAINS 141
Recoverability 91
Reduced form 73
349
350
Refining 194
Reserves 82, 334
Resources 82, 334
Risk management 128
Risk-service contracts 129
Scarcity value 121
Scenario analysis 22, 166, 221
Sectoral shift 27, 50
Sedimentary basin 111
Sensitivity, see elasticity
Shift effect 49
Short-term dynamics 196
Shovel bucket capacity 89
SIBILIN 185,221-9
Simulation 146, 223, 263
Soviet Union 50, 226, 207,287
SOx 146
Standardized projections 287
Standardized scenarios 239
Standards 315
State intervention 117
Steady-state 164
Strategies
OPEC 280
Substitution 38, 197
autonomous 329
dynamic function 264
geological 276
po ten tial 81
price-induced 329
Index
technology 326
Sustainability 230
Swedish economy 147
Synthetic fuel 335
Taxation effects 122
Technical progress 37-9
Technique choice 31
Technology 32, 93
Time and adjustment 302
Transformation 235
TRANSLOG 16
Transport fuel 65-77
Transportation 53
UK 179
Uncertainty 232, 315
Undiscovered resources 109
US geological survey 95
USA 57, 323
Vehicle characteristics 73
Vertical aggregation 306
Vintage 31, 162, 307
Wage indexation 171
WOMS 242, 264
World energy 221
World oil market 239, 263
Yield 91

International Energy Economics

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

International Energy Economics

Transféré par

Droits d'auteur :

Formats disponibles

INTERNATIONAL ENERGY ECONOMICS

INTERNATIONAL STUDIES IN ECONOMIC MODELLING

Economic Modelling in the OECD Countries

Modelling the Labour Market

Models of Disequilibrium and Shortage in Centrally Planned Economies

Economic Modelling at the Bank of England

International Commodity Market Models

Economic Models of Trade Unions

International Energy Economics

International Trade Modelling

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

1992 Springer Scien ce+Bu sine ss Medi a Dordrecht

12 Models and projections of energy use in the Soviet Union

Strategic Studies Department, PowerGen, UK

contributions and there were to be a couple of papers from authors in the

1.1 ENERGY MODELLING

parency, comparability and the analysis of how results are conditioned by

1.3 ENERGY SUPPLY AND THE ENVIRONMENT

Energy and the economy

Forecasts and environment

hypotheses. In Chapter 14, Dr Huntington has found an extraordinarily

Forecasts and environment

Forecasting industrial energy use

How models deal with fuel choice

forecasts, because lack of experience can eventually be overcome if coal is

2.3 HOW MODELS DEAL WITH FUEL CHOICE

Forecasting industrial energy use

H ow models deal with fuel choice

Forecasting industrial energy use

qu, P K , PL , PE, P F , PM)

Selected models of industrial energy use

Econometric models may include a technology (time) trend to capture other

2.4 SELECTED MODELS OF INDUSTRIAL ENERGY USE

Forecasting industrial energy use

Selected models of industrial energy use

Forecasting industrial energy use

The Energy Information Administration (EIA) within the US DOE created

Selected models of industrial energy use

Forecasting industrial energy use

Regional industrial growth by sector, energy, capital, labour and

State level industrial growth by sector, energy, capital, labour and

Industrial growth by sector, energy prices, energy equipment

Inputs and outputs of models

Regional energy demands and fuel choice

Federal region energy service demands and

Energy service demands and fuel choice

All 2-digit industries

All 2-digit manufacturing SIC

Major 2- digit manufacturing SIC

All 2-digit manufacturing SIC

Process and energy service (for

Process and energy service

ECDB and other

ECDB and other

Table 2.2 Other model characteristics

Forecasting industrial energy use

cogenerated electricity is sold to utilities, this extra source of electricity

Forecasting industrial energy use

Technique choice and energy demand in a vintage model

progress is represented by a Cobb-Douglas ex ante production function from

denote planned production at time v in vintage v, planned use of variable

where qi (0) is the initial price.

Technique choice and energy demand in a vintage model