Vous êtes sur la page 1sur 419

Investment Strategy

Collected Research Papers


2015 Edition

Editors: Pascal BLANQUÉ & Philippe ITHURBIDE

For professional investors only


Investment Strategy
Collected Research Papers
Pascal BLANQUÉ & Philippe ITHURBIDE
(Editors)

2015 Edition
Amundi’s research capabilities ranked 2nd
Earning in Europe by Extel*
your confidence Once again our first-class equity research and
analysis has been acknowledged in the Extel 2014
Pan-European Survey:
also means providing 14 of Amundi’s financial analysts place amongst
the top 100 European analysts,
equity analysis rivalling Amundi is in the top 5 for 20 of the 23 sectors
analysed.
the best in Europe. Another illustration of our commitment to serving
you by ensuring our investment teams have access
to the highest quality analysis and research available.

* Survey conducted among sellside, buyside and corporates between March 2014 and May 2014. 532 brokerage houses and 881 corporates took part in the Survey. Out of those
voted, 167 sellside firms and 479 corporates nominated buyside firms and individuals. The methodology and the full results are on the Extel website, www.extelsurveys.com.
This document in no way constitutes a purchase offer or sales solicitation and should not be construed as a solicitation which might be deemed unlawful nor as investment advice.
Amundi shall not be held directly or indirectly liable in connection with the use of the information contained in this document. Under no circumstances shall Amundi be held liable for
any decision taken on the basis of this information. The information contained in this document may not be copied, reproduced, modified, translated or distributed, without the prior
written approval of Amundi, to any third person or entity in any country or jurisdiction which would subject Amundi or any of its funds to any registration requirements within these
jurisdictions or where it might be considered as unlawful. The information contained in this document is deemed to be accurate as at 28 July 2014 and may be altered without notice.
Amundi, société anonyme with capital of €596,262,615 - Portfolio management company approved by the AMF under no. GP 04000036 - Registered office: 90 boulevard Pasteur,
75015 Paris - France - 437 574 452 RCS Paris. Photo: Getty Images. |
Note of the editors

The book you have in your hands


is the 2015 edition of Amundi Research
papers. This book is a collection of some
of the most illustrative research papers
of the last 12 months. It is emblematic
of what Amundi Research can provide,
in addition to macroeconomic research,
strategy and forecasts, as well as
research on asset classes. Some of the
papers have already been published
in academic reviews, others are scheduled
to be published and, last but not least,
some of them were written to provide
guidance on often complex issues.”

Pascal BLANQUÉ & Philippe ITHURBIDE


Table of Contents

Foreword
- Amundi Research: a multifaceted and a client-oriented business-line p. 9
- Amundi Research in a nutshell p. 14
- Amundi Research Center p. 15
- Cross Asset Line p. 16

Amundi Discussion Papers Series


- Long economic cycles and the asset markets p. 21
- Reallocating savings to investment: the new role of asset managers p. 65
- Allocating alternative assets: why, how and how much? p. 77
- The short investment cycle: your roadmap p. 97
- Physical real estate in long-term asset allocation: the case of France p. 131

Amundi Working Papers


- Global Excess Liquidity and Asset Prices in Emerging Markets:
Evidence from the BRICS p. 143
- Sovereign Default in Emerging Market Countries:
A transition Model Allowing for Heterogeneity p. 181
- The Asset- and Mortgage-Backed Securities Market in Europe p. 209
- Option Pricing under Skewness and Kurtosis using
a Cornish Fisher Expansion p. 223
- Modelling Tail Risk in a Continuous Space p. 241
- Does Regulation Matter? Riskiness and Procyclicality
in Pension Asset Allocation p. 261
- Is your Portfolio Effectively Diversified? Various Perspectives
on Portfolio Diversification p. 307

About the authors and editors p. 367

Research publications p. 377

Amundi Investment Strategy Collected Research Papers 5


Foreword

O ver the past few years, Amundi has chosen to expand its research
teams and to get them implicated in its investment decisions. Our
approach has consisted in expanding both our top-down and bottom-
up resources and involving them with the management teams and asset
allocation decisions on a regular basis. Research teams also participate
in the various investment committees and help promote new investment
processes. All this makes portfolio investment returns a shared objective.
This has also helped enhance Amundi’s visibility worldwide. It is important
to provide guidelines that explain and publicise Amundi’s views both
externally and internally. The new line of research products (the cross-asset
line) has made a big contribution to improving this “ duty to explain”. In
such a way, research work promotes the group’s investment strategies and
themes at all times.
Amundi has also organised its research team to achieve the key objective
of remaining close to the academic world. Publishing working papers and
financing research chairs, resulting in conferences and calls for papers,
help achieve this objective.
Amundi has therefore provided the research team with the resources
to foster the role that we have given it. In addition to proximity with
the management teams and clients, the team offers a broad diversity of
profiles and publications. This book shows clearly, if there was a need, how
important research is in Amundi’s set-up, as well as the depth and variety
of our research capabilities. It highlights both the role of research within the
investment process and its ability to make lasting analyses that form the
basis of future action.

We wish you all a good reading.

Yves Perrier, Chief Executive Officer, Amundi


June 2015

Amundi Investment Strategy Collected Research Papers 7


8 Amundi Investment Strategy Collected Research Papers
Amundi Research: a multifaceted
and a client-oriented business-line

Amundi has developed a platform of independent research


services supporting domestic and international investment
teams, covering all aspects of investment research.
This independent research platform boasts around 130
international experts who support both domestic and
international investment teams.

5 main strengths

TConviction-based, relevant research that seeks to provide


macroeconomic analysis, financial forecasts, cross asset investment
strategies as well as country and sector allocation. Our aim: provide
relevant insight on the new challenges faced by our clients. To
have views and convictions is our DNA: macroeconomic scenario,
financial forecasts, strategies, long term returns, country and sector
allocation, top picks, both top down and bottom up research are an
integral part of our duties.
TComplementary & multidisciplinary teams comprised of economists,
strategists, credit, equity, real estate & socially responsible investment
analysts. Such an array of experts enables us to provide quantitative
& qualitative research covering a wide range of areas. These teams 
contribute on advisory activities, partnerships with universities,
training programmes...
TA complete integration of research (especially bottom-up research)
into investment processes through systematic portfolio reviews
with asset managers, and, in some segments, through the necessity
of a recommendation from dedicated analysts.
TA wide range of publications spanning weekly, monthly and event-
driven macro-economic overviews to in-depth thematic research and
academic publications.
TBespoke research as part of our DNA: apart from regular publications,
Amundi prepares numerous tailor-made and confidential pieces of
research to answer specific requests from clients.

Amundi Investment Strategy Collected Research Papers 9


A wide range
TOP
DOWN RESEARCH
ECON
OM I S T S
of research

The team consists of


Y
EG

QU

economists, strategists, credit


STRAT

ANT

Forex,
Fixed income, research, equity research,
Credit, Equities,
Real estate socially responsible investment
research, quantitative research,
real estate research, and
N
A

AL
YS I contributes on advisory activities,
IS SR

BO
partnerships with universities,
TTOM CH
UP RESEAR training programmes...

10 Amundi Investment Strategy Collected Research Papers


Amundi Research:
a cross asset business line

TEconomic and Strategy Research: provides economic and financial


forecasts, sets out a fundamental and systematic approach to
markets using tools developed in-house, provides country and sector
allocation ideas and strategies …
TQuantitative Research: provides quantitative tools for fixed income,
equity & balanced portfolios management; provides a methodological
backbone to investment teams, academic research…
TEquity Research: provides an active input (bottom up analysis,
sector analysis) into Equity investment processes, internal ratings,
relative value trades …
TCredit Research: a dedicated team in charge of reviewing the
primary and secondary markets, as well as to provide issuer analysis
reports to fund managers, internal ratings, relative value trades …
TSocially Responsible Investment Research: integrates ESG
criteria (Environmental, Social and Governance) in the evaluation of
companies, provides internal rating on equities and bonds
TInvestment Advisory Research: tailor-made research for Clients’
specific topics (in line with portfolio management, hedging….),
provides academic research, training for clients.

The research team,


at the heart of investment management

Amundi’s research team works closely with in-house fund management


teams & advisory services contributing to portfolio reviews, sector
reviews, internal rating, target prices. As part of Amundi’s investment
committees, the team plays a key role in portfolio construction &
optimisation, asset allocation and relative value trades.
The Hub concept facilitates interaction among the providers of
expertise, between providers and fund managers, and with the rest
of Amundi.

Amundi Investment Strategy Collected Research Papers 11


Amundi Equity Research 2014
is ranked #2 in Europe: #2
a strong recognition

According to the 2014 Extel survey (which ranks firms based on the quality
of equity analysis services), Amundi Equity Research is ranked #2 in Europe.
This study brings in over 15,000 votes (banks, brokers and companies) on
more than 100 asset managers, a sample set of more than 2,500 analysts. Past
years’ rankings in Europe: #13 in 2010, #8 in 2011, #4 in 2012, #2 in 2013 and
2014. In addition, all the analysts’ rankings went up, and now 14 analysts (8
last year and 5 the year before), all sectors combined, are considered to be
among the top 100 European analysts. 20 of the 23 sectors covered are now in
the European top 5, and 12 in top 3.
All sectors combined, the 12 best analysts in France are all part of the Amundi
research team (the sample included just over 325 analysts).
These results are an asset for Amundi with regard to its clients. They solidify
all our efforts to raise our profile and garner recognition - goals that have
been top priorities.

12 Amundi Investment Strategy Collected Research Papers


Client servicing

Clients have access to Amundi research and to advisory services.


Clients have also access to advisory services. Here are some of the
issues being developed with different types of key clients in the past 2
years:
– Strategic asset allocation: SWFs, corporates, pension funds
– Risk budgeting: corporates
– Risk monitoring system: private banks
– Low carbon investment: pension funds
– Solvency friendly equity hedging: insurance companies
– Strategic allocation on European asset-backed securities: Central Banks
– Currency hedging strategy: corporates pension funds
– Extending allocation to unconventional bonds: Central Banks
– Advice for a suited benchmark: insurance companies
– Global allocation on asset- and mortgage-backed securities: Central
Banks
– Understanding Smart Beta strategies: performance drivers and portfolio
characteristics: SWFs, Central banks, pension funds
– Low-risk anomaly and interest rates environment: SWFs
– Building Risk-Parity portfolios on a “local currency” framework: pension
funds
– Risk monitoring tool for Emerging equity indices: insurance companies
– Tail risk reduction with volatility product: asset managers
– Integrating FX into investment process: SWFs
– Stress analysis: banks, pension funds
– Impact analysis of rising interest rates: pension funds
– Gold as an asset class in FX reserves: Central Bank.

In addition to top down research


and bespoke studies, the service Amundi Research
can provide to clients is multiform:
– Regular conference calls on market trends, asset allocation…
– Exchange views on methodology, modelling…
– Advisory on any kind of topic related to portfolio management, investment
processes, risk management…
– Training on various themes

Amundi Investment Strategy Collected Research Papers 13


Amundi Research in a nutshell

12
130 Nationalities

Research people (including JVs),


of which 40% based in Asia
4500
Meetings with companies each year
over the world (bottom up research,

2nd equity, credit and SRI)

2014 Ranking of Amundi equity research


in Extel – Thomson survey

Stockholm
Amsterdam Helsinki
Brussels Frankfurt
Luxembourg
London Warsaw
Montreal Paris Prague
Geneva Zurich
Yerevan Beijing
Milan
New York Madrid Seoul
Durham Tokyo
Athens
Shanghai
Casablanca
Hong Kong Taipei
Abu Dhabi
Bangkok
Mumbai Brunei
Kuala
Lumpur
Singapore

Santiago 20 Sydney

Spoken languages

International Investment Center Office dedicated to networkpartners Office dedicated to institutional


clients and third-party distributors

150+ 2000+
Number of conferences each year Meetings with portfolio managers
each year (both bottom up and top
down research): sector reviews,
portfolio reviews, weekly meetings…
2000+
Meetings with clients each year
in more than 40 countries

14 Amundi Investment Strategy Collected Research Papers


Amundi Research Center

Amundi has launched a website –


Amundi Research Center – dedicated
to providing you with our latest
economic outlook and research. It
seeks to bring us a step closer to
positioning Amundi as a thought
leader in investment management. The
Amundi Research Center publishes
leading edge research produced by
Find out more at: Amundi’s Research Team and delivers
www.research-center.amundi.com
up-to-date perspectives on current
financial issues.
The diversified online content
includes:
SCAN IT! ■ Commentaries and videos on recent
market events,
■ Fundamental research papers,
■ Weekly and monthly analysis,
■ In-depth academic research.

35+
Number of Working Papers published in
academic reviews in the past 4 years
(Applied Economics, Bankers
Markets & Investors, Economic
Policy, European Financial
Management, Finance, Financial
Analysts Journal, Journal of
Alternative Investments, Journal
of Asset Management, Journal
of Banking and Finance, Journal of
Business Ethics, Journal of Finance and Risk Perspectives, Journal of
Fixed Income, Journal of International Money and Finance, Journal
of Investment Strategies, Journal of Portfolio Management, Research
in International Business and Finance, Revue Française d’Economie…)

Amundi Investment Strategy Collected Research Papers 15


Cross Asset Line

Cross asset
investment
strategy # 03
Letter finalised at 3pm Paris time March 16 - 20, 2015
Research, Strategy and Analysis MONTHLY
Highlights of the week
March 2015 Ɣ Eurozone: Total employment rose slightly in 2014, but payroll costs decelerate. Good signals still coming from Ger-
many.
Ɣ United States: Figures on industry and real estate are disappointing.
Ɣ Markets: The Fed triggered a strong downturn in bond yields in the developed countries and stops dollar appreciation.
US credit markets have performed in recent days and the climb continues for equities.

Key focus

The Fed and the market: reality strike back


For some time now, the markets have been ruling out a rapid and, shall we say, "traditional" tightening of fed funds. Market expecta-
tions about the fed funds policy were vastly different from fed funds projections made by the FOMC members themselves (also known
as the "dots"). Now, this difference is much less pronounced, because the FOMC has clearly sent the message that fed funds
tightening would be very gradual. FOMC members projected a fed funds target of 0.50/0.75% (two 25 bp increases) at the end of
2015, in contrast to 1.00/1.25% previously, and of 1.75/2.00% by the end of 2016, in contrast to 2.375/2.625% previously. The "dots"
say the monetary tightening might even be far from finished by the end of 2017. Even though it has dropped the term from its
statement, the Fed will be very patient over the coming years.

It is no surprise that the Fed is retreating. In the past, we have regularly listed the reasons for doing so: a strong real appreciation of
the US dollar, weak inflation, low wages, a persistent slack on the labour market,
and faster-than-anticipated slowdown of China. If there is anything new in the The FOMC members' ("dots")
signals the Fed is sending, it is that the Fed is really being destabilised by
Asset Allocation how weak the impact of the improved labour market conditions has been on
projections vs. market expeta-
wage trends and therefore inflation: while FOMC members set their projection tions

ECB of the unemployment rate in the long run at around 5.3/5.4% for the whole year
2014, they have just abruptly lowered it to 5.1%. As certain FOMC members had
said in speeches recently, the unemployment rate below which inflation acceler-
4
3.5
3
ates is probably lower than it was in the past. While unemployment stood at 5.5%
Oil in February, there is still a bit further to go. 2.5
2

China In its statement, the FOMC explains that it will raise interest rates "when it has
seen further improvement in the labour market and is reasonably confident that
inflation will move back to its 2% objective over the medium term." This is a key
1.5
1
0.5
Quantitative point: whether or not the Fed begins its fed funds tightening in 2015, it will 0

Spot

JAN 16
MAR 16

JAN 17
MAY 15
JUL 15
SEP 15
NOV 15

MAY 16
JUL 16
SEP 16
NOV 16

MAR 17
MAY 17
JUL 17
SEP 17
NOV 17
need much more time to truly be confident that the inflation trend will return

easing to normal. Some of the best evidence is that the underlying inflation (i.e. the core
PCE) forecast has been adjusted downward for 2015 and 2016 (respectively to
1.3% and 1.7%). Now, inflation will be the key variable. Dots December 2014
Yuan Dots March 2015
Market expections 19 March 2015
Thus, the FOMC is giving the markets a little air, the US dollar has lost a little Market expections 13 March 2015

Fed ground, and US long-term yields have fallen a little. It's as if the American central Source : Bloomberg, Amundi Research

bankers had heard Christine Lagarde's warnings about the disastrous effect a Market expectations are proxied here by the fed
too-sudden monetary tightening would have on the emerging countries. funds futures.

Russia
Emerging countries
Luxury market
Document for the exclusive use of professional clients, investment service providers and other financial industry professionals. 1
Document finalised at March 9, 2015

Document for the exclusive attention of professional clients, investment services providers and any other professional of the financial industry

Monthly Weekly
This publication documents the outlooks, This publication provides a four-page
macroeconomic and financial forecasts, summary of the past week’s highlights,
risk factors, strategies and asset featuring an in-depth editorial on a
allocation of Amundi, covering equities, key development. It is not a listing of
debt and dive rsified portfolios. It also events, but a report on how some of
reports on the outcomes and topics of the events may impact our forecasts,
discussion from Amundi’s investment Amundi’s asset allocation and changes
committee. These topics are given special in risk factors identified to date. A
coverage in this monthly publication. quick overview of important events,
auctions and key indicators is provided
at the end of the publication.
ECB QE Monitor – March 24, 2015
The document you are reading is the first issue of a new publication which is aimed at monitoring the
ECB’s QE.

On credibility, the announcement of QE was a resounding success. Its scale, comprehensiveness and
suitability did not disappoint. The financial markets responded positively, as reflected in the fall in short
and long rates and the advance of asset classes such as Eurozone equities.

Implementation seems like a simple matter, at least in principle. The ECB will nonetheless be faced
with the challenge of purchasing €60 billion in securities each month in illiquid markets that are short of
willing sellers. QE encourages market players to buy or hold on to assets rather than selling them.
Meanwhile, the banks, which are major holders of government debt, retain these assets in portfolios for
regulatory purposes or simply out of liquidity considerations, as the securities can be used as collateral.
Given this backdrop, unless there is an explosion in issuance by governments, a rapid change in
regulators’ policy or “forced” sales by public funds (is this not what Japan demanded of public pension

ECB QE Monitor
funds?), the ECB’s drive to establish this programme will inevitably run into the realities of the market,
which will undoubtedly push down short and long-term interest rates even further. There will be at least
seven major impacts from this:

- Short term rates will remain in negative territory;

- The euro will be under downward pressure, partially counterbalanced by current account
surpluses;

- The long-term rate spectrum will see heightened contagion. Like the two-year and five-
year yields before it, the German seven-year yield has fallen into negative territory. Long-term rates will
inevitably stay very low;

- Government bond spreads will tighten from already exceedingly low levels. While they
can now offer little protection against a potential growth downturn or troubles in public debt or the
political situation, they are in the hands of the ECB, which is reassuring, at least in the short term;
This monthly publication is aimed at monitoring the
- Corporate bond spreads will tighten;

- European equities will continue to see gains;

- US long rates will remain low, all other things being equal: the slope of the curve, rate
ECB’s Quantitative Easing programme.
levels and the attractiveness of the USD are convincing arguments in favour of the US bond market.
Whereas the emerging markets were able to attract 20% of the liquidity from US QE, the US market

It monitors all the transmission channels of the


should attract some of the flows from Europe’s QE programme.

Ultimately, if everything proceeds as anticipated, the Eurozone can look forward to a more growth-
friendly environment. The final step for Mario Draghi involves the matter of effectiveness.

We have already alluded to the importance of transmitting QE to the real economy. Several transmission

monetary policy to examine the impact on the real


channels will have to be activated in order for growth to be revived:

- An “exchange rate effect”: any currency depreciation would contribute to competitiveness


and/or help restore business margins and/or lead to natural profit growth;

Document for the exclusive use of professional clients, investment service providers and other financial industry professionals. 1

economy.

16 Amundi Investment Strategy Collected Research Papers


Amundi Discussion Papers Series

DP-08-2014

November 2014

January 19 2015

Pre ECB Committee commentary


ECB’s QE. Which QE?
ALLOCATING ALTERNATIVE ASSETS:
PHILIPPE ITHURBIDE, Global Head of Research, Strategy and Analysis – Paris The ECB is expected to announce a large QE on
January 22, following its monetary policy
WHY, HOW AND HOW MUCH?
committee meeting.
The CREDIBILITY of the programme will Sylvie De LAGUICHE, Head of Quantitative Research
determine the impact on financial markets. The Eric TAZÉ-BERNARD, Chief Allocation Advisor
Introduction
ability of the ECB to announce a large QE and to
execute it successfully is crucial.
The ECB is expected to announce a large QE on January 22, following its monetary policy
A large QE would favour risky assets… if it is
committee meeting. This paper aims at explaining the rationale, the terms and the impact of
such a decision. Two key points are to be stressed first: substantial, well perceived, and if the ECB
demonstrates the capacity to execute it
¾ The CREDIBILITY of the programme will determine the impact on financial markets. The
ability of the ECB to announce a large QE and to execute it successfully is crucial. successfully.
Bond yields would decline further, while sovereign
¾ The EFFICIENCY of the programme will determine the impact of the real economy.
Some transmission channels will be at work: FX channel, Interest rate channel, liquidity spreads would narrow. Pay attention to the terms
and wealth effect channel and inflation expectations channel. of the QE, particularly if it excludes some
Credibility is the first round of the announcement, while efficiency will be second-round, to be sovereign debts (rating threshold, public finances
manifest in a few months. threshold…)
Without credibility, the impact of QE may be
disastrous, on economic sentiment on one hand
(GDP growth expectations, investment,
1 What would be the rationale behind a sovereign consumption, fiscal balances, debt sustainability),
QE? and on financial markets on the other hand
(sovereign spreads, equities, corporate bonds
The ECB’s current aim is to justify such a measure by appealing to the need to fight against particularly).
the menace of deflation, which threatens price stability—in other words, by fulfilling its sole The EFFICIENCY of the programme will determine
mandate (long-term inflation expectations dropped heavily over the last months). However, the impact of the real economy. Some
do not consider QE as a tool to solve debt problems… it is an indirect tool, not a direct one. transmission channels will be at work: FX channel,
This is not anecdotal should the Europeans want to come to an agreement with Germany. A Interest rate channel, liquidity and wealth effect
QE programme, based on capital key weight at the ECB is also a kind of debt mutualisation,
channel and inflation expectations channel.
and this is an issue for Germany. We will explore below different alternative solutions.
Credibility is the first round effect of the
announcement, while efficiency will be the second-
round, to be manifest in a few months.

2 Announcement

There are at least two options:

R Will the ECB announce a phased QE, a bit like in the US (Q1, Q2 and then Q3)... or
would it quickly announce a large-scale QE? In the first case, there could be some
disappointment (too little, too late). In the second, the hope of renewed optimism on
growth and a significant impact on equities, interest rates and spreads.
Amundi views: the announcement of a large QE is by far much better, taking into
䇾 Do not consider QE as a tool to
solve debt problems, but as a tool
to fight deflation


account the stakes, and market expectations. A EUR 500bln is highly likely.

Document for the exclusive attention of professional clients, investment services providers and any other professional of the financial industry
For professional investors only

Special Focus Discussion Papers


These reports, published intermittently, These do not cover the in-depth
deal with topics that are of current issues dealt with in the Special Focus
interest. They cover issues from a range publications, nor do they share the
of areas, including economics, politics academic angle of the Working Papers.
fixed-income markets, equities, credit, Rather, they offer an in-depth analysis
foreign exchange markets and asset on structural themes with long-term
allocation. implications and consequences on asset
allocation decisions.

Working Papers
Amundi Working Paper

These are research documents WP-046-2014


January 2015

intended for presentation at academic


conferences and for publishing in
theoretical and empirical research Option Pricing under Skewness and Kurtosis using a
Cornish Fisher Expansion
journals. The topics are always related Sofiane Aboura, Associate Professor – Paris-Dauphine University
Didier Maillard, Professor - Cnam, Senior Advisor – Amundi

to asset classes and asset management,


including long-term issues, portfolio
construction, and issues directly
related to the activities of our
customer segments and particular
methodologies. Recent topics of
publication have included asset and
liability management, pension funds,
the allocation of sovereign fund assets,
SMART benchmarking, strategies
asset allocation and inf lation-indexed
bonds. For professional investors only

Amundi Investment Strategy Collected Research Papers 17


18 Amundi Investment Strategy Collected Research Papers
Amundi Discussion Papers Series

Amundi Investment Strategy Collected Research Papers 19


20 Amundi Investment Strategy Collected Research Papers
DP-10

Long cycles
and the asset markets
ÉRIC MIJOT,
Head of Strategy Research, Amundi

May 2015

This article explains our interpretation of long cycles, sometimes


referred to as “Kondratiev cycles,” which informs our analysis of
the markets and our investment choices.
Our empirical approach, developed over the past thirty years, is
based first and foremost on an observation of asset trends since
the beginning of the industrial revolution and how these trends
match up with economic cycles.
There are four major cycles, which in turn can be broken down
into four periods (named after the seasons: spring, summer,
autumn, winter), which have repeated themselves for more than
two centuries now. A few special cases aside, generational breaks
are the rule rather than the exception.
Taking the US as our base example, we show this observation
to be true in the dozen different countries that we studied.
Rather than undermining this trend, demographic factors
tend to support it while providing indications to better predict
developments to come.
This analysis precedes the one laid out in our Discussion Paper
entitled “The Short Investment Cycle: a Roadmap,” which takes
a more detailed viewpoint in order to better seize opportunities.

Amundi Investment Strategy Collected Research Papers 21


LONG CYCLES
AND THE ASSET MARKETS

I. Kondratiev: father of the supercycle


Long economic cycles were first brought to light by Nicolaï Dimitrievitch
Kondratiev in 1926. When World War I drew to a close, the brand new Soviet
Republic had to institute a new economic policy. At the time, Kondratiev (1892-
1938) was working on a plan for the development of Russian agriculture as
Director of the Institute of Conjuncture and Professor at the Agricultural Academy
of Peter the Great. Since it takes a long-term view to plan effectively, Kondratiev
did empirical research to study the real operation of the economy since the
start of the industrial revolution, based on data from the UK, France, Germany
and the US. He used data on prices, wages, interest rates, commodities and
international trade, which he then verified by comparing them against data on
output and consumption.
He presented his work in 1926 before being dismissed from his position in
1928. Arrested in July 1930, he was sentenced to death by firing squad in 1938.
Based on his observations, he predicted the natural regeneration of capitalism
in cycles, which flew in the face of official Soviet views at the time. He noted that
two major complete cycles, each covering a roughly equivalent time period of
about 50 years, occurred back-to-back from the end of the 18th century to the
end of the 19th century. Another cycle and a half’s worth of observations have
been added since Kondratiev presented his work, lending further credibility to
his conclusions even as the great Communist experiment came to a definitive
end with the fall of the Berlin Wall in 1989.
According to Kondratiev, the length of these cycles is linked to the lifespan of
infrastructures in the broad sense of the term (transport and communication, but
also training, etc.). These major cycles cover all aspects of human organisation,
including the economy of course but also the very rules of operation of
our capitalist societies. At certain points in time, divides are created in the
established order that structures our way of life, thus generating changes in the

22 Amundi Investment Strategy Collected Research Papers


fabric of society itself. These cycles are also international in nature and therefore
embrace geopolitical factors. They possess certain innate characteristics. They
have regenerated at various levels of economic development and in different
economic environments. History always repeats itself, but in a different way
from one generation to the next, which is what makes economic and market
forecasts so uniquely probabilistic.
Kondratiev more specifically observed four recurring fundamental factors which
he dubbed empirical laws:

1.1 The first “law” pertains to inventions: they take place about 20
years before actually being rolled out on a large scale
The birth of new technologies deeply changes our way of life. Since the
investments required for their development are substantial and have to be made
profitable, the lobbies for these powerful industries work to make the movement
last. The associated changes gradually take shape and become the rule once
economic necessity and financial resources are combined, often after pushing
the previous state of affairs well beyond the acceptable limit. More than the
inventions themselves, it is the democratisation of these technologies - often
through “secondary” inventions - that truly shakes up the status quo.
Below are just a few major inventions or discoveries that marked the four past
Kondratiev cycles and hint at what’s to come in the next.
tCycle I (1783-1842): Coal - Steam Engine
James Watt invented the separate condenser in 1769 and the crankshaft in
1780, which enabled the widespread use of the steam engine. In addition, the
first power loom developed by Edward Carthwright in 1787 spread throughout
England in the 1830s.
tCycle II (1842-1896): Iron and Steel - Railway
Progress achieved in the use of coke to produce cast iron - a process established
in 1709 but not put into practice until the 1850s and 1860s - was the catalyst
for the technological revolution in this cycle. George Stephenson invented the
first steam locomotive engine in 1813 and the first railway was opened in 1830
between Manchester and Liverpool.
tCycle III (1896-1949): Electricity - Automobile
The first electricity transmission took place in 1883, between the cities of Vizille
and Grenoble, thanks to Marcel Deprez, and Thomas Edison invented the
incandescent light bulb, demonstrated in 1879. As for the internal combustion
engine, it owes its development to the designs of Beau de Rochas in 1862. After
an initial model was manufactured by German Nikolaus Otto, which was then
miniaturised by Frenchman Jules-Albert de Dion, German inventor Carl Benz
was able to create the first gasoline-powered automobile in 1886.

Amundi Investment Strategy Collected Research Papers 23


tCycle IV (1949- ?): Oil - Aviation
John D. Rockefeller founded Standard Oil of California in 1870 and Henry Ford
began mass producing the gasoline-powered engine in 1913 with the launch of
the Ford T. The two World Wars demonstrated the strategic role played by oil, now
the very life’s blood of our economies. In aviation, Alberto Santos-Dumont adapted
the internal combustion engine which before 1906 had been used exclusively in
automobiles. Once again, the World Wars played a key role in this development.
Aviation gave rise to the aerospace industry and accelerated the development of
electronics.
t Next cycle: Internet, Nano and Biotechnologies, Robotics, even Artificial
Intelligence
In 1965, Gordon Moore, co-founder of Intel, the world’s leading semiconductor
chip maker, predicted that the number of transistors contained in an integrated
circuit (electronic chip) would double roughly every two years, a prediction known
as Moore’s Law… a law that proved to be incredibly exact. This phenomenal
technological advance helped boost the information storage capacity and speed
of computers, and opened up new pathways to the future.
In 1962, Paul Baran proposed a distributed communications network to meet US
national security communication needs. This idea would not be further developed
until 1969 by US IT service provider Bolt Beranek and Newman Inc, which went
on to invent electronic mail in 1971. In 1989, Tim Berners-Lee invented the Web,
paving the way for the Internet boom largely beginning in 1995. The Internet, of
course, has already started to transform the way we trade goods and services. But
this is only the beginning.
Nano and biotechnologies are still in the fledgling stage, but can also be expected
to contribute to the transformation of our society. Miniaturisation allowed two
IBM researchers (Gerd Binnig and Heinrick Rohrer) to develop a scanning
tunnelling microscope in 1981, a much more powerful tool compared to the usual
microscopes in that it was capable of working at the nanometric scale. It thus
became possible to image surfaces at the atomic, molecular and macromolecular
level, where properties are very different from what we know about the macro level.
Following on from the coal and oil eras, carbon nanotubes are among the first of
these particles to have been produced, thus opening up a whole new world of
possibilities.
Textiles (impermeability, insulation) and cosmetics (sunblock) are two industries
that have fully embraced the use of nanoparticles, which have also contributed to
greater energy efficiency by improving insulation and increasing PV cell capacities.
LEDs, developed from nanoparticles, were first introduced through red stoplights
and the red brakelights of our cars, and are now lighting our homes. The porosity of
some of these materials also means that improvements can be made in hydrogen
storage, leading to fuel cell solutions for the clean cars of the future.

24 Amundi Investment Strategy Collected Research Papers


The pharmaceuticals industry, which has virtually exhausted the properties of
petrochemicals, also sees a lot of promise in this field. The mapping of the human
genome has opened another considerable area of exploration in medicine, a
field in the process of reinventing itself. In the future, we will no longer be treating
sickness, but the sick themselves. The cost of mapping out the human genome has
plummeted since the early 2000s: estimated at $95 million in 2011, it is currently
in the neighbourhood of $5,000, making this technology much more affordable.
Finally, robotics has also benefited from miniaturisation. Robots are now capable of
providing services, even in consumers’ daily lives. This technology can be expected
to revolutionise our way of life and our production methods alike.
Going yet another step further, artificial intelligence is also on the verge of
considerable advancement. Ray Kurzweil has shown that integrated circuits,
whose development was so prophetically described by Gordon Moore, are part
of a process initiated at the end of the 19th century with the aim of improving
processing power: electromechanics circa 1900, then solid state relays (late 1930s
to the end of World War I), vacuum tubes (from World War II to the late 1950s),
transistors (until the early 1970s) and finally the integrated circuits combining
these transistors. If we take the speed of advancement from the last twenty years
and extend it to the next twenty years, the combination of electronics, nano and
biotechnologies could pave the way for huge improvements in the medical field and
in robotics, to the extent we could begin speaking of artificial intelligence. Some
brain deficiencies notably due to ageing could then be treated (Alzheimer’s, hearing
and vision problems, etc.). What’s more, as the population ages, we could turn to
robotics to step up productivity.
These are just a few examples, of course, but they tend to indicate that, as with
the four previous major cycles, these types of inventions are likely to vastly change
our way of life. Their large-scale development will be organised around a new long
economic cycle.

1.2 The second “law” concerns social movements, revolutions and wars:
their frequency increases at cycle extremities
Wars can sometimes be stimulating for economies, when they are not being fought
within the country’s borders, because preparing for war gives a boost to industry.
No matter what the circumstances, however, they distort the economy and it
takes a post-war recession to purge the resulting artificial effects. This shows that
reversing a decades-long established order and direction calls for one or more
powerful catalysts, takes time and does not happen without tension.
Conflicts at the peak of long cycles lengthen expansion phases and are hegemonic
in nature. Examples include:
t the Napoleonic Wars in 1803-1815 (British continental embargo starting in
1806), the War of 1812,

Amundi Investment Strategy Collected Research Papers 25


t the Crimean War (1854-1856), the US Civil War (1861-1865), the Franco-
Mexican War (1864), the Battle of Könnigrätz during the Austria-Prussian
War (1866), the Franco-Prussian War (1870),
tthe Balkan Wars (1912-1913), World War I (1914-1918),
t the Vietnam War (1964-1975), the May 1968 civil movements in France.
During cycle troughs, conflicts revolve more around government refocusing efforts
and tend to be much more religious or ethnic in nature. Examples include:
tthe American Revolutionary War (1776-1781), the French Revolution (1789),
tthe widespread revolutionary movement in Europe (1848),
t the assassination of progressive Tsar Alexander II (1881), the Sino-Japanese War
(1895), the Greco-Turkish War over Crete (1897), the Spanish-American War (1898),
the Boer Wars (1880-1881 and 1899-1902), the Russo-Japanese War (1904-1905),
tthe Spanish Civil War, World War II (1939-1945),
t the War against Al-Qaeda (Afghanistan, Iraq) beginning on 9/11/2011,
the conflicts following the implosion of the USSR (Chechnya 1999-2000,
Georgia 2008, Belarus 2009-2010, Ukraine 2014-2015), Arab Spring in 2011
(Tunisia, Egypt, Libya, Yemen), Civil War in Syria since 2011.

1.3 The third “law” refers to the overlapping of long and short cycles
During the upward phase of a long cycle, short cycles have a longer expansion
phase then the subsequent correction. The opposite takes place during the
downward phase of a long cycle.
Each long cycle contains several short cycles, to which it gives a logical order. This
is not a deterministic statement, but rather the observation that each cycle has an
influence on the next. On the one hand, major recessions are ultimately regenerating.
On the other, growth cannot be exponential. It is also clear that trend reversals in
long cycles are naturally more complicated because the logic has to be reversed.
Long cycles can therefore be broken down into four phases, which will be the focus
of the next several pages: an upward phase, a more complex downturn phase,
then a downward phase and finally a more complex upturn phase.

1.4 The fourth “law” relates to deflation of agricultural prices during


downward phases
Agricultural depressions amplify industrial depressions. This statement might seem
out-of-date in today’s developed economies. However, it is important to consider that
in developing economies such as India for example, the majority of the population
still gains its livelihood from farming; inflation continues to be highly impacted by
agricultural prices in these countries, and governments incorporate them in their
economic policies if only to curb civil unrest.

26 Amundi Investment Strategy Collected Research Papers


II. Breakdown of major cycles into four phases: the
example of the United States
Many observers have taken up the baton from Kondratiev, drawing on his work,
some of which was published in German in 1926. In 1939, Austrian economist
Joseph Schumpeter who became a refugee in the United States, published
Business Cycles, an extremely comprehensive analysis of economic cycles from
the Industrial Revolution onward. After the 1973 crisis proved that the major
cycles discovered by Kondratiev and revealed by Schumpeter in the early 20th
century were also a reality for this century, other observers grew more closely
interested in long cycles. As mentioned above, we will break down these major
cycles into four phases each lasting ten to twenty years, which we will call
seasons: spring, summer, autumn and winter. Each “season” is also made up of
several short cycles consisting of alternating periods of growth and slowdowns
(if not recessions).

2.1 Long cycle observations based on financial data


Evidence of this breakdown can be clearly seen in the US market, where long rates
- which reflect inflation - reach a peak during each major cycle. They rise in the first
two phases and fall in the next two phases. The equities market alternates between
rallies and downturns. It climbs in the first phase, declines in the second, and so
forth (see diagram below).

1 - The principle

SPRING SUMMER AUTUMN WINTER

Interest
Rates

Equities

Source: Amundi Research

Share prices are presented in six different ways in the following charts p 12 and
p 13: nominal value, deflated by consumer prices, government bond prices, real
estate prices, gold prices and industrial commodity prices. The transition points
from one cycle to another are very consistent. Each approach sheds slightly
different light that helps validate the change in “season” when the time comes.

Amundi Investment Strategy Collected Research Papers 27


In terms of nominal values, for example, US equities are currently once again
at peak levels, which suggests that the spring phase has already begun in the
United States. But, once deflated by bond or consumer prices, this turns out
not be the case. Finally, deflated by real estate prices, gold prices or industrial
commodity prices (in other words by the prices of real assets), there is even
more ground to cover to top the record highs, and may call for several waves of
ups and downs to get there.
The ratio of industrial commodity prices to bond prices (see chart 7) is particularly
instructive. In winter, for example, bonds (reverse of bond yield) outperform
industrial commodities (the ratio increases), which in turn outperform equities
(excluding dividends) or at the very least perform in line with them. In spring,
however, equities outperform commodities, which in turn outperform bonds
(the ratio falls), and that is still not the case today. According to this approach,
and despite appearances as reflected in nominal values, the US market is still
apparently in the winter phase.
Industrial commodities are subject to the law of supply and demand. As such,
data on industrial commodities provides structural support to this approach
because their development is linked to economic growth more directly than
equities. As for government bonds, they primarily reflect inflation expectations.
Finally, since we can consider dividend yields on equities and rental yields on
real estate to be equivalent over the long term, the ratio of equities to real estate
(excluding dividends and rents) also has a lot to teach us (see chart 5 ).
This ratio delineates the economic seasons remarkably well. The bottom line
is, if we had to choose between these two asset classes, which outperform
over the very long term, real estate tends to be the best choice in summer
and winter (despite the significant price discrepancies observed in these two
seasons) and equities in spring and autumn.
On this subject, Reinhart and Rogoff noted that during banking crises, while
real estate prices decline by an average of -35% over six years (all countries
combined), equities shed an average of -56% over three and a half years. Price
declines are therefore steeper in equities and more long-term in real estate, but
even so this long term is much shorter than the average length of an economic
season (17 years).
In the case of a rental investment, rent naturally has to be taken into account;
the return partially offsets the annual drop in prices. Another factor to consider
is the borrowing rate. Winter and summer phases are conducive to lower - if
not negative – real interest rates (closer to early summer and late winter), which
improves the average rate of return on the deal. But investors must be ready to
hold their positions once the deal has been made because they are less liquid
during these tricky seasons.

28 Amundi Investment Strategy Collected Research Papers


US equity markets and long rates
- Equities and ond ields
I II III IV
3125 16
(100:1978, log)

14
625
12

ond ields
125 10
8
25 6
S P

4
5
2
1 0
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

- Equities onsu er Price Index and ond ields


I II III IV
1250 16
real (100:1900, log)

14
12
250

ond ields
10
8
6
50
4
2
S P

10 0
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

- Equities onds RI and ond ields


I II III IV 16
3125
onds (log)

14
625 12

ond ields
10
125
8
25 6
S P

4
5
2
1 0
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

Source: Shiller, Global Financial Data, Kondratieff, Amundi Research

These three charts show trends in long rates and equities during long cycles.
Equities are presented in three different ways: nominal value (chart 2), deflated by
consumer prices (chart 3) and by government bond prices (chart 4). Long rates
rise during the first two phases (spring and summer) of each long cycle (labelled
I, II, III and IV), then decline during the next two phases (autumn and winter).
Equities alternate between rallies and downturns. They rise in spring and autumn
and fall in summer and winter.

Amundi Investment Strategy Collected Research Papers 29


Equity markets and US long rates
- Equities Real Estate and ond ields
I II III IV 16
16
Real Estate

14
12
8

ond ields
10
(log)

8
4
6
2 4
S P

2
1 0
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

- Equities Gold and ond ields


I II III IV 16
4.0 14
12
Gold

ond ields
(in US$, log)

1.0 10
8
6
S P

0.3
4
2
0.1 0
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

- Equities Industrial co odities and Industrial co odities onds


I II III IV

odities
6.3 64
Industrial
odities (log)

32
onds (log)
1.3
16
Industrial co

0.3 8

4
0.1
S P
co

0.0 1
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

Source: Shiller, GFD, Kondratieff, R , Amundi Research

In this series of three charts, equities are deflated by real asset prices, by real estate
prices (chart 5), gold prices in USD (chart 6), then industrial commodity prices
(chart 7). The last chart also shows industrial commodity prices compared to
government bond prices. What we learn is that particularly in winter, commodities
outperform equities (excluding dividends) and that equities come out on top again
in spring. Similarly, commodities underperform bonds in winter and outperform
in spring.

30 Amundi Investment Strategy Collected Research Papers


To sum up our observation on financial data, the following charts show the average
annual performance of assets corrected for inflation, this time in total return, broken
down by economic season over the four long cycles, then the last two on which more
data are available. We will comment on these performances in the next section.

Real returns on the main asset classes during the four seasons
of long cycles for investors in USD
8- Real returns during cycles I, II III, IV (yearly averages)

Spring Summer Autumn Winter


20%
15%
10%
5%
0%
-5%
Equities Government Bonds Gold

- Real returns durin c cles III IV earl a era es

Spring
Printemps
Spring Summer
Summer
Été Autumn
Automne
Autumn Winter
Winter
Hiver
20%
15%
10%
5%
0%
-5%
Actions
Equities
Equities Oblig.
Corp. d'entreprises
Corp. Bonds
Bonds Gov. Oblig. d'Etat
Gov.Bonds
Bonds Money Monétaire
MoneyMarket
Market GoldOr
Gold Immobilier
Real
Real Estate
Estate

Source: Shiller, GFD, Kondratieff, omer S lla, ood s, Fed, Datastream, Amundi Research

The colours represent economic seasons. Returns are shown corrected for
inflation and as “total return” (dividends reinvested). Government and corporate
bonds are deemed to have a constant duration (7 and 5, respectively). Real estate
returns are deemed stable and equivalent to two-thirds of returns on equities
(expenses are considered to account for one-third of rent) and reinvested. Using
only Cycles III and IV (chart 9) gives more assets for our comparison.

2.2. These “seasons” also reflect an economic reality


This economic reality was notably underscored by the Canadian Ian Gordon, one
of the world’s leading references in the interpretation of long cycles.

Amundi Investment Strategy Collected Research Papers 31


Economic spring (1842-1852 / 1896-1907 / 1949-1966) is a period of sustainable
growth without inflationary excess.
The economy’s productive capacity is obsolete and requires new investments.
Interest rates and wages are at rock bottom. Surviving banks have sufficient
liquidity reserves. Economic sentiment starts very low and improves in pace with
reconstruction and job restoration. The major innovations developed since the
end of economic autumn and during the preceding winter are adapted through
other more secondary innovations aimed at meeting increasingly intense demand.
Real profits start climbing again and a virtuous circle is created. From a value
standpoint, realism, progress and liberalism take precedence. Former colonies
become independent. A case in point is Australia in 1901, India, Pakistan in 1947
and a large part of Africa in the 1950s and 1960s.
This period is especially supportive for equities. Real estate is resilient but is still
outdistanced. Bonds suffer due to higher interest rates and relatively low returns.
Money market investments hold up better but are still not very attractive. Gold
declines systematically.
Economic summer (1852-1866 / 1907-1920 / 1966-1982) is a period of excess
that marks the high point of inflation.
Productive capacities have been built back up. Though operating at full force, they
are not sufficient and wages rise faster than productivity gains. The virtuous circle
becomes a vicious price-wage circle. Company profits are hurt by rising wages,
commodity prices and cost of debt. Steep price rises are the main economic feature of
this season, which ends with a harsh recession that breaks the vicious circle (recession
from July 1981 to November 1982 during the last major cycle in the United States).
This period is unsupportive for equities, whose apparent performance is wiped out
by runaway inflation. Bonds also slide as interest rates climb. On the other hand,
money market investments are increasingly attractive. Central banks are inclined to
raise short rates to very high levels, above the level of inflation, and financial stability
is jeopardised as a result. Gold comes to the rescue as a safe haven investment,
alongside real estate. This period is potentially beneficial for inflation-linked bonds.
Economic autumn (1814-1835 / 1866-1873 / 1920- 1929 / 1982-2000) benefits
from a vast disinflation trend and the ongoing rise in debt. It ends with a major
international crash.
In the wake of the major recession that succeeded in breaking the inflationary spiral
at summer’s end, the central bank cranks up the cash-printing machine. This cash is
not invested in the production capacities that have grown sluggish due to the wave
of over-investment in recent years, particularly given how high interest rates have
climbed. On the plus side, households and companies are building their previously
decimated savings back up. Employment, plus wage and price rises, gradually begin
to decline. Increased asset prices create the illusion of a new world with people
convinced that “this time it’s different!”. The term “new paradigm” is bantered about.

32 Amundi Investment Strategy Collected Research Papers


This is the “new economy” of the late 1990s which, looking back, strangely resembles
the “Roaring 20s.” Domestic values make a comeback (Reagan, Thatcher).
This period is most supportive for equities and bonds thanks to lower interest rates.
Slightly riskier corporate bonds (rated BAA for example) do better than government
bonds. The money market brings up the rear behind fixed income products. Real
estate barely does any better than the money market and generally underperforms
financial investments. This period is especially unsupportive for commodities in
general and gold in particular. After a market euphoria phase, however, autumn
ends with a major crash that nothing appears capable of stopping.
Economic winter (1835-1842 / 1873-1896 / 1929-1949 / 2000- ?) is marked
by a major low point in inflation and a debt purge.
This new period is when the record is set straight. It is the period of creative
destruction, when debts are liquidated, competition is extremely intense and
protectionism is the order of the day. Competitive devaluations get in the way of
international trade, thus hampering growth. Investments are focused on streamlining
and companies merger in order to fight the competition. Companies on the fringe
simply fail. This period favours the economic emergence of new countries or
propels them to the international stage. Examples include the take-off of the US,
Germany and Japan from the 1870s and that of China in the 2000s.
Gold, the ultimate safe haven asset, once again outshines equities. Government
bonds fare well, especially relative to risk (volatility). Investment grade, and even
medium-grade (BAA according to Moody’s) corporate bonds beat government
bonds and especially the money market. Higher-risk bonds (speculative grade)
go through an intense stress phase, with default rates nearing 11% in 1932 for
example. As a result of deflation, money market investments temporarily become
attractive again, provided they are investments in quality vehicles that are not
susceptible to bank failures. Later in the period, however, short rates are taken to
a record low point - where they stay - causing them to lose their appeal. Though
weakened by a crash, residential real estate ends up turning around, generating
respectable average returns over the period.

2.3 Clarifications on real interest rates, inflation, debt and equity


valuations
We put forward the idea that although the US equities market is currently at a
record high in nominal value terms, it is likely that the winter we just described is
not yet over. So let’s take a closer look at other key parameters in the transition
from winter to spring, using the series of graphs p 18.
tLong rates and inflation
The chart 10 shows long rates and average consumer prices over five years, which
mirror inflation expectations. What we see is that the transition from winter to spring

Amundi Investment Strategy Collected Research Papers 33


coincides with the upturn in inflation during the transition from Cycles I to II and II to
III. The turnaround in inflation expectations tends to stimulate activity; it is no longer
justified to wait for prices to fall in order to consume. In 1896, Charles Gide, a highly
renowned French economist during the economy’s “long stagnation” period at the
time, called for the return of inflation because “price rises are a sort of economic
spring.”
During the transition to Cycle IV, the same mechanism was in place in 1942 before
World War II triggered very strong inflation and delayed the arrival of spring.
Excluding this exceptional period, we see that phases of very strong inflation take
place in summer, as they did in all four major cycles described above.
If World War II generated exceptional inflation in the price of goods at the end of
Cycle III winter, we can consider that the action taken by the Fed - an institution that
has matured since World War II - is encouraging exceptional inflation in the price
of assets this time around, mainly the price of fixed income investments but also
equities in nominal value terms, at the end of Cycle IV winter.
tReal interest rates and government debt
The second chart (11) shows real long rates and government debt relative to GDP.
In the three previous major cycles, government debt relative to GDP peaked during
the transition to spring, an extended economic growth phase. There is also a
relatively strong link between trends in public debt and trends in real interest rates,
which in turn are closely linked to changes in inflation. Before the 1950s, inflation
was indeed much more volatile than interest rates (see also previous chart).
This time around, real interest rates have fallen but skyrocketing price inflation is
unlikely in the short-term; the level of real interest rates will therefore not fall as
low as it has in the past. We can consider, however, that the Fed’s QE programme
served in some way to prolong the interest rate decline. Meanwhile, government
debt increased this winter (Cycle IV) just as it did in the 1930s. Right now a drop
in public sector debt relative to GDP would be an encouraging sign that spring is
on its way.
tGovernment debt and private sector debt
The third graph (12) refers to total United States debt, which can be broken
down into public and private sector debt over a shorter period (though it does
go back as far as 1916). It is clear that the total debt level has gone up a notch
in Cycle IV compared to Cycle III, as a direct result of the financialisation of the
economy. Another distinction is the back-and-forth between public and private
sector debt. To get out of the Great Depression in the 1930s, the State raised its
debt level on the heels of the private sector, which began deleveraging as soon
as the crisis began. The same thing also happened in Cycle IV starting in 2007.
An increase in private sector debt relative to GDP would indicate that savings are
being diverted from deleveraging to investment, which would also be a positive sign.

34 Amundi Investment Strategy Collected Research Papers


Real interest rates, inflation and debt
1 - In lation and ond ields
I II III IV
16 16

12 12
(5 year average, in %)

8 8

ond ield
PI

4 4

0 0

-4 -4

-8 -8
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

11- Pu lic e t and ond ields

140
I II III IV 15
120 10
Pu lic e t

100
(% of GDP)

Real ond ields


5
80
0
60
-5
40
20 -10

0 -15
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

1 - Total ars Pu lic and Pri ate li ns e t

400
I II III IV

Pu lic and pri ate


e t (ligns, i% of GDP)
350
300
(bar chart, % of GDP)
Total e t

250
200
150
100
50
0
1800 1814 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

Source: Shiller, GFD, Fed, S ureau of ensus, us o ernmentdebt us, Amundi Research

These charts show long rates and inflation (average consumer prices over 5 years, a
proxy for inflation expectations), real interest rates and public sector debt relative to
GDP and, finally, public and private sector debt. We can see that the transitions from
winter to spring coincide with an upward reversal of inflation expectations during the
first two cycles. World War II changed the game completely, triggering exceptional
consumer price inflation at the end of Cycle III. This time around (end of Cycle IV), the
Fed may have triggered exceptional asset price inflation. Furthermore, the transition
for spring coincides with a long-term downward reversal of public sector debt relative
to GDP with private sector debt taking up the baton.

Amundi Investment Strategy Collected Research Papers 35


t Equities market, profits and cycle-adjusted valuation
The equities market (Chart 13) is shown here based on real data (deflated by
consumer prices) with dividends reinvested. It mainly tends to rally in spring and
autumn, and consolidates its gains in summer and winter. Dividend reinvestment
evens out the market’s performance over time, especially in challenging periods.
Buying shares when they’re falling brings down the average buy price. At the end
of the 19 th century, dividend reinvestment even got equities to beat inflation over
the winter. With dividends reinvested, the transition from winter to spring went off
without a hitch...something to bear in mind when it comes to the US example in
the current period.
Corporate profits (chart 14) alternate between the same upward and downward
phases, following short cycles within the long cycles. The extremity of cycle-
adjusted PER trends (chart 15) coincides fairly well with seasonal changes in the
long cycle. Accordingly, it is a good idea to start being more cautious when the
PER climbs above 20x in spring or autumn. On the opposite side of the coin, when
it dips below 10x in summer or winter, it would be a great time to buy for the long
term.
It’s worth noting that the level of cycle-adjusted PER is currently very high for a
winter period and may appear inflated, especially given that corporate profits are
also at a historically high level. Today’s solid corporate profitability has more to do
with strict cost management than it does with leveraging or investment, which is
typical of winter periods.
This would tend to indicate that the Fed’s QE programme got the US equities
market looking ahead to spring somewhat early in terms of performance and thus
making a start on its potential. Either this potential will regenerate during the next
recession, which could involve a fairly strong market correction, or on the contrary
the transition to spring will go smoothly as it did at the end of the 19th century,
though with a slightly less generous spring than average as far as equities are
concerned.
t Additional equity valuation measures
There are other valuation measures that refer to economic activity and provide
the same type of information as cycle-adjusted PER. One is Tobin’s Q (Chart 16),
which is the ratio between a physical asset’s market value to its replacement value.
Another is the market cap of non-financial companies relative to GDP (Chart 17).
The risk premium (calculated here as [1/PER - (10y government bond yield -
average inflation over 5 years)]) is the only valuation metric that remains favourable
in the example of the US equities market. The lower the risk premium, the more
expensive the market compared to interest rate levels. Conversely, the higher the
risk premium, the more attractive the market compared to interest rate levels.
Currently close to its historic average, it does not show any overvaluation relative
to today’s record-low interest rates.

36 Amundi Investment Strategy Collected Research Papers


Equities market, profits and valuation

13- S&P500 Index in total real return

100000
II III IV

10000
(log scale)

1000

100
S P

10
average
1
1881 1896 1907 1920 1929 1949 1966 1981 2000

14- S&P500 Real Profits


II III IV
96
(log sccale)

48

24
Pro its

12

6
average
3
1881 1896 1907 1920 1929 1949 1966 1981 2000

15- S&P500 Cyclically Adjusted PER

80
II III IV
clicall Ad PER

40

20

10

5
--- average + ou - 1 standard deviation
3
1881 1896 1907 1920 1929 1949 1966 1981 2000

Source: Shiller, Amundi Research

The trend in the equities market performance based on real data with dividends
reinvested is around +6.5% per year, with weaker phases in summer and winter
and stronger ones in spring and autumn. Dividend reinvestment evens out this
performance. Profits mainly climb in spring and autumn, following short cycles
within the long cycles. Cycle-adjusted PER coincides fairly well with seasonal
turnarounds.

Amundi Investment Strategy Collected Research Papers 37


Equities market: other valuation approaches
1 - To in's
III IV
1.4
1.2
1.0
To in's

0.8
0.6
0.4
0.2
--- average + ou - 1 standard deviation
0.0
1896 07 20 29 1949 66 81 2000

1 - Mar et apitalisation ex- inancials No inal G P


III IV
1.8
G P

1.6
1.4
1.2
Mar et ap

1.0
0.8
0.6
0.4
0.2 --- average + ou - 1 standard deviation
0.0
1896 07 20 29 1949 66 81 2000

1 - Equit Ris Pre iu


III IV
25
20
Equit Ris Pre iu

15
10
5
0
-5
--- average + ou - 1 standard deviation
-10
1896 07 20 29 1949 66 81 2000

Source: Shiller, S , Fed, Datastream, Amundi Research

Tobin’s Q and market cap relative to GDP are at high levels, as is the cycle-
adjusted PER. Only the risk premium is currently close to its historic average.
This state of affairs can be attributed to the absence of inflation, low interest
rates and the Fed’s credibility.

38 Amundi Investment Strategy Collected Research Papers


III. Major international cycles
Our objective is to identify a replicable case that can serve as a guide to interpreting
the markets. In this respect, the US market has been an exemplary case up to
this point, but it is important to point out that this approach is not exclusive to the
United States. In fact, we can observe it in many other countries as well.

3.1 We reviewed ten countries presenting basically very different


features
Three major countries that were the first to industrialise:

Longest track record. Leading country in the first two industrialisation cycles.
UK
Colonial empire. Winner of two World Wars that did not take place within its borders.

Track record for best quality, most documented. Leading country in the last two
US long cycles. Winner of two World Wars that did not take place within its borders.
Imperial power.

On the winning side of two World Wars, but which took place within its borders.
France France has long been the most highly populated country in Europe. It suffered
massive population losses in both World Wars. Colonial empire.

Three countries that arrived on the international stage during the economic winter
at the end of the 19th century:

Truly unified in 1871 after its victory over France. Lost two World Wars with massive
Germany
destruction of property. Hyperinflation. Cut off from its colonies.

Reign of Emperor Meiji and industrialisation both began in 1867. Winner in World
Japan War I and loser in World War II. Supported the United States in the Vietnam War
(1964-1975).

Unified in 1861. On the winning side of World War I and the losing side of World
Italy
War II.

Two former British colonies that have long been part of the Pound Sterling
area:

Canada Former British colony. Close to the United States. Commodity-producing country.

Former British colony. Gold discovered in 1851. Commodity-producing country.


Australia
Supported the United States in the Vietnam War. Geographically close to China.

Amundi Investment Strategy Collected Research Papers 39


Two Nordic countries:

European oil-producing country. United with Sweden from 1814 to 1905. Neutral
Norway
during World War I. Occupied by the Germans during World War II.

Sweden Neutral during both World Wars.

As for the United States, there are periods that are repeated in many other
countries, despite a number of distinctive features. Europe, for example, was
beset by hyperinflation, a plague that hit Germany hardest. The collapse of the
Japanese equities market in the wake of Hiroshima and Nagasaki is also a prime
example of a rare phenomenon. Australia is another special case, one that is able
to come out on top in commodity-supportive periods (summer and winter).

Equity markets and long-term interest rates for a series of countries

1 - United in do
I II III IV 16
3200
14
Real Equit Index

12
(100:1783, log)

800

ond ields
10
8
200 6
4
50 2
1783 93 1815 25 1842 52 63 73 1893 12 20 29 1952 68 81 2000

Source: Fisher, Datastream, Amundi Research

- rance
II III IV 18
800 16
Real Equit Index

14
ond ields
(100:1848, log)

400 12
10
200 8
6
100 4
2
50 0
1840 51 63 71 81 1896 12 26 29 1950 62 81 2000

Source: Fri it, Datastream, Amundi Research

40 Amundi Investment Strategy Collected Research Papers


Equity markets and long-term interest rates for a series of countries
1- apan

625
III IV 12

10
125
Real Equit Index

ond ields
(100:1915, log)

25 6

4
5
2

1 0
1915 1924 1937 1947 1972 1981 1989

- anada
III IV 16
800
14
Real Equit Index

12
400

ond ields
(100:1913, log)

10
8
200
6
100 4
2
50 0
1913 1922 1929 1944 1963 1981 2007

- Australia
II III IV
18
625 16
Real Equit Index

PI and ond ields


14
125 12
(100:1872, log)

10
25 8
6
5 4
2
1 0
1850 1858 1873 1893 1912 1920 1937 1952 1968 1982 2007

- Ital

125
III IV 22
20
Real Equit Index

18
(100:1906, log)

25
ond ields

16
14
5 12
10
8
1
6
4
0.2 2
1929 1949 1961 1981 2000

Source: Australian Stoc chan e, GFD, omer S lla, Datastream, Amundi Research

Amundi Investment Strategy Collected Research Papers 41


Equity markets and long-term interest rates for a series of countries
- Nor e en
III IV
15
13
270
Real Equit Index

11

ond ields
(100:1920, log)

90 9
7

30 5
3
10 1
1900 1917 1921 1937 1953 1961 1982 2007

- S eden
III IV 16
14
270
Real Equit Index

12
(100:1906, log)

10

ond ields
90 8
6
30 4
2
10 0
1887 1913 1921 1929 1949 1965 1981 2000

- Ger an e ore 1 1

150
II III 7
Real Equit Index

120
6

ond ields
(100:1856)

90
5
60

4
30

0 3
1856 1867 1873 1893 1905 1921

- Ger an a ter 1
III IV
4
10
Real Equit Index

3
ond ields

8
(100:1856)

6
2
4
1
2

0 0
1924 1929 1949 1961 1981 2000

Source: or es an , Ri sban , GFD, omer S lla, Datastream, Amundi Research

42 Amundi Investment Strategy Collected Research Papers


Even though all these countries are located in different geographic areas, we
can see the same logical sequence each time (see chart p.29), which enhances
our confidence in this line of reasoning. Let’s start with how well the dates of
these cycles coincide increasingly from one country to the next and end up
converging toward the pertinent dates for the US, the leading country in the last
two long cycles. It is true that at the end of World War II, the Marshall Plan was
accompanied by a commitment that the beneficiary States would gear themselves
up for the market economy.
In fact, the innovations referred to earlier in this paper made possible the advances
from cycle to cycle in international communication and trade, each time reducing
the distances and time involved until the planet became one big global internet
village. This of course underscores that long capitalism cycles are indeed an
international phenomenon, as Kondratiev observed in his own time.

3.2 Distinctive behavioural features of these countries


Cycle I
The first long cycle began after the Treaty of Paris on United States independence
was signed in 1783. The United Kingdom focused its attention on its other
colonies, starting with India. Spring lasted until France’s campaign against Great
Britain and the Netherlands was declared during the French Revolutionary Wars.
Summer drew to a close with France’s defeat at Waterloo in 1815. Latin American
countries gained their independence, launching a tidal wave of investment flows
into these countries and triggering a huge crash in London in 1825. This brought
economic winter to Britain, but it wasn’t until the post-boom recession of 1835
that the United States was plunged into winter as well.
Cycle II
The second cycle began simultaneously in the UK and the US in 1842. The Entente
Cordiale (1843-1848) between France and the UK symbolised the international
détente and the transition from protectionism to free trade. France wouldn’t have
its own spring until a little later, however, after the Revolution of 1848 and the
proclamation of the Second Empire following the coup d’état of 2 December 1851
by the man who would become Napoleon III. With the discovery of gold in California
in 1847 and Australia in 1851, the expansion became a tad more inflationary. The
Crimean War (1853-1856) triggered economic summer in the UK and consequently
in the US. France trailed slightly behind, with summer arriving only in 1863. The
US Civil War created a shortage of cotton that same year, causing textile prices to
climb and plunging the entire textile industry into crisis throughout Europe.
While the US and UK enjoyed the clemency of autumn at the end of the US
Civil War in 1865 and after the Panic of 1866 (sparked by the failure of railway
companies and that of Overend, Guerney and Company in England), France had
to wait until the end of the Franco-Prussian War in 1870.

Amundi Investment Strategy Collected Research Papers 43


- The our lon c cles cles I II III and IV

RU
1
1
Supercycle I
USA 1 1 Coal
1 1 Steam engines
1 1

1 1 RA
1 1
1 1 1 Supercycle II

1 1 1 1 GER Steel
1 Railways
1 1 1 1 1 1

AUS 1 1 1 1
1
1 1 1 1 1 1 1 1 SWE NOR
Supercycle III
1 1 1 1
AN 1 1 1 1 1 1 1 1 1 AP Electricity
Automobile
1 1
1 1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1 1 ITA
1
1 1 1 1 1 1 1 1 1 1 1 1 1 Supercycle IV

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Oil
Electronic

Spring Summer Autumn Winter

Source: Amundi Research

Note: The colours represent economic seasons.


The same trends can be observed in very different countries. The dates of
the economic seasons coincide increasingly from one country to another and
gradually converge toward the pertinent dates for the leading country: the United
States. Innovations make advances possible in international communication
and trade from cycle to cycle.

The war indemnity paid by France from 1871 to 1873 enabled the now-incontestable
German empire to develop its industry so fast that it ended up causing a major

44 Amundi Investment Strategy Collected Research Papers


crash, marking the beginning of winter not only for Germany but for the UK and
US as well. France was impacted, but experienced a solo growth phase from 1878
to 1882 driven by a large-scale public works policy (Freycinet Plan of 1878 and
Railway Act of 1880), which in turn led to a major crisis that spread around the
world in 1882 (triggered by the collapse of Union Générale). Thus France joined its
fellow countries in winter.
Cycle III
The third cycle started in 1893 or 1896 depending on the country. Summer began
in the US with the 1907 Bank Panic, which hit Germany harder than the rest of
Europe, as Germany had been in a weakened state since 1905 and had already
raised its discount rate. Summer started for France, the UK and Australia (very
closely linked to the UK) in the wake of the German Steel Crisis of 1913. Summer
lasted until the end of the recession after World War I.
Most countries - first the Anglo-Saxon countries (UK, US, Australia, Canada), then
Italy and the Nordic countries - headed into autumn at the very beginning of the
1920s. Labouring under the weight of enormous war reparations imposed by the
Treaty of Versailles, Germany was assailed with now-legendary hyperinflation.
It finally abated in 1923, taking Germany into autumn. Finally, unlike the Anglo-
Saxon countries, France decided to raise rates after the war to combat the inflation
generated by the removal of exchange rate and price controls, dragging it in to
recession. It lifted itself out of this recession by significantly devaluing its currency
in 1925, marking the belated beginning of autumn in 1926. Winter came on the
heels of the 1929 Stock Market Crash, everywhere except Japan, Norway and
Australia. These three countries did not manage to escape the US recession in
1937, however, which became another textbook case, with the US authorities
prematurely withdrawing the massive support they had set up to combat the crisis.
Cycle IV
The fourth cycle began after a traditional post-war recession in the late 1940s/
early 1950s. Reconstruction paved the way for robust growth, later leading to the
monetary crises of 1961, 1963 and 1965, then the devaluation of the Pound Sterling
in 1967 which sent oil prices skyrocketing in 1968, and wrapping up with President
Nixon ending the gold standard on 15 August 1971. Summer began for all countries
during one or another of these crises. They came out of it together in 1981 on the
coattails of the US, which now held the world’s de facto reserve currency.
Autumn lasted from 1982 to 2000 and once again ended with a sweeping
international crash: the bursting of the Internet Bubble. This took us into winter,
where we’ve been since the year 2000. There are four exceptions to this rule:
Japan, where winter took hold with the collapse of the real estate bubble in
1989, and commodity-producing countries Australia, Canada and Norway, which
benefited from the emergence of the Chinese market and thus staved off winter
until 2007.

Amundi Investment Strategy Collected Research Papers 45


General comments
This brief historical overview clearly shows that the same pattern has repeated
in several countries despite their very strong distinctive features, giving us an
opportunity to point out some of the specific characteristics from one cycle to
another. For example, inflation hit record levels in the 20th century compared to the
19th century but the general trend was the same. It was the policies implemented
to stabilise the economy that were adapted.
Until World War I, exchange rates were the main focus when it came to stabilising the
economy. When a trade deficit got too big, often because agricultural commodities
had to be imported due to poor harvests, interest rates rose to repatriate gold,
which had the potential to generate crises, but the price decline ended up restoring
household purchasing power. As a result, the 19th century was generally a time of
low inflation.
With the advent of World War II, controlling inflation became priority number one in
order to stabilise the economy. Industrial and social progress gradually put power
in the hands of employees. The first strikes took place in the 1830s, which inspired
the 1848 social movements. The 1873 crisis was due in part to excessively high
wage rises. Thanks to Fordism, wages were raised to allow employees to buy
cars starting in the early 20th century, which became the era of hyperinflation (in
Germany in the early 1920s, in France during World War II, in the rest of the world
due to the oil shocks in the 1970s).
Ever since then, wages may not be indexed to prices. Inflation will return, but not
necessarily in the form of hyperinflation, which may remain unique to the 20th
century.
Another distinction is that the structural increase in debt during the 20th century
boosted potential growth. Such a phenomenon has a limit that can be reached, of
course. The cyclical mechanism of debt and deleveraging, however, is sustainable.

IV. The demographic factor plays a key role


Kondratiev made no reference to demographic factors, despite the fact that they
have become critical to understanding the developments of the 20th century.
These factors include increased life expectancy, the post-war baby boom and
the ageing of the population. In this respect, our research has shown however
that demographic factors only reinforce the consistency of the long cycles we
have described and provide some intriguing leads to pursue in order to improve
forecasts.
We know that an economy’s long-term growth potential is based in large part on
population growth and the productivity gains it generates. So let’s explore these
two factors with a brief review of population trends in the principal developed
countries.

46 Amundi Investment Strategy Collected Research Papers


4.1 Population trends in the principal developed countries
The following chart uses Angus Maddison statistics to demonstrate the
extraordinary demographic momentum of the United States and its capacity to
attract new populations for the past two centuries. Another significant event was
the German population’s growth spurt from the second half of the 19th century
to World War I, with Germans outnumbering the French just before the War of
1870. France, which has long boasted the largest population in Europe, saw its
birth rate drop unrelentingly from 1867 to the 1950s. It should also be noted that
population losses were much greater in France and Germany than anywhere
else during both World Wars. As for Japan, its population exploded from the
early 20th century until 1990, with its demographic decline coinciding with the
beginning of economic winter.

- E olution o the population o ain de elopped countries

320
I II III IV USA
illions

160
apan
Ger an
80
(log scale)

rance
Population in

U
40

20

10
1820 35 1842 52 66 73 1896 07 20 29 1949 66 81 2000

Source: An us addisson, Amundi Research


The different colours represent the seasons of the US cycle.

4.2 Growth in the economically active population


While the size of a population gives a general idea of a country’s economic and
political weight, growth in the economically active population has the biggest
impact on the economy’s potential growth. The annual growth of the global
economically active population (age bracket from 15 to 64) went from +1.9% ten
years ago to +1.6% over the last five years according to UN data (2012 revised
data) and will continue to lose steam to settle at +1.1% by 2025. Developed
countries (currently stable on average) will venture into negative territory. Less
developed countries (“emerging countries”) are also on the decline but have
maintained growth of +1.5%. Finally, the most economically underdeveloped
countries are the only ones that will see their economically active population
expand at an impressive rate of over +3%.
We must not draw any hasty conclusions, however. The economically active
population is a very broad concept in that it encompasses several generations,

Amundi Investment Strategy Collected Research Papers 47


each of which has its own characteristics. On average, young adults enter the
workforce, get married and have kids that grow up and leave the nest. Their
housing needs, for example (first home, then expansion of the main residence or
subsequent acquisition of a second home), evolve with the size of their family and
income. The same is true for the size and number of cars owned by a couple,
etc. Accordingly, the economically active population must be subject to a more
detailed analysis.

1- Acti e population ro th in

Least developed countries

Less developed regions


1 -
More developed regions - 1

WORLD

-0.5% 0.0% 0.5% 1.0% 1.5% 2.0% 2.5% 3.0% 3.5%

Source: nited ations the re ision , Amundi Research

4.3 Generational impact on the main asset classes


Economically active young adults (aged 20-35) have the greatest long-term
influence on real estate prices; first-time homebuyers are marginal buyers on
this market. Chart 32 clearly shows that the value of real estate assets relative
to financial assets closely follows the proportion of economically active young
adults in the adult population. Of course, this population wave will grow older to
become an intermediate generation (aged 35-44) before hitting the peak of its
consumption around age 46. It’s easy to understand why the real estate cycle
is often referred to as the “mother” of all cycles. Getting back to the chart, the
sharp rise in the relative value of real estate in the 2000s can be attributed to
the combined effect of the equity market collapse and excessive increase in
real estate prices linked to subprime loans. Everything seems to be falling into
place. Real estate has proven highly volatile during the winter (since 2000) but
ended up on an equal footing with financial assets, which is highly consistent
with the historic observations we presented earlier. According to this approach,
real estate should fare relatively poorer than financial assets as from the 2020s.
The intermediate generation (35-44s) is poised to be the next driving force of
the economy. The transition to the intermediate age bracket is when anyone’s
productivity increases the fastest. In Chart 33 we can see that the proportion of
35-44s in the adult population is very consistent with the trend in real interest

48 Amundi Investment Strategy Collected Research Papers


rates, which represents potential growth. Since by definition the generation just
after the baby-boomers was smaller, this was conducive to a drop in real interest
rates, which ended up encouraging the development of bubbles. According to
UN demographic data, real interest rates are now supposed to remain contained
for a long time, with a retracting force of just over 2%.

Impact of the proportion of different generations on real estate prices,


En lish
real interest Version
rates and equities

- Real estate inancial ealth and the proportion o irst ti e u ers


550 29%
500 27%
450 25%
400 23%
350 21%
300 19%
250 17%
200 15%
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030

US real estate / Financial assets 20-35 / 20+ (R.H.S.)

- Real ond ield and the proportion o - ear old


10 21%
8 19%
6 17%
4 15%
2 13%
0 11%
-2 9%
-4 7%
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030

Real 10Y bond yield 35-44 / 20+ (R.H.S.)

- Equit share in the inancial assets and the proprtion o - ear old
35% 41%
39%
30%
37%
25% 35%
20% 33%
31%
15%
29%
10% 27%
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030

US equity share / Financial assets 40-59 / 20+ (R.H.S.)

Source: nited ations re ision , Fed, Datastream, Amundi Research

Amundi Investment Strategy Collected Research Papers 49


Chart 32 shows that the value of real estate assets relative to financial assets
closely follows the proportion of economically active young adults in the adult
population. The proportion of 35-44s is very consistent with the trend in real
interest rates, while the proportion of 40-59s follows the same trend as the share
of equities in household financial assets. From this, we can conclude that trends
in real estate and financial assets should be fairly close going forward, that real
long rates could reach a balance around 2% in the coming years, and that a
structural turnaround in favour of equities should gain traction in the 2020s.

The previous generation (40-59s) does the most saving, so it’s not surprising to
see consistency with the proportion of equities in financial assets.
The 40-59 age bracket can be split into two: The 50-59s, who are actively saving
in preparation for retirement: individuals who continue to work in their fifties
dramatically increase their financial assets, on average. And the 40-49s, who are
at the peak in terms of consumption capacity, but also in terms of productivity: as
such, it is this generation that drives effective growth the most.
The chart 35 highlights the substantial increase in the number of Japanese 40-
49s starting in the 1960s (“Japanese Miracle”) followed by the sharp drop in
this age bracket as from the 1990s (real estate and stock market crash, start
of economic winter). Similarly in the United States, the decline in this age group
in 1966 coincided with the beginning of economic summer. Then the gradual
transition of the baby-boomers into this bracket marked the beginning of
economic autumn (as from 1981), while the peak in 2000 coincided with the dot-
com crash. The same thing happened in Europe, but further down the road, as
the European baby boom took place later and was less pronounced than in the
United States. Though it started in the US, the subprime crisis hit the euro zone
hard, just as the population of 40-49s had begun to decline.

- Proportion o the - ear old in the ore than ear old population

1
22%
Europe
21%
USA
USA
20%
Japan
19%
18%
17%
16%
15%
14%
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030

Source: re ision , Amundi Research

50 Amundi Investment Strategy Collected Research Papers


If the Japanese curve is completely offset from the US curve, it’s because the number
of births picked up more in the 1930s (as it did in Germany by the way) and not so much
after the war it lost. The Japanese Economic Miracle of the 1960s (economic spring)
nevertheless paved the way for a jump in the birth rate until 1973 (oil crisis and start
of economic summer), sparking the rebound in Japan’s 40-49 age bracket starting in
the mid-2000s. This effect promises to wear off in Japan when the “echo” of the US
baby boom (the children of the baby boomers) generate their effects in the 2020s.
The baby boomer wave was so powerful that their children were more numerous than
the intermediate generation. It’s not surprising, then, that Prime Minister Abe is taking
advantage of this respite to try to get his country out of deflation.
This just goes to show that a demographic shock creates a self-propagating wave.
It’s also consistent with the rhythm of economic seasons in that winter and summer
are conducive to crises and conflicts (and thus lower birth rates), as opposed to
spring and autumn. The decline in the French birth rate in the 19th century took
root in 1867, smack in the middle of economic summer.
The link we described between generational trends and asset trends is particularly
valid for developed countries. Development also has other consequences on
demographics, such as the structural decline in the number of children per woman,
which is not irreversible however. France was the first country to establish a policy
in this area and has now stabilised its birth rate. As for the United States, some
have even begun to speak of another possible wave of births that would be more
than just the “echo of the baby boom” and could be linked to the decision to
postpone having a first child due to the crisis, increased life expectancy, women
focusing on their careers (now an accepted part of the Western way of life) and
scientific advancement.

4.4. Productivity gains


In developed economies, in addition to the generational factors described above,
a critical role is also played by innovation, one of the four pillars mentioned by
Kondratiev. The US was a pioneer in this regard - if not to say it largely dominated
the classification. Based on World Bank 2011 figures (in value terms), the top ten
countries account for 75% of global R&D expenditures, but the US accounts for
30% all by itself. Japan takes second place. Despite its lost decade, it has kept
up considerable research efforts. Japan is also the most advanced country for
robotics, which may end up compensating for the more pronounced ageing of its
population compared to other countries. Germany has been surpassed by China,
and the UK by South Korea. France comes in fifth place.
According to Alan Greenspan, however, innovations are not capable of making
productivity gains increase by more than 3% per year over an extended period.
In the United States, for example, growth in Per capita GDP averages about 2%
per year over the long term (2.1% since 1929 according to the NBER). Even if
productivity gains were to partially offset the decline in the economically active

Amundi Investment Strategy Collected Research Papers 51


populations of developed countries, potential growth is still lower than in previous
decades, which is also consistent with the equilibrium real interest rate level
suggested above.

- The 1 irst contri utors to the Glo al R


United States
Japan
China
Germany
France
South Korea
United Kingdom
Canada
Brazil
Australia
0% 5% 10% 15% 20% 25% 30% 35%

Source: orld an data , Amundi Research

It was thus necessary for new powers to emerge in order to maintain global
growth at a decent level. At the end of the 19th century, during its winter phase,
the United Kingdom already had to make room for the rise of other economic
powers, beginning with Germany and the United States. With access to existing
technologies while initially skipping the fastidious innovation phase, these new
economies developed very quickly. This was the case for Japan from the end
of World War II to the 1980s, and for the Asian “tigers” (South Korea, Hong
Kong, Singapore and Taiwan) as from the 1970s; the increase in Per capita GDP
(see chart 37), which measures both wealth and productivity, was unequivocal.
Other countries followed suit: Thailand, Malaysia, Indonesia and the Philippines
all made strides in the 1980s, the Chinese cities of Shenzhen and Shanghai took
off in the 1990s, and Vietnam for example in the 2000s.

- G P per ha itant in 1 International Gear - ha is dollars


35,100
USA
30,100 Western Europe
Japan
25,100 4 Asian Tigers
China
20,100

15,100

10,100

5,100

100
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000

Source: An us addisson, Amundi Research

52 Amundi Investment Strategy Collected Research Papers


Per capita GDP, R&D and urbanisation rates
-G P apita and R G P
35,000
R² = 0.6991
30,000
GDP / Capita (in dollar)

25,000

20,000

15,000

10,000

5,000

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
R&D (%GDP)
Western Europe North America, Aust.,N.Z.
Eastern Europe Ex-USSR
Latam Eastern Asia
Middle East-Western Asia Africa

Source: An us addisson, orld an , Amundi Research

-G P apita and Ur anisation rate


35,000
R² = 0.4862
30,000
GDP / Capita (in dollar)

25,000

20,000

15,000

10,000

5,000

0
0 20 40 60 80 100
Urbanisation rate (in %)
Western Europe North America, Aust.,N.Z.
Eastern Europe Ex-USSR
Latam Eastern Asia
Middle East-Western Asia Africa

Source: An us addisson, re ision , Amundi Research

R&D stimulates the structural productivity of the economy. The lion’s share of
this effort is made by developed countries; this role is filled by urbanisation in
emerging countries.

Amundi Investment Strategy Collected Research Papers 53


In an emerging country, urbanisation is the main driver of Per capita GDP growth,
which implies a substantial investment in infrastructure. It also takes a certain
level of Per capita GDP for growth to pick up the pace ($3000 to $5000 in
the Maddison dataset, which refers to the Geary-Khamis dollar of the 1990s),
corresponding to an urbanisation rate of around 50% in Chart 39. Finally, there
is a clear link with the demographic approach described above, since younger
rather than older adults are the ones who migrate to the cities. This naturally
leads to demand for real estate, and so on.
It’s hard for a large country to achieve an urbanisation rate of more than 80%,
which is why North America, Western Europe and Japan are positioned so
high on the chart. We can also see that some emerging countries are already
significantly urbanised, but Per capita GDP has hardly benefited. Examples
include some countries in Latin America, the Middle East, Western Asia and
Africa. These countries therefore depend on the growth of their economically
active population, much like a developed country. Asia has a much better
positioning on this curve, however.

4.5 Positioning of developed and emerging countries based


on these criteria
Developed countries
The most advanced countries have been ranked in the chart 40 (p 38) according
to R&D effort (as a percentage of GDP) and growth in the economically active
population by 2025 (according to UN data). This graphic representation
is complemented by a table that adds the proportion of 40-49s within the
economically active population. The resulting classification equally takes
into account the ranking of R&D efforts and the average ranking of the two
demographic momentum criteria used.
This analysis is particularly useful in that it maps out countries but does not
take into account either the valuation of assets or the financial position of these
countries. It should also be noted that for “smaller countries,” the economy
should not be confused with the equities market because listed companies can
be very international. This classification thus more accurately reflect the long-
term attractiveness of these countries.
As a result, we can see that the US combines stronger R&D efforts and
demographics than average - position (1) on the chart. In absolute terms,
however, demographics are fairly neutral on a 2025 horizon, even corrected for
the proportion of 40-49s.

54 Amundi Investment Strategy Collected Research Papers


40 -Attractiveness of developed markets
Acti e population ro th and R

0.8
(3) Ireland
0.6 (1)
Active population growth
0.4
New Zealand
0.2
UK Australia
0.0
Norway USA Sweden
-0.2 France
Denmark
-0.4 Spain Belgium
Netherlands Finland
-0.6 Portugal Japan
Canada Switzerland
Greece Italy Republic of Korea
-0.8 Austria
Singapore
-1.0 (4) Germany (2)
-1.2
0.0 1.0 2.0 3.0 4.0 5.0
R&D (% GDP)
(1) Above average dynamism of the active population and more powerful R&D
(2) Below average dynamism of the active population but more powerful R&D
(3) Above average dynamism of the active population but less powerful R&D
(4) Below average dynamism of the active population and less powerful R&D

R&D Active pop 40-49/20+


Ranking
in % of GDP % YoY over 2010-2025
1 Finland 3.5 -0.5 0.7
2 Ireland 1.7 0.5 3.3
3 Sweden 3.4 -0.1 -2.3
4 South Korea 4.0 -0.9 -2.2
5 USA 2.8 -0.2 0.1
6 Australia 2.4 0.0 0.1
7 Japan 3.4 -0.6 -2.8
8 Denmark 3.0 -0.2 -3.5
9 France 2.3 -0.1 -2.6
10 Switzerland 2.9 -0.7 -2.9
11 Belgium 2.2 -0.4 -1.8
12 Austria 2.8 -0.8 -3.3
13 UK 1.7 0.0 -1.8
14 Germany 2.9 -1.0 -3.6
15 New Zealand 1.3 0.2 -1.8
16 Canada 1.7 -0.6 -0.1
17 Norway 1.7 -0.1 -2.6
18 Singapore 2.1 -0.8 -1.8
19 Netherlands 2.2 -0.5 -3.9
20 Spain 1.3 -0.3 -0.8
21 Portugal 1.5 -0.6 -0.8
22 Greece 0.7 -0.6 -1.3
23 Italy 1.3 -0.7 -4.4
Median 2.2 -0.5 -1.8
Source: Angus Maddisson, ONU (revision 2014), Amundi Research
This classification is sorted by ranking: we first combined the rankings of the two demographic factors
and then added that result to the result obtained on the R&D criterion.

The US occupies a solid position in terms of long-term attractiveness. Japan is a close


second thanks to its considerable R&D efforts offsetting its declining demographics.
Germany - another country known for R&D - has a lower ranking clearly due to its
demographic factor, which is one of the worst in our sample. Several countries are in
the same boat, such as Italy, the Netherlands, Austria and Switzerland. France is in
the middle of the ranking.

Amundi Investment Strategy Collected Research Papers 55


41 - Attractiveness of less developed markets
Acti e population ro th and ur anisation rate

2.5
Philippines
(3) (1)
Active population growth (in %)

2.0 Saudi Arabia

Venezuela Peru Indonesia


1.5
Colombia
Kuwait Mexico India
UAE Turkey
1.0 Argentina Malaysia
Brazil
0.5 Uruguay Vietnam
China
Chile
0.0
Czech Republic
Thailand
-0.5 Hungary Romania
(4) (2)
Russie Slovakia
-1.0 Poland
120 100 80 60 40 20 0
Urbanisation rate (in %)
(1) Above average dynamism of the active population and more potential of urbanisation
(2) Below average dynamism of the active population but more potential of urbanisation
(3) Already urbanised but above average dynamism of the active population
(4) Already urbanised and below average dynamism of the active population

Ranking Urbanisation % Active Pop. % YoY 2010-25


1 Philippines 45 2.0
2 India 31 1.4
3 Vietnam 30 0.7
4 Indonesia 50 1.4
5 Saudi Arabia 82 1.7
6 Peru 77 1.5
7 Mexico 78 1.4
8 Colombia 75 1.2
9 Turkey 71 1.0
10 China 49 0.1
11 Malaysia 71 0.9
12 Thailand 44 -0.3
13 Venezuela 89 1.4
14 UAE 84 1.1
15 Kuwait 98 1.4
16 Romania 54 -0.6
17 Brazil 84 0.6
18 Argentina 91 0.9
19 Slovakia 55 -0.6
20 Poland 61 -0.8
21 Czech Republic 73 -0.3
22 Chile 89 0.2
23 Uruguay 94 0.4
24 Hungary 69 -0.7
25 Russia 74 -0.7
Median 73 0.9
Source: Angus Maddisson, ONU (revision 2014), Amundi Research
This classification is sorted by ranking: we combined the demographic factor rankings with the R&D ranking.

Asia may well monopolise the first part of the table, but China only makes an appearance
toward the bottom of the Top 10, mainly because of its demographics. Three Latin
American countries (Peru, Mexico and Columbia) manage to slip in thanks to their
demographic momentum. At the bottom of the table, Eastern Europe, which is already
urbanised, is penalised by declining demographics. Russia comes in dead last.

56 Amundi Investment Strategy Collected Research Papers


Australia is slightly positive across all criteria and comes out very close to the
United States. Sweden and Denmark also come out very well at first glance, but
are penalised by the decline in the 40-49 bracket, bringing them down a few pegs
in the ranking, as opposed to Finland.
Japan, Germany and South Korea partially make up for declining growth in their
economically active population with a substantial R&D effort - position (2). The
United Kingdom, whose R&D expenditures are below average, can be expected to
benefit from relatively dynamic demographics - position (3). France is in a similar
situation, but with slightly higher R&D efforts.
Finally, southern Europe posts a low R&D effort and falling demographics - position
(4) - especially if we incorporate the proportion of 40-49, which is why Italy especially
but also Greece, Portugal and Spain are at the bottom of the ranking.
Emerging countries
The less advanced countries are ranked in the chart 41 (p39), which this time
examines urbanisation rates and growth in the economically active population by
2025. The resulting classification (see table) equally takes into account the ranking
of urbanisation rates and the momentum of growth in the economically active
population.
The most attractive countries hands down are the Philippines, India and Indonesia
- position (1) in the chart. Demographic momentum is prominent, while these
relatively rural countries have major investment efforts to undertake. The latter
criterion brings Vietnam into this group.
China has already made considerable country-wide urbanisation efforts. In theory,
it still holds some potential in this area but will have to contend with stable growth at
best in its economically active population from now to 2025 - position (2). Thailand,
which is not more urbanised than Indonesia or the Philippines, is faced with even
less buoyant demographics than China. At the opposite end of the spectrum, Latin
America and the Gulf countries are already very advanced in terms of urbanisation,
but they still have the demographic ace up their sleeve to keep their economies
moving forward - position (3) - some more than others: Peru, Mexico and Columbia
being the best-equipped.
Finally, Eastern Europe is severely handicapped by the demographic factor. Some
countries can still count on their urbanisation potential (Romania, Slovenia), but
others such as Russia are further along this road - position (4).

V. Consequences of ongoing transformations on


geopolitics and our growth model
There are at least two major consequences of the migration of the world’s centre
of gravity from the Atlantic to the Pacific and the technological changes underway:
one is geopolitical and the other impacts our growth model.

Amundi Investment Strategy Collected Research Papers 57


5.1 By sheer virtue of its size, China is destined for leadership
Ever since the Industrial Revolution, there have been two leading countries:
the UK and the US. The emergence of the Chinese behemoth this economic
winter suggests that it might one day dethrone the United States and reclaim
the position it held in the 18th century before the British industrial revolution
(see chart 42).

- rea do n o the Glo al G P 1 International Gear - ha is dollars

100%
Africa
90%
Middle East
80%
Japan
70% India
60% China

50%
Latam
Ex USSR
40%
USA
30%
UK
20%
Italy
10% Germany
0% France
1500 1550 1600 1650 1700 1750 1820 1870 1913 1950 2000

Source: addison, Amundi Research

If history does repeat itself, however, it will take several decades before China
achieves total supremacy and the road to this achievement is sure to be chaotic.
The United States surpassed the United Kingdom industrially in the 1890s, during
the winter of Cycle II, but only really consolidated its leadership in the following
winter with the advent of World War II (see Per capita GDP chart 37). It had already
taken a big step forward in summer with World War I, but was hit much harder than
the UK by the 1929 Stock Market Crash. And let’s not forget that US demographics
are expected to become more buoyant again in 2020-2025, which is not the case
for China. Finally, Japan - a key ally of the United States in Asia since World War
II - began to rival it on the economic front in the 1980s. After a decline of more than
20 years, Japan can no longer claim to rival the US, which has every interest in
promoting the balance of economic forces in Asia.

58 Amundi Investment Strategy Collected Research Papers


Another giant - India - could also return to its former glory, at least partially; its
population is much more dynamic than China’s. With its young population and as-
yet very low urbanisation rate, India is definitely an interesting candidate to keep an
eye on over the long term. Asia has taken its destiny in hand since the end of the
Cold War (of which it was one of the principal victims), putting it squarely back in
the centre of the global chess board.

5.2 The size of the global capitalist economy is changing


China’s awakening and the rapid development of other emerging countries are
responsible for this phenomenon.
The emergence of middle classes, which is the corollary to urbanisation, goes
hand-in-hand with increased revenues and thus growing demand for consumer
goods and healthcare expenditures, which are themselves boosted by the ageing
of the global population.
The biggest impact of the change in magnitude of global development, however,
is the awareness that the environment must now be factored in to our growth
model, especially considering that the world population recently passed the 7
billion mark and is expected to add another billion over the next 10 years. This
not only affects resources, which are by no means inexhaustible (energy, water,
etc.), but also the very equilibrium of our planet beleaguered by global warming.
The following chart 43 compares oil consumption with development, as measured
by Per capita GDP, and population size. It is obvious that, given the size of
emerging countries, a new means of growth will have to be invented.
Paradoxically, this challenge has every chance of becoming one of the drivers
of the next Major Cycle. The transformation of our infrastructures (construction,
transport, communication) in a bid for greater energy efficiency will require
investments over several years if not decades. Considerable technological
advances were made in the 1990s - so about twenty years ago - the same as in
previous major cycles; there are no technical impossibilities impeding them. The
successive crises of 2000, 2007 and 2011 drove interest rates to historically low
levels to fight deflation. There is no longer any such thing as a risk-free investment.
The cash is available to fund these investments once debt has been purged and
confidence restored.
Finally, we would point out that the advent of the information age is another societal
development that is radically changing our behaviour. The rise of cybercrime, the
power of the social networks which ushered in the so-called “Arab Spring”, and
the diminishing gap between developed and emerging countries that led the G7
to become the G20 are just a few examples. Global governance will also have to
be reinvented.
We may still be in the winter phase, but economic spring is just ahead!

Amundi Investment Strategy Collected Research Papers 59


- Oil consu ption and G P per capita

100
Saoudi Arabia
90
(per thousands of inhabitants and per day)

80
Number of barels of oil consumption

USA
70 Canada

60 Netherlands

South Korea Taiwan


50
Norwegen
Australia
40 Japan Hong Kong
Switzerland
Spain
30 Venezuela Italy Germany
Iran
Russia UK
France
20 Indonesia
Mexico
Brasil Polska
10 Inde Argentina
China Turkey Bubbles represent the size of
Philippines the population in 2008
0 Pakistan
0 10,000 20,000 30,000 40,000 50,000 60,000

GDP per capita in dollar (PPP)


Source: , F, Amundi Research

60 Amundi Investment Strategy Collected Research Papers


Conclusion
Long cycles spanning roughly 50-year periods embrace all economic activity
and are inherent in nature. Although each generation tends to challenge this
mechanism, it has nevertheless repeated itself since the dawn of capitalism
over two centuries ago.
These cycles contain a long upward phase, which can easily be read in interest
rate movements, followed by another phase in which interest rates decline over
an extended period. Equities alternate more quickly between uptrends and
downtrends. Thus, each approximately 50-year long cycle can be divided into
four phases called economic seasons: equities rise in spring and fall in summer,
rise again in autumn and consolidate their gains in winter.
“Spring” coincides with the start of the long cycle. It is especially supportive for
equities and slightly unsupportive for bonds, whose yields have to climb from
very low levels. Gold in particular and money market investments are no more
attractive. “Summer” is a time of excess marking the high point of inflation. It is
supportive for gold, commodities in general, the money market and inflation-
linked bonds. “Autumn” benefits from the vast disinflation movement and the
continued rise in debt. It ends with a major international crash. It’s a good
season for financial assets, starting with equities. Finally, “winter” is when
inflation hits a major low point and debt is purged. Fixed income investments
- at least the surest bets - and gold fare better than equities, which are highly
volatile.
These mechanisms are universal within market economies, despite time
differences of a few years. These cycles are even international by nature.
International trade progresses in line with technical advances achieved from
generation to generation. Trends in various countries tend to converge toward
that of the leading country, namely the United States in the 20th century. In our
view, the US is a relevant litmus test for the future. It may be an exceptional
case, but even so it has experienced fewer rare events such as the massive
population losses France was subjected to during both World Wars.
Demographic factors are key to understanding the changes that took place in
the 20th century. Rather than calling this approach into question, however, they
support it while offering leads to better grasp future trends. Each age group has
its own preoccupations and thus its own impact on different asset classes. The
proportion of young adults influences real estate prices over the long term, and the
intermediate generation the level of growth potential and thus real interest rates.
The 40-59s impact effective growth and financial saving and thus share prices.
For example, the post-war baby boom happened in spring; when this generation
reached consumer age, it drove the autumn economic expansion from 1981
onward. The Japanese crash of 1990, the 2000 crash in the US, and even the

Amundi Investment Strategy Collected Research Papers 61


European crisis of 2010 following on from the US subprime crisis all coincided
with the drop in the proportion of the baby boom generation in the total
population. The “echo” of the baby boom should not generate any beneficial
effects on the long cycle until around 2020-2025.
Globally, we are still in the winter phase, and major changes are taking shape
as they always do. These include geopolitical changes, with the world’s centre
of gravity shifting from the Atlantic to the Pacific. With the development of
new countries and changing size of the global economy, the technological
advancements of the 1990s boom will find applications in new areas on a
worldwide scale, for the fifth time since the dawn of the industrial age. After their
winter purge, banks will begin lending again to finance these developments.
This revival will transform the global economy and our way of life. It should lead
to a period of balanced growth: an economic spring.
Of course, each major cycle has its own distinctive features. The transition
from winter to spring in the early 20th century (transition to Cycle III) went
smoothly. The transition to Cycle IV was delayed by World War II, which had
exceptional inflationary consequences for a winter period. The Fed’s handling
of the current winter through its quantitative easing programme can also be
considered exceptional from a historical standpoint, leading to the formation of
bubbles. From that perspective, the US market may have been looking ahead
to the upcoming spring.
Obviously, the path back to more auspicious conditions is not a straight one.
Short cycles are contained within the long cycles. While positioning the current
period within the long cycle is critical to benefiting from the features of the
different asset classes, determining where we are in the short cycle is the best
way to seize opportunities. This was the topic of the Discussion Paper we
published in October 2014: “The Short Investment Cycle: Our Roadmap.”

62 Amundi Investment Strategy Collected Research Papers


Bibliography

Jean-François De Laulanié (2003): “Les placements de l’épargne à long terme” – Economica


Jérémy Siegel (1994): “Stocks for the long run”– Mac Graw Hill
Elroy Dimson, Paul Marsh, Mike Staunton (2002): “Triumph of the Optimists” – Princeton
University Press
Sidney Homer & Richard Sylla (2005 fourth edition): “A history of interest rates” - Willey
Finance
US Bureau of Census (1975): “Historical Statistic of the US from colonial time to 1970,
Bicententennial Edition, Part 2” - US Government Printing Office
Robert Flood and Nancy Marion (2004): “Stock Prices, Output and the Monetary Regime,
draft version” - IMF
Nicolaï Dimitrievitch Kondratiev (1992): “Long Cycles of Economic Conjuncture” – Edition
presented by Louis Fontvieille - Economica
Philippe Gilles (2009): “Histoire des crises et des cycles économiques”’ - Armand Colin
Charles P. Kindleberger (2000 fourth edition): “Manias, Panics, and Crashes, a History of
Financial Crises” - Willey investment classics
Yves Breton, Albert Broder, Michel Lutfalla (1997): “la grande stagnation en France” -
Economica
Christina D. Romer (1992): “What ended the Great Depression?” - The journal of economic
history
Reinhart and Rogoff (2010): “This Time it’s Different” - Person
Aymeric Chauprade (2003 second edition): “Constantes et changements dans l’histoire” -
Ellipses
Peter Berezin (2013): “‘Human Intelligence and Economic Growth from 50,000 B.C. To the
Singularity” – The Bank Credit Analyst
Harald Edquist (2006): “Technological Breakthroughs and Productivity Growth” - EFl, The
Economic Research Institute, Stockholm School of Economics
Ian Gordon (1998): “The Long Wave Analyst Volume 1, Issue 1” - Canaccord Capital
Paull Wallace (1999): “Agequake” - Nicola Brealey Publishing
Harry S. Dent (2011): “The Great Crash Ahead” – Free Press
Eric Mijot (2013): “Equities: Spring is in the air, but the winter blues still need shaking”
– Amundi Cross Asset Investment Strategy, June

Amundi Investment Strategy Collected Research Papers 63


64 Amundi Investment Strategy Collected Research Papers
DP-09

Reallocating savings
to investment:
The new role of asset managers
Y VES PERRIER,
Chief Executive Officer, Amundi

February 2015

The “Great Recession” brought significant changes into stark


relief: declining returns from factors in production as well as
technological innovation and productive investment insufficient
to drive future growth. Investing in future economic growth often
means investing in moderate-sized projects and companies. What
is needed first and foremost is i) for savings to be better allocated in
terms of supportsvehicles, in order to favour long-term investment
and “risky” assets. In this sense, taxation and education are
essential contributions; ii) for savings flows to circulated better at
the European level in order to better direct excess savings from the
North toward investment deficits in the South. Asset managers,
who manage private and institutional savings, are naturally at
the hub of these financial channels and play a major role between
(domestic and foreign) investors’ (domestic and foreign) yield return
needs and (small and large) companies’ (small and large) investment
needs. It is therefore crucial that companies developing new
technologies are not held back by a lack of understanding of their
activities, by inadequate regulatory frameworks or by allocating too
little capital. It is in this last point where the contribution from asset
managers is found.

Amundi Investment Strategy Collected Research Papers 65


REALLOCATING SAVINGS TO INVESTMENT
THE NEW ROLE OF ASSET MANAGERS

Introduction
How do we encourage productive investment? This is undoubtedly one
of the critical questions of the current economic climate. The following text
puts the current environment into perspective and develops solutions, as seen
by an asset manager in conjunction with long-term national and international
investors.

1 - Corporate investment:
twofold challenge - growth and jobs
Excess debt and deficits led Japan, then the United States and Europe to
implement highly expansionary monetary policies as well as very restrictive
budgetary and fiscal policies. Overall, budget austerity and monetary largesse
brought tranquillity to the financial world, but exposed weaknesses in the
economic realm:
t Unsustainable growth models, particularly with excess credit;
t Wide variations between countries, in a eurozone where convergence
should be the benchmark;
Without the leverage provided by credit for over a decade, reality has sunk
in, and it is not exactly pleasant:
t Investment is in the doldrums pretty well everywhere in the eurozone.
t Potential growth has been revised downward nearly everywhere: in the
US, Japan, the UK, Germany, France, the eurozone, and even China. It is
now 1% in the eurozone and Japan and a little over 2% in the United States;
t The unemployment rate remains high in Europe because there is not
enough economic growth;

66 Amundi Investment Strategy Collected Research Papers


t The unemployment rate has fallen sharply in the US, but the participation
rate is disappointing overall;
t Youth employment is extremely low, particularly in Europe;
Now more than ever, growth and employment are the main priorities of
European authorities. The Juncker Commission has set its top priority as “a
return to growth in Europe and job creation without taking on new debt”.

1.1. How did we get here?


Proponents of the “great stagnation” hypothesis have identified the “headwinds”
that are holding back growth and employment:
t The ageing population has resulted in a lower workforce participation
rate and weaker gains in productivity. Baby boomers are gradually exiting the
workforce, birth rates are often too low and longer lifespans are maintaining
pressure on economic activity. The older a population gets, the more pension
funds will quickly enter their distribution phase, where asset liquidity becomes
crucial and risk-taking is reduced. In other words, long-term and real asset
classes are naturally and gradually being abandoned for fixed-income assets.
t Globalisation is exerting downward pressure on wages in developed
economies due to competition from emerging economies and industrial
offshoring. The equalisation of input prices is inevitably detrimental to the
highest-wage countries, i.e., advanced economies. This downward pressure on
wages in the advanced countries is weighing down consumption and therefore
growth and investment.
t Dealing with the environmental crisis will gradually cut into the budget that
households devote to other consumer items.
t Deleveraging in the private sector and the need to stabilise public debt
are gradually reducing disposable income and consumer spending. Returning
public debt to a sustainable path will also weigh down GDP growth, because
it limits economic policies, as countries have fewer and fewer resources to
stimulate growth and investment.
t Improperly allocated education spending: We must promote greater
efficiency and target spending towards the sectors of the future in order to re-
establish the link between education and growth, which is sometimes hard to
demonstrate.
In addition, the “Great Recession” (the recession that followed the 2008 financial
crisis) has highlighted important changes: decreasing returns for production
factors and inadequate technological innovation, all of which lead to a “great
stagnation”. The consequences are clear:
t Weak investment and economic growth;
t Low inflation;

Amundi Investment Strategy Collected Research Papers 67


t Low short- and long-term interest rates;
t High asset prices;
t Higher financial volatility.
It is difficult now to identify a source of economic growth. The current low growth
situation could last a long time if we do not act. “There is a new normal: low interest
rates and high asset prices. People who think that interest rates will normalise are
almost surely wrong” (P. Krugman, Amundi International Forum, June 2014).

1.2 - Technological changes and innovations


When we talk about the “Great Stagnation”, we are referring to innovation deficits,
to insufficient productive investment or investment that is not enough to
stimulate future growth and to factors that are markedly structural, not just
cyclical. The answers therefore need to be structural and not just cyclical.
Investments spurring long-term growth should be encouraged. According to
a report prepared for the European Commission (2010), the most promising
technologies in terms of growth and innovation are:
t New communications technologies;
t Alternative energy development;
t Biotechnologies;
t Nanotechnologies;
t Automation;
t Development of new materials;
t Health technologies.
In this turbulent period on commodity markets, the development of new energy
storage technologies is particularly significant.
These innovative technologies are quite often developed by SMEs, for which
financing should be facilitated. They form the linchpin of the European
industrial recovery. Let’s remember that the new European Commission has
set a goal of raising the weight of industry in Europe’s GDP to 20% from its
current level of 16%. This is a crucial point for strategic sectors such as the
automotive, aeronautics, engineering and space sectors as well as the chemical
and pharmaceutical industries.

2 - How can we get out of this situation of low


investment/growth? The role of asset managers
To stimulate investment, there are, of course, the traditional government
responses, but there are also innovative, targeted private responses.

68 Amundi Investment Strategy Collected Research Papers


2.1 - Government strategies
Three (government) exit strategies are frequently cited:
Exit strategy # 1: returning to full employment. According to this strategy, it would
be crucial to return to full employment. To accomplish this, ultra-accommodative
monetary policies would need to be adopted for longer than strictly needed. Even
the existence of bubbles would have to be tolerated. Clearly followed by Alan
Greenspan in the past, this strategy resulted in devastating excesses. Along with
this strategy, there are also investment policies like the Juncker Plan, a plan that
aims to promote infrastructure investment. Note that these policies quickly reached
their limits: Even if the United States, the United Kingdom and Japan were able
to conduct large-scale stimulus policies, all of them ended up with government
deficits (above 10% of GDP) that Europe could not accept. As sensible as it is, the
Juncker Plan is betting heavily on private investment, since the contribution of new
public money is, all things considered, rather marginal.
Exit strategy #2: returning to inflation. Making economic policy effective
requires a return to inflation in order to definitively pull out of a deflationary spiral,
which encourages buyers to delay purchases as they know that prices will be
lower later. This subject, which has been mentioned on numerous occasions
by Olivier Blanchard (IMF) and Paul Krugman, is also a favoured topic of central
banks (Fed Vice Chair Stanley Fischer mentions it often) and governments. The
Bank of Japan is also in this camp.
Exit strategy #3: structural reforms. Carrying out structural reforms is proving
essential. The BIS and central bankers in general are often quick to remind us of
the need to make economies less expensive and more flexible. This is the only
way to avoid generating financial bubbles (strategy #1) or (strategy #2) letting
inflation skid out of control (one of the crucial goals for central banks and the
only goal for some). We know that reforms—when they are properly targeted
and carried out effectively—lead to more or less foreseeable gains in growth, but
they must not weigh too heavily on short-term growth in the current lacklustre
economic climate. Such are their limitations at the moment.
As legitimate as these three strategies may be (quite “traditional” and very “macro,
top down”), we believe they are inadequate. There are other strategies, and they
should not be underestimated. On the contrary, they should be promoted. Asset
managers have the capability of inserting themselves into these strategies, or
even being the driving force behind them.

2.2 - Reallocate savings to investment:


a key challenge for growth
The finding is clear: funding for the eurozone economy is suffering from a structural
deficit in allocating savings to long-term funding in equities. This is especially true

Amundi Investment Strategy Collected Research Papers 69


in France, where only 20% of managed assets are invested in equities, compared
to more than 40% in the United Kingdom. In all, this corresponds to a difference
in holdings of more than €1 trillion in equities. There are two reasons for this:
First, low equity holdings by French insurers, a situation aggravated by the
restrictions of Solvency 2. Insurers have less than 10% of their portfolio assets
invested in listed equities (of which half are invested in shares of corporate or
financial institutions), and three-quarters in bonds (of which a very low percentage
are securities issued by French companies).
Second, the lack of pension funds, long-term investors that are generally
heavily invested in long-term assets (equity and private equity), is a real handicap
in a country like France.
This €1 trillion should be placed in context with corporate financing requirements.
Capital requirements for French SMEs and intermediate-sized companies
apparently stand at €11 billion according to AFIC and €20 billion by 2020
according to the FINECO of Paris EUROPLACE. Requirements in terms of capital
and debt for all companies are apparently €80 billion, according to the FINECO
Report of Paris EUROPLACE, and €100 billion according to the Berger/Lefebvre
report (of which €15-25 billion just for SMEs/intermediate-sized companies).
In other words, measures and structures must be established that can
redistribute savings to the economy’s financing requirements. Banks
provide a significant share of financing for the economy via banking credit (more
than 90% of loans to businesses in the eurozone), but institutional investors are
also crucial for meeting requirements - debt or equity - of all economic actors,
be they SMEs, intermediate-sized companies, large corporations, government
or the public sector in general. Indeed, the asset management industry, which
provides third-party management for institutional investors and directly attracts
foreign capital and savings, actively contributes to financing the economy. There
are several distinct channels:
t Holdings that represent a significant portion of the market
capitalisation in equities (more than 20% in France’s case);
t Medium- to long-term financing: holdings that represent 20% of French
issuers’ debt inventory in medium- to long-term securities, 25% of the
inventory of non-financial corporate bonds and more than 20% of the
inventory of government bonds;
t Short-term financing: holdings that represent more than 30% of French
issuer’s debt inventory in short-term securities, 45% of certificates of
deposit (for financing banking liquidity), and more than 35% of corporate
treasury bills;
t The equity financing of unlisted SMEs, via private equity, is indispensable,
specifically in France where the lack of equity capital is glaring.

70 Amundi Investment Strategy Collected Research Papers


Long-term savings must be encouraged, which means taking several unavoidable
steps:
t Reviewing tax regulations on savings, by giving long-term savings the
benefits provided by “most favoured savings” status (at least as far as
taxation is concerned);
t Expanding and improving the range of platforms and products
contributing to the long-term financing of the economy, by developing
products such as diversified funds with restrictions on allocation (unlisted
equities 10%, listed equities 40%, bonds 30%, cash 20%), eligible for the
PEA [equity savings plan], the PEA PME [equity savings plan for small
and medium-sized enterprises], and unit-linked life insurance vehicles
at preferential rates, as well as Euro-Growth funds. It is also a matter
of offering long-term investment funds, as part of the ELTIFs (“European
Long-Term Investment Funds”).
t Supplement pensions with individual and group pension savings
by relying on existing platforms (life insurance, employee savings plan,
individual platforms, etc.). This inevitably requires tax incentives for each
of the employee savings product families.
t Establishing educational initiatives to raise awareness among potential
private investors of the collective and individual benefits of long-term investment,
and allowing them to better understand how the “Risk/Return” model works.
This ultimately means making the public’s currently fearful approach to risk
less subjective and more rational. Promoting and compensating risk-taking is
necessary, because growth lies in the ability to face risk.
In all, to move savings toward productive investment, we must promote long-term
savings, attract private capital using attractive tax regulations, and ensure greater
attraction of foreign assets: only 14% of managed assets in France come from
abroad, vs. 38% in London. To do this, we must break away from the dissuasive
nature of current regulation, and ease the measures that push traditional long-
term investors to “ignore” long-term risky assets. It is thanks to such measures
that it will be possible for asset managers to make a greater contribution to
financing the economy. In additional to playing this traditional role, we can add
more innovative actions.

2.3 - Asset managers and stimulating investment:


a new role
We mentioned above that we are living in a time of changing technology,
specifically digital technology and the energy transition. Investments in these
sectors will eventually be the greatest generators of growth, and a more “micro”
approach is both unavoidable and desirable.
It is important to preserve what we have, but preparing for the future is vital.

Amundi Investment Strategy Collected Research Papers 71


Investing in the economy to spur future growth often means investing in
medium-sized projects and companies. Why SMEs? There are several reasons:
t We must remember that, in 2012, SMEs employed 87 million people in Europe
(67% of the total workforce) and generated 58% of overall added value.
They accounted for nearly 80% of the workforce and 70% of added value
in Italy, Spain and Portugal. In these countries, the SME sector is dominated
by microbusinesses with fewer than 10 employees. Moreover, SMEs created
85% of net new jobs between 2002 and 2010. Without helping and investing
in SMEs, it is hard to imagine any recovery in employment.
t Furthermore, when we talk about investment, we are talking about
financing. Large corporations and listed companies that are well-known
and are rated have no trouble finding financing. They can raise funds on
the equity or bond markets. For example, in the last two years, there were
nearly 160 new high-yield issuers in Europe. Access to financing is
easy, given current interest rate levels and investors’ search for yield. For
SMEs, though, it is a different story: more than 95% of them depend
on financing from banks—which are now more cautious following the
financial crisis and the sovereign debt crisis. The latest ECB survey on
the topic of SME financing shows that the approval rate for bank loans to
SMEs remains low in Southern European countries (Spain, Italy, Portugal
and Greece), but also (and unexpectedly) in the Netherlands.
t The third reason involves the current economic climate. It is no longer
simply a struggle between the “small” and the “large” companies,
but also (and particularly in some areas of innovation), a struggle
between the “quick” and the “slow”. SMEs, which are naturally faster
and more agile, should be prioritised.
Thus, executing strategies that affect SMEs and future sectors is crucial.
Institutional investors and asset managers have a role to play in this context,
because beyond direct investment and job creation, and beyond the major
government plans (Juncker Plan), there are three additional paths:
– Solution # 1: Financing SMEs by developing ABS. SMEs’ dependence on
banks for financing is a disadvantage in a situation where banks are lending
less and taking less risk. The ECB hopes to facilitate SMEs’ access to credit
by reviving the ABS market. Banks would therefore be able to dispose of
some or all loans granted through securitisation products, which would be
purchased by market players capable of carrying the risk over the medium
and long term. Some asset managers, including Amundi, are very active
in this area through two channels: first, through the management of
ABS portfolios, a type of management that was not undermined by
the financial crisis. Second, through the management and analysis
partnership for the ECB itself.

72 Amundi Investment Strategy Collected Research Papers


– Solution # 2: Investments in partnership. This involves bringing together
productive investment and financial investment, companies and long-term
investors, be they French, European or global. In this regard, very concrete
relationships may be implemented between borrowers and asset managers,
as was recently established between EDF and Amundi. The goal is to build
an investment portfolio around the theme of the energy transition, an issue
that is emblematic of changes in technology and long-term growth.
– Solution # 3: developing intra-European financing. One of the characteristics
that currently defines the eurozone is the extent of divergence between
countries. There is an apparent convergence (return to current account
surpluses) that actually reflects one of the most troublesome divergences:
excess savings in Northern Europe and a lack of investment in Southern
countries. The eurozone is unable to ensure that excess savings in the North are
recycled into investments in the South: perceived risk, custom, fragmentation
of economies—all factors that discourage investment. Excessive savings in the
northern countries should help to finance businesses in southern Europe that
are struggling to finance themselves. Since it does not happen on its own,
the development of intra-European financing must be promoted, specifically
in equity, and where applicable by adding a European guarantor like the EIB.
Asset managers know long-term investors and are likely to be the driving
forces of intra-European financing projects.
In all, the eurozone’s good health does not just involve its financial solvency. It
also involves its ability to engage in the investments that represent the growth and
jobs of the future. In turn, this will demonstrate its ability to repay its debt, which is
certainly a current issue.
To encourage investment and business, a change of course is in order:
t Reduce direct or indirect incentives for investing in real estate: two-
thirds of French consumers’ financial wealth is placed in real estate.
t Be able to dissuade investments in public debt. Of course, we
understand the motives of government pushing for these investments
(financial repression and regulation, reduced dependence on non-resident
investors, easy and stable debt financing), but this is all at the expense of
long-term growth. Governments shoulder this responsibility.
t Do not depend too heavily on public investments. The answer will not
come from governments. Given the scale of debt, many governments can
now play only a marginal role in terms of investments. The private sector is
and will remain the driver of growth.
t Do not bet too much on infrastructure investments. They are not
worthless, but they are not the key issue in Europe, in contrast with
elsewhere in the world.

Amundi Investment Strategy Collected Research Papers 73


t Domestic long-term investors exist, but they are not active enough.
Attracting foreign savings should not prevent development of long-term
savings in the eurozone
t Finally, we must be ready to do a better job of supporting business
development (financing, tax regulations), a major difference between
France and Germany, for example. Creating a business is rather easy;
growing and transferring it is much less so, especially in certain
countries.

Conclusion
Returning to growth and reaching full employment levels are undoubtedly the top
issues for the eurozone. For the sake of greater eurozone uniformity and to make
it possible to execute common policies more effectively, it is also necessary to
reduce the current unprecedented economic divergences:
t Different growth models, which led to the excesses of the 2000s;
t Different growth drivers (exports in Spain and Germany, consumption in
Germany, and public spending in France, etc.);
t Stark variations in terms of competitiveness;
t Solid budget balancing in Germany, but fragile in peripheral countries
and deteriorated in France;
t Public debt under control (Germany), fragile (France) or still vulnerable
(Spain, Italy, etc.);
t Current account surpluses everywhere but France, but which reflect
different realities: excess savings in the Northern eurozone and lack of
investment in the South. Worse still is the eurozone’s inability to recycle
excess savings from the North toward the investment deficits of the South.
The ECB cannot solve everything. Through its actions and statements, it
has been able to ease the financial crisis and to ensure much of the financial
convergence.
Governments as well as budget and fiscal policies must be the drivers in terms of
economic convergence. Three major levers should be used:
Lever 1: cost of labour: reducing and harmonising this will allow the eurozone’s
cost-competitiveness to grow. Reducing public spending to reduce taxes on
labour would kill two birds with one stone: improve public finances on one hand,
and restore competitiveness and employment on the other.
Lever 2: technological competitiveness: without setting the new and old
economies against each other, economic development is always achieved
through technological advances that must be harnessed effectively in order to

74 Amundi Investment Strategy Collected Research Papers


improve margins, competitiveness, employment levels and potential growth. At
the moment, it is the more resourceful and flexible US that has demonstrated its
ability to exploit new technologies and technological changes.
Lever #3: reallocation of savings to investment. More than ever, investment
and technological innovation have a major role to play in growing the European
economy of tomorrow. For this to happen:
Savings must be better allocated in terms of vehicles, to promote long-term
savings and “risky” assets. In this regard, taxes and education are essential
factors;
Savings flows must circulate better on a European scale, to better guide surplus
savings from the North toward investment deficits in the South.
Asset managers, who manage private savings and institutional savings, are
naturally at the intersection of these financial channels, and they play a major role
between investors’ yield requirements and businesses’ investment requirements,
be they large or small. As such, it is crucial that companies developing new
technologies are not held back by a lack of understanding of their activities,
by inadequate regulatory frameworks or by allocating too little capital. It is
particularly in this last point that asset managers such as Amundi can make
a contribution.

Amundi Investment Strategy Collected Research Papers 75


Bibliography

European Commission (2010): “The regional impact of technological change in 2020,”


September
Ernst and Young (2013): “The power of three: The EY G20 entrepreneurship barometer
2013”
Gordon R. J. (2012): “Is US economic growth over? Faltering innovation confronts the six
headwinds,” CEPR, Centre for Economic Policy Research, September
Juncker J. C. (2014): “A new start for Europe: my agenda for jobs, growth, fairness and
democratic change,” 15 July
Paris EUROPLACE (2010): “Le développement de l’épargne longue [Developing Long-
Term Savings],” June
Paris EUROPLACE (2013): “Redonner sa compétitivité au Pôle Investisseurs de la Place
de Paris [Making the Investor Division of the Paris Stock Exchange competitive again],”
Final report, September

76 Amundi Investment Strategy Collected Research Papers


DP-08

Allocating alternative assets:


Why, how and how much?
SYLVIE DE LAGUICHE,
Head of Quantitative Research, Amundi
ÉRIC TAZÉ-BERNARD,
Chief Allocation Advisor, Amundi

November 2014

Global institutional investors are showing a growing interest in


alternative investments (hedge funds and unlisted assets: real
estate, private equity, infrastructure and loans), which is further
accentuated by low interest rates. In so doing, they hope to
improve their portfolios’ returns while diversifying their assets.
We propose a method for classifying these assets into segments
based on their sensitivity to macroeconomic factors. This is
because the traditional relationship between return and risk
is a poor fit for these assets, for various reasons: infrequent
data, smoothed valuation methods and asymmetrical return
distributions.
Next, we advise investors on how to calibrate these assets
within their portfolios, and recommend a pragmatic approach,
which takes into account the investor’s liability constraints,
investment horizon, current asset allocation. When it comes
to risk assessment, we suggest to combine different methods:
representative listed assets, regulatory parameters and probable
maximum loss in stress scenarios. Finally, the targets set for
calibrating these assets should be expressed as risk ranges
rather than as weights, and the investor must proceed in stages
to achieve them.

Amundi Investment Strategy Collected Research Papers 77


ALLOCATING ALTERNATIVE ASSETS:
WHY, HOW AND HOW MUCH?

Introduction
Asset allocation traditionally refers to the way a portfolio is divided between equity,
fixed income and money market products, but the investment universe used to
build an institutional asset portfolio has expanded over the past few years to so-
called alternative investments: investments in hedge funds and unlisted assets
- direct real estate investment, private equity, loans or infrastructure projects-.
High-net-worth private investors, as well as certain pension funds and American
university foundations, have pioneered this type of assets, which have more
recently been the subject of increasing interest from supranational investors and
sovereign wealth funds. For example, large sovereign investors like the ADIA or
GIC devote 20% (half in real estate) and 26% of their asset allocation to alternative
investments, respectively. The graph below gives more details on the GIC’s
allocation as of late 2013:

1- Trends in so erei n asset allocation exa ple o GI

7% 11% 7%

23% 27% 26 %

26% 17% 21 %

44% 45% 46 %

2008 2012 2013

Public equities Fixed Income Alternatives Cash and Others


Source: Source: GIC reports, Amundi Research

78 Amundi Investment Strategy Collected Research Papers


2- Trends in so erei n asset allocation exa ple o GI detail
2008 (%) 2013 (%)
34 Developed Equity Markets 31
20 Nominal Bonds 19
10 Emerging Equity Markets 15
10 Real Estate 10
8 PE/YC & Infrastructure 11
7 Cash and Others 7
6 Inflation-linked Bonds 2
3 Absolute Returns Strategies 3
2 Natural Resources 2
Source: Reports GIC, Amundi Research

The purpose of this paper is to use Amundi’s expertise in research, asset


allocation and specialised management of alternative investment classes as a
basis for providing clarity to questions frequently asked by institutional investors
about this subject:
t What advantages do these investments have?
t How should they be classified to study their inclusion within an asset
allocation?
t What weight should they be assigned within a portfolio and what criteria
should serve as a basis for determining that weight?
t Finally, what operational rules may be adopted to manage these
investments?

Amundi Investment Strategy Collected Research Papers 79


1 - Definition, Objectives
and Investment Performance
1.1 - A Highly Diverse Investment Universe
It is hard to define a general asset allocation for these assets because
of their high degree of heterogeneity, both between and within each of the
alternative classes.
To describe this universe, a distinction is commonly made between:
– listed assets (mostly hedge funds): this category is even more important given
that many hedge funds since the crisis have increased their transparency and
more frequently updated their valuations and reduced their leverage, thereby
converging towards the same standards that apply to traditional funds.
– unlisted assets, which are sometimes divided into real assets (real estate
and infrastructure, as well as land and forests which will not be considered
in the development that follows) and private equity. A distinction may also
be made between the underlying equity or fixed-income risks - which is the
case for infrastructure investments - and between different categories of
borrowers.
The distinction between listed and unlisted assets also reflects a difference in
terms of imperfect liquidity. Hedge funds offer the liquidity inherent in listed assets,
but some types of strategies may be subject to the risk of liquidity suddenly
drying up in times of crisis: the problem for them is therefore the variability of
liquidity. Unlisted assets, meanwhile, are always illiquid in the short-term; they
automatically become liquid over rather long investment horizons, such as when
cash flows are distributed to investors in the case of loan repayment, or after
shares are sold in the case of a private equity fund. Their liquidity therefore
depends on the investment horizon.
Within each group, the performance of these investments and their level of
risk, depend greatly on:
– for hedge funds, the type of strategy (credit arbitrage, merger arbitrage,
futures funds, etc.) and the amount of leverage
– for unlisted assets, their geographic location (which is much more important
than global factors)
– their intended purpose: e.g. offices, shops, or homes in the case of real
estate; venture capital or development capital for private equity
– the investment method adopted: for private equity, there are direct investments
in unlisted companies, investments in private equity funds and even funds of
funds. In infrastructure, it is possible to invest in funds, in projects or in the
debt issued to finance a project.

80 Amundi Investment Strategy Collected Research Papers


– the type of project being financed (in the context of private equity or
infrastructure).
For this reason, there is much greater diversification between an investment in
Paris office space and residential real estate in Hong Kong, or between Californian
venture capital and an LBO in the British market, than between two stocks listed
on different developed exchanges 1.
This pronounced idiosyncratic nature will often justify assessing alternative
investments asset by asset, rather than using indices representative of the asset
classes analysed, when they exist.

1.2 - Investor Objectives


These investments meet objectives based on return or diversification:
– they make it possible to access an attractive source of return: these
assets have inefficiencies as defined by financial theory, which vary
depending on the asset class in question, which investors will want to
exploit. For private equity, infrastructure, or loans, it’s about access to unlisted
companies or projects: in this case, issuers that want to diversify their sources
of funding beyond simply issuing listed shares or bonds (or that cannot access
these markets) are ready to diversify by offering a premium to investors willing
to invest in unlisted assets. For hedge funds, it’s about access to added-value
managers who exploit uncommon sources of arbitrage, are subject to lower
investment restrictions and benefit from a longer investment horizon or higher
performance incentives.
– most provide the benefit of an illiquidity premium. According to financial theory,
the low liquidity of private equity or real estate investments, which investors are
required to keep in their portfolios for a certain number of years, is compensated
by a greater return. Some institutional investors may benefit even more from this
premium, given that they normally have a long enough time horizon.
– they offer the investor enhanced portfolio diversification, owing to the
expanded investment universe, their supposedly lower dependency with
regard to traditional asset classes and the highly idiosyncratic nature of the
underlying risks within those investments as mentioned above.
– in more specific circumstances, they make it possible to generate return
against a backdrop of low interest rates, with a smoother short-term risk
than for listed assets. Some institutional investors therefore look at their
investments in real estate, infrastructure, or direct private sector loans as
replacing a portion of their bond portfolio, which currently offers returns that
are too low for their long-term needs.

1 See Hauss for a quantification of systemic risk's share in the total risk of real estate
investments

Amundi Investment Strategy Collected Research Papers 81


1-3- How Should These Assets Be Classified?
The simplest approach to segmentation consists of supplementing the traditional
asset classes (equity, fixed income and cash) with a “diversification” category that
includes all alternative assets. However, we believe that this simplistic approach
is only acceptable if the weight of these assets is very marginal in the investor’s
portfolio. Otherwise, it is useful to identify at a granular level, meaning for every
sub-segment in the world of alternative investments, these assets’ sensitivity to
risk factors: equity (and potentially style factors within equity), real interest rates,
inflation, credit, etc. in order to better estimate the risk allocation of the portfolio
as a whole.
The investment universe may also be segmented based on the investment
objective, for example:
– Participation in economic growth: private equity investments logically fall
into this category, just like investments in publicly traded stocks.
– Generation of recurring income: real estate and infrastructure meet this
objective, which makes them similar in that regard to bond investments,
although they also have an equity profile owing to their exposure to growth,
as we shall see later on. This shows that the various defined segments are
not mutually exclusive and that one asset may belong to multiple categories.
– Protection from certain macroeconomic and financial risks: Investments in
commodities, which some investors consider part of the alternative category,
are therefore sought out in order to protect a portfolio from a risk of rising
inflation. The DAMS methodology developed by Amundi Quantitative Research2
(Pola [2013]) makes it possible to estimate the polarisation of a large number of
asset classes and therefore that of a portfolio, using three factors considered
to significantly affect their return: growth, inflation and stress. The polarisation
coefficient may be interpreted as the asset’s likelihood to perform favourably in
the event of an upward or downward variation in the factor.
When applied to alternative assets, whose performance has been simulated in
this context by that of the corresponding listed assets, this approach shows that
real estate, private equity and infrastructure are negatively affected by a decrease
in growth or increase in stress, thereby manifesting behavior similar to that of
equity. Their inflation polarisation, however, is less marked than for equity. In
fact, positive (though non significant) polarisation is observed for infrastructure
investments when inflation increases, a performance in sharp contrast with
that of equity. These asset classes therefore have beneficial properties for
protection from inflation, most likely due to their regular yield, which is generally
indexed to inflation.

2 See Gianni Pola: Rethinking strategic asset allocation in terms of diversification across
macroeconomic scenarios, Cross Asset Investment Strategy Special Focus, May 2013.

82 Amundi Investment Strategy Collected Research Papers


ASSET SEGMENTATION DEPENDING ON INVESTORS’ OBJECTIVES
Participation Generation Protection
Illiquidity
to economic of stable against
premium
growth revenues certain risks
PRIVATE ASSETS
Private Equity Strong No No Yes
Real Estate Moderate Significant Moderate Yes
Infrastructure Projects Significant Moderate Moderate Yes
Infrastructure Debt Moderate Significant Moderate Yes
Loans Significant Significant Moderate Yes
TRADITIONAL ASSETS
Developed market Equities Strong Moderate No No
Cash No Moderate Moderate No
Government bonds No Significant Moderate No
Indexed bonds No Significant Moderate No
Source : Amundi Research

2 - Return, Risk and Correlation


2.1 - A Disrupted Relationship between Risk and Return
In terms of methodology, the traditional asset allocation approach consists
of defining return, risk and correlation forecasts for the various asset classes
under consideration and of using an optimisation technique to propose a
portfolio structure that makes it possible to best meet the investor’s return or
risk objectives. The method that we recommend 3 for long-term strategic return
in the absence of market views relies on a constant Sharpe ratio hypothesis for
all assets included in the broad category. Studies conducted at Amundi on the
performance of asset classes (de Laguiche, Pola [2012]) have shown that a Sharpe
ratio of 0.3 is consistent with historic observations over very long periods of time.
This method, which does not involve any forecast exercises, has the advantage
of simplicity, because the desired return is simply the result of adding the risk-
free rate 4 to the ex ante volatility multiplied by the Sharpe ratio. In this context,
risk is represented by volatility. This simple relationship between volatility and
return is disrupted when it comes to alternative assets. First, for illiquid assets,
some of the additional return is compensation for greater liquidity risk, which
cannot be measured via volatility. Second, volatility is an imperfect measurement

3 On this subject, see G Pola and S de Laguiche:  "Forecasting returns on assets in an


environment of uncertainty", Amundi Cross Asset Investment Special Focus, January 2013.
4 See S. de Laguiche, "Risk free assets: whant long term normalized return?", Amundi
Discussion Paper, March 2014

S
Amundi Investment Strategy Collected Research Papers 83
of market risk, because especially in the world of hedge funds, the extreme risks
that are observed are substantially greater than those of conventional assets with
equivalent volatility, making the assumption of symmetrical returns required for
the estimate unsuitable. Performance-related fees also contribute to asymmetry
in the return distribution of these assets.

2.2 - Nature and Frequency of the Data Series


It is often difficult to obtain definitive price series in order to estimate the return
and risk indicators needed for this allocation work5. In the world of private equity,
observed historic returns may differ greatly depending on the sources used.
This is particularly true, as indicated in Phalippou and Gottschalg [2009], for the
difference that may exist between the valuation of the residual, still-unliquidated
portion of private equity funds and the amount that will actually be paid back
to investors when the fund is liquidated, which is the source of a temporary
improvement in the performance of the still-unliquidated funds. Additionally,
the sometimes substantial weight of management fees tied to performance –
particularly in the field of private equity – may alter the investment’s return profile
compared to shares held within the fund.
For hedge funds, the historical data is subject to biases related to how
performance is reported. The literature shows a survivorship bias: funds which,
due to insufficient performance, were liquidated do not always show up in the
numbers. The literature also mentions a “backfill bias” that leads managers to
report the performance of successful funds that, during their incubation period,
succeeded in getting good results, which were then retroactively included in the
reported history.
To take real estate as an example, significant differences exist between different
price indices, depending on whether they are calculated based on expert
assessments or transaction values 6. For example, it is possible to question the
ability of investors to actually sell their property at the market price estimated at a
given time by experts. Furthermore, indices that rely on transactions incorporate
the prices of transactions that were made over a given interval of time; however,
those prices have often been set in agreements that took place several months
before the transactions were finalised. In such a case, this type of index exhibits
delays and smoothing phenomena compared to what an exchange-traded price
would show.
If a decision is made, despite these limits, to use the conventional risk/return
framework, the question arises of how to determine return and risk of those
investments.
5 See in particular Cremers [2013] for a discussion of the limits inherent in the data for
alternative investments
6 On this subject, see S de Laguiche, A Russo, and C Blanchard: "Physical real estate in long
term asset allocation: the case of France", Amundi Discussion Paper, July 2014.

84 Amundi Investment Strategy Collected Research Papers


2.3 - Expected Return Estimation Methods
To forecast the strategic return of alternative assets, different estimation
methods may be considered:

Simple Extrapolation from Historical Returns 


This method is generally not recommended without assessing the usefulness of
historical data with respect to the anticipated context for the forecast period. The
price series must also match the assets included in or under consideration for
the investor’s portfolio. Finally, is the available data reliable and available across a
long enough time span that an average can be meaningfully calculated?

Relying on Expert Opinions 


Because alternative investments are heavily idiosyncratic by nature and there is
no obvious method for predicting their returns, why not rely on the estimates of
specialists in those asset classes? This solution has the advantage of operational
effectiveness in large asset management organisations that have multiple skills,
but it often poses some problems.
The overconfidence bias of the managers and investors who select them is well
documented and it may also be observed in the field of alternative investments.
Hedge fund or private equity fund managers sometimes anticipate returns of at
least 10% or even 15% per year, which are only observed over time for the best-
performing investment vehicles. This optimism may be because the estimate of
projected returns is tied to a core scenario that assumes the reasons that justify
the choice are borne out: acceptable timing and proper vehicle selection. To get a
genuine expected return in the statistical sense of the term, the forecast should be
corrected by taking into account alternative scenarios along with the likelihood of
their occurring. In private equity, however, the literature documents a certain degree
of persistence in manager performance, but it is not always possible to put to use,
as the best private equity managers cannot keep increasing their assets under
management indefinitely. Within this class, it should also be noted that managers
often go by the IRR of equity investments, while the return achieved by the investor
must also take into account fees and “cash drag”, meaning the negative impact of
private equity funds holding cash during the investment period.
The graph below therefore illustrates the difference between the return of equity
investments and the return (IRR) for the investor. It is assumed that the fund
gradually invests the initially committed capital in equity over five years.
The IRR is calculated after the cash drag effect (due to the fact that the funds
raised from investors are invested in shares only gradually) and fees (2% fixed fees
+ variable fees equal to 20% of the stocks’ performance with a 7% “hurdle”). The
return of cash in this simulation is assumed to be equal to 1%.

Amundi Investment Strategy Collected Research Papers 85


3- IRR or in estor in PE und as a unction o Equit IRR
25

20

15
IRR PE und

10 IRR PE fund
5 Equity IRR

0
-10 -5 0 5 10 15 20 25
-5

-10
IRR equit Source: Amundi Research

Note that, particularly in the event that the return of the equities is high, the cash
drag and more importantly the variable management fees substantially reduce
the investors’ share of that profit.
Lastly, before using these return forecasts in a risk/return optimisation, we must
check their consistency with those of conventional listed asset classes and based
on the risk adjusted return used in the optimisation. Otherwise, in the absence of
a uniform method for estimating returns among different asset classes, experts’
degree of optimism or realism will constitute the primary foundation for the asset
classes’ apparent attractiveness.

Defining Normalized Returns


Recommended by some major institutional investors, the opportunity cost model7
consists of representing an alternative investment as the sum of fundamental
building blocks of exposure to traditional asset classes, justifying diversification into
those assets based on the surplus return anticipated from them. As an example,
based on the segmentation method mentioned in section 1-3, the alternative
assets may be represented as follows:
t Hedge funds: by a dynamic composition over time of elementary betas in
the main stock, bond, currency and volatility indices (Lezmi et al [2013]).
However, this approximation is more useful for indices that are representative
of alternative strategies but will rarely lead to stable betas on the individual
hedge fund level, for which the alpha component is deemed to be of utmost
importance.
t Private equity: by equity (with a beta generally greater than 1, particularly for
its riskiest components, such as venture capital), corrected by style effects
- size and value - or sectoral effects.
7 On this subject, see Sung C. C.: Trends in Sovereign Wealth Management: Emergence of
New Investment Models, Presentation to Nomura Central Bank Seminar, April 2014

86 Amundi Investment Strategy Collected Research Papers


t Real estate: by an equity component – the strategic return of real estate
investments can actually be considered to be tied to the long-term growth
of the underlying economy - and an interest rate component tied to
rental yields. If we follow the logic of net return for the investor, we must
remember to deduct estimated fees for maintaining and managing the
assets from these standard expected returns.
t Infrastructure investments: by an equity component, an indexed bond
component - representative of the actual return offered by the asset class
- and fixed-rate financing.
t Loans: by high-yield bonds and a base effect that includes a spread
differential and an illiquidity premium.
An illiquidity premium may also be incorporated into the expected return of real
estate, infrastructure, or private equity. Quantifying this premium, however, is
complicated and has been studied in the financial literature, which goes beyond
the scope of this article (Ang et al, De Jong et al, Kinlaw et al, Phalippou et al).

2.4 Estimating Volatilities and Correlations


Volatility and correlation is another area where different approaches may
be considered.

Using Return Series Related To Listed Alternative Assets


The volatilities and correlations of alternative assets can be approximated
using return series related to listed alternative assets 8. This is a rough estimate of
course, particularly for real estate, in which the performance of listed securities
is often much closer to the performance of equity in general than to physical
assets. However, if these investments represent only a small percentage of the
investor’s allocation, the impact of this approximation on the risk parameters of
the portfolio as a whole will remain slight. When estimating long-term risk and
correlations, it is also necessary to reprocess the data, either by “unsmoothing”
it (the apparent volatilities of alternative classes calculated based on low
frequencies underestimate the actual risk) or by taking into account time lags
and autocorrelations.
The chosen valuation frequency also influences the observed annualised volatility,
as shown in the graph below, which depicts a much larger volatility gap between
real estate securities and physical real estate – measured by the CBRE index –
when the frequency of the observations is quarterly rather than annual.

8 Gilfedder and Sheperd show the usefulness of looking at similar listed securities to
represent the risk of infrastructure investments

Amundi Investment Strategy Collected Research Papers 87


provided by the EIOPA 10, it is possible, at least for real estate, which is one of the
risk pillars adopted by insurance regulations, to estimate its correlation with the
asset classes included in the investor’s portfolio, by breaking them down into risk
pillars (interest rates, equity, real estate, spread, currency) and by using those
pillars’ variances and correlations.

ESTIMATED
EIOPA CORRELATION MATRIX ASSET CORRELATIONS
WITH REAL ESTATE

Correlation
Interest Real
Equity Spread Currency with real
rate estate
estate

Interest rate 1 0 0 0 0,25 Cash 0

Government bonds
Equity 0 1 0,75 0,75 0,25 0
Eurozone

Credit IG bonds
Real estate 0 0,75 1 0,5 25 0,43
Eurozone

High yield bonds


Spread 0 0,75 0,5 1 0,25 0,49
short duration

Emerging bonds
Currency 0,25 0,25 0,25 0,25 1 0,47
hard currency

Emerging local debt 0,27

Equities 0,75
Source: EIOPA, Amundi Insurance Solutions

– Finally, this analysis may be conducted in view of the specific features of


the investments made or planned in the institution’s portfolio. It is therefore
necessary whenever possible to use “bottom-up” thinking based on
the characteristics of the individual investments in question. The
main providers of portfolio risk analysis models also offer increasingly finer
segmentation of alternative investments and a granular breakdown of their
risks.
All of this information allows us to conclude that despite the many methodological
pitfalls involved in estimating profitability, risk and correlations of alternative asset
classes, it is possible, by combining different methods, to provide an order of
magnitude estimate for the risk presented by alternative assets and their
correlation with traditional assets.

10 European Insurance and Occupational Pensions Authority

88 Amundi Investment Strategy Collected Research Papers


3 - Inserting Alternative Investments into an Allocation:
Amundi’s Recommendations
Amundi’s analyses of alternative asset classes and whether to include them in
the portfolio have allowed us to outline an allocation philosophy for such assets
and to provide preliminary answers to the key questions that investors ask: how
to segment these assets; what criteria to use to determine a portfolio’s exposure
to these assets; and how to formally set out these exposures.

3.1 Recommended segmentation method


First of all, the institution must take care to define how each so-called alternative
asset meets its objectives and fits into its investment philosophy. It must also
rely on segmenting its investment universe using an approach that we believe is
founded on analysing the performance of alternative asset classes based on
macroeconomic and financial factors. At all times, the investor must be aware
that the performance of private equity, real estate and a large share of hedge
funds is closer to the performance of equities than that of bonds, particularly in
terms of sensitivity to growth and market stress. Meanwhile, loans, infrastructure
debt and credit arbitrage strategies are similar to high-yield bond assets, which
are generally classified as bonds even though they are sensitive to the business
and credit cycle and therefore also economically similar to equities.
Regulation may also guide segmentation choices. As already mentioned, the
Solvency II directive distinguishes real estate risk from interest rate, credit and
equity risks and considers all non-”look through” vehicles, such as private equity
funds and hedge funds, to be equity risks. Furthermore, certain regulations make
a distinction between investment-grade bonds, equities and real estate; all other
assets fall into a leftover category (sometimes called “junk”) whose weight is
generally capped in the investor’s total assets.
It should be noted, however, that this sort of detailed segmentation work is only
helpful if the alternative investments form a non-negligible share of the investor’s
portfolio (at least 5%, for example).

3.2 The Importance of Institution-Specific Factors


We feel that determining the weight of alternative investments in a portfolio
must rely more on a qualitative judgment than on a traditional mean/variance
optimisation approach11, which has the shortcoming of a large margin of error
due to uncertainties in those assets’ risk and correlation parameters. We suggest
incorporating the following factors into this analysis. Many are specific to each
institution and allow the response to be adapted to its own particular environment.

11 The importance of including judgment in assessing the risk of alternative investments is


also emphasized in Pedersen et al [2014]

Amundi Investment Strategy Collected Research Papers 89


Investment Horizon and Liability-Related Constraints
Taking the institution’s investment horizon into account as well as its liability-
related constraints helps ensure that there is no gap between the asset’s liquidity
and the schedule of disbursements (Buscombe, McNally [2014]). This horizon
must be long enough - at least 10 years - to justify a substantial allocation - above
5% - to such assets. The analysis of any liquidity gaps must also incorporate
known commitments and contingent ones, such as the possibility that a sovereign
wealth fund might need to support public spending or the underlying economy’s
banking sector in the event of a crisis.

Asset Allocation Structure


The investor’s asset allocation structure is also a decisive factor. Analysing
correlations between alternative and traditional assets mentioned in section 2-4
helps distinguish the impact of adding these assets to a portfolio based on its
initial allocation between equities and bonds.
In the specific case of physical real estate, we have observed that this
asset class provides better diversification in bond-heavy than stock-heavy
portfolios, due to a near-zero correlation between real estate and bonds (though it
is positive with stocks). The resulting diversifying power, however, depends on the
horizon in question: it is higher over a short horizon (particularly quarterly) owing
to the observed risk smoothing effect over the short term. On the other hand,
physical real estate appears to be less diversifying in a stock-heavy portfolio, due
to the tendency of investors to underestimate the correlation between real estate
and equities. This comment also holds true for private equity, whose systemic risk
is clearly consistent with that of publicly traded stocks.

Suitability for Alternative Investments


Although this criterion appears unscientific, the institution’s suitability for the
alternative assets – in particular, how readily those investments will be accepted
by the administrators – and the usefulness of those assets to its investment
philosophy, cannot be ignored. For this reason, a diversification using physical real
estate (or private equity, respectively) will be more natural for institutions involved
in the construction sector (or technology sectors, respectively).

Available Resources for Making and Managing These Investments


Finally, in practical terms, the specifics and complexity of these assets require
investors to ensure that they have the necessary means to select the most
attractive investments that meet their specifications and to manage them over
time. Some studies12 have shown that in private equity, experienced investors are

12 See in particular Dyck and Pomorski [2011]

90 Amundi Investment Strategy Collected Research Papers


the ones who add value, while newcomers must expect disappointment while
learning the ropes of these asset classes.

3.3 How Can You Formally Define Exposure to These Assets?


Target levels may be formally defined for these assets in terms of risk-
adjusted weight ranges. Recent work summarised in Ang and Sorensen show
that their illiquidity needs to be addressed by defining ranges rather than targets.
An expression in terms of weight alone is of little use because the level of risk may
vary greatly depending on the type of product.
The definition of these targets must also rely on the impact of the alternative asset
component on the portfolio’s total risk. Depending on the segmentation system
adopted, it may be necessary to also take into account the risk contribution related
to a more focused risk source - financial risk, overall alternative risk, equity risk
- and to take into account the specific risk characteristics of alternative assets
included in the allocation.
With regard to risk parameters, we recommend going by VaR or probable
maximum loss over a long enough horizon (one year), an indicator that may also
apply to the other asset classes included in the investor’s portfolio and therefore
sufficient to uniformly analyse the risk for the asset portfolio as a whole. As already
mentioned, volatility is not an appropriate risk measure for these alternative asset
classes due to liquidity risk and non-normal return distribution.
Correlation estimates between assets are necessary to assess risk, but in this
case it is possible:
– to use whenever feasible (for real estate) correlations with other major asset
classes defined by Solvency II regulations
– to adopt correlations with equities of close to 1: the resulting overestimation of
risk is acceptable for bond-heavy portfolios and when the weight allocated to
alternative assets remains limited.
The allocation’s risk may also be analysed using stress tests that model
unfavourable market conditions, including a risk scenario that is qualified (for
example, a premature increase in US interest rates) and quantified in its impact on
major asset classes.
It may be desirable to assess risk over different horizons. This is because the
relationship between risk and horizon is different from the one that exists for listed
assets: given equivalent long-term risk, short-term volatility is limited by a smoother
evaluation mode than for listed assets. This advantage is counterbalanced by lower
short-term liquidity. Rebalancing from listed to unlisted assets may also be justified
for investors who have very long horizons but are nonetheless subject to high
short-term volatility constraints.

Amundi Investment Strategy Collected Research Papers 91


3.4 Ramp-Up Timing
The matter of timing cannot be ignored. For these assets, it is necessary to
be a trailblazer and have the courage to invest when it is uncomfortable rather
than when the targeted asset is already being sought out by a large number of
investors. Additionally, particularly for hedge funds, the best asset managers have
a limited investment capability that often benefits only those first to the table; in
a market boom, increased demand from less experienced investors can only be
invested in lower-quality funds. This is also true of the best-known private equity
funds, which larger investors have easier access to. However, the difference in
performance between the top and bottom funds is extremely high in this sector
(Jagger [2012]) and only by being reasonably sure that one is able to choose a
fund that falls within the top quartile can better performance be generated than
by investing in liquid assets (see below).

DISPERSION OF ACTIVE MANAGEMENT


Performances over 20 years 1982-2012

Asset class Top quartile Median Bottom quartile Delta


Global bonds 7,3% 6,5% 6,3% 1%
Global Equities 10% 8% 6,1% 3,9%
Private Equity 12,7% 3,1% (2,29)% 15,6%
Source: Thomson One, Amundi Research

Consequently, for an institution that initiates investments in these asset


classes, the target ranges may differ depending on the horizon, with a
distinction between the ramp up period and permanent regime. For this reason,
small as they may be during the investor’s learning phase, the investments must
ultimately be large enough to have a real impact on the expected return of the
investor’s portfolio and justify the increased complexity of monitoring them.
These investments may only be ramped up gradually, taking into account:
– market access (the size of the “investable” universe)
– the liquidity of the asset classes considered relative to the size of the investor’s
portfolio.
The ramp-up timing of these assets also depends on more tactical considerations,
such as their position in the business and financial cycle and their valuation.

92 Amundi Investment Strategy Collected Research Papers


Conclusion
In the current context of historically low interest rates and persistent fears regarding
equity investments, institutional investors are demonstrating an increasing appetite
for alternative investments, a disparate group that includes hedge funds and
unlisted assets: real estate, private equity, infrastructure and loans. This appetite
relies on their ability to diversify away from traditional asset classes and to improve
return for investors who are able to select the top-performing vehicles and have
a long enough investment horizon to capture the illiquidity premium of unlisted
assets.
The size of these investments within a portfolio raises many questions for which
this article provides some initial insights.
We have highlighted the numerous methodological difficulties that arise in applying
a traditional mean/variance asset allocation approach to these assets and which
particularly rely on the frequency and specificity of the ways in which unlisted
assets are valued.
The limited usefulness of applying the notion of volatility to these asset classes has
led us to recommend that their risk and correlation with other asset classes be
estimated through a combination of various methods: use of representative listed
assets, regulatory parameters and probable maximum loss in stress scenarios.
The institution must also ensure that the illiquidity of these investments does not
generate an unacceptable liquidity gap between assets and liabilities.
More generally speaking, the investor must adopt a pragmatic approach to
determining how much of its portfolio is allocated to these assets, by clearly
defining the economic objective intended with each of these investments and by
identifying their sensitivity to key macroeconomic and financial factors. Finally, we
recommend expressing the allocation to these assets within the portfolio in terms
of a target risk range rather than in terms of weight and to manage a gradual ramp-
up to that target, taking into account investment vehicles available for accessing
those assets.
The various options proposed in this paper should be explored further in multiple
directions. How can the illiquidity premium be integrated into the expected return of
unlisted alternative investments? How can liquidity variations be taken into account
in estimating risk and correlations? The breakdown of these assets in terms of their
exposure to macroeconomic or financial risk factors may also be refined. Finally,
the risk/return analysis of these investments must be continued for homogeneous
subsets, segmented by geographic area and the purpose of those assets (for
example offices/housing/retail for real estate). We will have the opportunity to
report on these subjects in the coming months.

Amundi Investment Strategy Collected Research Papers 93


Bibliography

Ang A., Kaplan A., Columbia Business School and NBER, Estimating Private Equity Returns
from Limited Partner Cash Flows, Netspar, January 2014
Ang A., Sorensen M., Risk, returns and optimal holdings of private equity : a survey of
existing approaches, Netspar panel paper 39
Buscombe T., McNally S., Mercer, Spending your Illiquidity Budget, Presentation to Global
Investment Forums, 2014
Cremers M., The Performance of Direct Investments in real Assets : Natural Resources,
Infrastructure and Commercial Real-Estate, Deutsche Asset and Wealth Management Global
Financial Institute, June 2013
De Jong F., Driessen J., The Norwegian Government Pension Fund’s Potential for capturing
illiquidity premiums, Report to the Norwegian Ministry of Finance, February 2013
De Laguiche S., Blanchard C., Russo A., Physical real estate in long-term asset allocation:
the case of France, Amundi Discussion Paper, September 2014
De Laguiche S., Pola G., Unexpected Returns : Methodological Considerations on Expected
Returns in uncertainty, Amundi Working Paper 032-2012, November 2012
Gilfedder N., Shepard P., Stocks, Bonds and Airports : Infrastructure Assets in Pension
Plan Portfolios, MSCI Market Insight, January 2014
Government of Singapore Investment Corporation, GIC’s New Investment Framework,
2012-2013 Annual report
Hauss H. The Role of International Property Investments in the Global Asset allocation
Process, University of South Australia
Jagger S, A few caveats attached to private equity statistics, FT, October 2012
Kinlaw W., Kritzman M.,Turkington D., Liquidity and Portfolio Choice : A Unified Approach,
The Journal of Portfolio Management, Winter 2013
Lezmi E., Lou K., Zhao S., Hedge Funds Replication : Factor models, Amundi internal
document, November 2013
Pedersen N., Page S., He F. 2014, Asset Allocation : Risk Models for Alternative Investments,
Financial Analysts Journal, May-June
Phalippou L, Gottschalg O, The performance of private equity funds, The Review of
financial studies vol 22 issue 4, April 2009
Pola G., Managing uncertainty with dams. Asset segmentation in response to macroeconomic
changes, Amundi Working paper, May 2013
Santiso J., Sovereign Wealth Funds, 2013 Report, ESADE Business School
Santiso J., Sovereign Funds 3.0, Asset allocation strategies, 2010-2020, Report to Amundi
Shephard P and Liu Y, The Barra Private Equity model, MSCI Research note, August 2014
Sung C. C., Trends in Sovereign Wealth Management : Emergence of New Investment
Models, presentation to Nomura Central Bank Seminar, April 2014

94 Amundi Investment Strategy Collected Research Papers


Acknowledgements

We would like to thank all of our colleagues at Amundi, in particular from Amundi
Insurance Solutions, Amundi Real Estate and Amundi Private Equity, whose insights
helped enhance these analyses. Bernard Arock’s comments on Private Equity were
particularly useful to us in fleshing out the sections related to such assets.

Amundi Investment Strategy Collected Research Papers 95


96 Amundi Investment Strategy Collected Research Papers
DP-07

The short investment cycle:


our roadmap
ÉRIC MIJOT,
Head of Strategy Research, Amundi

October 2014

This article examines the short investment cycle. It is based on


our historical observations and asset management experience.
This approach complements the one described in our Discussion
Paper entitled “Long Cycles and the Asset Markets.” It serves as
a guide to interpreting the markets and is regularly presented at
Investment Committee meetings.
Like long cycles, short cycles can be broken down into four
phases which we describe while underscoring trends in assets,
the indicators preceding each phase, the most relevant forms of
analysis, smart strategies and a preliminary approach to asset
allocation. We demonstrate that these trends were visible in
recent and supposedly atypical cycles.
We show that monetary policy is very instructive, but also
that analysis of stock prices is a highly valuable contribution.
Finally, in order to navigate efficiently in a short cycle and
seize opportunities, we focus primarily on consistent trends
between and within asset classes and on the complementary
characteristics of different forms of analysis.

Amundi Investment Strategy Collected Research Papers 97


THE SHORT INVESTMENT CYCLE:
OUR ROADMAP

I. The four phases of the investment cycle


An economy’s ability to grow depends on the growth of the population of working
age and the productivity gains that it generates. During the course of a cycle,
economic growth fluctuates around this trend, which corresponds to its potential.
We can therefore distinguish 4 phases of the economic cycle: i) At the beginning,
growth is lower than its potential and inflationary risks are small; ii) Subsequently,
growth accelerates and exceeds its potential, which generates inflationary
tensions; iii) Then growth decelerates while at the same time remaining above its
potential; inflationary tensions appear since they are a belated function of growth;
iv) Finally, growth moves back below its potential and inflationary tensions end up
disappearing.
The difference in economic growth in relation to its potential (Output Gap) is
obviously very closely monitored by central banks, whose role is precisely to
ensure long-term economic stability. In practice, for the investor it is easier to
monitor the trend in the interest rates of central banks (and more generally their
monetary policy) than the Output Gap itself. The central bank adopts a tougher
monetary policy when inflationary risks increase and relaxes monetary policy when
inflationary tensions decline and recession risks emerge (see diagram below).

Markets are mainly influenced by…


Economic growth Decline in interest rates

Phase ii Phase iii


GDP
trend
Phase i Phase iv
GDP

Increase in Central Bank rates


Decline in Central Bank rates
Source: Amundi Research

98 Amundi Investment Strategy Collected Research Papers


During the first 2 phases, the trend in assets is influenced positively by the growth of
profits whereas interest rates gradually act as a brake. Conversely, during the last 2
phases, investors cling to a decline in interest rates and the hopes of an economic
recovery that they generate while corporate profits are given a rough ride.

The short investment cycle

The Asset cycle

Sell Stocks Sell Commodities


Æ Æ

Sell Bonds Cash


Æ É
Phase ii Phase iii
GDP
trend
Å
Phase i Phase iv Å
Buy Buy
GDP Commodities
Bonds Å
Buy Stocks

Sectors

Energy Non cyclical


and Basic Resources consumption

Investment Telecommunication
and Technology Phase ii Phase iii and Public Utilities
GDP
trend
Phase i Phase iv
GDP
Cyclical Banking
consumption and Insurance

Strategies

Relative Value Secure Return

Phase ii Phase iii


GDP
trend
Phase i Phase iv
GDP

Contrarian Protection

Increase in Central Bank rates


Decline in Central Bank rates

Source: Amundi Research

Let us now break down these four phases more finely, highlighting for each of them:
the market behaviour, the indicators that prefigure this phase, the most appropriate
forms of analysis, opportune investment strategies and an initial asset allocation
approach.

Amundi Investment Strategy Collected Research Papers 99


1.1 Phase i:
Equity outperformance
The first phase is that of an upturn in the cycle. Volatility, a fearsome short-term
enemy of the investor, is very strong. Risk is high but tremendous opportunities
emerge.

T Asset performance
This initial phase of the cycle is favourable to equities generally and particularly to
pro-cyclical stocks and markets. With the worst having been avoided in economic
terms, we are ready to take more risks and even to invest in less liquid assets.
After collapsing, small-cap stocks, for example, now start to soar. In terms of
sectors, sectors that are the most sensitive to a decline in interest rates and a
decline in energy prices, now on a real downtrend, are the first to pick up. This is
the case of the banks and more generally financials, stocks related to residential
construction, transport and automobile stocks.
Corporate bonds, which react positively to the reduction in the default risk, rally
before equities, which prevail in the end due to hopes of a resumption in growth.
Government bonds underperform equities, even if their prices do not necessarily
decline. Commodities, which are very volatile, have sometimes reached their low
point before equities but are generally still very risky.

T Indicators
At this stage, the flow of bad news is at its peak. Companies, battling with
recession, deliver cautious or even gloomy messages on their prospects. Financial
analysts therefore downgrade their company earnings forecasts.
Investor sentiment in surveys is at its most pessimistic. Individual investors throw
in the towel and sell their shares to well-informed professionals who decide to look
beyond the valley which is now not very far away. We notice, consistently over
time, that the low point of the US equity market is situated two quarters before the
low point of economic growth, as defined by the NBER. However, unfortunately
we notice it a posteriori! Investor doubts, related to the previous crisis (excessive
debt, overinvestment, overstocking, etc.) end up clearing at the same time as
the rise in share prices, and subsequently the gradual improvement in company
results. We therefore say that the market climbs “the wall of worries”.
The decline in the central bank’s interest rates accelerates. The decline in short
rates is now faster than for long rates. The steepening of the yield curve enables
banks, which finance a large part of the economy, to recover their health. This
is essential to ensure the sound and sustainable resumption of a new growth
cycle. If short rates go below the level of inflation, monetary policy becomes very
“accommodative” and that generally constitutes a good signal to return to equities.
The pick-up in bank loans constitutes confirmation, therefore a posteriori, that

100 Amundi Investment Strategy Collected Research Papers


the recovery is sustainable. A simple positive shift in the deterioration of loan
volume (Credit Impulse) is sufficient for the equity markets to endorse it. There is a
reduced decline in the number of building permits and new housing starts, which
are the most leading indicators of the economic cycle.

Decline in central bank rates


and reaction of the euro zone’s equity market

Datastream index of euro zone equities


(base 100 in 1973)

3200

1600

800

400

200

100

ECB/Bundesbank rates (as a %)

10
9
8
7
6
5
4
3
2
1
0
Oct. 1974 Oct. 1982 Feb. 1993 Sept. 2001 Nov. 2008

ŷŷ Downturn in the rate of the ECB / Bundesbank


Source: Amundi Research

The accelerated decline in interest rates is a powerful buy signal with regard to
equity markets, especially when the decline is synchronised at a global level.

Amundi Investment Strategy Collected Research Papers 101


the recovery is sustainable. A simple positive shift in the deterioration of loan
volume (Credit Impulse) is sufficient for the equity markets to endorse it. There is a
reduced decline in the number of building permits and new housing starts, which
are the most leading indicators of the economic cycle.

Decline in central bank rates


and reaction of the euro zone’s equity market

Datastream index of euro zone equities


(base 100 in 1973)

3200

1600

800

400

200

100

ECB/Bundesbank rates (as a %)

10
9
8
7
6
5
4
3
2
1
0
Oct. 1974 Oct. 1982 Feb. 1993 Sept. 2001 Nov. 2008

ŷŷ Downturn in the rate of the ECB / Bundesbank


Source: Amundi Research

The accelerated decline in interest rates is a powerful buy signal with regard to
equity markets, especially when the decline is synchronised at a global level.

102 Amundi Investment Strategy Collected Research Papers


Another point of reference in a globalised world: economies influence each other.
The United States, the world’s leading economy, has been systematically ahead
of Europe since the 1960s except for the period of German reunification. The
decline in the US Federal Reserve’s interest rates occurs around 2 quarters
before the decline implemented by the European Central Bank, the ECB, and
historically the Bundesbank. Therefore, when the ECB joins its forces to combat
the global recession (or the slowdown), the equity markets are ready to react
rapidly upwards. This illustrates that the synchronised accelerated decline in
interest rates at a global level is a powerful buy signal with regard to equity
markets (see graph).

T Relevance of analyses
In this phase, the market is more governed by the psychology of investors than
by the fundamentals of economies. Consequently, the technical analysis will be
essential to pinpoint a reversal pattern in equity markets and therefore help in the
timing for re-weighting them. Stocks that rise fastest are also initially simply those
that have fallen the most previously (hedging of short positions), whatever their
quality. Financial analysis will gradually become more relevant for pinpointing
deep value stocks that the patient investor can include in their portfolio. On the
other hand, quantitative analysis is generally ineffective in this market phase,
which is difficult to pinpoint through modelling.

T Strategies within asset classes


In terms of strategy, it is a phase during which investors have to be contrarian.
Directional strategies are gradually implemented on equities; in other words, this
involves re-investing cash. Systematic factors (beta) are more important than
specific factors (alpha). In fixed income markets, it’s time to bet on a decline in
the risk premiums of bonds related to companies or governments with weaker
fundamentals (spread strategies). On currencies, a pro-cyclical approach is
required, preferring those associated with more reactive economies.

T Asset allocation
In terms of asset allocation, an allocation predominantly invested in equities is
essential. Indeed, most of the performance of equities takes place during the
rebound phase. Given the strong volatility, we will gradually switch to equities
when the reversal pattern is validated. During the construction phase of this
pattern we will tend to be invested evenly between equities and fixed income
products. Credit, which is inexpensive (the spreads are very high) and less volatile
than equities is a good way of starting to re-weight portfolio risk.

Amundi Investment Strategy Collected Research Papers 103


1.2 Phase ii:
Outperformance of equities and commodities
The second phase corresponds to core growth. It is an extension of the previous
phase. There is no clear rupture. Risk aversion seems to have disappeared and
volatility declines (or at least does not increase). Risky assets all rise and the
differences in performance between them diminish.

T Asset performance
This phase is still favourable for equities. The rise becomes widespread. There are
fewer and fewer discounted small-cap stocks. While their margins have already
been rebuilt, companies are also more confident and will have to invest in order
to, in the future, satisfy this demand, which is in the process of exceeding their
production capacity. As a result, investment-related stocks gain the upper hand
over those related to consumption. Those that have the best pricing power, such
as luxury goods stocks, remain in the running longer. Finally, mining companies and
more generally companies related to raw materials end up outperforming due to the
general increase in commodity prices.
With economic growth now exceeding its potential, demand for commodities exceeds
supply and they even end up doing better than equities. Within commodities, oil
and pro-cyclical industrial metals such as copper, zinc and nickel perform better
than gold (except in cases of hyperinflation). Among the precious metals, silver
outperforms gold.
Generally, safe haven stocks underperform. This is the worst phase for government
bonds. Within fixed income markets, corporate bonds with higher yields remain
attractive longer while company default rates tend to diminish. The pick-up in
corporate debt to finance investment or mergers and acquisitions increases paper
supply but the very strong demand from yield-seeking investors results in a maximum
reduction in spreads. The spreads of Investment Grade credit are the first to shrink,
in parallel with bond volatility. The spreads of high-yield bonds tend rather to reflect
equity volatility, which also remains contained.

T Indicators
The Fed ends up raising its rates, always after the unemployment rate has reached
its peak and before corporate margins have finished increasing. Since 1950, the
median top margin occurs 18 months before the top of the economic cycle (see
graph and table).
The accelerated rise in bond yields anticipates the reversal in monetary policy and
the outperformance of industrial commodities in relation to government bonds
constitutes an excellent confirmation of the transition to phase ii. This is an initial
major warning sign for risky assets. However, there is no need to worry unduly. It
is not a sufficient obstacle to halt the rise of equities, at least apart from extreme
periods of hyperinflation or deflation.

104 Amundi Investment Strategy Collected Research Papers


In phase ii, profits growth is the most important factor. The messages from companies
are encouraging and financial analysts systematically upgrade profit forecasts.
Confidence is self-sustaining with the rise in prices and the sentiment of profits that
are easy to achieve. The press echoes this sentiment and through unconscious
imitation the least well-informed investors return to the market. Low volatility also
flatters the risk budgets of pension funds and insurers, which sometimes allow
themselves to be belatedly tempted. Equities end up anticipating the top of the cycle,
but only 3 months in advance.

T Relevance of analyses
As for the forms of analyses, they all function well in this phase. Financial analysis
is very relevant to forecast profits growth and provide a fundamental valuation of
companies, which also helps to further reassure investors. Quantitative analysis,
which systematises the fundamental approach of financial analysts also comes into
its own and is at its height. Often, a few well calibrated ratios are sufficient to pinpoint
cheap stocks. Technical analysis, which excellently captures the momentum of
prices, also works very well. The leitmotif of technical analysts therefore becomes
“the highest highs are bullish”.

T Strategies within asset classes


Initially, a simple but effective strategy for all asset classes consists in monitoring
the momentum of prices; in other words, purchasing what is in the process of rising.
However, very quickly “relative value” strategies are required. The rebound phase
is now behind us and it is necessary to be increasingly subtle in order to maximise
performance; risk becomes more specific than systematic. The absolute performance
of equities is not as strong as in the previous phase and it becomes more relevant to
focus on the relative performances of one sector in relation to another for example.
In terms of exchange rates, which are traded in relative terms by definition, the well-
informed investor can capitalise on the substantial decline in volatility to put in place
carry trade strategies; i.e. purchasing currencies at very high interest rates and
financing them through the sale of currencies at very low interest rates. This makes
it possible to play on the economic time-lags between countries more directly than
between their equity markets, since the latter often move in the opposite direction
to their currency, which is likely to wipe out part of the gain that is now smaller if the
forex risk is not hedged.
Finally, in fixed income markets, spread strategies can be retained and yield curve
flattening strategies, preferring short maturities to long maturities (bear flattening),
can be initiated.

T Asset allocation
In terms of asset allocation, bonds can make way for commodities. Since commodities
are very volatile, they will nevertheless be contented with a smaller percentage.

Amundi Investment Strategy Collected Research Papers 105


Margins of companies, equity markets and anticipation of recessions
19200 25%
23%
9600
21%
4800 19%

2400 17%
15%
1200 13%
600 11%
9%
300
7%
150 5%
1950

1953

1957

1960

1963

1967

1970

1974

1977

1980

1984

1987

1991

1994

1998

2001

2004

2008

2011
DOW JONES Gross margin Recession
(scale. L) (scale. R) Source: Datastream, Amundi Research

Duration in months between the tops


Top Margin Top Dow Top NBER Margin-NBER Dow-NBER
Q4 1950 Q1 1953 Q3 1953 11 2
Q1 1955 Q4 1956 Q3 1957 10 3
Q2 1959 Q1 1960 Q2 1960 4 1
Q1 1966 Q1 1969 Q4 1969 15 3
Q1 1973 Q1 1973 Q4 1973 3 3
Q1 1978 Q4 1980 Q1 1980 5 1
Q4 1980 Q2 1981 Q3 1981 3 1
Q4 1988 Q3 1990 Q3 1990 7 0
Q3 1997 Q1 2000 Q1 2001 14 0
Q3 2006 Q4 2007 Q4 2007 5 0
Median 6 1

The median companies’ margin decline occurs 18 months before the recession.
When margins decline, the equity market continues to rise. It ends up capitulating
only one quarter before the recession.

1.3 Phase iii:


Outperformance of the money market and inflation-indexed bonds
This third phase prepares for the reversal of the cycle which is mature. Volatility is
extremely low. Everything apparently seems to be as previously, and yet the crisis
is brewing. The rise in equities is less and less consistent. Performance dispersion
reduces considerably.

T Asset performance
Growth is still apparently abundant but valuations tend to converge, whatever the risk
of the underlying assets. The performance of equities in relation to volatility begins
to decline. Small-cap stocks have more difficulty than large-cap stocks. In terms
of sector, the leading sectors of this cycle no longer outperform; more defensive

106 Amundi Investment Strategy Collected Research Papers


sectors make up part of their time-lag. They start to outperform even before the
crisis begins. The sectors involved are food processing, pharmaceuticals as well
as telecommunications and services utilities. The same applies to geographical
regions; for example, the outperformance of European equities, which are more
cyclical, in relation to US equities diminishes.
Central bank rates have now reached their cyclical high. They will generally be
maintained on a plateau for several months. It is therefore at that point that money
market rates are the most attractive in the cycle. The yield curve is flat, or even
inverted. Government bond yields are more hesitant. Company default rates,
a belated function of the cycle, are at their lowest for several years, as are the
spreads between high-yield bonds and government bonds.
Meanwhile, commodities continue to rise for a little while and fuel inflationary fears
before collapsing; inflation-indexed bonds (linkers) outperform equity indices.
This preparation phase for the correction of the cycle can very well take place while
the equity markets are still rising strongly, which obscures the picture even more. It is
also in this phase that bubbles are formed (except in 1987 where the crash occurred
in phase ii). Whereas the valuation is already expensive and the cycle is coming to an
end, flows are faster than profits growth. They initially focus on an asset that fires the
imagination (internet stocks in 1998-2000) and that is not necessarily very liquid (a lot
of new stock market listings) before extending to the rest of the market.

T Indicators
In addition to this symptomatic behaviour which raises questions, it is worth noting
that investors are often hypnotised by the very low volatility and think that the growth
of the last few months is sustainable; it is the calm before the storm. Optimism is at
its height. We therefore hear comments that “this time is different!”. However, if the
central bank stops raising its rates, it undoubtedly has good reason for doing so,
even if it is not infallible. It is a difficult phase for the central bank as well: inflation,
which is a belated function of economic growth, continues to rise and deserves
all its attention. Conversely, the central bank risks causing a recession if it waits
too long to lower its guard. Moreover, the rate of growth of bank loans is already
starting to fall.

T Relevance of analyses
At this stage, financial and quantitative analyses become less relevant. Companies
rarely see the economic downturn looming and guide analysts towards a continuing
increase in profits. The consensus adopts these data and the quantitative analysts
that refer to it often get caught by the downturn that will follow. Technical analysts
that are too focused on the continuing momentum are also partially blinded, but
others notice signs of trend exhaustion and securities’ less consistent participation
in the rise and then gradually detect reversal patterns: double or triple top or even
a head and shoulders.

Amundi Investment Strategy Collected Research Papers 107


T Strategies within asset classes
In strategy terms, the secure return needs to be given priority. In fixed income markets,
high-yield bonds should be sold and money market assets and government bonds
reinforced, even if the latter may temporarily be less attractive than money market
assets. Inflation-indexed government bonds are a good compromise. Equities
should be sold in order to generate liquid assets and take up positions in defensive
sectors. In terms of currencies, priority should be given to those whose underlying
economies exhibit the most solid fundamentals: positive current account, positive
net long-term investment flows, low total debt to GDP, low external debt, low short-
term debt with regard to forex reserves, etc. Finally, commodities will experience a
sharp reversal during phase iv. It is better to have sold them prematurely since the
decline will be sudden and will not leave sufficient time to react.

T Asset allocation
Consequently, in terms of allocation, it is wise to give priority to money market
investments and inflation-indexed bonds: the latter will benefit from the fall in real
rates in the phase iv to come. This is the time to be invested as far as possible in
cash, while waiting to see which way the wind blows. The share of risky assets
should be reduced or even sold before the next phase.

108 Amundi Investment Strategy Collected Research Papers


Cyclical Downturns since 1950

FED RATES
20
p of Penn
Collapse
Square r Bank
B Collapse
15 + 1 Mex
st
Mexican crisis
M of Continental Illinois
ollapse
s
Collapse
klin National
of Franklin N
Natio Bank
Savings & Loans
10 crisis Internet
Collapse
a o
of
Penn
n Central
t 2nd Mexican bubble
American credit crisis Subprimes
crisis
5 crisiss

0
CBR INDUSTRIALS
700
Emergence
600 of China
500
400
300
200
100
0
US COMPANY PROFITS
125
Vietnam War
Korean War

25

1
1954 1958 1960 1967 1970 1974 1981 1984 1989 1995 2001 2007

ŷŷ Decline in Fed rates Source: Amundi Research

The contraction of economic growth is often revealed through a crisis that forces
the Fed to lower its rates, whereas industrial commodity prices have already
passed their cyclical high. This arrives at the top of corporate profits which are a
lagging cycle indicator.

Amundi Investment Strategy Collected Research Papers 109


1.4 Phase iv:
Outperformance of government bonds
This phase corresponds to the correction of the cycle. It sees the materialisation
of the nascent fears of the previous phase. It marks a rupture. The movement is
sudden. Capital has to be protected. The correction can be more or less marked
and longer or shorter but, in any event, government bonds put up a good show.
The money market becomes less attractive than government bonds whose prices
rise when rates fall. The crisis is established and it is generally the flight to quality
and liquidity that prevails.

T Asset performance
The most risky assets decline or even collapse. They involve industrial commodities,
high-yield bonds, convertibles, equity markets generally and in particular less
liquid stocks and stocks that were fashionable in this cycle and whose valuation
has become too generous. In terms of sectors, telecommunications and services
utilities prove more resilient to the decline in the markets. These stocks are very
liquid, less sensitive to economic activity. Often indebted, they even profit from
the decline in rates under way. Finally, they often have a more domestic profile,
which is reassuring. The other defensive sectors such as food processing and
pharmaceuticals also perform very well relative to the rest of the market.

T Indicators
The contraction of economic growth is often revealed through a crisis that forces
the Fed to lower its rates (first graph). This corresponds to the top of US company
profits (third graph) which is often only detected a posteriori. The central bank only
accelerates its rate decline when it is certain that the supply/demand equilibrium has
been broken. In this respect, the decline in industrial commodity prices provides an
invaluable indication (second graph). The shaded periods on the following graphs
correspond to periods of war (Korea and Vietnam) which are slightly less relevant
in terms of analysis. It is worth noting that in 2007, the emergence of China and
speculation on the shortage of commodity prolonged the rise in commodity prices
and sent a false signal to central banks.

T Relevance of analyses
During phase iv, the financial or quantitative analyses of companies no longer
function. The lack of liquidity accentuates volatility, which is at its height. Investors
that have held out until now reduce their risks. Markets are grappling with a
psychology that is very rapidly deteriorating. Technical analysis, which takes
account of psychological effects, is again one of the most effective approaches
for analysing this phase of the asset cycle. It is often impressive to observe how,
during their decline, markets “stop off” at levels corresponding to Fibonacci ratios.
Then, when share prices have finished declining, they do not really pick up either.

110 Amundi Investment Strategy Collected Research Papers


Technically, a reversal pattern gradually emerges: double or triple bottom or
inverted head and shoulders. This can take several months.

T Strategies within asset classes


It becomes difficult to be very constructive since the volatility is frightening in the
short term. It is necessary to think about protecting portfolios. It is wise to hedge
a proportion of equity positions with futures or ETFs and to maintain a defensive
position on the component that is not hedged. In fixed income markets, preference
should be given to long maturities. In other words, extend the average duration of
the portfolio. As for currencies, reserve currencies are often protective; they play an
“insurance” role in international portfolios, starting with the dollar.

T Asset allocation
The exposure on the money market should have been reduced in favour of
government bonds. Equities will only gradually be re-weighted when market
sentiment is at its gloomiest, ideally during the reversal pattern that marks the
transition to the next phase.

Performance of the principal dollar assets


according to the Fed rate position since 1950

EQUITIES CRB INDUSTRIAL - GOVERNMENT BONDS


(performance différencial)
Low point Increase Last increase Decline Low point Increase Last increase Decline
of Fed rates in Fed rates in Fed rates in Fed rates of Fed rates in Fed rates in Fed rates in Fed rates
25% 15%

20% 10%

5%
15%
0%
10%
- 5%
5%
- 10%

0% - 15%

- 5% - 20%
S&P500 real RI CRB Industrial ‐ Government bonds

CRB INDUSTRIAL - GOLD


BONDS (performance différencial)
Low point Increase Last increase Decline Low point Increase Last increase Decline
of Fed rates in Fed rates in Fed rates in Fed rates of Fed rates in Fed rates in Fed rates in Fed rates
12% 15%

10%
10%
8%
5%
6%

4% 0%

2%
5%‐
0%
- 10%
- 2%

- 4% - 15%
Sovereign Bonds real RI CRB Index - Gold
Corporate bonds real RI
Money Market real RI Source: Datastream, F D, GlobalFinancial Data, Shiller data set, Amundi Research

Amundi Investment Strategy Collected Research Papers 111


We have recorded (charts p.17) the 4 phases of Fed rates since the 1950s: 1) the low
point of rates which can remain unchanged for some time, 2) upward phase, 3) last
increase in rates that can mark a plateau phase, and then 4) downward phase.
It emerges that, on average, since the 1950s, equities have performed well especially
in the downward rate phase and before rates go back up, even if they also prove
resilient for some time to the rate rise. Fixed-income investments prefer a situation
when the Fed stops tightening its monetary policy. Credit outperforms other fixed-
income investments in the same phases as equities. As for industrial commodities,
they do better than government bonds and gold when Fed rates are low, but
especially in their upward phase.
These observations correspond to the stylised roadmap on the performance of
assets in relation to the cycle presented on page 6.

Performance of style, size and sector indices from the equity low point
in the cycle and the Fed rate position since 1950
SMALL - LARGE VALUE - GROWTH

Equity increase last increase Decline Equity increase last increase Decline
25% low point in Fed rates in short rates in Fed rate 25%
low point in Fed rates in short rates in Fed rate
20% 20%
15%
15%
10%
10%
5%
5%
0%
-5% 0%

-10% -5%
Petites-Grandes valeurs Value-Growth

LI ALS E ENSIVES ENERG INAN IALS


di erence o per or ance di erence o per or ance
Equity increase last increase Decline Equity increase last increase Decline
low point in Fed rates in short rates in Fed rate low point in Fed rates in short rates in Fed rate
12% 15%
10%
8% 10%
6%
4% 5%
2%
0% 0%
-2%
-4% -5%
-6%
-8% -10%
Cyclicals - Defensives Energy-Financials

SE TORS IIN USTRIALS URA LE ONSUMER GOO S


di erence o per or ance di erence o per or ance
Equity increase last increase Decline Equity increase last increase Decline
low point in Fed rates in short rates in Fed rate low point in Fed rates in short rates in Fed rate
60% 15%

50% 10%
5%
40%
0%
30%
-5%
20%
-10%
10% -15%
0% -20%
-10% -25%
-20% -30%
Financial, Durable consumption Tech, Manufacturing, Energy Defensives Industrials - Durable consumer

SE TORS NON- URA LE - URA LE ONSUMER GOO S


per or ance di erence o per or ance
Equity increase last increase Decline Equity increase last increase Decline
low point in Fed rates in short rates in Fed rate low point in Fed rates in short rates in Fed rate
60% 20%

50% 10%
40% 0%
30%
-10%
20%
-20%
10%
-30%
0%
-10% -40%

-20% -50%
Non durable - Durable consumption
Financial Tech, Industrial
Energy Non cyclical consumption
Telecom, Utilities Source: Fama French data set, Amundi Research

112 Amundi Investment Strategy Collected Research Papers


Here (charts p.18) we consider that the Investment cycle begins (phase i) with a major
equity low point that occurs during the downward phase of Fed rates. Phase ii begins
when the Fed raises its rates, phase iii when it stops raising them and phase iv when
it reduces them.
In phase i and ii, small-cap stocks do better than large-cap stocks, “Value” stocks
better than “Growth” stocks, Cyclicals better than Defensives, but more so in phase
i than ii. Phase i is favourable for financials and consumer discretionary. Phase ii
benefits technology and industrials and then energy ends up prevailing. In phase iii
and until the new equity low point, defensives (non-cyclical consumption as well as
telecoms and public utilities) outperform.
These observations correspond to the stylised roadmap on the performance of
sectors in relation to the cycle on page 6.

1.5 Conclusion on target strategies


To conclude this description of the four-stage investment cycle, we propose the
following diagram (p.20) which recaps the asset classes to favour and the relevant
strategies within these asset classes. We have introduced two axis compared with
the previous diagrams – the return observed in the equity market year-on-year and
the corresponding volatility. Volatility declines in phase i and ii and increases during
the 2 following phases, while annual performance decreases in phase ii and iii.

Equity performance and Volatility change since 1950


Equity Increase Last increase Decline
Low point Fed rates Fed rates Fed rates
60%
50%
40%
30%
20%
10%
0%
-10%
-20%
Equity Perf Vol change Source: Données Shiller, Datastream, Amundi Research

To avoid overload, we have removed references to the economic cycle, which are
no longer represented except through the name of phases i,ii,iii,iv. Similarly, we
have not included strategies within commodities either, given that oil and industrial
commodities tend to outperform precious metals including gold in phase i and ii,
while conversely precious metals, which are more defensive, outperform in phase
iii and iv.

Amundi Investment Strategy Collected Research Papers 113


Target Strategies

LOW VOLATILIT

RELATIVE SE URE
VALUE STRATEGIES RETURN STRATEGIES
RELATIVE E ENSIVE

ARR TRA E UN AMENTAL


SPREA S LATTENING ARR TRA E

E UITIES MONE MAR ET


AN AN IN LATION-IN E E
OMMO ITIES ON S

HIGH ii iii LOW


RETURN RETURN
i i

E UITIES ON S
SPREA S STEEPENING LATTENING

LI ALS RESERVE

IRE TIONAL HE GING

ONTRARIAN PROTE TION


STRATEGIES STRATEGIES

HIGH VOLATILIT

Priority asset class Currency strategies


Bond strategies Equity strategies
Source: Amundi Reserach

At the bottom of the investment cycle (bottom of the diagram), volatility is strong. It
is necessary to be contrarian. Phase i is the phase where the annual return of risky
assets will be the highest. Subsequently (phase ii), it is necessary to be more subtle
and implement “relative value” strategies. When volatility becomes very low, returns
have to be made even more secure (phase iii). Finally, whereas the return of risky assets
continues to weaken, it is necessary to have switched to “protection” mode (phase iv).

The sequence of these four phases is very consistent. However, they can be longer
or shorter according to the cycles. Therefore, we are now going to focus on the
duration of short investment cycles, observing firstly economic cycles, and then
stock market cycles.

114 Amundi Investment Strategy Collected Research Papers


II. The duration of short investment cycles
2.1 The duration of economic cycles
Here, we have drawn, in particular, on the work of the NBER referring to this
subject. This US body, officially responsible for measuring the cycles in the United
States, has also worked on the cycles of other countries in the world and especially
in Europe. We have drawn 3 major conclusions from this work:

T Firstly, the cycles and the idea we have of them have changed over time
The biggest rupture since industrialisation was marked by the Second World War.
The majority of studies date back to this period which has the attraction of being
much better supplied in terms of statistics. However, observing previous cycles is
not devoid of interest:
Prior to the Second World War, cycles were much more marked, both upwards
and downwards. Shocks concerned first and foremost infrastructures. After the
Second World War and the increasing momentum of our consumer society, shocks
mainly marked durable consumer goods. Moreover, they subsided. The expansion
from 1961 to 1969 was, at the time, the longest in history; so much so that during
the 1960s, some observers even began to doubt that the cycles still existed. The
same phenomenon appeared during the 1990s. In both cases, the absence of
prolonged volatility in economic activity was nevertheless ultimately transformed
into recession!
In 1969, Mintz (NBER) gave birth to a new concept: “growth recession”, i.e. a
slowdown in growth in relation to its trend, but which did not turn into recession.
One, or even two re-accelerations occur, prolonging the cycle accordingly. However,
the outcome is always the same: recession. This therefore involves considering
cyclical movements in relative terms and no longer in absolute terms. Monetary
policy espouses these cyclical movements, or even provokes them; monitoring
its different stages (rise in rates, stability on a plateau, decline, new stability) is
therefore fundamental.

T Secondly, the duration of cycles is relatively constant at around 47 months


In this respect, the study of what occurred prior to the Second World War is very
instructive. We observe that a cycle lasts 47 months if we consider the median
since 1854. The modifications since the Second World War have related more to the
extension of the expansion period and the reduction of the length of consolidation
within this period of time rather than on the duration of the cycle itself.
In order to go into a little more detail on this observation, let us refer to and comment
on the conclusions of a study by Geoffrey H. Moore and Victor Zarnowitz (“The
development and role of the NBER’s Business Cycle Chronologies”-1986) which
highlights the structural changes of US cycles:

Amundi Investment Strategy Collected Research Papers 115


Firstly, the average total duration of cycles lengthened by one year after the major
crisis of 1933. From 1790 to 1933, there were 34 cycles, or 1 every 4 years on
average. From 1933 to 1982, there were 10, of around 5 years on average. This
is not a major difference, especially if we take into account the phenomenon of
“growth recessions”. It is even almost surprising that the duration of average cycles
is as constant in time over some two hundred years of history!
On the other hand, the most important change has taken place within cycles.
Expansion phases have lengthened by 2 years, still on average, and contraction
phases have shortened by one year to last 11 months.
Contraction phases have become much more uniform. The differences in relation
to this 11-month average have narrowed considerably. From 1790 to 1855, the
difference in duration around the average was 18 months. From 1855 to 1933, it
fell to 14 months. From 1933 to 1982, it was no more than 3 months. They have
therefore become easier to anticipate once under way.
As for expansion phases, they, on the contrary, have become more complex
and more uncertain. From 1855 to 1933, the typical difference was 9 months. It
increased to 27 months from 1933 to 1982. This can be explained primarily by the
phenomenon of “growth recessions”.
Since 1982, the NBER has recorded three additional cycles. The first lasted 100
months from the low of November 1982 to the low of March 1991, the second
128 months from March 1991 to November 2001 and the third 91 months from
November 2001 to June 2009. However, on average the correction phase of these
3 cycles lasted 11 months! These cycles were marked by intermediary pauses
(growth recessions) which complicate the interpretation of the rising phase. It is
worth noting that generous monetary policies are, to a large extent, at the origin of
the greater longevity of expansion phases.

T Lastly, global cycles have tended to converge towards US cycles


In the past, cycles were slightly longer in the United Kingdom, the breeding ground
of industrialisation. The same is true in France and Germany. In the United States,
the 4-year US presidential cycle has merged with the inventory cycles pinpointed by
Kitchin, especially as the State has assumed an increasingly important role since the
crisis of the 1930s.
With internationalisation, the cycles of developed countries are converging towards
US cycles. Movements of financial assets, and notably equities, are even more
sensitive to this situation, major listed stocks being more international than the
average.

116 Amundi Investment Strategy Collected Research Papers


Duration (months) between 2 cyclical trough in the USA since 1854
Duration from trough to WURXJK Average
140
120
100
80
60
40
20
0

Duration trough to peak of cyclical


Duration, trough to peak Avearge 1854-1933 Average 1933-2009
140
120
100
80
60
40
20
0

Duration peak to trough of cycle


Duration, peak to trough Average 1854-1933 Average 1933-2009
70
60
50
40
30
20
10
0

Source: NBER, Amundi Research

An economic cycle as defined by the NBER lasts 47 months, based on the median
since 1854. Changes since the Second World War have been greatest within the
cycle. They have related to the extension of the expansion period and the reduction
of the consolidation period. The duration of the consolidation has become more
uniform. As for expansion phases, they have become more complex and can
include “Growth Recessions”.

Amundi Investment Strategy Collected Research Papers 117


2.2 The duration of “stock market” cycles
We have seen that central banks are a very practical reference for positionning
where we are within the investment cycle. We propose a second reference point
which consists in solely looking at the price of equity indices, such as the CAC 40
(France), the S&P 500 (United States), the FTSE 100 (United Kingdom), the NIKKEI
(Japan), etc.
It may be surprising to use changes in the prices of equities themselves in order
to anticipate their own trend. However, this is very logical. The stock market
anticipates and accentuates economic changes. Economic data merely validate
or invalidate market anticipations. Moreover, equity markets are particularly prone
to psychology, more than commodities which respond first and foremost to the
law of supply and demand, and more than interest rates which are partly driven by
central banks.
Let us therefore observe stock market cycles:
We have analysed all the stock market cycles of the US market since the end of the
19th century. Working on US data has several advantages: apart from the fact that
the behaviour of other markets, notably European, converges towards it, we have
precise dating of cycles by the NBER, as we have just seen. This makes it possible
to put stock market cycles and economic cycles perfectly into perspective.
T Firstly, what is a stock market cycle?
Index prices fluctuate around their 200-day moving average. This duration is very
consistent with the annual frequency of corporate results. This moving average is
one of the most used by technical analysts since index prices only rarely cross it. It
acts as a wall, a white line (red on the diagrams below) which seems impassable for
some time at least.
There are bull cycles and bear cycles. In a bull cycle, the index reaches a major low
point below its 200-day moving average, then moves above it, can come and test it
on several occasions before moving back below it. In a bear cycle, the index starts
above its moving average, moves below it, can come and test it on several occasions
before moving back above it.
The index does not progress upwards or downwards in a straight line. It is driven
by impulses. As already indicated by Charles Dow more than a century ago, stock
market movement breaks down into 5 legs. The breakdown of this movement within
a stock market cycle was highlighted by Robin Griffiths, a renowned technical analyst
with whom I have been fortunate to have many discussions over the last 20 years.
He began working on the markets in 1964, which gives him a certain hindsight. His
book “Mapping the markets” summarises his approach very well and I invite you to
refer to it for more details. We therefore find 6 possible profiles: an average bull cycle
consists of 3 rising legs and 2 declining legs (profile 1, see graph). It may also have an
upward bias and consist of 4 rising legs and only 1 declining leg (profile 3). It is said

118 Amundi Investment Strategy Collected Research Papers


to be a secular bull cycle. In contrast, it may also have a downward bias and have
only 2 rising legs and 3 declining legs (profile 2). And conversely for bear cycles.

Stock Market Cycles


Profil 1 Profil 2 Profil 3
AVERAGE CYCLE DOWNWARD BIASED CYCLE UPWARD BIASED CYCLE

Source: inspired by Robin Griffiths (Mapping the Markets)

T Characteristics of stock market cycles


We can observe thirty or so consecutive stock market cycles, which begins to
become significant. Each consists of an upward phase and a downward phase.
The stock market cycle lasts an average of 46 months. When a stock market cycle
is shorter, the next one is longer and the average of the two is again closer to 46
months. The average duration of the upward phase is 32 months, the correction lasts
14 months. Conversely, once the cycle is under way, the market goes up for at least
14 months.

Duration of stock market cycles


Duration from bottom to bottom Average
80
70
60
50
40
30
20
10
0
1896

1900

1903

1907

1910

1914

1917

1921

1926

1932

1935

1949

1953

1957

1962

1966

1970

1974

1978

1982

1987

1990

1994

1998

2003

2006

Source: Shiller Data, Datastream, Amundi Research

Average duration of two consecutive stock market cycles


Average duration of 2 consecutive market cycles Average
60

50

40

30

20

10

0
1896

1900

1903

1907

1910

1914

1917

1921

1926

1932

1935

1949

1953

1957

1962

1966

1970

1974

1978

1982

1987

1990

1994

1998

2003

2006

Source: Shiller Data, Datastream, Amundi Research

Amundi Investment Strategy Collected Research Papers 119


The stock market cycle lasts an average of 46 months, very close to the average
economic cycle, especially if we take account of the phenomenon of “Growth
Recessions” (see first graph). When a stock market cycle is shorter, the next one is
longer and the average of the two is again closer to 46 months (see second graph).

As for the extent of the movements, the majority of the time, the rise in a S&P
500 cycle ranges between +50% and +100%, whereas the average correction
is around -25%. These proportions are very similar to the figures we observe on
average on corporate profits.
The high point of the stock market cycle can be situated at the end of 2, 3 or 4
rising waves. We therefore notice 2 phenomena:
The first is easy to understand. The more bullish the cycle, the more upward
stages the market goes through. It is a little like a diver who has to return to the
surface in stages.
The second is counter-intuitive and full of information. The more bullish the cycle,
the more the correction that follows is limited. This can be explained by the duration
of the stock market cycle. The fourth upward phase, when it occurs (profile 3),
anticipates the continuation of a favourable profits or valuation movement; the
following cycle having at least 2 rising legs. In contrast, if the cycle has only 2
rising legs (profile 2), the correction is therefore the strongest. Moreover, it is
frequent for cycles that have only 2 rising legs to occur after cycles with 4 rising
legs, such as during the formation and bursting of bubbles;1998-2003 and 2006-
2009 are good examples of this situation.

2.3 Comparison between stock market cycle and economic cycle:


T Evidently, the duration of a stock market cycle is on average incredibly similar
to that of the economic cycle.
This reinforces the idea that the data of 46-47 months is an “almost constant” figure.
The duration of the stock market cycle correction comes out on average at 14 months
and is similarly very close to the economic cycle correction of 11 months plus or minus
3 months.

T Moreover, the equity market anticipates the trend in the economy.


This is a common event but our research has also led to more unexpected
conclusions. When we superimpose equity cycle and economic cycle, we
observe that the equity market anticipates the top of the cycle and the bottom
of the cycle dissymmetrically. The top of the cycle is generally anticipated one
quarter in advance, i.e. the equity market reaches its high point one quarter before
the economy’s high point as recorded by the NBER. Meanwhile, the bottom of
the cycle tends to be anticipated two quarters in advance. This highlights the
psychological dimension of the markets in relation to the real economy.

120 Amundi Investment Strategy Collected Research Papers


In this respect, it is worth noting that the equity investor is optimistic by nature,
unlike the bond investor who is more sensitive to risk. This can be easily
understood in relation to the investment vehicle: in purchasing equities, the
investor takes the risk of losing all their outlay but in the hope of an infinite gain.
In purchasing bonds, the investor also takes the risk of losing all their outlay but
only has the hope of a limited gain.

T Finally, the stock market cycle accentuates the economic reality.


This phenomenon can be explained by the crowd psychology. When the market
goes up or down, it gradually carries along firstly the opinion leaders and then the
followers, so that there comes a time where everybody is positioned in the same
direction, generating excesses in relation to the reality. This is why the market
does not need a recession to correct, a “growth recession” can be sufficient. The
duration of stock market cycles would therefore be even more consistent with
the duration of economic cycles. The analysis of stock market cycles therefore
constitutes an invaluable additional tool to the analysis of the economic cycle
and, what is more, easy to access.
We therefore have two key reference points to find our way around short
investment cycles: central banks’ monetary policy, starting with that of the Fed,
and the stock market cycle. According to our experience, the stock market cycle
generally starts when Fed rates are at their lowest. If it starts when rates are high
and have even already plateaued, then it is probable that it has a downward bias
(profile 3) and that phase iv which we have described will be deep.

Definitions of the notions of cycles that we use:

R The economic cycle extends from one recession to another. In the United
States, the NBER is responsible for dating recessions. An economic cycle
defined accordingly can contain several “growth recessions” and therefore
several investment cycles.

R The investment cycle, determined by us, corresponds to recurrent asset


performance. To simplify, we characterise the stages with the policy of central
banks. Stage i starts with a major equity low point, stage ii is characterised
by the tightening of monetary policy (rise in rates), stage iii by stable monetary
policy (rates on a plateau), stage iv by the start of a decline in central bank rates.

R The stock market cycle is also determined by us. It starts with a major
low point, below the 200-day moving average, and ends with the last major
high point before a significant break in the 200-day moving average. It is a
tremendous addition to the analysis of central banks’ monetary policies in order
to find our way around the investment cycle.

Amundi Investment Strategy Collected Research Papers 121


III. The most recent short cycles perfectly
illustrate this logic
As Mark Twain said “history does not repeat itself, but it rhymes”. The economic
winter (since 2000) currently includes 2 complete investment cycles, that clearly
correspond to the logical process that we have observed historically, but
obviously each with its own specific features. In practice, this logical sequence
is extremely long-lasting.

Short cycles during this economic winter

i i ii iii i i ii iii i i ii
7

6 ed und rates
5

0
2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014
i i ii iii i i ii iii i i ii
8000

E er in ar ets

4000
S all-cap stoc s

2000
S P

1000
2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2000
i i ii iii i i ii iii i i ii
1800
1600
1400 Gold
1200
1000
800
600
400 R Industrials
200
0
2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

Source: Amundi Research

122 Amundi Investment Strategy Collected Research Papers


i i ii iii i i ii iii i i ii
1550

1350
DS Marchés Emergents
1150
orporate onds
E er in Mar et onds US
950 e i es a eurs
Go ern ent
onds
750

550 S P Mone Mar et

350
2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014
Source: Amundi Research

The first investment cycle (October 2002-March 2009) starts during the decline in
Fed rates. It develops fairly classically and corresponds to Fed rate movements,
except that it results in the subprime bubble. The second (March 2009-October
2011) is more unusual since it corresponds to quantitative easing and no longer to
interest rate phases. However, the sequence of asset class performance clearly
complies with our past observations.

The last two investment cycles’ common points with our roadmap
Phase i is logically marked by a very sharp equity recovery and phase ii by an
acceleration in the price of commodities whereas bonds pale into insignificance.
Phase iii is characterised by a decline in the return of risky assets and less dispersion
of returns. Lastly, phase iv is particularly destructive of value. Government bonds
and gold are therefore the best safe haven investment. Finally, volatility decreases
in the first 2 phases and increases during the last 2; we note, in passing, that this
is valid for the volatility of both equities and bonds.
If we go into more detail, the best phase for the dollar is phase iv where it serves
as a safe haven currency and the worst is phase ii when volatility is weakest.
Silver beats gold in phase i and ii, and then it is the reverse in phase iii and iv. The
same is true for small-cap stocks which outperform large-cap stocks in phases i
and ii and conversely underperform in phases iii and iv. Idem for value stocks in
relation to growth stocks.

Specific features of these two investment cycles

T First cycle: October 2002 – March 2009


The cycle begins in October 2002 at the equity market’s bottom and ends in
March 2009 with another major low point.
In this case, the specific feature stems from the fact that the emergence of new
countries resulted in a structural rise in commodity prices. Moreover, fears of

Amundi Investment Strategy Collected Research Papers 123


Average cycle performance since 2000

EQUITIES, VOLATILITY (VIX and MOVE)


i ii iii iv
120%
Oct 2002 - June 2004 June 2004 - June 2006 June 2006 - Aug 2007 Jan 2001 - Oct 2002
100% Aug 2007 - March 2009
March 2009 - June 2010 June 2010 - March 2011 March 2011 - July 2011
80% July 2011 - Oct 2011

60%
40%
20%
0%
-20%
-40%
-60%
S&P 500 VIX MOVE Source: Datastream, Amundi Research

USD (Real effective exchange rate)


i ii iii iv
8%
Oct 2002 - June 2004 June 2004 - June 2006 June 2006 - Aug 2007
6% June 2010 - March 2011
March 2009 - June 2010 March 2011 - July 2011
4%
2%
0%
-2% Jan 2001 - Oct 2002
Aug 2007 - March 2009
-4% July 2011 - Oct 2011
-6%
-8%
-10%
USD (JPM Real broad index) Source Datastream, Amundi Research

RAW MATERIALS
i ii iii iv
100%
Oct 2002 - June 2004 June 2006 - Aug 2007 Jan 2001 - Oct 2002
80% March 2009 - June 2010 March 2011 - July 2011 Août 207 - Mars 2009
July 2011 - Oct 2011
60%

40%

20%

0%

-20% June 2004 - June 2006


June 2010 - March 2011
-40%
CRB Industrials Silver-Gold Source: Datastream, Amundi Research

124 Amundi Investment Strategy Collected Research Papers


MONEY MARKET, BONDS, CREDIT
i ii iii iv
0,6
0,5
0,4
Jan 2001 - Oct 2002
0,3 Aug 2007 - March 2009
July 2011 - Oct 2011
0,2
0,1
0
Oct 2002 - June 2004 June 2004 - June 2006 June 2006 - August 2007
-0,1 March 2009 - June 2010 June 2010 - March 2011 March 2011 - July 2011

-0,2

Money Market Bond Gov Credit IG Bond IL Bond EM Credit HY Source:Datastream, Amundi Research

CYCLICALS - DEFENSIVES (net up of performance)


i ii iii iv
0,4
June 2004 - June 2006 June 2006 - Aug 2007 Jan 2001 - Oct 2002
0,3 June 2010 - March 2011 March 2011 - July 2011 Aug 2007 - March 2009
July 2011 - Oct 2011
0,2

0,1

0
Oct 2002 - June 2004
-0,1 March 2009 - June 2010

-0,2
Cyclicals-Defensives Source:Datastream, Amundi Research

STYLES et SIZES (net up of performance)


i ii iii iv
0,25
0,2 Jan 2001 - Oct 2002
June 2004 - June 2006 June 2006 - Aug 2007 Aug 2007 - March 2009
0,15 June 2010 - March 2011 March 2011 - July 2011 July 2011 - Oct 2011
0,1
0,05
0
-0,05 Oct 2002 - June 2004
March 2009 - June 2010
-0,1
-0,15
-0,2
Small-Large Value-Growth Source: Datastream, Amundi Research

Amundi Investment Strategy Collected Research Papers 125


Average cycle performance since 2000

EQUITIES, VOLATILITY (VIX and MOVE)


i ii iii iv
120%
Oct 2002 - June 2004 June 2004 - June 2006 June 2006 - Aug 2007 Jan 2001 - Oct 2002
100% Aug 2007 - March 2009
March 2009 - June 2010 June 2010 - March 2011 March 2011 - July 2011
80% July 2011 - Oct 2011

60%
40%
20%
0%
-20%
-40%
-60%
S&P 500 VIX MOVE Source: Datastream, Amundi Research

USD (Real effective exchange rate)


i ii iii iv
8%
Oct 2002 - June 2004 June 2004 - June 2006 June 2006 - Aug 2007
6% June 2010 - March 2011
March 2009 - June 2010 March 2011 - July 2011
4%
2%
0%
-2% Jan 2001 - Oct 2002
Aug 2007 - March 2009
-4% July 2011 - Oct 2011
-6%
-8%
-10%
USD (JPM Real broad index) Source Datastream, Amundi Research

RAW MATERIALS
i ii iii iv
100%
Oct 2002 - June 2004 June 2006 - Aug 2007 Jan 2001 - Oct 2002
80% March 2009 - June 2010 March 2011 - July 2011 Août 207 - Mars 2009
July 2011 - Oct 2011
60%

40%

20%

0%

-20% June 2004 - June 2006


June 2010 - March 2011
-40%
CRB Industrials Silver-Gold Source: Datastream, Amundi Research

126 Amundi Investment Strategy Collected Research Papers


deflation as during the economic winter of the 1930s justify a gold rush. Finally,
remember that since the Russian crisis in 1998, in order to avoid deflation, central
banks have flooded the planet with liquidity and generated several asset bubbles
(internet bubble in 2000, property bubble in 2006-2007).
It is worth observing that this economic cycle corresponded to 2 S&P 500 stock
market cycles:
The first lasted 44 months from October 2002 to June 2006, i.e. very similar to the
historical average. Therefore, a new stock market cycle commences at that time.
Occurring after a secular bull cycle (profile 3) and when Fed rates are at their
highest (phase iii), the probability that the rising phase has a downward bias and
only lasts 14 months is considerable and this is what happens. The S&P 500 will
reach its high point in October 2007, i.e. 15 months after its low point in June
2006. The following correction will beat all the standards in terms of the extent
of the correction. Only the corrections of 1929 on US equities, after the Second
World War in Germany and Japan and 1973 in the United Kingdom were greater.
That certainly explains why the following stock market cycle (June 2006-March
2009) lasted only 33 months, i.e. 12 months less than the historical average,
which also remains in the norm of our past observations.

T Second cycle: March 2009 – October 2011


The specific feature of that particular cycle stems from the fact that the Fed,
which could no longer lower rates, continued to relax its monetary policy
through quantitative easing. Apart from the liquidity injected, the central bank’s
communication played an even more important role than in the past. Quantitative
easing tapering is therefore considered by the market as a reversal of monetary
policy. In this unprecedented universe, points of reference therefore had to be
found not only in the central bank’s communication but also to a large extent in
terms of the performance of assets themselves, which was incredibly consistent
with our historical observations.
The cycle begins in March 2009 (phase i), at the equity low point. In June 2010,
commodities start to rise (phase ii). The Fed pre-announces its second quantitative
easing programme (QE2) at Jackson Hole at end-August 2010. Bonds fall. It is
not until the Arab spring that equities give up some ground in 2011, just before the
end of QE2 scheduled for August the same year. Therefore, linkers do better than
equities (phase iii). The end of these tensions in North Africa causes the correction
of commodities. Whereas the US Congress has difficulty agreeing budget cuts in
August 2011, this gives the rating agencies a good reason to withdraw the AAA
rating from US debt, prompting investors to protect themselves by turning to both
gold and the dollar (phase iv).

Amundi Investment Strategy Collected Research Papers 127


Conclusion
Asset rotation is well-organised during the short investment cycle. Equities have to be
purchased when we still hold bonds, then government bonds have to be reduced in
order to purchase commodities alongside equities. We subsequently reduce equities
and finally commodities. A fairly short phase results in preference being given to the
money market and inflation-indexed bonds over all other assets before returning to
long-term government bonds. And so on.
The importance of identifying the current phase is obviously being able to anticipate
the following phase. Pinpointing transitions is a complex exercise. There are often false
starts. However, we have several instruments to achieve this. Two of them are key:
monetary policy and the stock market cycle.
The low point of the equity cycle gives us the starting point of the investment cycle.
When the index rebounds and carries along with it its rising 200-day moving average,
we have confirmation that a new cycle is well under way. It will end around 46 months
later. The upward phase will last at least 14 months. Subsequently, it is the central
banks that give the key indications to characterise the investment cycle phase in which
we are situated. This is because central banks have the information before the markets.
Therefore, in a way they are insiders. Moreover, they are themselves players since they
decide in particular the level of interest rates. The rise in Fed rates, and then the halt in
the rate rise and finally their decline confirm that we are in phase ii, iii and then iv. In a
period where there is a struggle to avoid falling into deflation, the rate decline signal is
prolonged by the quantitative easing signal and then by its withdrawal.
If the stock market cycle starts when Fed rates have already reached their high point,
then it is probable that phase iv which follows marks a sharp correction, as was the
case during the last stock market cycle in 2006-2009. These mechanisms, highlighted
by analysing thirty or so consecutive cycles, have again proved their robustness.
Apart from these two fundamental reference points, it is first and foremost the
consistency of the movements that we have described in this discussion paper that
enables us to form a conviction; consistency with economic indicators which are
often lagging but provide confirmation data. Consistency between the behaviour of
the different asset classes and associated flows. Consistency in geographical terms.
Internal changes within each asset class also provide considerable information.
Finally, the first two phases see volatility gradually reduce. It is very weak at the
beginning of phase iii whereas it soars during the fourth phase. Volatility is a belated
function of the economic cycle. It changes with a time-lag of 18 months to 2 years
with central bank rates. It is a fearsome short-term enemy for the investor. Volatility
expresses risk. In the short-term, it reduces the relevance of more “rational” analyses,
prices therefore tend to be more governed by psychology, which technical analysis
on the other hand knows how to harness. Here, we measure the considerable
complementarity of the different forms of analysis.

128 Amundi Investment Strategy Collected Research Papers


References

Geoffrey H. Moore et Victor Zarnowitz (1986): “ The development and role of the NBER’s
Business Cycle Chronologies”- NBER Studies in Business Cycle Vol 25
Deborah Owen and Robin Griffith (2006): “ Mapping the Markets”, Bloomberg Press
Éric Mijot (2009): “Où en est-on dans le cycle d’investissement ?”, Investir, 8 March
Éric Mijot (2009): “Comment réagissent les secteurs au cours du cycle d’investissement ?”,
Investir, March 29
Éric Mijot (2009): “Quel est le comportement des actions au cours des cycles longs”,
Investir, 21 March
Andrew Garthwaite (2011): “Margins: higher for longer” Credit Suisse Research, May 6
Éric Mijot (2013): “Une hirondelle annonce le printemps mais le ne fait pas !”, Amundi Cross
Asset Invesment Strategy,June
Dough Peta (2013): “ Stocks and the Fed Fund rate Cycle”, BCA Research, December
Mark McClellan (2014): “Bonds and the Fed Funds rate Cycle”, BCA Research, May

Amundi Investment Strategy Collected Research Papers 129


130 Amundi Investment Strategy Collected Research Papers
DP-05

Physical real estate


in long-term asset allocation:
the case of France
CÉCILE BLANCHARD,
Real Estate Research, Amundi
SYLVIE DE LAGUICHE, ALESSANDRO RUSSO,
Quantitative Research, Amundi

September 2014

This paper analyses for France, indices that represent physical


real estate and shows that they differ from one another in terms
of assets which are included or their valuation methods. They
are constructed in such a way that comparisons with listed
asset indices must be done with care. In particular, they exhibit
smoothing and time lag. We investigate the hierarchy of risks
and we show that residential is less risky than offices, prime real
estate in Paris’s “Golden Triangle” is the riskiest.
The long-term risk is greater than what a simplistic calculation
would show, and may be as high as the risk level of equities.
Over the long term, there is a positive correlation with equities
and a negative correlation with government bonds. Again, the
phenomenon may be concealed through careless work on series
with too high a frequency.
Consequently, the diversifying power of physical real estate is
very attractive in the long term for bond-heavy portfolios, but less
so for equity-heavy portfolios. The smoothed valuation method,
however, has a favourable optical effect on short-term risk.

Amundi Investment Strategy Collected Research Papers 131


PHYSICAL REAL ESTATE
IN LONG-TERM ASSET ALLOCATION:
THE CASE OF FRANCE

Introduction
The benefit of an asset class is explained by the expected return given its risk,
as well as its impact on the total risk of the portfolio via diversification. It is hard
to extrapolate from past returns, but observing historical series is helpful for
quantifying risk and correlations. Here, we are interested in the risk of physical real
estate and its correlation with the listed assets to be taken into account for long-
term allocation in the case of France.

I - The available historical data and how to interpret it


Real estate has unique characteristics that make it stand apart. Its scope is
limited by definition – DTZ has estimated the amount of investment in the French
real estate market to be €534 billion, for a total investable scope of €883 billion
in late 2012.
Rental income and rental yields make it possible to reconstruct the price per m²
and calculate an overall performance index, which relies on rental yields and capital
gains for the asset in question, given its location and the current business cycle.
This profitability calculation does not take into account the expenses inherent in the
use of the property, nor does it consider vacancies, both of which may adversely
affect rental income. Neglecting rental expenses affects yield uniformly, and has
no significant impact on risk. With respect to the vacancy rate, the impact on risk
depends on the segments used. It is much lower than the risk from price variations
in the prime market, and when buildings and tenants are diversified; but this is less
so for lower-quality market segments.
In order to ensure that the results achieved are as useful as they can be, while
reflecting both the dependence of real estate markets on past events and the
gradual acceleration of cycles in terms of both their duration and their extent, real

132 Amundi Investment Strategy Collected Research Papers


estate indicators have been selected with great care. There are three varieties:
contribution, expertise, and transaction based indicators.

1.1 – CBRE

CBRE, a real estate broker, offers a contribution index based on rent and yield rate.
This indicator provides long series, which are updated quarterly and therefore more
often than the IPD index, for example. However, it exists in this form only for the
so-called “prime” segment of the office space market. Its restriction to this specific
market area means it does not necessarily have the same ability to benefit from the
diversification specific to the real estate asset class, in terms of both products and
geographic locations.
It therefore represents the rent achievable for available office surface area in an
office building with particularly good technical assets when leased by a user for
the long term (nine years in France). This index is an estimate for this specific and
especially narrow market segment over a given period of time.
Given their frequency, the CBRE data set is nonetheless the only one that can be
used to estimate correlations with other assets, but it remains much more volatile
than the office market as a whole.

1.2 – IPD

IPD, meanwhile, offers an index based on annual inspection of real estate


properties, including all properties (offices, shops, warehouses and housing),
owned by institutional investors. This index therefore covers a broader market
than the “prime” sample estimated by CBRE, but offers highly smoothed values
based on expert opinions. It is substantially more useful for offices, but less so for
residential space, which is a real estate product for which the portfolio allocation
changes very slowly due to heavy regulatory restrictions.
The values expressed in the IPD index therefore appear more inert, because only a
small share of the portfolios experience a transaction, so they have a longer delay
than CBRE with regard to market changes.
It is nonetheless possible to use those values to rank the risks of different market
segments and attempt to define an adjustment coefficient between the volatility of
CBRE’s “prime” data and office market data derived from the IPD index calculated
solely for the Parisian Golden Triangle.

1.3 – EDHEC IEIF

The EDHEC IEIF index is an index that is transaction-based. It corresponds to


“an index weighted by the capitalisation measuring the change in performance of
REITs (Real Estate Investment Trusts) that invest in business real estate (offices,

Amundi Investment Strategy Collected Research Papers 133


storefronts, logistics platforms, etc.)”1. This performance is calculated based on the
value of REIT shares, which are traded on the secondary market. This is a hedonic
index, which makes it possible to bypass the difficulty of analysing available data, with
the main difficulty resulting from their heterogeneity, given that REITs may be of different
types (variable-response or fixed-capital) and have varied investment strategies (single
or multiple real estate products, locations, etc.). This index erases any leverage, but
takes into account any discount or premium owing to the SCPI itself, which adds
noise on top of the price changes of the physical real estate. On the other hand, its
frequency has been monthly since June 2008, and was every six months before then.
Its investment scope covers a widely diverse range of products and market areas.

II - Long-term risk
When estimating the risk associated with a real estate investment, available data
must be used carefully. Depending on the type of indices used, the orders of
magnitude observed for volatility may be quite different. The graph shows volatilities
in yearly data for reconstructed index series from the providers listed in section 1.

Volatility according to real estate type


annual frequency 1991-2013
25%

20%

15%

10%

5%

0%
IEIF Edhec IPD offices IPD residential IPD golden Equities Equities Reits
triangle SBF250
Source: Amundi Research

Note that depending on the index type, CBRE, Edhec or IPD, the apparent volatility
is different and substantially stronger for CBRE than for the series drawn from IPD.
Naturally, some of this can be explained by the fact that the CBRE data is drawn
from observations of the prime market in Paris, the Golden Triangle, which is more
volatile, but even in this segment, the IPD data is much less volatile, a sign that a
significant share of the difference in volatility is due to the indices’ different natures:
the IPD data is much smoother. It should also be noted that when examining the
IPD data, residential is 30% less volatile on average than office. Although the orders
of magnitude of IPD volatilities are excessively low, it appears interesting to us to
consider the hierarchy between the risk levels available through the IPD data.

1 EDHEC IEIF Immobilier d’Entreprise France index rules, EDHEC-RISK Institute, September
2012.

134 Amundi Investment Strategy Collected Research Papers


The data smoothing problem leads to risk being underestimated when working with
quarterly data. This is why, for CBRE data, we have recalculated the annualised
volatility by using lower frequencies, up to two years. The graph shows the results
achieved, compared to the situation for listed shares in REITs.

Annualised volatility
according to frequency of observations
0,25

0,2

0,15

0,1

0,05

0
quaterly semiannual annual biennial
CBRE real estate Equity Reits Source: Amundi Research

At a biennial frequency, a slight increase is observed in the annualised volatility


of listed shares, but for CBRE data relating to physical real estate, the increase is
much larger and much more regular as the frequency decreases.
This means that the long-term risk of physical real estate is being substantially
underestimated if using a quarterly frequency to calculate volatility. It is necessary
to reprocess the serial correlations, unsmooth the data, or work with the lowest
possible frequencies. The important thing to remember is that in the “prime”
segment, the long-term risk of the physical real estate is fairly close to that of
equities.
Looking at a portfolio made up half of residential and half offices, a ratio between
the volatility of that composite and the volatility of Golden Triangle prime real estate
can be calculated from IPD volatilities. It comes out to 74%. Using this ratio, it is
possible to correct the CBRE figure and achieve volatilities more representative of
the real estate market as a whole.
Note however that these risk measurements ignore the risk caused by the variability
of the vacancy rate that is higher in these segments than in prime. It is interesting
to examine the ratio of this adjusted real estate volatility to equities volatility.
This gives a risk hierarchy that can be compared to that of regulators. In Solvency
II, real estate is counted for an SCR of 25%, while OECD equities are counted for
39% in the absence of a dampener.
The following graph shows that after discounting the misleading optical effect
of smoothing, the observed risk hierarchy between equities and real estate is
consistent on average with what regulators produce. However, as Solvency II does

Amundi Investment Strategy Collected Research Papers 135


not make a distinction between market segments, this regulation may therefore be
considered lenient for offices but strict for residential.
Ratio real estate observed risk/equity risk
depending on frequency of observations
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
quarterly semiannual yearly Solvency II
Source: Amundi Research

III - Correlation with other asset classes


The impact of frequency
Once again, we have calculated the correlation between physical real estate and
different listed asset classes: bonds, equities, and shares in real estate investment
trusts. The results are depicted below.

Correlation between real estate and listed asset classes

real estate CBRE and reits IEIF

real estate CBRE and equities SBF250

ƌĞĂůĞƐƚĂƚĞZĂŶĚďŽŶĚƐ

bonds and equities SBF 250

equities SBF 250 and IEIF Reits

-0,80 -0,60 -0,40 -0,20 0,00 0,20 0,40 0,60 0,80


biennial yearly quarterly Source: Amundi Research

Real estate has positive correlations with equities, and negative with government
bonds. As with volatility, extending the frequency exacerbates these correlations,
which are therefore more significant in the long than in the short term.
Smoothing causes an apparent reduction in risk, but also a delaying effect,
because the index’s value on a date includes average price indications for a
previous period. Additionally, the information transmission period also creates

136 Amundi Investment Strategy Collected Research Papers


a delay compared to the instant flow of information reflected in publicly quoted
prices. We have therefore calculated the correlations over rolling 10-year periods
both synchronously and with two- or four-quarter delays. For equities, the results
are given in the graph below.

10 years rolling correlation quarterly returns equities


real estate CBRE 1991-2013 France
50%

40%

30%

20%

10%

0%

-10%

CBRE vs SBF 250 synchrone CBRE vs SBF 250 (T-4)


CBRE vs SBF 250 (T-2) Source: Amundi Research

Looking just at synchronous correlation, it is around 40% today, but in the past, it
was practically zero. However, this is deceptive, because in the past the delayed
correlations were also on the same order, between 30% and 40%. This is therefore
a relatively stable correlation if time lags are taken into account, but those lags
have gotten smaller. To estimate a correlation over long time periods, the existence
of variable time lags has only a moderate effect. This may explain why computing
the figure using a biennial frequency gives a substantially higher result than when
using a quarterly frequency, if examining the period as a whole. Based on these
findings, we believe that over a long time period, the equities-real estate correlation
should be considered significantly positive. However, it is conceivable that working
with CBRE indices that only cover the “prime” Paris segment exacerbates this
correlation as a result of the increased sensitivity of this particular real estate
segment to economic conditions.

IV - Consequences for portfolio allocation


4.1 - An example

We shall start with an initial allocation of 25% equities and 75% government bonds,
and examine the impact of an increase in the riskiest asset classes (equities
and physical real estate) and reduction in bonds. To begin with, we increase the
share of equities by 10%. Next, instead of changing equities, we add 12.5% real
estate. This 12.5% figure was calculated in such a way as to take into account the
difference in volatility between the two asset classes (calculated based on the
average volatilities over both frequencies).

Amundi Investment Strategy Collected Research Papers 137


Impact on risk when increasing allocation to equities or real estate

10%
9%
8%
7%
6%
5%
4%
3%
2%
1%
0%
quarterly biennial
initial 25% equity-75% bonds +10% equity +12,5% real estate Source: Amundi Research

Note that the increase in risk is much smaller for real estate than for equities; the
result is somewhat less clear-cut with respect to the biennial frequency than for the
annual frequency. The diversifying power of physical real estate therefore appears
attractive, but is it being overestimated?

4.2 - Diversifying power of physical real estate depending


on the initial portfolio

We have examined the impact on risk when 10% cash is replaced with 10% real
estate. The results are different depending on the starting allocation and the
frequency.

Increase in risk when adding real estate


depending on initial allocation and frequency
10% cash-90% equity

10% cash-75% equity-15% bond

10% cash-50% equity-40% bond

10% cash-25% equity-65% bond

10% cash-90% bond

100% cash

-1,00% -0,50% 0,00% 0,50% 1,00% 1,50% 2,00% 2,50%


biennial quarterly Source: Amundi Research

Doing this with an initially 100% cash portfolio does not show the effects of
interaction with other asset classes. The volatility increase represents 10% of the
volatility of the added class. In other cases, the increase of risk that is obtained is
always significantly lower, because the risk that is introduced is diluted by the effect
of correlations. For bond-heavy portfolios, adding physical real estate might even
reduce long-term risk. Note that a simplistic calculation based on quarterly data

138 Amundi Investment Strategy Collected Research Papers


always gives a very moderate increase in risk, showing a good diversifying power.
However, for equity-heavy portfolios, examining the data calculated with a biennial
frequency shows that the increase in risk is substantially greater. In the long term,
the diversifying power of physical real estate is clear for bond-heavy portfolios, but
smaller for equity-heavy portfolios that might be suspected using higher-frequency
data. It should be noted that these findings relate to prime real estate, and are
probably less pronounced in other segments of the physical real estate market.

Conclusion
Physical real estate is a diverse asset class in which the risk and interaction with
other classes must be measured carefully. The results are very sensitive to the
types of historical data: expert valuation, or broker estimate. Due to the smoothing
and lags present in physical real estate indices, long-term risk is higher than a
typical volatility-based calculation would imply. The dependency with listed asset
classes is also more pronounced in the long term. The long-term risk of the “prime”
office segment in Paris is comparable to that of equities. It is lower for other offices,
and especially for residential. The correlation with equities is positive, and is
negative with government bonds. In allocating assets, the diversifying power of real
estate is most pronounced for bond-heavy portfolios in which adding real estate
may actually lower long-term risk. In equity-heavy portfolios, the positive long-term
correlation between real estate and equities greatly reduces the benefits in terms
of diversification. However, the smoothed valuation method of physical real estate
creates a favourable optical effect on short-term risk.

Amundi Investment Strategy Collected Research Papers 139


140 Amundi Investment Strategy Collected Research Papers
Amundi Working Papers

Amundi Investment Strategy Collected Research Papers 141


142 Amundi Investment Strategy Collected Research Papers
WP-050

Global Excess Liquidity


and Asset Prices
in Emerging Markets:
Evidence from the BRICS

JULIEN MOUSSAVI,
Ph.D. Student, Strategy and Economic Research, Amundi

April 2015

Since the early 2000s, global liquidity has experienced very


strong growth. Emerging Markets (EMs) have accumulated
large foreign exchange reserves while developed markets have
dramatically eased their monetary policies. Global excess
liquidity has resulted in an increase in the size of international
capital inf lows, especially toward EMs and may significantly
impact their financial stability. In this paper, we examine the
impact of global excess liquidity on asset prices for the well-
known BRICS countries. Using vector autoregressive and error
correction frameworks, we estimate the interaction between
global excess liquidity, economic activity and asset prices.
Despite mixed results for commodity prices, we show that global
excess liquidity causes significant increases in equity and bond
prices, a real appreciation of exchange rates, a decrease in 10-
year sovereign interest rates and a spread compression.

Amundi Investment Strategy Collected Research Papers 143


1. Introduction

Since the early 2000s and the bursting of the internet bubble, global liquidity has
experienced very strong growth and became excessive. We distinguish two different regimes
of global excess liquidity. Firstly, the saving glut in Emerging Markets (EMs) has fuelled
global excess liquidity, notably via the large accumulation of foreign exchange reserves.
Secondly, and in response to the global financial crisis of 2007-08 and the European
sovereign debt crisis of 2010, the central banks of the main Developed Markets (DMs) have
considerably eased their monetary policies by lowering interest rates and through successive
rounds of quantitative easing, mainly undertaken by the Federal Reserve (Fed), the Bank of
England (BoE) and the Bank of Japan (BoJ). More recently, the European Central Bank
(ECB) has also decided to increase the size of its balance sheet, to stop sterilising its
Securities Markets Programme and to launch its own quantitative easing. Global excess
liquidity has not resulted in a resurgence of inflation on a global scale, but rather in the
increase in the size of cross-border capital flows, especially towards EMs. However, these
capital flow surges, linked to global excess liquidity, are reversible and often end up in
sudden stops. Moreover, the risks of macroeconomic and financial imbalances in EMs have
been raised by many economists including Christine Lagarde1, the current IMF’s Managing
Director:

“Accommodative monetary policies in many advanced economies are likely to


spur large and volatile capital flows to emerging economies. This could strain the
capacity of these economies to absorb the potentially large flows and could lead
to overheating, asset price bubbles, and the build-up of financial imbalances.”

In the monetary sense, global liquidity is defined by the Bank for International Settlements
(BIS, CGFS, 2011) as the funding provided unconditionally to settle claims through the
monetary authorities. Excess liquidity can be measured by different aggregates such as the
money supply, domestic credit or also the foreign exchange reserves in excess of GDP.
Global excess liquidity appears to play a buffer role in the DMs’ deleveraging and is a
catalyst for growth in EMs. In the post-Lehman era, the Zero Interest Rate Policies (ZIRP)

1
Annual Meeting of the International Monetary Fund (IMF) and the World Bank in Tokyo, World Economic
Outlook, October 2012.

144 Amundi Investment Strategy Collected Research Papers


pursued by the Fed, the BoE, the BoJ and the ECB have fuelled massive capital inflows,
notably via some carry trade operations, mainly in Brazil and Russia. Furthermore, added to
these ZIRP, the non-conventional monetary policies have exacerbated the procyclical nature
of capital inflows towards EMs (Fratzscher et al., 2012) and raise concerns about the
emergence of bubbles in asset prices (Sidaoui et al., 2011), even though the emerging
financial markets are growing, i.e., they are more liquid, larger and deeper. In this context, the
EMs offer attractive returns for some risks which are still poorly assessed by investors.

In this paper, we explore how best to deal with global excess liquidty and to what extent it has
caused a rise in equity prices, a larger decline in EMs real interest rates than in the United
States, i.e., a spread compression, and a real appreciation of exchange rates, which is a new
issue that has not yet been discussed for EMs. Most studies have focused on the DMs and on
the impact of monetary expansion on GDP growth, consumer price inflation, short-term
interest rates or equity prices (Baks and Kramer, 1999; Gouteron and Szpiro, 2005; Rüffer
and Stracca, 2006; Giese and Tuxen, 2007; Sousa and Zaghini, 2007 and 2008; Belke et al.,
2010b). Some of them have broadened the scope to include more assets, e.g., bond, real
effective exchange rate, commodity or real estate markets (Sousa and Zaghini, 2008; Belke et
al., 2010a and 2013). More recent literature has transposed this issue to EMs (Rüffer and
Stracca, 2006; Hartelius et al., 2008; Brana et al., 2012). The large majority of authors who
have worked on this topic have used Vector AutoRegressive (VAR) models to analyse the
links between global excess liquidity and asset prices. They also have studied the Impulse
Response Functions (IRFs) to know more precisely how a shock on global liquidity could
affect asset prices.

Our first contribution is to build three different global excess liquidity aggregates to better
understand the contemporary relationship between global excess liquidity and asset prices.
Our second and main contribution is to analyse the impact of the rise in global excess
liquidity on different asset classes such as equities, interest rates, spreads, exchange rates and
some commodities, within VAR and Vector Error Correction (VEC) frameworks. Regarding
the results, according to the global excess liquidity aggregate and the models held, the IRFs
analysis leads us to conclude that there is a genuine link between global excess liquidity and

Amundi Investment Strategy Collected Research Papers 145


asset prices, notably on the BRICS’ real effective exchange rates. Overall, we find that global
excess liquidity causes significant increases in equity and bond prices, a real appreciation of
exchange rates, a decrease in 10-year sovereign interest rates and a spread compression.
However, the results about the impact of global excess liquidity on commodity prices are
more mixed.

The paper is organised as follows: As background, Section 2 focuses on the existing literature
pertaining to global excess liquidity, its measures and its links with the asset prices. Section 3
introduces the economic and financial data as well the different indices of global excess
liquidity we use. Section 4 presents our main findings, interprets them, and briefly points to
some robustness checks. We conclude our study in Section 5.

2. Literature review

Here, we address both theoretical and empirical foundations of global liquidity, its
excess as well as the links that may exist between global excess liquidity and asset prices.

2.1. Global excess liquidity and its measures

Global liquidity is a multifaceted and complex concept, which has often been suggested as an
explanation for financial developments. Here, we lean on two definitions of global liquidity
which are relevant both for policy makers and asset managers:

(i) Monetary liquidity, which is defined as the ease of converting monetary assets into
goods and services;
(ii) Financial market liquidity, which is defined as the ease with which large volumes of
financial securities can be bought or sold without affecting the market price.

According to the BIS (CGFS, 2011), monetary liquidity refers to the concept of “official” or
“public” liquidity and is defined as the funding provided unconditionally to settle claims
through the monetary authorities, comprising central bank money and foreign exchange
reserves. Concerning financial market liquidity, it refers to the concept of “private” liquidity,

146 Amundi Investment Strategy Collected Research Papers


i.e., created by the financial and non-financial sectors through, inter alia, cross-border
transactions. Chatterjee and Kim (2010) argue that financial market liquidity at the micro
level is related to a broader measure of liquidity at a macro level, i.e., monetary liquidity.
Adrian and Shin (2008) suggest that financial intermediaries raise their leverage during asset
price booms and lower it during downturns, procyclical actions that tend to exaggerate the
fluctuations of the financial cycle. They argue that the growth rate of aggregate balance sheets
may be the most fitting measure of liquidity in a market-based financial system. Moreover,
they show a strong correlation between balance sheet growth and the easing and tightening of
monetary policy. Monetary liquidity and financial market liquidity are similar notions and
their own dynamics interact in a coordinated way, notably through domestic credit 2.

The academic literature on this topic allows us to identify several indicators of global
liquidity. The most commonly used measures are the monetary and credit aggregates. In this
line, Baks and Kramer (1999) as well as Sousa and Zaghini (2007) propose different global
measures based on narrow (M1) and broad (M2 and M3) monetary aggregates for the G7 and
G5 countries respectively. Gouteron and Szpiro (2005) and Alessi and Detken (2011) suggest
using the domestic credit as it can be viewed as the main counterpart of monetary creation.
Another stream of the literature focuses on the foreign exchange reserves to assess global
liquidity. Indeed, this measure takes into account the increasing role of the liquidity created
by EMs (De Nicolo and Wiegand, 2007; Darius and Radde, 2010). Beyond these quantitative
indicators, price indicators can be used. Gouteron and Szpiro (2005) propose measuring
global excess liquidity from the short-term real interest rate (three-month interbank rate)
minus the natural interest rate3 and also from risk premiums4. However, we do not pursue
these price indicators further, because we prefer to use volume-based measures to explain
changes in asset prices.

2
Glocker and Towbin (2013) suggest that private liquidity shocks have a substantially larger effect on key
financial and macroeconomic variables, than public liquidity shocks. Moreover, they also show that global
liquidity shocks are more important on a medium-term horizon, than domestic liquidity shocks
3
The natural interest rate may be defined as the interest rate that establishes the equilibrium between supply and
demand on the goods and services market. It may notably be measured by the long-term economic growth.
4
The thinking behind this proposal is that excess liquidity could reduce the investors' risk aversion. Thus, the
spread between government and corporate bonds would constitute a measure of liquidity conditions.

Amundi Investment Strategy Collected Research Papers 147


In order to define more precisely the concept of global excess liquidity, we are using the
quantity theory of money5. This theory specifies that money supply has a direct and
proportional relationship with the price level. According to this theory and the liquidity
measures that are listed above, we can draw several normative conclusions: there is excess
liquidity when the money supply or the credit supply is too high in relation to transactions by
volume (goods and services or even assets). Baks and Kramer (1999) consider the average
growth rate of nominal GDP as a norm for money growth. In other words, this is the level of
liquidity that is consistent with the price stability. In the quantity theory of money, velocity is
assumed to be relatively constant and given the real GDP growth and the money supply
growth expectations, we can easily deduce the price trends. Following this hypothesis, the
following relationship can be established:

‫ܯ‬ ͳ (1)
‫ܯ‬ήܸ ൌܲήܳ ฻ ൌ ൌ ݇
ܲήܳ ܸ

After linearisation and differentiation, we obtain:

෥ ௧ ൌ ݉௧ െ ݃௧ 
݉ (2)

where ‫ ݐ‬denotes time, ݉


෥ denotes the observed excess liquidity, ݉ denotes the growth rate of
the chosen liquidity measure and ݃ denotes the nominal GDP growth rate.

5
The quantity theory of money specifies the causal relationship between the quantity of money in circulation and
the general price level. The first formulation of this theory goes back to the work of Jean Bodin in 1568 in which
he studied the inflationary effects of the large influx of gold from Latin America; this influx caused a price
increase across the European continent. The formalisation of this assumption is made in 1907 by Irving Fisher
who, through an accounting identity, linked the money supply, its velocity, the general price level and the
volume of transactions of goods and services. This theory is based on two presuppositions: (i) the change in the
quantity of money induces price changes in nominal terms. In other words, the source of inflation is
fundamentally derived from the growth rate of the money supply; (ii) Economic agents are supposed to be
rational, i.e., they know relative prices and are concerned only slightly in nominal prices. The accounting identity
is written as follows: ‫ ܯ‬ή ܸ ൌ ܲ ή ܳ where ‫ ܯ‬is the total amount of money in circulation on average in an
economy during a period, ܸ is the velocity of money in final expenditures, ܲ is the general price level and ܳ is
the real output which equals real expenditures in macroeconomic equilibrium.

148 Amundi Investment Strategy Collected Research Papers


2.2. The links between global excess liquidity and asset prices

By analogy with the quantity theory of money, we may reasonably assume that a surplus
of money that is not spent on the market of goods and services is highly likely to be spent on
the financial markets. However, even if we have clarified the concept of global excess
liquidity, the existence of links between rising global liquidity and rising asset prices via
higher transactions remains to be demonstrated. In addition to the quantity theory of money,
we need to find out more evidence on the links between monetary liquidity, funding liquidity6
and financial market liquidity. The following theories could explain these links:

According to Keynesian theory (Keynes, 1936), money demand satisfies three motives.
Transactions and precautionary motives are an increasing function of the income and
speculative motive is a decreasing function of the interest rate. Speculation takes the form of a
trade-off between holding money and holding long-term bonds. Incurring debt to buy
securities is particularly revealing of a process feeding bullish self-fulfilling expectations on
asset prices. Based on this assumption, the existence of a positive relationship between
liquidity and asset prices might seem almost trivial. With this in mind, we can easily realise
that the credit channel is a financial accelerator encouraging all the agents to indebtedness,
causing a cumulative process characterised by an increase in prices and debt especially to
acquire assets. In other words, we can establish that liquidity promotes the dynamics of
accumulation and thus the valuation of assets.

As stated by Stiglitz and Weiss (1981), the role of money may also be studied through its
counterpart of credit granting to the economy. Within this framework, a situation of abundant
liquidity is equivalent to low interest rates. Given that interest rates represent the cost of
capital, when interest rates are low, profitability is low and investors are willing to invest in
riskier assets resulting, de facto, in an increase of the price of these assets. This allows us to
establish that there is a negative relationship between interest rates and asset prices, and thus a
positive relationship between liquidity and asset prices.

6
Funding liquidity is defined as the ease with which market players can obtain funding.

Amundi Investment Strategy Collected Research Papers 149


According to the seminal contribution of Friedman (1988), the holding of money, which is
considered as an asset among others, is related to portfolio allocation. Thus, an increase in the
money supply leads to a portfolio reallocation. Therefore, if we assume that the quantity of
traded securities is fixed and the money supply increases, the price of other assets, i.e.,
equities, bonds, commodities, etc., is expected to rise in the same proportions as the price of
liquidity.

Furthermore, a common factor may lead to a simultaneous trend in monetary aggregates and
asset prices. This shock, whether positive or negative, is viewed as a signal, e.g., better
economic prospects lead to better expectations about future profits (Baks and Kramer, 1999).
Thus, the link between lower interest rates and an increase in the fundamental value of asset
prices follows from all the monetary policy transmission channels. Indeed, an accommodative
monetary policy informs agents on the willingness of financial authorities to support growth.
Investors therefore see this as a better outlook for future profits and start buying greater
amounts of risky assets.

Finally, Brunnermeier and Pedersen (2009) show that, under certain conditions (mainly boom
vs. bust cycles and monetary easing vs. tightening), financial market liquidity and funding
liquidity are mutually reinforcing, leading to liquidity spirals. They empirically explain that
market liquidity (i) can suddenly dry up, (ii) has commonality across securities, (iii) is related
to volatility, (iv) is subject to “flight to quality”, and (v) co-moves with the market. Without
loss of generality and given the link between the different liquidity concepts, we can
extrapolate these procyclical stylised facts to monetary liquidity.

2.3. Previous empirical contributions

In recent years, global excess liquidity has been mostly induced by ultra-
accommodative and non-conventional monetary policies conducted by the central banks of
the major developed countries, i.e., United States, United Kingdom, Japan and the Eurozone.
These monetary policies have had, inter alia, the effect of lowering the cost of liquidity to
international investors. This has led investors to search for yield by turning towards higher-
return, and therefore riskier, assets as argued by the IMF (2010) and Matsumoto (2011). This
resulted in massive capital inflows towards EMs notably through carry trade operations, with

150 Amundi Investment Strategy Collected Research Papers


Brazil, Russia, India, China and South Africa7 topping the list. In addition, the post-crisis
surge in capital flows has raised fears about the emergence of bubbles in asset prices (Sidaoui
et al., 2011), potential currency crises and the excessive growth of foreign exchange reserves,
while, at the same time, the emerging financial markets are growing, i.e., larger, deeper and
more liquid. Indeed, the ZIRP pursued by the Fed, the BoE, the BoJ and the ECB in the post-
Lehman era coupled with non-conventional monetary policies at the zero lower bound, i.e.,
quantitative easing, credit easing and signalling, have exacerbated the procyclical nature of
capital inflows towards EMs (Fratzscher et al., 2012).

We may wonder to what extent the abundance of global liquidity is responsible for upward
pressures on asset prices, especially in EMs. Few studies have directly investigated this issue.
Most studies focused on DMs and about the impact of money growth on GDP trends,
inflation, interest rate dynamics and equity prices. More recent literature transposes this
problematic to EMs and broadens the spectrum of the relevant assets, i.e., bonds, real
effective exchange rates and commodities. The vast majority of researchers who have worked
on this topic have used VAR models and have analysed IRFs.

Baks and Kramer (1999) study this issue for the G7 countries and conclude that global excess
liquidity has a negative impact on real interest rates and a positive impact on equity prices.
They also find some evidence for spillover effects from the volatility of money growth to the
volatility of equity prices across countries. By contrast, Gouteron and Szpiro (2005) find that
there is no common trend in asset prices, which is not supportive of a global effect of excess
liquidity. Rüffer and Stracca (2006) also examine the cross-border transmission channels of
global excess liquidity in fifteen DMs and EMs and find an expansionary effect in the
Eurozone and in Japan, though not in the United States. Furthermore, they highlight that
global excess liquidity is a useful indicator of inflationary pressure at a global level. Giese and
Tuxen (2007) show that the global excess liquidity has a positive impact on real estate prices
but not on equity prices for six major DMs. Sousa and Zaghini (2007 and 2008) identify that a
shock on global liquidity in the G5 countries has a positive impact on real GDP only in the
short term and a positive lagged impact on aggregate prices. They also find a temporary

7
These five EMs are better known by the acronym BRICS.

Amundi Investment Strategy Collected Research Papers 151


appreciation of the real effective exchange rate of the euro. Hartelius et al. (2008) highlight a
recent issue facing emerging bond markets: the spread compression with the United States.
They conclude that the convergence of bond yields in EMs as a whole to those of the United
States is largely due to improvement in fundamentals in EMs. However, they show that global
excess liquidity plays an important role in spread compression. Belke et al. (2010a) study
eleven major OECD countries and find that monetary aggregates may convey useful leading
indicator information on real estate prices, gold prices, commodity prices and the GDP
deflator at the global level. In contrast, they emphasise that equity prices do not show any
positive response to a liquidity shock. Brana et al. (2012) find support that global excess
liquidity generates significant spillover effects for sixteen major EMs taken as a whole.
Global excess liquidity contributes to the increase in GDP and in consumer prices in these
EMs. However, they conclude that the relationship between global liquidity shocks and equity
prices or real estate prices is weaker. Finally, Belke et al. (2013) find that a positive long-term
relationship exists between global liquidity and the trends in food and commodity prices.

3. Data

In this study, we gather data for eight countries and one monetary union, representing
nearly 70% of world GDP in Purchasing Power Parity (PPP) in 2014. This set of countries is
composed of the G4 countries, i.e., United States, United Kingdom, Japan and Eurozone, and
the well-known BRICS countries, i.e., Brazil, Russia, India, China and South Africa. The data
are collected for each country or monetary union on a quarterly frequency over a sample
period from Q1 1998 to Q1 2014, or 65 quarters.

3.1. Economic and financial data

We use economic and financial data from different sources across variables and
countries; they include the IMF, the World Bank, the OECD, the Bank for International
Settlements, Eurostat, Oxford Economics and Datastream databases. More formally, the data
we use are:

152 Amundi Investment Strategy Collected Research Papers


(i) Fundamental economic data: nominal GDP in local currencies and in USD, PPP GDP8
and consumer price indices;
(ii) Monetary and financial data: exchange rates against the USD, broad based real and
nominal effective exchange rates9, M2 monetary aggregates, domestic credit
aggregates and foreign exchange reserves;
(iii) Market data: MSCI in local currency10, EMBI Global11 and 10-year sovereign interest
rates;
(iv) Different indices and prices of the main commodities: GSCI12, CRB13, LMEX14,
gold15 and Brent crude oil16.

Some data are seasonally adjusted using the X-12-ARIMA procedure17 if necessary.
Furthermore, we use ex post revised data for most of the economic, monetary and financial

8
Depending on the paper, the weights used to build the aggregates of global excess liquidity are done either with
nominal GDP or with PPP GDP. In this study, we use nominal GDP weights to not underweight EMs in the
aggregates but using PPP GDP weights leads to similar results.
9
Real and nominal effective exchange rates indices (REER and NEER hereafter) are provided by the Bank for
International Settlements and cover 61 economies including individual Eurozone countries and, separately, the
Eurozone as an entity. REER and NEER are calculated as geometric weighted averages of bilateral exchange
rates, adjusted by relative consumer prices in the case of REER. The weighting pattern is time-varying, and the
most recent weights are based on trade from 2008 to 2010.
10
The Morgan Stanley Capital International are the indices most regularly followed by market participants. They
measure the performance of equity markets in countries or the aggregate of countries to which they refer. We
also retain the MSCI BRIC in local currency. This index is a free float-adjusted market capitalisation weighted
index that is designed to measure the equity market performance of the following four EM country indexes as a
whole: Brazil, Russia, India and China.
11
The Emerging Markets Bond Index Global are indices of JPMorgan Chase which track the total returns of debt
securities traded abroad in EMs. The EMBI Global indices are an expanded version of the EMBI+ indices.
12
The Goldman Sachs Commodity Index is an index originally developed by Goldman Sachs and which the
ownership has been transferred to Standard & Poor's. It serves as a benchmark for investment in the commodity
markets and comprises 24 commodities from all commodity sectors.
13
The Commodity Research Bureau is an index of listed commodities on New York Mercantile Exchange,
London Metal Exchange and Chicago Mercantile Exchange. It comprises 24 commodities from all commodity
sectors.
14
The London Metal Exchange Index is the benchmark for the listing of six main nonferrous metals, i.e., copper,
tin, lead, zinc, aluminum and nickel. In recent years, the LME has become a speculative market. Indeed, the
share of commodities actually delivered after establishing a contract on the LME fell below 1%.
15
Gold spot price in USD per ounce.
16
We chose Brent crude oil rather than West Texas Intermediate crude oil because Brent crude oil serves as a
major benchmark price for purchases of crude oil worldwide. It is used to price two thirds of the world's
internationally traded crude oil supplies. However, both kinds of crude oil are traded in a narrow range.
17
The procedure is performed on ex post revised data. Nevertheless, some variables are already seasonally
adjusted. Market data do not need to be seasonally adjusted.

Amundi Investment Strategy Collected Research Papers 153


variables. Each variable, other than interest rates, were log-transformed. This especially
allows a return to variables integrated of order one (cf. Section 4.2 for more details) and
results to be analysed more easily: the estimated coefficient can be interpreted as elasticities.

The mechanism we seek to highlight in this study may be interpreted differently depending on
whether we consider the nominal or real terms approach. From a theoretical point of view, the
valuation of assets is related to their nominal incomes, which in turn depend on the level of
prices of produced goods and services. This is the reason why we make this study on real
data. To do this, we multiply the variables of interest by the GDP deflator of the country or
monetary union and over the period being considered18. However, working with nominal data
amplifies the highlights that emerge from this study.

3.2. The different global excess liquidity indices

In order to account for global excess liquidity, we proceed in two steps. The first step is to
hold different measures of nationwide monetary liquidity. In a second step, we aggregate
these indices to establish a snapshot of global excess liquidity. We hold three indices of
excess liquidity on criteria such as economic relevance, data availability and homogeneity: (i)
M2 monetary aggregate, (ii) domestic credit and (iii) foreign exchange reserves. Each of these
measures is expressed as a share of GDP in local currencies for the first two indicators and in
USD for the third one. Moreover, each of these aggregate measures were log-transformed to
take into account the liquidity in excess of GDP. The aggregation of national series at a global
level raises some issues from an economic standpoint. Indeed, such aggregate measures
cannot be used for monetary and fiscal policy decisions at a global level19. However, the
purpose of this study is to better understand how monetary liquidity behaves and interacts
globally. There are different methods of aggregation but the non-stationarity of these time
series and structural breaks imply that no optimal aggregation method exists (Giese and
Tuxen, 2007). Nevertheless, Beyer et al. (2001) discuss various criteria in order to get a
18
In order to get real interest rates, nominal interest rates have been deflated by the annual average of domestic
inflation.
19
However, in an environment of global excess liquidity, and thus surges in capital flows, it is important for
EMs to ensure financial and economic stability through improved financial regulation and other policy measures.
Azis and Shin (2014) explore the range of policy options that may be deployed to address the impact of global
liquidity on domestic financial and socio-economic conditions.

154 Amundi Investment Strategy Collected Research Papers


useful aggregate measure of the historical Eurozone data. To this end, the authors propose the
following three criteria:
(i) A unique price series should be obtained in the sense that the aggregate of the
individual price deflators coincides with the price deflator of the aggregates;
(ii) When a variable increases or decreases in each country, then the aggregate measure
should not move in the opposite direction;
(iii) Aggregation should work correctly when different local currencies are used and, a
fortiori, when a common currency is used.

The method suggested by Beyer et al. (2001) uses variable weights to aggregate growth rates
and proceeds in the following four steps (cf. Methodological Appendix for more details):

(i) Calculate weights based on the relative share of the country or monetary union as
regards the variable at each date, in a common currency, e.g., in USD in this study;
(ii) Calculate within country or monetary union growth rates of each variable at each date,
in local currency;
(iii) Aggregate growth rates of the second step using weights of the first step;
(iv) aggregate growth rates to obtain aggregate levels. We use the Q1 1998 as the base to
anchor the aggregate measures over time.

Figure 1. Trends in the different global excess liquidity indices


Note: The figure plots the M2 over GDP index (continuous line), the domestic credit over GDP index
(dashed line) and the foreign exchange reserves over GDP index (dotted line). We distinguish two
different regimes of global excess liquidity. Firstly, and after the financial crisis that followed the
Internet bubble bursting, the saving glut in EMs has fuelled global excess liquidity, notably via the
large accumulation of foreign exchange reserves. Secondly, and in response to the global financial
crisis of 2007-08 and the European sovereign debt crisis of 2010, the central banks of the main DMs
have considerably eased their monetary policies by lowering interest rates and through non-
conventional tools, mainly undertaken by the Fed, the BoE, the BoJ and more recently by the ECB.
These highly accommodative monetary policies have led to a drastic increase in the monetary base M0
and consequently in the monetary aggregates like M2 since the end of 2008 and the announcement by
the Fed of its first round of quantitative easing. Moreover, the period of credit crunch that started in
2001 seemed to find an ending at the dawn of the first announcements of quantitative easing.
Unfortunately, the announcement effect only lasted a short time and domestic credit contracted once
again even though the M2 monetary aggregate continued to increase. For more details on the trends in
the different global excess liquidity indices by regional aggregates, cf. Appendix 1.

Amundi Investment Strategy Collected Research Papers 155


Figure 1 informs us on the global liquidity trend since Q1 1998 through three different
indices, namely the M2 over GDP index, the domestic credit over GDP index and the foreign
exchange reserves over GDP index. It shows two distinct regimes of global excess liquidity.
Firstly, and after the financial crisis that followed the Internet bubble bursting, the saving glut
in EMs has fuelled global excess liquidity, notably via the large accumulation of foreign
exchange reserves. Secondly, and in response to the global financial crisis of 2007-08 and the
European sovereign debt crisis of 2010, the central banks of the main DMs have considerably
eased their monetary policies by lowering interest rates and through non-conventional tools,
mainly undertaken by the Fed, the BoE, the BoJ and more recently by the ECB. These highly
accommodative monetary policies have led to a drastic increase in the monetary base M0 and
consequently in the monetary aggregates like M2 since the end of 2008 and the announcement
by the Fed of its first round of quantitative easing. Moreover, the period of credit crunch that
started in 2001 seemed to find an ending at the dawn of the first announcement of quantitative
easing. Unfortunately, the announcement effect only lasted a short time and domestic credit
contracted once again even though the M2 monetary aggregate continued to increase. For
more details on the trends in the different global excess liquidity indices by regional
aggregates, cf. Appendix 1.

156 Amundi Investment Strategy Collected Research Papers


4. Impact of global excess liquidity on BRICS’ asset prices

In this context, it seems interesting to investigate the potential impacts of global excess
liquidity on EM asset prices, and more particularly in the BRICS countries, as well as on
commodities which are mostly exported by these same EMs. After describing the economic
and financial environment in which this study is conducted, we develop the framework of the
model and look into the main results.

4.1. Economic and financial analysis

From an economic standpoint, the BRICS countries alone represent more than 28% of
world GDP in PPP in 2014 for more than 3 billion people, almost half of the Earth’s
population. According to the IMF, this group of EMs will account for nearly a third of world
GDP in PPP in 2020. These five EMs, i.e., Brazil, Russia, India, China and South Africa are
respectively the seventh, sixth, third, second and twenty-fifth largest economies in the
world20. Table 1 provides information about the average weights of each country or monetary
union over the whole sample period as a percentage of GDP in PPP in each of the three
aggregates considered. In the light of this table, we can say that China and India are in the
BRICS countries aggregate what the United States and Eurozone are in the G4 countries
aggregate, i.e., the largest contributors to global growth. The GDP per capita of the BRICS
countries is growing very rapidly, but it is expected to remain far below the standards of DMs
even on a very long-term horizon. The BRICS countries are currently strengthening their
economic and financial cooperation. Indeed, we can mention for example the New
Development Bank, formerly referred to as the BRICS Development Bank, which is a
multilateral development bank operated by the BRICS countries as an alternative to the
existing United States-dominated World Bank and IMF. This New Development Bank was
agreed by BRICS countries leaders at the fifth BRICS summit held in Durban, South Africa
on March 2013 and was ratified at the sixth BRICS summit held in Fortaleza, Brazil on July
2014. It is in this way, i.e., by creating multilateral supervisory and regulatory agencies, that
the EMs and, a fortiori, the BRICS countries are becoming the most attractive financial
markets in the world. Brazil and Russia produce and export crude oil and natural gas in large

20
This ranking stems from the IMF list of countries by GDP in PPP in 2013.

Amundi Investment Strategy Collected Research Papers 157


quantities, while China and India are undergoing an accelerated industrialisation process,
which requires a lot of energy. Meanwhile, South Africa extracts metals and minerals from its
mines.

Table 1. Average weights of each country or monetary union over the entire sample period
Note: The table provides information about the average weights of each country or monetary union
over the entire sample period as a percentage of GDP in PPP in each of the three aggregates
considered. Standard deviations are in parentheses. In light of this table, we can say that China and
India are in the BRICS countries aggregate what the United States and Eurozone are in the G4
countries aggregate, i.e., the largest contributors to global growth. Moreover, if we look at how the
weights have changed over time, it appears that the weight of the G4 countries have tended to decline
in favour of those of the BRICS countries. Indeed, the weight of China rose from slightly more than
9% in early 1998 to nearly 24% in early 2014. During the same period, the weights of the United
States and the Eurozone fell from about 34% and 26% to 28% and 19%, respectively.

Aggregate G4 countries BRICS All


Country/Area countries
United States 45.9% 31.5%
(0.9%) (2.2%)
Eurozone 33.6% 23.2%
(0.5%) (2.3%)
United Kingdom 6.8% 4.7%
(0.1%) (0.4%)
Japan 13.7% 9.5%
(0.5%) (1.1%)
Brazil 13.6% 4.1%
(2.6%) (0.1%)
Russia 13.6% 4.1%
(1.8%) (0.3%)
India 21.6% 6.7%
(0.4%) (1.2%)
China 47.8% 15.2%
(5.2%) (4.6%)
South Africa 3.4% 1.0%
(0.6%) (0.0%)

Although the responsibility of central banks in global excess liquidity that has fed speculative
bubbles in the DMs has often been mentioned, it is not trivial that the same phenomenon
occurred in EMs. Indeed, in a global economy with a structurally high savings rate, low
employment rate and where the global excess liquidity has no impact on the prices of goods

158 Amundi Investment Strategy Collected Research Papers


and services, we may wonder if there is an inflation of asset prices in EMs and if it is actually
fuelled by global excess liquidity. According to Artus and Virard (2010), at the end of 2009,
the root causes of the financial imbalances have not disappeared because the two liquidity
making machines, i.e., the very accommodative monetary policies of major central banks in
DMs and the accumulation of foreign exchange reserves in EMs, continued to run at full
speed. Moreover, we can say that these mechanisms are still at work in 2014 even though they
are of different forms. At the present time, even though the Fed and the BoE have stopped
their non-conventional monetary policy, the BoJ continues to inject a lot of liquidity and the
ECB has recently launched a major quantitative easing coupled with an Asset-Backed
Securities Purchase Programme and a Covered Bond Purchase Programme. Regarding EMs,
the People’s Bank of China joined their developed counterparts in boosting liquidity to
address weakening growth and promote credit expansion. In addition and in response to the
appreciation of the dollar induced by the tightening of the Fed’s monetary policy, the EMs
will have to resume their policy of accumulating foreign exchange reserves to protect
themselves against the depreciation of their currencies. As we have seen above, we
distinguish two different regimes of global excess liquidity. Firstly, and after the financial
crisis that followed the Internet bubble bursting, the saving glut in EMs has fuelled global
excess liquidity, notably via the large accumulation of foreign exchange reserves. Secondly,
and in response to the global financial crisis of 2007-08 and the European sovereign debt
crisis of 2010, the central banks of the main DMs have considerably eased their monetary
policies by lowering interest rates and through non-conventional tools, mainly undertaken by
the Fed, the BoE, the BoJ and more recently by the ECB.

The first regime of global excess liquidity is typical of a global economy where distortions in
terms of liquidity are exacerbated. Indeed, during the first regime of global excess liquidity,
i.e., from 2001 to 2008, we see a significant accumulation of foreign exchange reserves. This
increase in the foreign exchange reserves to GDP ratio mainly comes from the BRICS
countries. Over this period, Russia saw its foreign exchange reserves to GDP ratio multiplied
by almost four, China and India by more than three, Brazil and South Africa by around two
and a half. The G4 countries increased their foreign exchange reserves sparingly, Japan at the
top of the list. Japan adopted, through its central bank, a highly accommodative monetary
policy in order to support its own currency.

Amundi Investment Strategy Collected Research Papers 159


The second regime of global excess liquidity is characterised by a jump in the global M2 to
GDP ratio. This ratio has increased very rapidly in both EMs and DMs and has resumed a
more sustainable trend thereafter. However, there is an apparent dichotomy between the G4
countries and the BRICS countries. Indeed, although the acceleration of the increase in M2
over GDP indices is fairly similar in the two groups of countries, the G4 countries have higher
ratios than the BRICS countries. According to the World Bank, in 2013, the M2 to GDP ratios
are quite disparate for the G4 countries but very high: around 90% for the United States and
the Eurozone, 160% for the United Kingdom and nearly 250% for Japan. The M2 to GDP
ratios for the BRICS countries are relatively lower: 56% for Russia, between 70% and 80%
for South Africa, India and Brazil and nearly 200% for China. With regard to China, this very
high M2 to GDP ratio reflects the excessive monetisation of the financial system and the
indebtedness promoted by the Chinese authorities, notably to control their currency.

4.2. Model specification

In order to study the dynamic contemporary relationships between our aggregates of


global excess liquidity and the BRICS’ asset prices21, we follow the standard practices of time
series analysis assuming that the properties of linear regressions are biased for non-stationary
variables. We therefore start by testing the stationarity of our three aggregates of global
excess liquidity, the real GDP of each of the five EMs, the different asset prices, yields,
spreads and exchange rates of each of these same five EMs and some commodity prices with
Augmented Dickey-Fuller (1981, ADF hereafter) and Phillips-Perron (1987 and 1988, PP
hereafter) unit root tests22. The unit root tests results show us that in more than 85% of cases,
the series are integrated of the same order, namely the order one, i.e., ‫ܫ‬ሺͳሻ. In addition, all of
our global excess liquidity aggregates and real GDP of each of the five EMs are ‫ܫ‬ሺͳሻ. Only a
handful of EMs’ interest rates and spreads are stationary in level, i.e., ‫ܫ‬ሺͲሻ. Since the

21
As we have seen before, each variable, other than interest rates, were log-transformed but we deliberately omit
to specify that our variables are transformed for the sake of convenience.
22
The use of several tests to conclude on the nature of stationarity of the studied variables is essential to
disambiguate on some test results. Indeed, the PP unit root tests diơer from the ADF tests mainly in how they
deal with serial correlation and heteroskedasticity in the errors. In particular, where the ADF tests use a
parametric autoregression to approximate the ARMA structure of the errors in the test regression, the PP tests
ignore any serial correlation in the test regression.

160 Amundi Investment Strategy Collected Research Papers


overwhelming majority of our series are ‫ܫ‬ሺͳሻ, it is highly possible that these series are
cointegrated.

Then, to test whether the series are cointegrated and, if so, how many cointegrating
relationships exist, we use the Johansen procedure (Johansen, 1991). After having used
Akaike and Schwarz information criteria to determine the optimal number of lags that would
need to be considered23, we conclude that, in more than 60% of cases, at least one
cointegration relationship exists. Then, we perform Granger non-causality tests (Granger,
1969) on the remaining 40% to find out if the different global excess liquidity aggregates
Granger-cause the different asset prices. According to these tests, in almost 10% of cases,
some short-term relationships exist as opposed to the long-term relationships of cointegrated
models. Finally, in about 30% of cases, we do not estimate any model to avoid spurious
regressions, either because the variables which are integrated of a different order cannot be
cointegrated, or because no causal relationship exists.

In this study, we use the standard time series modelling taking into account the results of the
preliminary tests explained above, i.e., ADF, PP, Johansen cointegration and Granger non-
causality tests. We use two different models to better capture the nature of the relationships
between our time series. In the case where at least one cointegration relationship exists, we
estimate a VEC model as in (3) and in the case where no cointegration relationship exists but
that the global excess liquidity appears to be causal, in the Granger sense, for asset prices, we
estimate a VAR model as in (4):

ȟܻ௜௧ ൌ ܿ௜ ൅ ෍ ߛ௜௟ ȟܻ௜௧ି௟ ൅ ߜ௜ ൫ߙ௜ ൅ ܾ௜ ή ‫݀݊݁ݎݐ‬௜ ൅ ߚ௜ ܻ௜௧ିଵሻ ൯ ൅ ߝ௜௧ (3)


௟ୀଵ

ȟܻ௜௧ ൌ ܿ௜ ൅ ෍ ߛ௜௟ ȟܻ௜௧ି௟ ൅ ߝ௜௧ (4)


௟ୀଵ

23
In most cases, minimising Akaike and Schwarz information criteria leads us to conclude that the optimal
number of lags is one. In some cases, this optimal number goes up to two or three.

Amundi Investment Strategy Collected Research Papers 161


where ݅ denotes the different BRICS countries, ‫ ݐ‬denotes time and ݈ denotes the optimal
number of lags with ‫ ܮ‬ൌ ሼͳǡʹǡ͵ǡͶሽ. ܻ denotes a vector containing the endogenous variables of
the system, i.e., the different assets (alternatively equity prices, bond yields, spreads,
exchange rates and some commodity prices), the real GDP and the global excess liquidity
(alternatively one of the three global excess liquidity aggregates). For each of the two
different models, ܿ denotes the constant term and ߝ the error term. In the VEC model in (3),
൫ߙ ൅ ܾ ή ‫ ݀݊݁ݎݐ‬൅ ߚܻ௧ିଵሻ ൯ represents the cointegration relationship and eventually includes a
constant term and/or a linear trend. In addition, a long-run relationship exists only if ߜ, which
measures the speed of adjustment of the endogenous variables towards the equilibrium, is
significantly negative.

All in all, over the 123 estimable models24, and according to the preliminary tests results,
more than 70% are indeed estimated. In order to validate the stationarity and stability of these
models, we propose a kind of robustness check. It is well known in the literature on VAR and
VEC models that the stationarity and stability properties depend on the roots of the lag
polynomial. In particular, if all the inverted roots of the lag polynomial are strictly inside the
unit circle, then the VAR process is stationary. For the VEC process, ݇ െ ‫ ݎ‬roots should be
equal to the unity and so ݇ሺ‫ ݌‬െ ͳሻ ൅ ‫ ݎ‬inverted roots should be strictly inside the unit circle,
where ݇ is the number of endogenous variables, ‫ ݎ‬is the number of cointegration relationships
and ‫ ݌‬is the largest lag. According to this robustness check, only one VEC model is unstable
and hence, this estimate is excluded from the study.

4.3. Global excess liquidity promotes the search for yield

Here, we want to highlight the positive impacts of global excess liquidity in some
BRICS’ assets. Depending on the countries and assets, responses to a shock on liquidity have
the expected sign in more than half of cases. Finally, to better identify their sensibility to
some economic and econometric changes, we propose two additional robustness checks. First,

24
The 123 estimable models break down as follows: five EMs, each with seven different assets, plus six
different global assets, i.e., MSCI BRIC, some commodities and some commodity indices, and three different
global excess liquidity aggregates for a total of ሺͷ ή ͹ ൅ ͸ሻ ή ͵ ൌ ͳʹ͵.

162 Amundi Investment Strategy Collected Research Papers


we compare the results of our estimates in real terms with estimates in nominal terms and
second, we estimate our model in a panel approach.

4.3.1. Impulse response functions and variance decompositions

We want to see how the different BRICS’ assets are impacted by the increase in global
liquidity as measured by our three different indicators of global excess liquidity. To do this,
we look at how the assets react to a positive one standard deviation shock on the logarithm of
each liquidity aggregate. We focus on reviewing the Impulse Response Functions (IRFs).
According to the common practices, we estimate the IRFs with their confidence intervals. We
compute these confidence intervals using Monte-Carlo simulations in the case of the VAR
model and using the bootstrap method25 in the case of the VEC model. If confidence intervals
do not contain 0, the IRF is significant. If confidence intervals contain 0, the IRF is not
significant but we keep the sign of the IRF as a result.

Table 2. Summary results of the IRFs


Note: The table provides information about the results of the simulated IRFs based on the
estimated VAR and VEC models. For each BRICS country, we analyse the IRF of each asset
price or exchange rate to a positive one standard deviation shock on the logarithm of each
liquidity aggregate, except for the MSCI BRIC and commodity prices which are dealt with
more globally. The symbol “--” denotes a negative and significant impact to a given asset
price of a one standard deviation shock on a given liquidity aggregate; “-” denotes a negative
and non-significant impact; “0” denotes no impact; “+” denotes a positive and non-significant
impact; “++” denotes a positive and significant impact; an empty cell denotes that no model
has been estimated according to the preliminary unit root and cointegration tests. For
example, in the case of the MSCI Brazil, a one standard deviation shock on the M2 over GDP
aggregate is associated with an increase in equity prices but the impact is non-significant
according to the 95% confidence interval. However, a one standard deviation shock on the
domestic credit over GDP aggregate or on the foreign exchange reserves over GDP aggregate
(FX Reserves in the table below) is associated with a significant increase in equity prices. We
can therefore conclude that the global excess liquidity has a positive impact on the price of
Brazilian equities as reflected by the MSCI Brazil.

25
Theoretically, it is possible to compute analytical confidence intervals using an asymptotic approximation, but
this may lead to misleading confidence intervals because asymptotic formulas are known to give a poor
approximation of the finite-sample properties.

Amundi Investment Strategy Collected Research Papers 163


Country Brazil Russia India China South
Asset Class / Asset / Liquidity Aggregate Africa
M2 + + +
MSCI Credit ++
Equity

FX Reserves ++ ++ ++ ++
M2
MSCI BRIC26 Credit
FX Reserves ++
M2 - + - - +
10Y Interest Credit - - -
Fixed Income

Rate
FX Reserves - - - -- --
M2 -- -- 0 0 --
EMBI Global Credit - + 0 +
Spread
FX Reserves - + 0 0
M2 - + +
Exchange Rate
Exchange Rate

vs. USD Credit ++ + +


FX Reserves - + ++ 0 ++
M2 - - - ++
REER Credit + +
FX Reserves - ++ + -
M2 --
GSCI Credit
Commodity

FX Reserves ++
M2 --
Brent Credit
FX Reserves ++

As we can see in Table 2, global excess liquidity has played an important role in the evolution
of some asset prices. Overall, global excess liquidity pushed up equity prices and exchange
rates, while it brought down the fixed income rates and has more or less tightened the interest
rate spreads depending on the countries. With more granularity, several highlights appear:

26
For the MSCI BRIC, we can only consider the nominal terms approach because it could be difficult and
misleading to deflate an aggregated equity index.

164 Amundi Investment Strategy Collected Research Papers


(i) The asset class which is the most impacted by global excess liquidity is the BRICS’
equity markets. All countries except South Africa have seen their MSCI indices
increase with global liquidity. We obtain the same results for the MSCI BRIC in local
currencies. It is the variation in the foreign exchange reserves over GDP aggregate
which has the greatest impact on the equity prices.

(ii) On the fixed income side, the foreign exchange reserves over GDP aggregate and, to a
lesser extent the M2 over GDP aggregate, contributed to lower sovereign 10-year
interest rates in all the BRICS. According to the M2 over GDP aggregate, spread
compression is significant for Brazil, Russia and South Africa.

(iii) Concerning the foreign exchange markets, the currencies have globally appreciated
against USD and in real effective terms. Except for Brazil, once again, it is the foreign
exchange reserves over GDP aggregate which is the most significant measure of
global excess liquidity for EMs.

(iv) Finally, as for the three different global excess liquidity aggregates, the results for the
main commodity prices, i.e., the GSCI and the Brent, are more mixed. They were
positively impacted during the first regime of global excess liquidity, when EMs
accumulated some large foreign exchange reserves from early 2001 to mid-2008, by
the yardstick of the global financial crisis; whereafter the second regime of global
excess liquidity takes place. In this second regime, the commodity prices started to fall
since the developed central banks have injected significant liquidity until 2015.

While IRFs evaluate the effects of a shock to one endogenous variable on the other variables
in VAR or VEC models, variance decomposition separates the variation in an endogenous
variable into the component shocks to VAR or VEC models. Thus, the variance
decomposition provides information about the relative importance of each random innovation
in affecting the variables in VAR or VEC models. In order to remain consistent in our
approach, we will consider only the variance decomposition for the models we discussed
above.

Amundi Investment Strategy Collected Research Papers 165


By analysing the relevant variance decompositions (some examples are available in Appendix
2), we can draw several conclusions. First, after sixteen quarters, more than 80% of the
BRICS’ asset innovations are explained by their own innovations in about two thirds of cases.
Second, in the remaining one third, the BRICS’ asset innovations are mainly explained by the
different global excess liquidity aggregates innovations. Third, the more significant the IRF,
the more the variance decomposition is explained by the different global excess liquidity
aggregates innovations. Lastly, within the global excess liquidity aggregates, it is the foreign
exchange reserves and, to a lesser extent, the M2 aggregate which explain the BRICS’ asset
innovations.

4.3.2. Robustness checks

In order to ensure the robustness of our results, we propose two robustness checks.
First, we follow the same time series methodology replacing data in real terms by data in
nominal terms. Second, we estimate a Panel Dynamic Ordinary Least Squares model (PDOLS
hereafter) with country fixed effects. This PDOLS model, introduced by Kao and Chiang
(2000) and refined by Mark and Sul (2003), involves augmenting the panel cointegrating
regression equation with cross-section specific lags and leads of the explanatory variables in
first difference to eliminate the asymptotic endogeneity and serial correlation.

For our first robustness check, we apply exactly the same methodology to the data in nominal
terms27. The conclusion is that the same aggregates of global excess liquidity lead the same
assets upward or downward whether in real or nominal terms (for more detailed results, cf.
Appendix 3). The main difference between this two estimates is the amplitude of the IRFs to a
shock on global liquidity. Indeed, in the broader sense, global excess liquidity causes a
significant increase in equity and bond prices, an appreciation of exchange rates, a decrease in
10-year sovereign interest rates and a spread compression both in real and nominal terms.
Moreover, the IRFs in nominal terms are more significant than the IRFs in real terms. In the
case of Russia and compared with the real terms approach, a one standard deviation shock on
27
We use non-deflated economic and financial data. For example, for each country and each asset class, we use
the nominal GDP growth, the 10-year nominal sovereign interest rate, the nominal effective exchange rate, etc.
Obviously, our three global excess liquidity aggregates are also used in nominal terms for consistency

166 Amundi Investment Strategy Collected Research Papers


the foreign exchange reserves aggregate leads to a higher increase in the NEER. In the case of
India, the IRF on the NEER to a shock on the foreign exchange reserves aggregate remains
positive, as for the REER, but becomes significant. The same is true in other settings, e.g., for
the Brazilian and Chinese equity markets, for the Russian and Chinese 10-year sovereign
interest rates, and for the Brazilian and South African spread compressions. Regarding the
variance decompositions in the nominal terms approach, we get the same qualitative
conclusions as for the real terms approach. In addition, after sixteen quarters, the BRICS’
asset innovations are more explained by the different global excess liquidity aggregates
innovations in the nominal terms approach than in the real terms approach. Overall, this first
robustness check attests to the relevance of our main results.

Our second robustness check consists in the estimation of a PDOLS model. We chose this
panel model because it has several advantages. First, the panel approach, with its structure in
two dimensions, provides more complete information than in the time series approach. More
precisely, we can better understand our issue and provide a more global answer together with
more granularity on the question of the different global excess liquidity regimes. Second,
according to Kao and Chang (2000) and Mark and Sul (2003), the PDOLS estimators appear
to outperform all other panel estimators for non-stationary panel data, e.g., the Panel Fully
Modified OLS. In order to avoid some statistical bias in the estimates of the links between
global excess liquidity and the EMs asset prices, we add a control variable that reflects the
implied volatility of S&P 500 index options, i.e., the VIX index. Well-known as a “fear
index” for worldwide asset markets, it reflects both stock market uncertainty and a variance
risk premium. The first step of the panel analysis is to investigate the statistical properties of
our stacked data. Hence, we perform some panel stationarity and unit root tests and we
reasonably conclude that our variables are non-stationary in level and ‫ܫ‬ሺͳሻ (for more detailed
results, cf. Appendix 4). Then, we perform some panel cointegration tests28 in order to verify
the presence of a long-run relation between the variables in our dataset and we conclude that
our series are cointegrated in more than 85% of cases29. However, we do not find evidence

28
Here, we use the well-known panel cointegration tests proposed by Pedroni (1999).
29
Because of the huge number of cointegration tests and space limitation, the panel cointegration tests results are
not reported but available upon request.

Amundi Investment Strategy Collected Research Papers 167


that there could be some cointegration relationships on the two sub-periods that characterise
the two global excess liquidity regimes, i.e., from Q3 2000 to Q2 2008 and from Q3 2008 to
Q1 2014. After having highlighted the presence of cointegration relationships in the full
sample period, we estimate a PDOLS model as in (5):
௅మ

ܻ௧ ൌ ܿ௜ ൅ ෍ ߛ௜௝௟ ȟ௝௧ା௟ ൅ ߜ௝ ܺ௝௧ ൅ ߝ௧ (5)


௟ୀି௅భ

where ݅ denotes the different BRICS countries, ‫ ݐ‬denotes time and ݈ denotes the optimal
number of lags and leads30 with ‫ܮ‬ଵ ൌ ሼͳǡʹǡ͵ǡͶሽ and ‫ܮ‬ଶ ൌ ሼͳǡʹǡ͵ǡͶሽ. ܻ denotes the different
dependent variables, i.e., the different assets (alternatively equity prices, bond yields, spreads,
exchange rates and some commodity prices). ܺ௝ denotes the ݆ different explanatory variables,
i.e., the real GDP, the VIX index and the global excess liquidity (alternatively one of the three
global excess liquidity aggregates). Country fixed effects are denoted by ܿ and ߝ denotes the
error term. In addition, the short-run dynamics coefficients ߛ are allowed to be cross-section
specific.

Table 3. Summary of PDOLS estimates of the links between global excess liquidity and
asset prices
Note: The table presents the results of the PDOLS estimates in (5) which reflect the links
between global excess liquidity and asset prices. Standard errors are in parentheses. *, **, and
*** denote statistical significance at the 10%, 5% and 1% level of confidence, respectively.
According to the panel cointegration tests proposed by Pedroni (1999), our series are not
cointegrated in only three cases, denoted by “No cointegration”. Regarding the models with a
cointegration relationship, we conclude that the global excess liquidity aggregates are
significant in about 90% of cases and are in the expected direction in all these cases. On the
VIX index, even though it is significant in more than half of cases, we highlight that the VIX
index is rather weakly significant or not significant to explain the changes in BRICS’ asset
prices, i.e., excluding commodity prices. The ܴ ଶ should be interpreted only within the
estimates and we observe that our PDOLS models fit better for equity prices, exchange rates
and commodity prices than for bond prices and spreads.

30
In most cases, minimising Akaike and Schwarz information criteria leads us to conclude that the optimal
number of lags and leads is often the same. In some cases, this optimal number of lags and leads may be
different.

168 Amundi Investment Strategy Collected Research Papers


Dependent Variable: Asset
Q2 1998 – Q1 2014

Variable VIX Liquidity Number of Adj. R-


Real GDP
Asset/Liquidity Aggregate Index Aggregate Observations Squared
0.542*** -0.364 0.474
M2 (0.210) (0.298) (0.559)
309 0.89
-0.319* -0.503** 5.866***
MSCI Dom. Credit (0.171) (0.207) (0.846)
304 0.94
-0.243 -0.343* 0.801***
FX Reserves (0.174) (0.179) (0.142)
302 0.95
2.381 4.322 -22.888***
M2 (2.289) (2.808) (5.957)
280 0.70
10Y Interest 4.037* 5.274** -53.371***
Dom. Credit (2.262) (2.514) (10.937)
280 0.73
Rate
6.363** -0.232 -12.793***
FX Reserves (2.749) (2.486) (2.004)
274 0.78

M2 No cointegration
EMBI Global
Dom. Credit No cointegration
Spread
0.432 1.430*** -1.090***
FX Reserves (0.323) (0.341) (0.264)
303 0.81
-0.361** -0.221 0.894***
M2 (0.144) (0.157) (0.325)
307 0.98
Exchange -0.232* -0.133 0.803*
Dom. Credit (0.132) (0.156) (0.420)
297 0.98
Rate vs. USD
-0.386*** 0.175 0.469***
FX Reserves (0.091) (0.115) (0.091)
298 0.99
0.137 0.002 0.484*
M2 (0.109) (0.136) (0.281)
311 0.60

REER Dom. Credit No cointegration


0.263*** -0.038 -0.011
FX Reserves (0.090) (0.102) (0.081)
306 0.74
0.685*** -0.432*** 0.916***
M2 (0.120) (0.157) (0.307)
319 0.82
0.327*** -0.739*** 3.421***
GSCI Dom. Credit (0.106) (0.113) (0.501)
319 0.86
0.295*** -0.616*** 0.557***
FX Reserves (0.083) (0.089) (0.070)
303 0.94
0.870*** -0.568*** 1.614***
M2 (0.162) (0.213) (0.416)
319 0.84
0.461*** -1.029*** 4.878***
Brent Dom. Credit (0.148) (0.159) (0.701)
319 0.87
0.612*** -0.770*** 0.616***
FX Reserves (0.132) (0.144) (0.114)
299 0.93

Amundi Investment Strategy Collected Research Papers 169


5. Conclusion

Over the last fifteen years, global liquidity has become overabundant in different forms
and encouraged the search for yield by investors who may have access to this excess liquidity.
In this paper, we have examined the impact of global excess liquidity on asset prices for the
well-known BRICS countries. First, we built three global excess liquidity aggregates based on
the foreign exchange reserves, the M2 money supply and the domestic credit, we estimated
the interaction between global excess liquidity, economic activity and asset prices through
vector autoregressive and error correction models. We focused on a wide range of asset
classes, such as equities, interest rates, spreads, exchange rates for BRICS and some
commodities.

Overall, global excess liquidity pushed up equity prices and exchange rates, while it brought
down the fixed income rates and has more or less tightened the interest rates spreads
depending on the countries. Regarding exchange rates, global excess liquidity is a factor that
explains the appreciation trend both against the dollar and in real effective terms. Moreover,
we found that foreign exchange reserves have a genuine link with asset prices considering the
overall results of this paper. Indeed, this key measure of the first global excess liquidity
regime explains the trend in asset prices in the desired direction in almost two thirds of cases.
The global money supply M2 is the measure of the second global excess liquidity regime and
explains the trend in asset prices in the desired direction in more than four out of ten cases,
while it is only in about one third of cases for the global aggregate of domestic credit. Country
by country, the Brazilian, Russian and Indian assets have been the most impacted by the
global excess liquidity, whatever the regime. For China, the growth of domestic credit and
M2 money supply reflects the excessive monetisation of the financial system and the
indebtedness promoted by the Chinese authorities, notably to control their currency. The
results for South Africa are less eloquent. Last but not least, according to our robustness
checks, the results are broadly weakly sensitive to some economic and econometric changes.

170 Amundi Investment Strategy Collected Research Papers


Acknowledgement
In preparing this paper, I benefited from discussions with Gaëlle Le Fol and Didier Borowski,
to whom I am grateful. I also wish to thank Jean Barthélemy and Kambiz Mohkam for their
constructive comments.

Amundi Investment Strategy Collected Research Papers 171


Methodological Appendix
The method suggested by Beyer et al. (2001) uses variable weights to aggregate growth rates
and proceeds in the following four steps:

(i) Calculate weights ߱௜௝௧ based on the relative share of the country on monetary union ݅
as regards the variable ܻ௝ (݆ ൌ ‫ʹܯ‬ǡ ‫ݐ݅݀݁ݎܥܿ݅ݐݏ݁݉݋ܦ‬ǡ ‫ )ݏ݁ݒݎ݁ݏܴ݁݊݃݅݁ݎ݋ܨ‬at time ‫ݐ‬,
in a common currency, e.g., in USD in this study:

ܻ௜௝௧ (6)
߱௜௝௧ ൌ 
σଽ௜ୀଵ ܻ௜௝௧

(ii) Calculate within country or monetary union growth rates ܺ௜ of each variable ܼ௝ (where
௒ೕ
ܼ௝ ൌ σ೟ ) at time ‫ݐ‬, in local currency:
ೖస೟షయ ீ஽௉೔ೖ

ο݈‫݃݋‬൫ܺ௜௝௧ ൯ ൌ ο݈‫݃݋‬൫ܼ௜௝௧ ൯Ǥ ‫ݎ݋ݐ݈݂ܽ݁ܦܲܦܩ‬ (7)

(iii) Aggregate growth rates of (7) using weights of (6):


ଽ (8)
തܺ
തതఫ௧
തത ൌ ෍ ߱௜௝௧ ȟŽ‘‰൫ܺ௜௝௧ ൯
௜ୀଵ

(iv) Cumulate aggregate growth rates to obtain aggregate levels. We use the Q1 1998 as
the base to anchor the aggregate measures over time:

ܻ௝௧ ൌ ܻ௝௧ିଵ ൫ͳ ൅ തܺ
തതఫ௧
തത൯ (9)

with ܻ௝ǡொଵଵଽଽ଼ ൌ ܺ௜௝ǡொଵଵଽଽ଼ ൌ ͳͲͲ ‫ ݅׊‬and ‫݆׊‬.

172 Amundi Investment Strategy Collected Research Papers


Appendix 1 – Evolution of the different global excess liquidity indices by regional
aggregates
Note: The figures plot the different global excess liquidity indices by regional aggregates, namely the
All aggregate (continuous line), the G4 countries aggregate (dashed line) and the BRICS countries
aggregate (dotted line). As in Figure 1, we distinguish two different regimes of global excess liquidity.
The first one occurs between Q3 2000 and Q2 2008 while the second one takes place since Q3 2008 to
nowadays.

Amundi Investment Strategy Collected Research Papers 173


Appendix 2 – A bunch of variance decompositions
Note: The figures plot the variance decompositions of some of the most impacted asset class by global
excess liquidity. Light grey represents the own innovations of the asset, grey represents the GDP
innovations and dark grey represents the global excess liquidity innovations. For instance, after sixteen
quarters, Brent innovations are explained by around 62% by M2 innovations, 2% by GDP innovations
and 36% by its own innovations. Overall, within the global excess liquidity aggregates, it is the
foreign exchange reserves and, to a lesser extent, the M2 aggregate which best explain the BRICS’
asset innovations.

174 Amundi Investment Strategy Collected Research Papers


Appendix 3 – Summary results of the IRFs in nominal terms
Note: The table provides information about the results of the simulated IRFs based on the estimated
VAR and VEC models in nominal terms. For each BRICS countries, we analyse the IRF of each asset
price or exchange rate to a positive one standard deviation shock on the logarithm of each liquidity
aggregate, except for the MSCI BRIC and commodity prices which are dealt more globally. The
symbol “- -” denotes a negative and significant impact to a given asset price of a one standard
deviation shock on a given liquidity aggregate; “-” denotes a negative and non-significant impact; “0”
denotes no impact; “+” denotes a positive and non-significant impact; “++” denotes a positive and
significant impact; an empty cell denotes that no model has been estimated according to the
preliminary unit root and cointegration tests.

Country Brazil Russia India China South


Asset Class / Asset / Liquidity Aggregate Africa
M2 ++ + +
MSCI Credit ++
Equity

FX Reserves ++ ++ ++ ++ +
M2 +
MSCI BRIC Credit
FX Reserves ++
M2 - + - -- +
10Y Interest Credit - - -
Fixed Income

Rate
FX Reserves - -- - -- --
M2 -- -- 0 0 --
EMBI Global Credit - 0
Spread
FX Reserves -- + 0 0 -
M2 - + +
Exchange Rate
Exchange Rate

vs. USD Credit ++ + + -


FX Reserves - + ++ 0 ++
M2 - - - ++
REER Credit + +
FX Reserves - ++ ++ -
M2 --
GSCI Credit
Commodity

FX Reserves ++
M2 --
Brent Credit
FX Reserves ++

Amundi Investment Strategy Collected Research Papers 175


Appendix 4 – Panel stationarity and unit root tests results
Note: The table presents the results of the most commonly used panel stationarity and unit root tests,
i.e., Hadri (2000), Levin, Lin and Chu (2002, LLC hereafter), Im, Pesaran and Shin (2003, IPS
hereafter), Maddala and Wu (1999) for Fisher-type tests using ADF and PP tests. The figures reflect
the test-statistics in level (in first difference) of the panel stationarity and unit root tests with country
fixed effects. *, ** and *** denote rejecting the null hypothesis at the 10%, 5% and 1% level of
confidence, respectively. In Hadri (2000), the null hypothesis is the stationarity of the variable. In
LLC, the null hypothesis assumes a common unit root. In IPS, ADF Fisher and PP Fisher, the null
hypothesis assumes an individual unit root. According to the Hadri (2000) stationarity tests, all the
variables are ‫ܫ‬ሺͳሻ. According to LLC, IPS, ADF Fisher and PP Fisher unit root tests, the results are
more mixed for some variables, i.e., the 10-year sovereign interest rates, the exchange rates against the
USD and the VIX. Nonetheless and in the light of all these tests results, it is reasonable to conclude
that our variables are non-stationary in level and ‫ܫ‬ሺͳሻ.

Tests
Hadri LLC IPS ADF Fisher PP Fisher
Variables
7.02*** -0.53 -0.97 15.21 16.02*
MSCI
(-0.62) (-17.28***) (-14.84***) (172.30***) (172.04***)
10Y Interest 9.24*** -1.02 -2.67*** 23.96*** 38.62***
Rate (1.76*) (-8.29***) (-11.18***) (123.49***) (106.23***)
EMBI Global 6.00*** -0.83 -1.63* 18.27* 12.66
Spread (-1.15) (-15.95***) (-13.09***) (148.99***) (129.20***)
Exchange Rate 3.45*** -1.67** -3.42** 64.07*** 45.35***
vs. USD (1.80*) (-10.76***) (-10.14***) (109.48***) (107.21***)
8.03*** 0.64 -0.31 11.03 13.20
REER
(0.14) (-14.66***) (-15.31***) (179.21***) (178.32***)
11.61*** -1.76* -0.15 6.89 4.07
GSCI
(-1.29) (-15.72***) (-12.67***) (144.85***) (102.95***)
11.97*** -2.33** -0.69 9.30 5.55
Brent
(0.57) (-14.91***) (-12.97***) (149.07***) (125.26***)
12.98*** -1.04 0.77 6.89 2.81
Real GDP
(-0.62) (-13.47***) (-12.99***) (148.82***) (182.55***)
1.51** -2.10** -1.80** 16.06* 15.98
VIX
(-1.30) (-20.77***) (-17.94***) (214.88***) (230.05***)
12.68*** 1.00 3.87 0.47 0.20
M2 Aggregate
(-0.81) (-1.88**) (-4.26***) (36.15***) (40.19***)
Dom. Credit 12.40*** 0.12 2.75 1.11 1.95
Aggregate (-1.04) (-7.34***) (-8.57***) (86.48***) (92.10***)
FX Reserves 11.76*** -2.31** -1.00 11.25 6.78
Aggregate (3.52) (-6.61***) (-9.71***) (101.67***) (102.00***)

176 Amundi Investment Strategy Collected Research Papers


References

Adrian, T. and H. S. Shin, 2008, “Liquidity, Monetary Policy and Financial Cycles,”
Current Issues in Economics and Finance, Federal Reserve Bank of New-York, Vol. 14(1),
pp. 1-7, February.
Alessi, L. and C. Detken, 2011, “Quasi real time early warning indicators for costly
asset price boom/bust cycles: A role for global liquidity,” European Journal of Political
Economy, Elsevier, Vol. 27(3), pp. 520-533, September.
Artus, P. and M. P. Virard, 2010, “La liquidité incontrôlable : Qui va maîtriser la
monnaie mondiale,” Edition Pearson, February.
Azis, I. J. and H. S. Shin, 2014, “Managing Elevated Risk: Global Liquidity, Capital
Flows, and Macroprudential Policy–An Asian Perspective,” Asian Development Bank and
Springer Ed., December.
Baks, K. and C. Kramer, 1999, “Global Liquidity and Asset Prices: Measurement,
Implications, and Spillovers,” IMF Working Paper No. 99/168, International Monetary Fund,
December.
Belke, A., I. G. Bordon and T. W. Hendricks, 2010a, “Global Liquidity and
Commodity Prices - A Cointegrated VAR Approach for OECD Countries,” Applied Financial
Economics, Taylor & Francis Journals, Vol. 20(3), pp. 227-242, February.
Belke, A., W. Orth and R. Setzer, 2010b, “Liquidity and the Dynamic Pattern of Asset
Price Adjustment: A Global View,” Journal of Banking and Finance, Elsevier, Vol. 34(8), pp.
1933-1945, August.
Belke, A., I. G. Bordon and U. Volz, 2013, “Effects of Global Liquidity on
Commodity and Food Prices,” World Development, Elsevier, Vol. 44(C), pp. 31-43, April.
Beyer, A., J. A. Doornik and D. F. Hendry, 2001, “Constructing Historical Euro-Zone
Data,” Economic Journal, Royal Economic Society, Vol. 111(469), pp. 2-21, February.
Brana, S., M. L. Djigbenou and S. Prat, 2012, “Global excess liquidity and asset prices
in emerging countries: A PVAR approach,” Emerging Markets Review, Elsevier, Vol. 13(3),
pp. 256-267, September.
Brunnermeier, M. K. and L. H. Pedersen, 2009, “Market Liquidity and Funding
Liquidity,” Review of Financial Studies, Society for Financial Studies, Vol. 22(6), pp. 2201-
2238, June.
Chatterjee, U. and Y. Kim, 2010, “Monetary Liquidity, Market Liquidity, and
Financial Intermediation,” Asia-Pacific Journal of Financial Studies, Fifth Annual
Conference on Asia-Pacific Financial Markets, August.
Committee on the Global Financial System (CGFS), 2011, “Global Liquidity -
Concept, Measurement and Policy Implications,” CGFS Papers No. 45, Bank for International
Settlements, November.

Amundi Investment Strategy Collected Research Papers 177


Darius, R. and S. Radde, 2010, “Can Global Liquidity Forecast Asset Prices?,” IMF
Workin Paper No. 10/196, International Monetary Fund, August.
De Nicolo, G. and J. Wiegand, 2007, “What is Global Liquidity?,” World Economic
Outlook, “Globalization and Inequality,” International Monetary Fund, October.
Dickey, D. and W. Fuller, 1981, “Likelihood Ratio Statistics for Autoregressive Time
Series with a Unit Root,” Econometrica, Econometric Society, Vol. 49(4), pp. 1057-1072,
June.
Fratzscher, M., M. Lo Duca and R. Straub, 2012, “A Global Monetary Tsunami? On
the Spillovers of US Quantitative Easing,” ECB Working Paper Series, European Central
Bank, October.
Friedman, M., 1988, “Money and the Stock Market,” Journal of Political Economy,
University of Chicago Press, Vol. 96(2), pp. 221-245, April.
Giese, J. V. and C. K. Tuxen, 2007, “Global Liquidity, Asset Prices and Monetary
Policy: Evidence from Cointegrated VAR Models,” Manuscript, University of Oxford,
Nuffield College and University of Copenhagen, Department of Economics, November.
Glocker, C. and P. Towbin, 2012, “Global Liquidity, Domestic Credit, and Capital
Flows,” Mimeo.
Gouteron, S. and D. Szpiro, 2005, “Excès de liquidité monétaire et prix des actifs,”
Notes d’études et de recherche No. 131, pp. 1-73, Banque de France, September.
Granger, C. W. J., 1969, “Investigating Causal Relations by Econometric Models and
Cross-Spectral Methods,” Econometrica, Econometric Society, Vol. 37(3), pp. 424-438, July.
Hadri, K., 2000, “Testing for Stationarity in Heterogeneous Panel Data,” The
Econometrics Journal, Vol. 3(2), pp. 148-161, December.
Hartelius, K., K. Kashiwase and L. E. Kodres, 2008, “Emerging Market Spread
Compression: Is it Real or is it Liquidity?,” IMF Working Paper No. 08/10, International
Monetary Fund, January.
Im, K. S., M. H. Pesaran and Y. Shin, 2003, “Testing for Unit Roots in Heterogeneous
Panels,” Journal of Econometrics, Vol. 115(1), pp. 53-74, July.
International Monetary Fund (IMF), 2010, “Global Liquidity Expansion: Effects on
‘Receiving’ Economies and Policy Response Options,” Global Financial Stability Report,
Chapter 4, April.
Johansen, S., 1991, “Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Vector Autoregressive Models,” Econometrica, Econometric Society, Vol. 59(6),
pp. 1551-1580, November.
Kao, C. and M. H. Chiang, 2000, “On the Estimation and Inference of a Cointegrated
Regression in Panel Data,” in Baltagi, B. H. et al. eds., Nonstationary Panels, Panel
Cointegration and Dynamic Panels, Vol. 15, pp. 179–222.

178 Amundi Investment Strategy Collected Research Papers


Keynes, J. M., 1936, “The General Theory of Employment, Interest and Money,”
Macmillan, London, Chapter 13.
Levin, A., C. F. Lin and C. S. J. Chu, 2002, “Unit Root Tests in Panel Data:
Asymptotic and Finite-Sample Properties,” Journal of Econometrics, Vol. 108(1), pp. 1-24,
May.
Maddala, G. S. and S. Wu, 1999, “A Comparative Study of Unit Root Tests with Panel
Data and a New Simple Test,” Oxford Bulletin of Economics and Statistics, Vol. 61(0), pp.
631-652, November.
Mark, N. C. and D. Sul, 2003, “Cointegration Vector Estimation by Panel DOLS and
Long-run Money Demand,” Oxford Bulletin of Economics and Statistics, Vol. 65(5), pp. 655-
680, December.
Matsumoto, A., 2011, “Global Liquidity: Availability of Funds for Safe and Risky
Assets,” IMF Working Paper No. 11/136, International Monetary Fund, June.
Pedroni, P., 1999, “Critical Values for Cointegration Tests in Heterogeneous Panels
with Multiple Regressors,” Oxford Bulletin of Economics and Statistics, Vol. 61(0), pp. 631-
652, November.
Phillips, P. C. B., 1987, “Time Series Regression with a Unit Root,” Econometrica,
Econometric Society, Vol. 55(2), pp. 277-301, March.
Phillips, P. C. B. and P. Perron, 1988, “Testing for a Unit Root in Time Series
Regression,” Biometrika, Vol. 75(2), pp. 335-346, June.
Rüffer, R. and L. Stracca, 2006, “What is global excess liquidity, and does it matter?,”
ECB Working Paper No. 696, European Central Bank, November.
Sidaoui, J., M. Ramos-Francia and G. Cuarda, 2010, “Global liquidity, capital flows
and challenges for policymakers: the Mexican experience,” BIS Papers chapters, Vol. 57, pp.
239-260, October.
Sousa, J. M. and A. Zaghini, 2007, “Global Monetary Policy Shocks in the G5: A
SVAR Approach,” Journal of International Financial Markets, Institutions and Money,
Elsevier, Vol. 17(5), pp. 403-419, December.
Sousa, J. M. and A. Zaghini, 2008, “Monetary Policy Shocks in the Euro Area and
Global Liquidity Spillovers,” International Journal of Finance and Economics, John Wiley &
Sons, Ltd., Vol. 13(3), pp. 205-218, July.
Stiglitz, J. E. and A. Weiss, 1981, “Credit Rationing in Markets with Imperfect
Information,” American Economic Review, Vol. 71(3), pp. 393-410, June.
World Economic Outlook (WEO), 2012, “Coping with High Debt and Sluggish
Growth,” International Monetary Fund, World Economic and Financial Surveys, October.

Amundi Investment Strategy Collected Research Papers 179


180 Amundi Investment Strategy Collected Research Papers
WP-049

Sovereign Default in Emerging


Market Countries: A Transition
Model Allowing for Heterogeneity
ANNE-CHARLOTTE PARET,
Ph.D. Candidate at AMSE,
Strategy and Economic Research, Amundi
GILLES DUFRÉNOT,
AMSE, CEPII, Banque de France

April 2015

In this paper, we use a transition model to study the determinants


of the amount of debt defaulted by the emerging countries,
going a step further than the usual estimation of a probability
of sovereign default. The empirical framework is a panel smooth
transition model that allows to capture the heterogeneous
effects, across time and countries, from threshold variables
defining different regimes of vulnerability to sovereign default.
We highlight four variables able to discriminate country-
year observations into different vulnerability levels, and find
that countries located in the same geographical area do not
necessarily present the same vulnerability profile.

Amundi Investment Strategy Collected Research Papers 181


1 Introduction
Are the emerging countries that defaulted in the past condemned to remain in a debt spiral and are they more
prone to default again in the future ? Up until recently, the literature provided a positive answer to this question. The
emerging countries’ debt situation has often been referred to as "serial default" (see, for instance Reinhart and Rogoff,
2004, [13]). This denomination gives the impression that they are condemned to their unsustainable debt situation
without a possible trip back. Reinhart, Rogoff and Savastano (2003, [15]) argue that the historic dimension of past
default and inflation is enough to give a good insight on emerging countries’ default risk. Thus, they suggest that
their past default and inflation history are important warning indicators of debt intolerance in some emerging market
countries. This pessimistic view has been echoed extensively in many papers, the authors putting forward several argu-
ments. As a matter of fact, redundant default episodes over the past decades have probably contributed to exacerbate
the economic policies’ lack of credibility, due to inefficient institutions. In fact, emerging countries were inclined to
use pro-cyclical economic policies (notably because of weak automatic stabilizers, unemployment benefits being for
example not much significant). Once a budgetary crisis has occurred, it is therefore more difficult to go over it through
counter-cyclical policies, leading emerging market countries remain stuck in the so called "debt trap" (Sachs, 2002,
[16]), from where they struggle to go out because of a debt burden that yet became to heavy.

Another argument that has been frequently put forward is the "original sin". This expression refers to a situation
in which countries need to resort to short-term maturity external debt. In fact, the detention of external debt ties the
hands of the emerging countries’ governments which have to back money creation on foreign currency reserves if they
want to maintain a peg on their exchange rate (or at least some currency stability). If, by contrast, money depreciates,
an asymmetry appears between interest expenditure (essentially in foreign currency) and revenues (in local currency).
This "currency mismatch" explains the importance of exports’ revenues for emerging markets.

However, since the mid 2000s, the hypothesis of a "debt trap" as a consequence of past defaults seems to be
undermined by the historical facts. Indeed, it seems that the share of the emerging countries’ domestic debt over
GDP has slightly increased in the last decade, linked to a better managed debt and their willingness to hedge against
capital flows sudden stops (Mehl and Reynaud, 2005, [10]). Lower inflation and healthier fiscal and monetary policies
(becoming more countercyclical) have helped following this trend (Mohanty, 2012, [9]). Moreover a new risk manage-
ment strategy, taking into account the liquidity inflows from investors (willing to diversify their portfolios by buying
emerging countries’ local debt) have enhanced their financial stability and their integration into the world financial
markets (Blommestein and Santiso, 2007, [4]). On the whole, these changes should have reduced their exposition to
currency risk and lowered rollover risk.

Now, the general conclusions in terms of a typical default behavior must be nuanced, even if we consider the epi-
sodes of defaults since the 1980s. In this paper, we support the idea that, in line with the emerging countries’ diversity,
the origins of the important defaults on sovereign debt have been very heterogeneous across the emerging countries
and over time. Thus, we overcome a drawback of the existing empirical literature by permitting some heterogeneity
in the slope parameters of a model that relates sovereign debt default to several determinants. This allows taking
a step forward with respect to the existing studies by proposing early warning indicators specific to each country,
in spite of the fact that the forewarning indicator is based on panel data. We are able to see which macroeconomic
and financial variables explain the weight of sovereign default, depending on the country’s economic features. And
for the variables that matter, we show that their impact on sovereign default quantitatively differs across countries
and through time. The sources of heterogeneity come from the fact that the variables causing debt default follow
regime-switching dynamics that characterize different degrees of vulnerability. These different vulnerability levels are
defined inherently by threshold values estimated for some highlighted variables.

Moreover, we choose to look beyond the probability of default and consider the amount of default. In fact, we
think that the occurrence of a default isn’t in itself the sole important aspect, as the non-payment of a tiny or of a
large part of one’s debt or debt service commitments are not equivalent at all, notably in terms of market impacts.
This is the reason why we focus, not on the incidence of a default, but on its heaviness compared to the country’s GDP.

More precisely, our contribution to the literature is threefold : i) we use a new database from the Bank of Canada
(2014, [3]) that reports information on outstanding amounts of their debt being in default by countries between 1975
and 2013. We work on the amount of sovereign default for 50 frontier and emerging countries, as of 1980 ; ii) our
empirical analysis is based on a nonlinear panel data model that allows for regime-switching dynamics and hetero-
geneous effects of the determinants on sovereign debt being in default ; iii) we highlight four variables that are able
to determine distinct vulnerability regimes and provide evidence that the variables influencing the amount of debt
default vary across these regimes.

The remainder of the paper is organized as follows. Section 2 presents the empirical framework (data and model).
In Section 3 we comment our main results. Finally Section 4 concludes the paper.

182 Amundi Investment Strategy Collected Research Papers


2 Empirical framework
2.1 Data
2.1.1 Sources of data
In order to get a panel which presents sufficient diversity, we consider the most important emerging countries that
are either in the list of the emerging market countries of the IMF 1 or in the MSCI coverage 2 , but also less important
ones 3 , some of which could be qualified as frontier-countries. The observations for the 50 countries considered, taken
on an annual basis, span from 1980 to 2013, depending on availability. Nevertheless, as some variables are not available
until 1990, the observations used in many estimations only begin at this date.

Sovereign default data (yit in our regressions) stems from a dataset built by D.T. Beers and J.-S. Nadeau from
the Bank of Canada (2014, [3]). In this database, the authors do not only reference episodes of sovereign default from
1975 to 2013 worldwide (merging data previously published by the Paris Club, the IMF, the World Bank and other
institutions), but also report the amount of debt concerned by the non-payment for each episode of sovereign default,
distinguishing between different types of creditors 4 . They consider "that a default has occurred when debt service is
not paid on the due date (or within a specified grace period), payments are not made within the time frame specified
under a guarantee, or, absent an outright payment default, in [...] circumstances where creditors incur material eco-
nomic losses on the sovereign debt they hold" (including agreements reducing interest rates or extending maturities
on outstanding debt and government exchange offers where existing debt is swapped or re-denominated, leading to
the detention of new debt or equity on less-economic terms). Our aim is to analyse the determinants of the amount of
debt in default using this data (as a share of the country’s GDP), highlighting the existence of different vulnerability
regimes regarding default risk, depending on the country’s economic characteristics.

The variables we make use of to explain the amount of sovereign default and determine the vulnerability regimes are
similar to those used in the literature, which essentially focuses on the linear explanation of default events (whatever
their significance as regards the country’s activity). They consist of :
– capital and current account variables (yearly change in the terms of trade and in the exchange rate, and portfolio
equity flows as a share of GDP) from the World Bank ;
– economic variables (real GDP growth, inflation, gross domestic savings as a share of GDP, total reserves on
short-term external debt, external debt on exports, external debt as a share of general government debt, general
government debt, public balance, interest expenditure as a share of public revenues) from the IMF, the World
Bank and Oxford Economics ;
– institutional variables (World Global Indicators from the World Bank and Corruption perception index from
Transparency International) ;
– financial variables (spreads versus the United States as regards treasury bill interest rates, blended spread
component of the Emerging Market Bond Index, overnight interbank rates, lending minus deposit rates and
Standard & Poor’s ratings) from various sources, even if their availability is not so obvious for some of the
considered countries.
For more details regarding the sources of the data, see Appendix 1, Tables A.2 to A.6.

2.1.2 Heterogeneity in terms of vulnerability to sovereign default


As said before, we do believe that all countries do not react the same way to changes in their economic environment,
in terms of making default. As a matter of fact, even though it is right to observe that default episodes really differ
from one geographic area to another (in terms of temporality, but also regarding their importance, see Figure 1), we
do think that countries belonging to a same geographic area can be more or less prone to default due to a specific
shock, depending on their economic profiles.
1. Argentina, Brazil, Bulgaria, Chile, China, Colombia, Estonia, Hungary, India, Indonesia, Latvia, Lithuania, Malaysia, Mexico, Pa-
kistan, Peru, Philippines, Poland, Romania, Russia, South Africa, Thailand, Turkey, Ukraine and Venezuela.
2. Brazil, Chile, Colombia, Mexico, Czech Republic, Egypt, Greece, Hungary, Poland, Qatar, Russia, South Africa, Turkey, United Arab
Emirates, China, India, Indonesia, South Korea, Malaysia, Philippines, Taiwan and Thailand.
3. Algeria, Croatia, Hong Kong, Israel, Kenya, Kuwait, Lebanon, Mauritius, Morocco, Nigeria, Saudi Arabia, Singapore, Slovakia,
Slovenia, Tanzania, Tunisia and Uruguay.
4. We do not use this information here.

Amundi Investment Strategy Collected Research Papers 183


Figure 1 – Debt in default (% of GDP)

Note : The reported percentage corresponds to the mean of the amount of debt being in default (as a % of GDP), for
each geographic area. See Table A.1, in Annex, for the repartition of the countries into geographic areas.

In fact, if one looks at the economic characteristics of the countries when they face default and when they do not, it
can be observed that there really is a difference. Table 1 shows that, compared to a country whose situation is sound,
a country being in default is more likely to have a deteriorated economic environment (local currency depreciating
hugely, low equity net inflows, low growth and high inflation), less healthy public finances (high debt burden, high
general government debt), a more fragile financing basis (low domestic savings, high share of external public debt,
less foreign reserves compared to short-term financial commitments), a more stressed financial environment (higher
interest rates) and institutions of lower quality (lower WGI and CPI indicators). This observation strengthens our
belief that a model allowing for heterogeneity can be a good framework to highlight the explanatory factors of the
potential amount of debt being in default.

No default Default
Mean N Mean N
Current Terms of trade (YoY % Change) 0,4 801 0,5 388
account Exchange Rate (YoY % Change) 6,4 982 68,9 450
variables Portfolio Equity Net Inflows (% of GDP) 0,8 673 0,4 219
Real GDP growth (YoY % Change) 4,4 1078 3,0 497
Intern Inflation (YoY % Change) 10,2 1064 110,5 497
economic Central government interest payments (% of general government revenues) 13,3 708 23,3 210
variables General government debt 45,3 711 53,8 173
General government balance -1,9 979 -2,9 343
Domestic Savings (% of GDP) 26,3 1010 19,9 463
External Reserves (% of total short-term external debt) 301,1 671 280,7 445
debt External debt (% of Exports) 126,7 864 231,0 382
variables Share of external debt in General Government debt 40,2 467 47,9 73
Corruption perception Index 45,5 659 28,5 183
Institutionnal WGI Rule of Law 0,3 705 -0,5 195
variables WGI Government Effectiveness 0,3 705 -0,3 195
WGI Control of Corruption 0,2 705 -0,6 195
EMBI 366,3 659 644,7 158
Financial Interbank rate (overnight) 5,2 312 15,0 52
variables S&P Rating 14,3 871 10,3 329
Lending minus deposit rate 7,0 768 24,9 390
Sovereign interest rate spreads (3Months, vs US) 6,0 556 15,8 196

Table 1 – Descriptive statistics

Note : Data stretches from 1980 to 2013, depending on availability for each variable and each country. Means (for the
available observations whose number is mentionned in column N) are computed in two sub-samples : countries not being in
default and countries whose amount of debt being in default is strictly positive. For S&P rating data, notations have been
rescaled, AAA (resp. D) rating being equivalent to 22 (resp. 0). For more details regarding the data, see Tables in Appendix
A.1.

2.2 Model and econometric methodology


Though our modelling methodology differs from the approaches considered so far, our background is the literature
on early warning indicators to model sovereign risk. We briefly recall how the empirical studies usually proceed to

184 Amundi Investment Strategy Collected Research Papers


predict sovereign risk.

Sovereign risk is often evaluated using vulnerability indicators based on economic fundamentals. These models are
employed to explain the occurrence of extreme events (debt restructuring, sovereign default, late payments episodes,
stress episodes in terms of exchange rate or inflation, banking crises). See, for instance, Reinhart, Goldstein and Ka-
minsky, (2000, [12]).

Many predictors of sovereign default take the form of a z-score (Baldacci, Mc Hugh and Petrova, 2011, [2], Ra-
bobank, 2011, [11]). Early-warning indicators are then resulting from the link done between such indicators and the
occurrence of default events. These warnings trigger once some thresholds are exceeded : that is why they are called
"signals approach". The choice of the threshold (which can for example be based on the minimisation of some "wrong
warnings") favors either predicting more crises correctly (at the expense of more noise) or being less frequently wrong.
Nevertheless, as underlined by Andréou, Dufrénot and alii. (2009, [1]) as regards financial crises, noise due to an
early warning indicator which would trigger more easily is not necessary a bad thing, as it can highlight a very de-
teriorated economic situation (which is an interesting information in itself), even if default does not occur immediately.

Another common way to describe the determinants of sovereign default is to estimate a default probability through
a binary model (logit or probit), with similar explanatory variables as the ones used for early warning indicators
(i.e. macroeconomic variables, fiscal variables, financial variables and institutional variables). Manasse, Roubini and
Schimmelpfennig (2003, [8]), as Cohen and Valladier (2011, [5]) estimate the default probability (including important
IMF supports for the former, and also Paris Club supports and other punctual events regarding debt servicing and
payment arrears for the latter) with a logit model. Cohen and Valladier (2011, [5]) then classify the countries with
respect to the quintile they belong to, in order to differentiate them by risk level. Kraay and Nehru (2004, [7]) resort
to a probit model to highlight the determinants of events relative to payment arrears as regards external debt, Club
de Paris debt rescheduling and non conventional IMF loans.

To the best of our knowledge, the literature seems to have focused until now on models or indicators describing the
origins (economic and financial, but also political and institutional ones) of the occurrence of sovereign defaults (taking
various forms). Most frequently, the developed models are linear, as they assume that all the observations present the
same sensibilities to the explanative variables to explain this default occurence. Finally, they usually derive different
degrees of vulnerability from the distribution of the indicator. We attempt to enrich the existing models by working
with a model that allows for heterogeneity in the response to explanative variables to explain this amount of default.
Moreover, our model enables us to directly identify the different vulnerability regimes, as the thresholds dividing in
different regimes are estimated endogenously, and thus, are inherent to the model.

We consider the panel smooth transition regression model (PSTR proposed by Gonzalez, Teräsvirta, van Dijk, 2005,
[6], henceforth GTD) 5 . It enables us to estimate, for various emerging market countries, the heterogeneous effects on
the amount of debt being in default of the explanatory variables (presented herebefore). The heterogeneous effects,
across time and countries come from threshold variables defining different regimes of vulnerability to sovereign default.

Formally, the panel smooth transition regression model (PSTR) is as follows :


r
 (j)
yit = μi + β0 xit + βj xit gj (qit ; γj , cj ) + uit (1)
j=1

where i = 1, ..., N , t = 1, ..., T , μi corresponds to the country fixed effects, r is the number of transition functions gj
(j)
defined by the threshold variables qit (for j ≤ r) ; cj are the corresponding thresholds, γj are the parameters defining
the degree of smoothness of the transition from one regime to another and uit the error term.

The transition function is continuous, bounded between 0 and 1 :

(j) 1
gj (qit ; γj , cj ) = m (j)
(2)
1 + exp(−γj k=1 (qit − cjk ))

with γj > 0, and cj1 < cj2 < .. < cjm the thresholds associated to the transition function gj . m is the number
of returns between the two extreme regimes associated to this transition function, these returns taking place at the
thresholds cj1 , cj2 , .., cjm . In the extreme regimes associated to each transition function, the vectors of parameters take
the values β0 and β0 + βj .

In Equation (1) two important parameters are m and r. m describes the shape of the transition function gj . For
m = 1, the shape is that of a standard logistic function. For m = 2, the transition function is described by a "V curve"
with a middle regime and two identical outer regimes. For m > 2, the transition function has a more complex shape.
5. The estimations were done using the MATLAB code provided by Christophe Hurlin.

Amundi Investment Strategy Collected Research Papers 185


The parameter r defines the number of transition functions (or, equivalently, the number of regimes).

To estimate Equation (1), we proceed in two steps.

Step 1

We test the null assumption of a linear model against a PSTR model (r = 0 against r = 1). This is equivalent to
testing the null assumption of homogeneous effects against heterogeneous effects across countries and years. We test

 
H0 : γ = 0 or H0 : β1 = 0 against H1 : γ = 0 or H1 : β1 = 0

To overcome the problem of nuisance parameter (β1 is unidentified under H0 and γ is unidentified under H0 ), we
consider the following auxiliary regression, resulting from the first-order Taylor expansion around γ = 0 of Equation
(1) (for r=1) :
m
 j
yit = μi + βj∗ xit qit + u∗it (3)
j=0

Our null assumption is then equivalent to H0∗ : β1∗ = .. = βm∗


= 0. Considering the result of a LM test, if the null
hypothesis is not rejected, then we conclude that the model reduces to an homogeneous model. If the null is rejected,
then we proceed to test the null H0 : r = 1 against H1 : r = 2. If the null is not rejected, then we conclude that the
model is a PSTR model with one transition function. If the null is rejected, then we test H0 : r = 2 against H1 : r = 3.

Step 2

For a given r, we estimate the PSTR model. This is done by applying a non linear least squares estimator after
appropriately subtracting the individual means from the variables in the model (see GTD for details). First, we consi-
der some initial values for the slope parameter γj and the threshold values cj 6 . Next we estimate the slopes β of the
model conditional to these values, by ordinary least squares. Then, we go on iteratively, finding new values for γj and
cj through non linear least squares optimization and then re-estimating the slopes.

3 Results
3.1 Results based on individual variables
We begin applying the PSTR model by showing that several macroeconomic and financial variables are potential
candidates that can explain time-varying and cross section differences in the impact of vulnerability factors on the
amount of potential debt default.

To begin with, we estimate Equation (1) for m = 1 and m = 2 with qit (the threshold variable) being the different
explanatory variables that are potential candidates to explain the amount of the sovereign default. We run as many
regressions as potential candidate variables for qit . For each of them, xit is a vector of qit and a dummy variable
describing past default 7 .
The estimated equation is

Debt in default (% of GDP) = μi + β0 xit + β1 xit g(qit ; γ, cj ) + uit (4)

with xit = (dummy, qit )

The results of the tests are shown in Table 2. We also report the p-value corresponding to the Fisher statistic of
the test H0 : r = i against H1 : r = i + 1 focusing on the number of transition functions (see Table 2, columns 4 to
7) 8 . The idea is then to select the regressions accepted with the highest p-value, in order to identify the variables that
could be relevant for splitting observations into regimes differing in the effects of explanatory variables (among which
past default) on the amount of sovereign debt in default.

Then, for the best models, we report the PSTR estimations when the explanatory variables are considered indivi-
dually. The results are shown for m = 1 and for m = 2 (in Table 3). Over the 21 variables originally considered, we
present the results of the estimations making sense (i.e. where there is convergence of the parameter’s estimation and
6. We use the method proposed by GTD, to determine the initial values. It consists in computing the concentrated sum of squared
residuals for a "grid" of possible values for γ and c, finally taking as initial values for the NLS estimation those minimising this target.
7. This dummy equals 1 if sovereign default has occurred within the last 5 years, and 0 otherwise.
8. For m=1 (resp. m=2) we go on until rmax = 3 (resp. rmax = 2).

186 Amundi Investment Strategy Collected Research Papers


Dependent variable : amount of debt being in default (% of GDP)
Fisher Statistic : p-value
Threshold variable m N r=0 vs r=1 r=1 vs r=2 r=2 vs r=3 r=3 vs r=4
Terms of Trade m=1 1112 0,124 0,845
m=2 1112 0,025 0,295 -
m=1 1339 0,002 0,033
Exchange rate
m=2 1339 0,000 0,083 -
Equity net inflows m=1 892 0,762 0,503
m=2 892 0,951 1,000 -
Growth rate of GDP m=1 1465 0,000 0,121
m=2 1465 0,000 0,004 0,000 -
m=1 1451 0,009 0,089
Inflation
m=2 1451 0,001 0,365 -
m=1 918 0,002 0,115
Interests payments on public revenue
m=2 918 0,012 0,023 0,025 -
m=1 871 0,000 0,000 0,000 0,000
General government debt
m=2 871 0,000 0,000 0,000 -
m=1 1322 0,004 0,303
Public balance
m=2 1322 0,011 0,017 0,014 -
m=1 1390 0,152 0,449
Gross domestic savings
m=2 1390 0,034 0,023 0,076 -
m=1 1018 0,000 0,264
International reserves on short-term external debt
m=2 1018 0,000 0,051 -
m=1 1246 0,000 0,066
External debt on exports
m=2 1246 0,000 0,000 0,000 -
External debt share in general government debt m=1 536 0,007 0,456
m=2 536 0,035 0,035 -
Corruption perception index m=1 826 0,367 0,387
m=2 826 0,644 0,560 -
m=1 882 0,904 0,020 0,551
WGI Rule of Law
m=2 882 0,599 0,253 -
m=1 882 0,483 0,975
WGI Government effectiveness
m=2 882 0,686 0,793 -
WGI Control of corruption m=1 882 0,964 0,024 0,958
m=2 882 0,291 0,742 -
EMBI m=1 801 0,000 0,000 0,000 0,000
m=2 801 0,000 0,000 0,000 -
m=1 364 0,019 0,097
Interbank rates
m=2 364 0,000 0,337 -
m=1 1176 0,000 0,009 0,491
S&P Ratings
m=2 1176 0,000 0,013 0,389 -
Lending minus deposit rates m=1 1069 0,295 0,526
m=2 1069 0,057 0,462
Spreads m=1 713 0,000 0,101
m=2 713 0,000 0,002 0,000

Table 2 – First PSTR estimations - Tests regarding the number of regimes

Note : For each threshold variable tested, the estimation is done for m=1 and m=2 (respectively constraining to rmax = 3
and rmax = 2), and the countries whose required information was available less than 3 years were not taken into account.
The number of observations included is mentionned in column 3. For each threshold variable, we test H0 : r = i against
H1 : r = i + 1 until the p-value of the Fisher Statistic (presented in columns 4, 5, 6 and 7) is superior to α. α is set to 0.05 for
r=0 and divided by 2 at each new iteration, to favour parcimony (doing this, we follow a method suggested by the authors of
GTD). The bold p-values correspond to the estimations the more likely accepted. Moreover, it is worth mentionning that the
estimations do not converge for the following threshold variables : Equity net inflows (m=2), General goverment debt (m=1),
EMBI (m=1), and Spreads (m=2). The results are therefore not presented for these estimations.

Amundi Investment Strategy Collected Research Papers 187


where each regime contains enough observations).

Our main findings are the following.

The dummy variable capturing past defaults within the recent five years appears to be statistically significant in
almost all regressions. Thus, a country which defaulted in the past is likely to be confronted to a sovereign default
today. However, a default is of particularly high significance, only if the country faces a macroeconomic and financial
environment that becomes more risky. Whether or not the country is more sensitive to past default depends upon
some threshold values taken by the variables in the transition function.

The first column of Table 3 contains the transition variables qit , the second column reports the selected value for
r, columns 3 and 4 show the estimated values of the slope γj of the transition function and of the threshold value cj
for the associated transition variables. In column 6, we write the exogenous variable considered in the left-hand side of
the estimated equation. Columns 7 and 11 show the estimated values of the coefficients in the extreme regimes, while
columns 8 and 12 respectively report the p-value of the estimates to see whether they are statistically significant. In
the last column, we report the share of observations in the first regime 9 (the share of observations in the second regime
is 100 minus the percentage reported for the first regime). The estimates in Table 3 lead to the following conclusions.

9. i.e. the observations characterized by the slope β0 .

188 Amundi Investment Strategy Collected Research Papers


Dependent variable : amount of debt being in default (% of GDP)
r γ c Iterations β0 p-value β1 p-value β0 + β1 p-value N P0
Estimations done with m= 1
0,1 43,3 Past default 0,761 0,289 14,732 0,000 15,493 0,000 1339 93%
Exchange Rate 1 31
Exchange Rate 0,048 0,161 -0,045 0,185 0,003 0,048
21,4 -3,4 Past default 17,055 0,000 -13,014 0,001 4,041 0,000 1465 7%
Real GDP growth 1 36
Real GDP growth -0,081 0,585 0,077 0,633 -0,004 0,908
267,0 -10,4 Past default 16,854 0,003 -13,271 0,017 3,583 0,000 1322 4%
Public balance 1 58
Public balance 0,194 0,227 -0,307 0,089 -0,113 0,023
31,6 11,3 Past default -0,176 0,915 6,156 0,004 5,980 0,000 918 53%
Interest exp. (% of public revenue) 1 40
Interest exp. (% of public revenue) 0,389 0,090 -0,297 0,188 0,091 0,002
0,1 97,5 Past default 10,641 0,000 -9,065 0,000 1,577 0,047 1018 25%
Reserves (% of STED) 1 32
Reserves (% of STED) 0,027 0,033 -0,028 0,030 0,000 0,120
0,0 336,9 Past default -1,741 0,072 30,206 0,000 28,465 0,000 1246 92%
External Debt (% of exports) 1 35
External Debt (% of exports) 0,013 0,092 -0,003 0,775 0,011 0,157
14,5 59,0 Past default 0,037 0,974 7,620 0,044 7,658 0,043 536 77%
Share of ext. debt in public debt 1 38
Share of ext. debt in public debt 0,033 0,152 -0,017 0,332 0,016 0,584
0,3 63,3 Past default 2,487 0,038 5,935 0,350 8,421 0,190 713 98%
Sovereign interest rate spreads 1 31
Sovereign interest rate spreads 0,335 0,001 -0,316 0,001 0,019 0,168

Estimations done with m= 2


1 547,2 -5,9 6,7 31 Past default 2,578 0,004 4,045 0,000 6,623 0,000 1112 30%
Terms of trade
Terms of trade -0,044 0,557 -0,012 0,880 -0,056 0,030
1 1135,1 -0,3 29,4 29 Past default 2,174 0,001 10,828 0,000 13,002 0,000 1451 14%
Inflation
Inflation 0,170 0,004 -0,167 0,005 0,003 0,001
1 24,7 0,9 7,9 122 Past default -0,647 0,292 3,961 0,008 3,314 0,007 364 38%
Interbank rate
Interbank rate -0,021 0,455 -0,026 0,180 -0,047 0,069

Table 3 – First PSTR estimations - Estimations results for m=1 and m=2
Note : Here, we only report the results of the estimations of Table 2 that make sense, i.e. where there is convergence of the parameter’s estimation and where each regime contains enough
observations. For each estimation presented, the number of observations is reported on column 13, the number of transition functions on column 2, the shape γ of the transition function
in column 3 (smooth if γ → 0, abrupt if γ → ∞), the transition thresholds c in column 4 and the number of iterations needed to estimated this parameters in column 5. The slopes
of each explanatory variable are reported on columns 7, 9 and 11 (with the corresponding p-values in columns 8, 10 and 12). βi is the slope corresponding to the explanative variable
xit = (dummy; qit ). For m=1, the observations for which the threshold variable is inferior to the threshold c (whose part in the sample is reported on column 14) are broadly characterized by
the slope β0 , and others by the slope β0 + β1 (slopes corresponding to the extreme observations). For m=2, c is of dimension 2. In this case, the observations for which the threshold variable

Amundi Investment Strategy Collected Research Papers


is between the two threshold values (whose part in the sample is reported on column 14) are broadly characterized by the slope β0 , and others by the slope β0 + β1 (slopes corresponding to
the extreme observations). In actual fact, each observation is characterized by a specific slope, which varies smoothly between the slopes associated to the extreme observations.

189
Countries that are likely to be confronted to a current sovereign default of a higher amount are those which already
defaulted at least once during the 5 preceding years and whose macroeconomic environment is more fragile, i.e. with
the following characteristics (for m=1) :

– a huge depreciation of their currency by more than 43% (changes measured on a year-on-year basis) ;
– a deep recession with a real GDP growth below -3.4% ;
– a fiscal deficit higher than 10.4% ;
– a ratio of debt service over fiscal revenues above 11.3% ;
– a high exchange rate risk (measured by total international reserves as share of total short-term debt below 97%) ;
– a high currency mismatch (with total external debt accounting for more than 337% of total exports) ;
– the external debt representing more than 59% of the sovereign debt ;
– increased short-term interest rate spreads (63.3% above the US three-months rate).

When we consider the estimates with m = 2 (Table 3), three additional variables are able to discriminate into
significantly different regimes : the terms of trade, inflation and interbank rates 10 . Table 3 provides evidence that a
country that was confronted to a default during the past five years is likely to increase the current amount of debt in
default even more if it faces adverse conditions characterized by :

– extreme change in the terms of trade (with either a decrease by more than 5.9% or an increase by more than
6.7%) ;
– extreme inflation conditions (a year on year change of less than -0.3% or above 29%). This finding leads us
to nuance Reinhart and Rogoff’s argument according to which inflation and past default are causes of "serial
defaults" in the emerging countries. Here, past defaults seem to importantly increase the potential of a current
sovereign default, especially if a country experiences a situation of deflation or hyperinflation. Inflation pressures
within a range of moderate inflation rates are a "pushing" factor to a new default, but only to a lesser extent.
It can be seen that, in our sample, only 3% of the observations are in the deflation regime 11 , which means that
hyperinflation is what really matters ;
– low or high interbank rate (below 0.9% or above 8%).

After having confirmed that a PSTR provides a consistent model allowing for heterogeneity in the explanation of the
amount of potential sovereign default, we focus on determining which variables are able to discriminate observations
into distinct vulnerability regimes, in a more comprehensive framework. We therefore consider a combination of
different explanatory variables (instead of determinants taken individually). This gives a better picture of the key
factors that influence the amount of debt default.

3.2 Four main sources of vulnerability


As mentioned above, we now consider models with several explanatory variables in the right-hand side of our
equations. As we are interested in highlighting different regimes of vulnerability along some discriminating threshold
variables, we focus on models with m=1. We ran regressions with different transition variables. Among them, four
appeared to be robust (in terms of the relevance of the estimated coefficients and distribution of observations within
the different regimes) : the ratio of debt service over fiscal revenues, domestic savings as share of GDP, internatio-
nal reserves as share of short-term external debt, and external debt over exports. These four lines of approach are
brought out to be the ones able to discriminate between groups of observations presenting different features in terms
of sovereign debt vulnerability. The fact that they all refer to the capability of emerging market countries to protect
themselves against default is therefore not surprising.

Table 4 shows the results of the tests for these four threshold variables, while Table 5 contains our main regressions.
We now comment the main conclusions from Table 5.

10. We do not consider the regressions for which the identification of regimes was meaningless (for example because the estimated
thresholds were very close), because this implied a situation with very few observations in one regime (for example exchange rate, domestic
savings and EMBI). We also neglect the regressions for which the estimated values of γ was small so that they behave like linear homogenous
models (for example, S&P rating)
11. which is characterized by an annual inflation rate inferior to -0.3%.

190 Amundi Investment Strategy Collected Research Papers


Dependent variable : amount of debt being in default (% of GDP)
Fisher Statistic : p-value
m N r=0 vs r=1 r=1 vs r=2 r=2 vs r=3 r=3 vs r=4
Interests payments on public revenue m=1 612 0,000 0,000 0,019
Gross domestic savings m=1 755 0,000 0,000 0,043
International reserves on STED m=1 624 0,000 0,189
External debt on exports m=1 764 0,000 0,000 0,160

Table 4 – PSTR estimations - Tests regarding the number of regimes (m=1)

Note : For each threshold variable tested, the estimation is done for m=1 (constraining to rmax = 3), and the coun-
tries whose required information was available less than 3 years were not taken into account. The number of observations
included is mentionned in column 3. For each threshold variable, we test H0 : r = i against H1 : r = i + 1 until the p-value of
the Fisher Statistic (presented in columns 4, 5, 6 and 7) is superior to α. α is set to 0.05 for r=0 and divided by 2 at each
new iteration, to favour parcimony (doing this, we follow a method suggested by the authors of GTD). The bold p-values
correspond to the estimations the more likely accepted.

Amundi Investment Strategy Collected Research Papers 191


192
Dependent variable : amount of debt being in default (% of GDP)
Estimations done with m= 1
r γ c Iterations β0 p-value β1 p-value β2 p-value β0 + β1 p-value β0 + β2 p-value β0 + β1 + β2 p-value N P0 P1 P2
Interest exp. (% of public revenue) 7429,9 9,0 Past default -0,792 0,407 -0,879 0,502 0,999 0,581 -1,672 0,129 0,207 0,914 -0,673 0,482 612 47% 18% 35%
3,7 13,8 Exchange rate -0,074 0,023 0,327 0,000 -0,257 0,000 0,253 0,000 -0,332 0,000 -0,004 0,820
Inflation 0,179 0,005 -0,055 0,614 -0,057 0,595 0,124 0,143 0,122 0,361 0,067 0,163
2 140
Public debt 0,168 0,000 -0,022 0,503 -0,048 0,231 0,146 0,000 0,120 0,032 0,098 0,005
Ext.debt/Exports 0,035 0,016 -0,040 0,007 0,006 0,376 -0,004 0,546 0,041 0,007 0,002 0,713
S&P rating -0,519 0,004 0,241 0,100 0,122 0,431 -0,278 0,063 -0,398 0,071 -0,156 0,286
Domestic Savings (% of GDP) 3203,1 21,7 Past default -0,834 0,187 -1,836 0,204 1,454 0,280 -2,669 0,032 0,620 0,686 -1,215 0,025 755 44% 10% 46%
21,7 24,6 Exchange rate 0,002 0,916 0,047 0,288 -0,027 0,552 0,049 0,246 -0,026 0,593 0,021 0,255
Inflation 0,030 0,345 0,361 0,000 -0,150 0,119 0,391 0,000 -0,120 0,241 0,241 0,000
2 135
Public debt 0,041 0,106 0,078 0,023 -0,071 0,009 0,119 0,000 -0,030 0,386 0,048 0,000
Ext.debt/Exports 0,000 0,925 0,075 0,000 -0,086 0,000 0,076 0,000 -0,086 0,000 -0,011 0,016
S&P rating -0,221 0,038 -1,007 0,000 0,977 0,000 -1,227 0,000 0,757 0,000 -0,250 0,001
Reserves (% of STED) 121,9 78,5 Past default 4,966 0,153 -5,494 0,127 -0,528 0,408 624 12% 88%
Exchange rate 0,111 0,225 -0,119 0,192 -0,008 0,604
Inflation -0,107 0,355 0,239 0,055 0,131 0,004
1 23
Public debt 0,262 0,002 -0,189 0,017 0,073 0,000
Ext.debt/Exports 0,001 0,906 0,001 0,939 0,002 0,625
S&P rating -0,705 0,004 0,545 0,007 -0,160 0,169
External Debt (% of exports) 23,0 385,0 Past default -0,695 0,064 20,292 0,000 -6,491 0,028 19,597 0,000 -7,186 0,014 13,106 0,000 764 91% 6% 3%
126,8 258,3 Exchange rate 0,027 0,162 0,367 0,007 -0,342 0,014 0,394 0,004 -0,315 0,023 0,052 0,037
Inflation 0,108 0,019 0,097 0,751 0,261 0,236 0,205 0,508 0,369 0,085 0,465 0,006
2 114
Public debt 0,042 0,000 0,009 0,796 0,225 0,000 0,051 0,179 0,267 0,000 0,276 0,000
Ext.debt/Exports -0,005 0,235 0,016 0,453 -0,001 0,918 0,011 0,609 -0,006 0,626 0,010 0,571
S&P rating -0,069 0,295 -1,928 0,013 -0,586 0,122 -1,997 0,010 -0,655 0,091 -2,583 0,000

Table 5 – PSTR estimations - Estimations results (m=1)


Note : For each estimation presented, the number of observations is reported on column 19, the number of transition functions on column 2, the shape γ of the transition function in column
3 (smooth if γ → 0, abrupt if γ → ∞), the transition thresholds c in column 4 and the number of iterations needed to estimated this parameters in column 5. The slopes of each explanatory
variable are reported on columns 7, 9, 11, 13, 15 and 17 (with the corresponding p-values in columns 8, 10, 12, 14, 16 and 18). βi is the slope corresponding to the explanative variable xit
refered to in column 6. The observations for which the threshold variable is inferior to the lowest threshold c (whose part in the sample is reported on column 20) are broadly characterized
by the slope β0 (slope corresponding to the extreme observations). If ci and cj are respectively being the lowest and the highest threshold, the observations for which the threshold variable
is superior to the lowest threshold but inferior to the highest (whose part in the sample is reported on column 21) are broadly characterized by the slope β0 + βi and others by the slope
β0 + βi + βj (slopes corresponding to the extreme observations). In actual fact, each observation is characterized by a specific slope, which varies smoothly between the slopes associated to
the extreme observations.

Amundi Investment Strategy Collected Research Papers


Threshold variable 1 : Debt service as share of fiscal revenues

The model distinguishes between three regimes : one is characterized by a low level of debt service (which amounts
for less than 9% of fiscal revenues), a regime of high level of debt service (where the latter weighs more than 14%
of fiscal revenues), and an intermediate regime. We indicate below the explanatory variables that seem to be key
determinants of the amount of debt default within each regime :

Regime 0 Regime 1 Regime 2


debt service lower than 9% debt service between 9% and debt service higher than 14%
of public revenues 14% of public revenues of public revenues
Exchange rate (-) Exchange rate (+)
Inflation (+)
Public debt (+++) Public debt (++) Public debt (+)
Ext.debt on exports (+)
S&P rating (-)

Table 6 – Dependent variable : amount of sovereign debt being in default


Note : For each regime, defined by the thresholds presented in Table 5, the significant determinants of the amount of debt being
in default are reported here. When the associated slope is positive (resp. negative), they are reported with a + (resp. -) sign.
When a determinant is significant in different regimes with the same sign, the number of +/- signs reported informs on the
importance of the slope from one regime to another.

47% of the observations are located in the low debt burden regime. In this regime, the following factors yields an
increase in the amount of potential sovereign default : a nominal appreciation of the domestic currency, an increase
in inflation, in public debt and in the ratio of external debt as share of total exports, a poor S&P rating. 35% of the
observations are located in the high debt burden regime and there, public debt is the main driver of higher default.
In the intermediate regime (18% of the observations), public debt and the exchange rate are the main determinants
of sovereign default (with for the former an influence that is lower compared to the low debt burden regime). For
the observations in this regime, a currency depreciation induces a currency mismatch which makes it more likely to
default importantly. It is no longer the case once debt service on public revenues is inferior to a certain threshold (in
regime 0). Finally, it is seen from the table above that the number of determinants that significantly affect the amount
of debt decreases as we go from the low to the high debt burden regimes. This illustrates the fact that, when debt
service is under control, the determinants of debt default are clearly identified. This is not the case when debt service
becomes too heavy.

Threshold variable 2 : Domestic savings ratio

Along the domestic savings’ axis, the three vulnerability regimes and the key determinants of debt default in each
regime can be represented as follows :

Regime 0 Regime 1 Regime 2


domestic savings lower than domestic savings between domestic savings higher than
22% of GDP 22% and 25% of GDP 25% of GDP
Past default (- -) Past default (-)
Inflation (++) Inflation (+)
Public debt (++) Public debt (+)
Ext.debt on exports (+) Ext.debt on exports (-)
S&P rating (-) S&P rating (- -) S&P rating (-)

Table 7 – Dependent variable : amount of sovereign debt being in default


Note : For each regime, defined by the thresholds presented in Table 5, the significant determinants of the amount of debt being
in default are reported here. When the associated slope is positive (resp. negative), they are reported with a + (resp. -) sign.
When a determinant is significant in different regimes with the same sign, the number of +/- signs reported informs on the
importance of the slope from one regime to another.

44% of the observations in the estimation sample are located in the low savings ratio regime, characterised by a
domestic savings ratio lower than 22% of GDP. In this regime, having a weak S&P rating is the key determinant of
the amount of debt default. 46% of the observations belong to the high savings ratio regime (when the saving ratio is
above 25%) where the following variables are found to rise the amount of sovereign default : a high inflation and debt
ratio, a low external debt over exports ratio, a bad S&P rating. Interestingly, we also find that a country that did
default during the preceding five years tends to decrease its current amount of debt in default. This could illustrate a
"learning process", which goes against the "serial default" view : when they are able to (i.e. when their domestic savings
are high enough), countries having recently defaulted are less inclined to default again because they learn from their
errors. We obtain a similar conclusion for the intermediate regime. Finally, we also see that external debt tends to
increase the amount of potential default when domestic savings are intermediate, but do not anymore once domestic

Amundi Investment Strategy Collected Research Papers 193


savings are high enough to build a safety buffer.

Threshold variable 3 : International reserves as a share of short-term external debt

Regime 0 Regime 1
Reserves lower than 79% of Reserves higher than 79% of
STED STED
Inflation (+)
Public debt (++) Public debt (+)
S&P rating (-)

Table 8 – Dependent variable : amount of sovereign debt being in default


Note : For each regime, defined by the thresholds presented in Table 5, the significant determinants of the amount of debt being
in default are reported here. When the associated slope is positive (resp. negative), they are reported with a + (resp. -) sign.
When a determinant is significant in different regimes with the same sign, the number of +/- signs reported informs on the
importance of the slope from one regime to another.

Here we have two regimes. 12% of the observations are assigned into the regime 0 when the reserve ratio falls below
79%. In this regime, we identify a high debt ratio and a weak S&P rating as the main factors of higher sovereign
default. 88% of the observations lie in the regime 1 (when the international reserves account for more than 79% of
total exports) in which a high inflation rate and public debt ratio are found to increase sovereign debt.

Threshold variable 4 : External debt as ratio of total exports

Regime 0 Regime 1 Regime 2


external debt lower than external debt lower between external debt higher than
258% of exports 258% and 385% of exports 385% of exports
Past default (-) Past default (+)
Exchange rate (-) Exchange rate (+)
Inflation (+) Inflation (++)
Public debt (+) Public debt (++) Public debt (+++)
S&P rating (-)

Table 9 – Dependent variable : amount of sovereign debt being in default


Note : For each regime, defined by the thresholds presented in Table 5, the significant determinants of the amount of debt being
in default are reported here. When the associated slope is positive (resp. negative), they are reported with a + (resp. -) sign.
When a determinant is significant in different regimes with the same sign, the number of +/- signs reported informs on the
importance of the slope from one regime to another.

The observations switch between three regimes. A large majority (91%) are classified into a regime in which
the external debt ratio remains below 258%, while a minority (3%) are in the extreme regime characterized by the
highest external debt vulnerability with a ratio jumping above 385%. The remainder 6% are located in an in-between
regime (external debt ratio between 258% and 385%). Comparing the two extreme regimes, the estimates suggest that
more variables matter to impact sovereign default when external debt is high compared to total exports. Specifically, a
currency depreciation, higher inflation and public debt ratio, a poor S&P rating and past defaults during the preceding
five years, all concur to increase the weight of sovereign debt default when external debt ratio exceeds 385%. In the
low external debt regime, only inflation and public debt ratio are key determinants of default (to a lesser extent than
in the other extreme regime).

3.3 Time-varying effects and countries’ heterogeneity


Our estimates are also to be interpreted in terms of time-varying dynamics of the vulnerability regimes across
time (by seeing how the countries move from a regime to another, or by pointing possible situations characterized by
statu-quo). Making cross-country comparisons is also worthy to point for potential heterogeneous situations, notably
between countries located in the same geographical area.

Figures A.1 to A.4 in Appendix 2 show, for each country, the dynamics of the vulnerability regimes for the four
vulnerability axes presented above. Different numbers (from 0 to 2) 12 and colours (green, orange and red) are used to
signal whether a given year a country was situated in an priori low, medium or high vulnerability regime according to
the vulnerability axes highlighted in the previous section. For each of the four threshold variables, the threshold values
estimated above define the regimes in which each observation can be classified. Here, we consider that a country is "a
priori" more vulnerable (in red) if debt burden (as a share of public revenue) is higher, domestic savings (as a share
of GDP) are lower, international reserves (as a share of short-term external debt) are lower and external debt (as a
12. corresponding to the regime numbers mentioned in Tables 6 to 9.

194 Amundi Investment Strategy Collected Research Papers


share of exports) is higher.

Considering debt service, it seems that, since 2000, the countries have succeeded in hedging themselves against
states where sovereign debt is less under control (in the sense that the determinants of sovereign default are less iden-
tified). However, there remains some differences in trajectories, with some countries now completely insulated from
high debt burden (for example Chile and Morocco), while others are still in a debt spiral (for instance, Egypt, India
and Israel).

Turning to domestic savings, Figure A.2 strongly suggests that the vulnerability regimes are "absorptive" in the
sense that rarely does a country move durably from a regime to another. Therefore, if a country is highly vulnerable
in terms of having a too low saving rate, it remains stuck to this situation and moves to a less vulnerable situation
very slowly.

Concerning external debt and international reserves, it is seen that only a few countries experience a situation of
high vulnerability that does not improve over the years (for instance, Argentina, Lebanon, Pakistan).

1. Latin America 2. Middle East & Emerging Europe

3. Asia 4. Africa

Figure 2 – Vulnerability mapping

Note : For a given variable, the smaller the surface, the higher the degree of a priori vulnerability in terms of that va-
riable. A country is considered to be a priori more vulnerable if debt burden (as a share of public revenue) is higher, domestic
savings (as a share of GDP) are lower, international reserves (as a share of short-term external debt) are lower and external
debt (as a share of exports) is higher. They are classified on a scale of a priori vulnerability (0 to 2) according to their location
based on the estimated threshold values.

Next, we consider cross-country heterogeneity. Figure 2 portrays the multidimensional aspect of countries’ vulne-
rability, by selecting for purpose of illustration some countries in different geographical areas. For a given variable, the
smaller the surface, the higher the degree of a priori vulnerability in terms of that variable (as defined above). The
countries are classified on a scale of a priori vulnerability (0 to 2) according to their location based on the estimated
threshold values 13 . A graph could be made for each year, but we show here a figure based on the last year of available
observation.

It is noteworthy that, in terms of vulnerability, countries located in the same geographical area do not necessarily
share the same characteristics. For instance, Russia, Hungary and China are not in any of the a priori vulnerability
13. For interest payments and external debt, the values of the regimes have been inversed for this example, so as to have a regime 0
corresponding to the a priori most vulnerable regime and a regime 2 to the a priori less vulnerable one, for each of the four variables of
interest.

Amundi Investment Strategy Collected Research Papers 195


regimes identified. Brazil, Turkey and South Africa share similar characteristics (sound situation in terms of external
debt and external reserves, but more fragile if we consider domestic savings and to a lesser extent debt service). A
similar argument applies for Argentina, Morocco and Tunisia.

These results go in the direction of Reinhart and Rogoff (2013, [14]), who argue that we can’t consider developed
countries to be completely different from emerging ones anymore. In fact, more than the broad classifications that
are currently used ("developed" countries versus "emerging" ones), these vulnerability profiles and the classification
of countries into them shows that what really matters are the economic features of the countries to identify the
determinants of the potential default the country is exposed to.

4 Conclusion
In this article, we contribute to the literature according to three aspects. First, we focus, not only on the occur-
rence, but also on the amount of sovereign debt being in default, relative to the country’s GDP. Second, our empirical
analysis is based on a nonlinear panel data model that allows regime-switching dynamics and heterogeneous effects of
the determinants on sovereign debt being in default, whereas the literature essentially assumes homogeneous effects of
the determinants of sovereign default. Finally, we manage to highlight four variables that are able to determine distinct
vulnerability regimes and provide evidence that the variables influencing the amount of debt default vary across these
regimes.

It appears that, taken together, debt service over fiscal revenues, domestic savings, international reserves on short-
term external debt, and external debt on exports give an accurate insight on the emerging country’s determinants
of sovereign default. These lines of approach allow i) to see if some emerging countries, reinforcing their economic
fundamentals, have succeeded in moving from one regime to another through time, therefore evolving in their behavior
in terms of potential default ; ii) to gather countries together according to their similar profile in terms of sensibilities
to specific economic variables.

Ultimately, this article allows to nuance the conclusions regarding the determinants of sovereign default in emerging
countries. As this "group" of countries precisely stands out through a great diversity, our findings enable to highlight
some subgroup’s specificities. One possible extension to this work could be to investigate further, looking at the reaction
to specific shocks, depending on each group’s features. Some institutional aspects could also be added, as they do not
significantly come out here, also they are one of the important aspects through which emerging countries differ from
developed ones.

196 Amundi Investment Strategy Collected Research Papers


A Appendix
A.1 Data

Number Country Code Geographic area


1 Algeria AA Africa
2 Argentina AG Latin America
3 Brazil BR Latin America
4 Bulgaria BL Emerging Europe
5 Chile CL Latin America
6 China CH Asia
7 Colombia CO Latin America
8 Croatia CT Emerging Europe
9 Czech Republic CZ Emerging Europe
10 Egypt EY Africa
11 Estonia EO Emerging Europe
12 Greece GR Emerging Europe
13 Hong Kong HK Asia
14 Hungary HN Emerging Europe
15 India IN Asia
16 Indonesia ID Asia
17 Israel IS Middle East
18 Kenya KN Africa
19 Korea KO Asia
20 Kuwait KW Middle East
21 Latvia LV Emerging Europe
22 Lebanon LB Middle East
23 Lithuania LN Emerging Europe
24 Malaysia MY Asia
25 Mauritius MU Africa
26 Mexico MX Latin America
27 Morocco MC Africa
28 Nigeria NG Africa
29 Pakistan PK Asia
30 Peru PE Latin America
31 Philippines PH Asia
32 Poland PO Emerging Europe
33 Qatar QA Middle East
34 Romania RM Emerging Europe
35 Russia RS Emerging Europe
36 Saudi Arabia SI Middle East
37 Singapore SP Asia
38 Slovakia SX Emerging Europe
39 Slovenia SJ Emerging Europe
40 South Africa SA Africa
41 Taiwan TW Asia
42 Tanzania TN Africa
43 Thailand TH Asia
44 Tunisia TU Africa
45 Turkey TK Middle East
46 Ukraine UR Emerging Europe
47 United Arab Emirates UA Middle East
48 Uruguay UY Latin America
49 Venezuela VE Latin America
50 Vietnam VI Asia

Table A.1 – Selected Emerging markets countries

Amundi Investment Strategy Collected Research Papers 197


198
Item Source Description => Treat- Definition Time period
ment
Terms of trade World Bank, World Deve- Trade Indexes, Net barter Net barter terms of trade index is 1981-2013 except for BL, CT, CZ,
lopment Indicators terms of trade index (2000 = calculated as the percentage ratio EO, GR, HN, IS, KW, LV, LB, LN,
100) => Y/Y % change of the export unit value indexes to PO, QA, RM, RS, SI, SX, SJ, UR,
the import unit value indexes, mea- UA, VI (81-01). TW non available.
sured relative to the base year 2000.
Exchange rate World Bank, World Deve- Exchange Rates & Prices, Official exchange rate refers to the 1981-2013 except for BL, TK (86-
lopment Indicators Official exchange rate (LCU exchange rate determined by na- 13), VI (87-13), AG (89-13), PE
per US$, period average), tional authorities or to the rate (90-13), SJ (92-13), BR, LV, LN,
USD => Y/Y % change determined in the legally sanctio- CT (93-13), EO (94-10), CZ, RS,
ned exchange market. It is calcu- SX, UR (94-13), SJ (06-13), SX (08-
lated as an annual average based 13) and GR (81-00). TW non avai-
on monthly averages (local currency lable.
units relative to the U.S. dollar).
Portfolio Equity World Bank, World Deve- Capital & Financial Ac- Portfolio equity includes net inflows 1980-2013 except for IS (82-13), SA
flows lopment Indicators count, Portfolio equity, from equity securities other than (85-13), IS, PK, TU (88-13), CL,
net inflows (BoP), Current those recorded as direct invest- MX, TK (89-13), KO (90-13), IN
Prices, USD => % of GDP ment and including shares, stocks, (91-13), AG, VE (92-13), KN (93-
depository receipts (American or 12), CT, CZ, EO, HN, ID, KN, LN,
global), and direct purchases of MC, PE, SX (93-13), CO, MU, RS
shares in local stock markets by fo- (94-13), UR, PO (95-13), UR, PH,
reign investors. Data are in current LV, BL (96-13), CH, RM, SJ (97-
U.S. dollars. Missing values have 13), GR (99-13), UY (00-13), HK
been replaced by linear interpola- (98-13), MY (02-09), LB (03-13),
tion when there was a lack of less NG (05-12), VI (05-13), KW (06-
than 5 years. 13), EY (97-12), QA (11-13). AA,
SI, TN, TW, UA non available.

Amundi Investment Strategy Collected Research Papers


Table A.2 – Sources of data : Capital account and foreign exchange variables
Item Source Description => Definition Time period
Treatment
Real GDP International Mone- Gross domestic pro- IMF/WEO general metho- 1980-2013 except for CT, RS, UR, LV, SJ, SX (93-13), EO, SX
growth tary Fund, World duct, constant prices, dological notes : Annual (94-13), CZ, LN (96-13).
Economic Outlook Chg Y/Y percentages of constant
price GDP are year-on-year
changes.
Nominal GDP Oxford Economics GDP, nominal, local In billions for AG, CL, CH, 1980-2013 except for AA, CO, EY, EO, IS, KN, LV, LB, KW,
currency CO, CZ, GR, HN, IN, I, KO, LN, MU, MC, NG, PK, PE, QA, SJ, TU, UR, UY, VI (88-13),
MX, RM, RS, SX, TH, VE, PO (89-13), RM, RS (90-13), CT, HN (91-13), CZ, SX (93-13).
VI. In millions for the other TN non available.
countries.
Nominal GDP Oxford Economics GDP, millions USD 1980-2013 except for AA, CO, EY, EO, IS, KN, LV, LB, KW,
LN, MU, MC, NG, PK, PE, QA, SI, SJ, TU, UR, UY, VI (88-
13), PO (89-13), RS (90-13), HN (91-13), CT (92-13), CZ, SX
(93-13). TN non available.
CPI International Mone- Inflation, average IMF/WEO general metho- 1980-2013, except for CH (90-13), RS, LV, UR, SJ (93-13), EO,
tary Fund, World consumer prices, dological notes : Annual per- SX, CT (94-13), CZ (96-13), LN (00-13)
Economic Outlook Chg Y/Y centages of average consu-
mer prices are year-on-year

of GDP) was inferior to -100 or superior to 100.


changes.
Interest pay- Oxford Economics Central government, In billions for AG, CL, CH, 1980-2013 except for PH (86-13), BL, CL, ID, PK, PE, SA, UY,
ments interest expenditure, CO, CZ, GR, HN, IN, I, KO, TU (90-13), KN, EY (91-13), AG, CT, RS (92-13), KW (93-07),
local currency => % MX, RM, RS, SX, TH, VE, CZ, SJ (93-13), LV (94-13), EO, GR, HN (95-13), BR (97-13),
of public revenues VI. In millions for the other UR (99-13), LN, IS, LB (00-13), PO (01-13), MC, PH, RM (02-
countries. 13), CO, SX (03-13), QA (04-13), HK (05-13), AA (06-13), MU,
TH (09-13). NG, SI, TN, UA, VI non available.

Table A.3 – Sources of data : Economic variables (intern)


General go- International Mone- General government IMF-WEO general metho- 1990-2013 except for GR (80-13), IN, AA, TU (91-13), AG (93-
vernment tary Fund, World gross debt, % of GDP dological notes : Gross debt 13), PK, PH (94-13), PO, CZ, SJ, CH, EO (95-13), TH, CO,
debt Economic Outlook consists of all liabilities that MX (96-13), UR, HN, TW, SX (97-13), VE, KN (98-13), RS,
require payment or pay- LV, UA, SI (99-13), BR, ID, BL, LN, PE, SA, RM, BL, MU,
ments of interest and/or NG, IS, LB (00-13), TN, TK, HK, UY, VI (01-13), EY, CT
principal by the debtor to (02-13).
the creditor at a date or
dates in the future.
Public ba- Oxford Economics Government balance, 1980-2013 except for MU, PE, NG, QA, PK, MC, CO, KN, IS,
lance % of GDP KW, TU, UY, VI (88-13), LV (90-13), EY, HN, LN (91-13),
AG, CT, LB, SI, UR (92-13), RS, SX, SJ, CZ, AA (93-13), UA
(94-13), EO (95-13), BR (97-13). TN non available.
Eurostat General government For all countries except PO. 1994-2013

Amundi Investment Strategy Collected Research Papers


deficit/surplus, % of
GDP
General go- Oxford Economics General government 1980-2013 except for CO, KN, MU, MC, NG, PK, QA, IS, KW,
vernment revenues (% of GDP) SI, TU, UY, VI (88-13), PE, CL (89-13), LV, RM (90-13), EY,
revenues HN, LN (91-13), AG, UR CT, LB (92-13), CZ, RS, SX, SJ,
AA (93-13), UA (94-13), EO, PO (95-13), BR (97-13). TN non
available.
General go- Oxford Economics General government 1980-2013 except for CO, KN, MU, MC, NG, PK, PE, QA, IS,
vernment expenditure (% of KW, SI, TU, UY, VI (88-13), CL (89-13), LV, RM (90-13), EY,
expenditure GDP) HN, LN (91-13), AG, UR CT, LB (92-13), CZ, RS, SX, SJ,
AA (93-13), UA (94-13), EO, PO (95-13), BR (97-13). TN non
available.
Domestic sa- World Bank, World Shares of GDP & Gross domestic savings are 1980-2013 except for LV (80-11),VE (80-12), NG (81-13), CH
vings Development Indica- Other, Gross domes- calculated as GDP less fi- (82-13),VI (86-13), RS, UR (89-13), LN (90-11), CZ, RM, TN,
tors tic savings (% of nal consumption expendi- LB, PO, SX (90-13), QA (94-11), CT, EO, SJ (95-13), UA (01-
GDP) ture (total consumption). 13). TW non available.

199
Note : As some observations were clearly aberrant, we deleted them from the raw data described above. It was for example the
case for observations whose information on GDP growth and GDP was not consistent, and for those where public balance (%
200
Item Source Description => Treat- Definition Time period
ment
Sovereign Bank of Canada Million US$ Estimates of stocks of government obliga- 1980-2013 except for MU (85-13), MC, NG,
debt in de- (David T. Beers and tions in default, including bonds and other CO, EY, KN, LB, PE, TU, UY, VI (88-
fault Jean-Sébastien marketable securities, bank loans, and offi- 13), PO (89-13), TN (13). No default on
Nadeau) cial loans in default, valued in U.S. dollars, this period for CH, CZ, EO, HK, HN, IS,
for the years 1980 to 2013. KO, LN, MY, QA, SP, SI, SX, TW, UA.
International World Bank, Interna- Total reserves (includes Total reserves comprise holdings of mone- 1980-2013 except for CL, NG, VE, UY (80-
reserves tional Debt Statistics gold), Current Prices, tary gold, special drawing rights, reserves 11), CH (82-13), PO (86-08), LN (92-11),
millions USD => % of of IMF members held by the IMF, and hol- BL, UR (92-13), CT (93-08), LV, RS (93-
STED dings of foreign exchange under the control 11), EO, CZ (93-13), SX (93-06), SA (94),
of monetary authorities. The gold com- VI (95-13), SI (96-13), HN (00-13), SP (03-
ponent of these reserves is valued at year- 13), IS (06-13). GR, HK, KW, QA, SJ,
end (December 31) London prices. Data TW, UA non available.
are in current U.S. dollars. Aggregation
Method : Sum. Millions USD except for
UA, SJ, SP, SI, QA, KW, IS, HK, GR, CT,
CZ, EO, HN, SX, PO, KO (US $).
Total short- World Bank, World Debt Outstanding, External Short-term external debt is defined as debt 1980-2013 except for CL, UY (80-11), CH,
term external Development Indica- debt stocks, short-term that has an original maturity of one year or HN (82-13), BL (85-13), PO (86-08), VI
debt (STED) tors (DOD), Current Prices, less. Available data permit no distinction (87-13), RS, LN (92-11), UR (92-13), LV
USD between public and private nonguaranteed (93-11), SX (93-06), CT (93-07), SA (94-
short-term debt. 13). GR, QA, SJ, UA,HK, KW non avai-
lable.
Oxford Economics External debt, short-term, 1992-2013 except for KO, TW (80-13), EO,
millions USD CZ (93-13), SI (96-13), SP (03-13), IS (06-
13).

share of external debt in general government debt was higher than 100%.
External Oxford Economics External debt, total, share of 1980-2013 except for CH, BL (81-13), ID
debt (% of exports (83-13), SA (87-13), PK, MC, NG, CO,
Exports) EY, PE, MU, KN, AA, LB, TU, UY, VI
(88-13), PO (89-13), RS (90-13), HN (91-
13), LV, EO, UR (92-13), SX, CZ, LN, CT
(93-13), IS (95-13), SI (96-13), QA, KW
(97-13), UA (99-13), HK (02-13), GR, SP
(03-13). TN, SJ non available.
General go- Oxford Economics External debt, millions USD For all countries except AA,CO,EY, EO, 1980-2013 except for CH (81-13), PO (89-
vernment => % of General Govern- IS, LV, LN, MU, MC, NG, PE, RM, SP, 13), HN (91-13), RS (92-13), CZ (93-13),

Table A.4 – Sources of data : Economic variables relative to default and external debt
external debt ment debt SX, SJ, TU, UR, and UY. SA (94-13), CT (98-13), BL (99-13), SJ

Amundi Investment Strategy Collected Research Papers


(01-13), HK (02-13), GR (03-13), CO (04-
13). UA, KN, PK, QA, TN, KW, LB, SI,
SP, VI non available.
World Bank QEDS Gross External Debt, Ge- 1999-2013 except for EO, IS (98-13), TU
neral Government, Overall, (03-13), CO, PE, LV, LN, SX, UR, UY (04-
Total, Current Prices, mil- 13), EY (05-13), RM (09-13), MC (10-13),
lions USD => % of General NG, MU (12), AA (13).
Government debt

case for observations whose information on reserves on short-term external debt was not consistent, and for those where the
Note : As some observations were clearly aberrant, we deleted them from the raw data described above. It was for example the
Item Source Description => Treat- Definition Time period
ment
Rule of law in- World Bank, World Rule of Law (RL) : capturing perceptions of the 1996-2013. For 1997, 1999 and 2001,
dicator Development Indica- extent to which agents have confidence in and means of the years before and after
tors abide by the rules of society, and in particular the have been taken for each country.
quality of contract enforcement, property rights,
the police, and the courts, as well as the likelihood
of crime and violence.
Government World Bank, World Government Effectiveness (GE) : capturing per- 1996-2013. For 1997, 1999 and 2001,
effectiveness Development Indica- ceptions of the quality of public services, the qua- means of the years before and after
indicator tors lity of the civil service and the degree of its in- have been taken for each country.
dependence from political pressures, the quality
of policy formulation and implementation, and
the credibility of the government’s commitment
to such policies.
Control of World Bank, World Control of Corruption (CC) : capturing percep- 1996-2013. For 1997, 1999 and 2001,
corruption Development Indica- tions of the extent to which public power is exer- means of the years before and after
indicator tors cised for private gain, including both petty and have been taken for each country.
grand forms of corruption, as well as "capture" of
the state by elites and private interests.
Corruption Transparency Missing values have been replaced by linear in- 1995-2013 except for BL, EO, LV,
Perception terpolation when there was a lack of less than 5 MU, EY, MC, PE, TN, UR (98-13),
Index years. CZ, EY, HK, KN, NG, PO, RS, TW
(96-13), IS, RM, UY, VI (97-13),
TU (98-13), CT, LN, SJ (99-13), SX
(00-13), AA, LB, KW, QA, SI, UA

Amundi Investment Strategy Collected Research Papers


Table A.5 – Sources of data : Institutional variables
(03-13), QA (04-13).

201
202
Item Source Description => Treat- Definition Time period
ment
Emerging JP Morgan JPM EMBI, blended Bond index. CL is given the Latin America EMBI ; CZ, 1994-2013 except for BR, BL, CH, SA, PO
markets bond spreads EO, GR, LV, LN, RM, SX, SJ, the Europe EMBI ; KN, (95-13), CT, MY, TK (97-13), AA, CL,
index (blen- MU, MC, TU, AA, TN the African EMBI ; QA, UA, IS, CB, CZ, EO, GR, KN, LV, LN, MU, MC,
ded spreads) SI, KW the Middle East EMBI ; and IN, KO, TW, TH, PE, RM, RS, SX, SJ, PH, TN, TH, TU
SP, HK the Asia EMBI. (98-13), HN, IS, KW, LB, QA, SI, UA (99-
13), UR (01-13), EY, PK, UY (02-13), ID
(05-13), VI (06-13).
S&P rating Credit rating, average, (0=D Rating on foreign currency long-term debt. Countries 1990-2013
and 22=AAA) which were first rated only after 1990 get a constant ra-
ting before (corresponding to their first rating). AA gets
the rating of TU, TN the rating of KN, UA the rating of
QA, MU the rating of SA.
Interbank Essentially Natio- Interbank overnight 1988-2013 except for SP (88), PH (90), TH
rates nal/Central Banks (91), CB (92-07), CZ (93), MY, PO (94),
CL, HN, PE (96), ID (97), LV (98), IN,
LN (99), RM (00), CH (02), BL (04), HK
(06), CT, TK, EY (07), IS (08). AG, BR,
EO, GR, KN, KO, MU, MX, MC, PK, QA,
RS, SX, SJ, SA, UA, VE, SI, LB, KW, AA,
TU, UY, VI non available.
Lending rate World Bank, World Interest Rates, Lending in- Missing values have been replaced by linear interpolation 1980-2013 except for GR (80-03), MC (80-
Development Indica- terest rate (%) => lending when there was a lack of less than 5 years. 05), PO (80-06), MU (81-13), LB (82-13),
tors rate - deposit rate VE (84-13), CB, ID, PE (86-13), MY (87-
13), HN (89-13), HK (90-13), SJ, BL (91-
13), EO, CT (92-13), LN (93-10), CZ, LV,
MX, SX, UR, VI (93-13), AG, RM, AA
(94-13), RS (95-13), BR (97-13), PK (04-
13), SX (08-13), SJ (09-13), UA (80-84 &
91-01), QA (80-94 & 04-13), TU (80-89).
TW, TK, SI non available.
Deposit rate World Bank, World Interest Rates, Deposit in- Missing values have been replaced by linear interpolation 1980-2013 except for UA (80-84 & 99-01),
Development Indica- terest rate (%) => lending when there was a lack of less than 5 years. TU (80-88), GR (80-05), MC (80-91 & 98-
tors rate - deposit rate 13), SJ (80 -09), MU (81), LB (82), IS (83-
12), VE (84), CB (86), SI (87-08), PE (88),
PO (89-06), HK (90), BL, SJ (91), CT (92),
LN (93-10), SX (93-08), CZ, LV, UR, VI

Table A.6 – Sources of data : Financial variables


(93), EO, RM (94), RS (95), PK (04). TW,
IN non available.
Sovereign in- International Mone- Interest Rates, Government Missing values have been replaced by linear interpolation 1980-2013 except for TH (81-88 & 02-13),
terest rates tary Fund, Interna- securities, Treasury bills when there was a lack of less than 5 years. TK (87-92 & 00-07), IS (84), HN (88-13),

Amundi Investment Strategy Collected Research Papers


tional Financial Sta- PO (92-11), BL, HK, NG, PK (92-13), CZ,
tistics and National TN (93-13), UY (94-13), BR, LV, LN, RS
sources (95-13), RM (96-13), EY (97-13), SJ (99-
13), VI(00-13), MU (01-13), RS (03-13), IN
(08-13),SI (10-13), SP (80-12), CL, ID, VE
(12-13). AG, CH, CO, EO, KO, MC, QA,
PE, SX, TW, UR, UA, TU, CT non avai-
lable.
A.2

27
Figure A.1 – Vulnerability regimes - Central government interest payments (% of public revenues)

Amundi Investment Strategy Collected Research Papers


Note : The table reports the regime to which the observations belong to, for each country and each year. For the "interest payments" model, the three regimes (0, 1 and 2) are delimited by
Evolution of the countries within the vulnerability regimes

interest payments of 3,7% and 13,8% of public revenues (see Table 5). Green (respectively red) cells correspond to observations characterized by lower (respectively higher) interest payments
as a share of public revenues."NA" observations are those for which information on interest payments is not available.

203
204
Amundi Investment Strategy Collected Research Papers
Figure A.2 – Vulnerability regimes - Domestic savings (% of GDP)
Note : The table reports the regime to which the observations belong to, for each country and each year. For the "domestic savings" model, the three regimes (0, 1 and 2) are delimited by
domestic savings of 21,7% and 24,6% of GDP (see Table 5). Green (respectively red) cells correspond to observations characterized by higher (respectively lower) domestic savings as a share
of GDP."NA" observations are those for which information on domestic savings is not available.
Amundi Investment Strategy Collected Research Papers
Figure A.3 – Vulnerability regimes - Foreign reserves (% of short-term external debt)
Note : The table reports the regime to which the observations belong to, for each country and each year. For the "foreign reserves" model, the two regimes (0 and 1) are delimited by foreign
reserves of 78,5% of short-term external debt (see Table 5). Green (respectively red) cells correspond to observations characterized by higher (respectively lower) foreign reserves as a share
of short-term external debt."NA" observations are those for which information on foreign reserves is not available.

205
206
Amundi Investment Strategy Collected Research Papers
Figure A.3 – Vulnerability regimes - Foreign reserves (% of short-term external debt)
Note : The table reports the regime to which the observations belong to, for each country and each year. For the "foreign reserves" model, the two regimes (0 and 1) are delimited by foreign
reserves of 78,5% of short-term external debt (see Table 5). Green (respectively red) cells correspond to observations characterized by higher (respectively lower) foreign reserves as a share
of short-term external debt."NA" observations are those for which information on foreign reserves is not available.
Amundi Investment Strategy Collected Research Papers
Figure A.4 – Vulnerability regimes - External debt (% of exports)
Note : The table reports the regime to which the observations belong to, for each country and each year. For the "external debt" model, the three regimes (0, 1 and 2) are delimited by
external debt of 126,8% and 258,3% of exports (see Table 5). Green (respectively red) cells correspond to observations characterized by lower (respectively higher) external debt as a share
of exports."NA" observations are those for which information on external debt is not available.

207
References
[1] I. Andréou, G. Dufrénot, A. Sand-Zantman & A. Zdzienicka-Durand. A forewarning indicator system for financial
crises : the case of six-central and eastern European countries. Journal of economic integration 24, 1 (2009), pp.
87-115, June 2008.
[2] E. Baldacci, J. McHugh & I. Petrova. Measuring Fiscal Vulnerability and Fiscal Stress : A Proposed Set of
Indicators. IMF Working Paper, WP/11/94, April 2011.
[3] D. T. Beers & J-S. Nadeau. Introducing a New Database of Sovereign Defaults. Bank of Canada Technical report
N0.101, February 2014.
[4] H. J. Blommestein & J. Santiso. New strategies for emerging domestic sovereign bond markets. OECD Develop-
ment Centre, Working Paper No.260, April 2007.
[5] D. Cohen & C. Valadier. The Sovereign Debt Crisis that Was Not. Sovereign Debt and the Financial Crisis : Will
This Time Be Different ? (Edited by Carlos A. Primo Braga and Gallina A. Vincelette), Part 1, Chapter 1, 2011.
[6] A. Gonzalez, T. Terasvirta & D.v. Dijk. Panel Smooth Transition Regression Models. Quantitative Finance
Research Centre, Research Paper 165, August 2005.
[7] A. Kraay & V. Nehru. When is external debt sustainable ? World Bank Policy Research Working Paper 3200,
February 2004.
[8] P. Manasse, N. Roubini, A. Schimmelpfennig. Predicting Sovereign Crises.IMF Working Paper, WP/03/221,
November 2003.
[9] M. S. Mohanty. Fiscal policy, public debt and monetary policy in EMEs : an overview. Bank of International
Settlements Papers, No 67, October 2012.
[10] A. Mehl & J. Reynaud. The determinants of « domestic »original sin in emerging market economies. European
Central Bank Working Paper Series No.560, December 2005.
[11] Rabobank, Economic Research Department. The Rabo Sovereign Vulnerability Index. Special Report, 2011/07,
April 2011.
[12] C. M. Reinhart, M. Goldstein, G. Kaminsky. Assessing financial vulnerability, an early warning system for emer-
ging markets : Introduction. Munich Personal RePEc Archive Paper No. 13629, 2000.
[13] C. M. Reinhart & K. S. Rogoff. Serial Default and the « Paradox »of Rich-to-Poor Capital Flows. American
Economic Review, 94(2), 53-58, 2004.
[14] C. M. Reinhart, K. S. Rogoff. Financial and Sovereign Debt Crises : Some Lessons Learned and Those Forgotten.
IMF Working Paper, WP/13/266, December 2013.
[15] C. M. Reinhart, K. S. Rogoff & M. A. Savastano. Debt Intolerance. Brookings Papers on Economic Activity, No.
1 (2003), pp. 1-62, Vol. 2003.
[16] J. D. Sachs. Resolving the Debt Crisis of Low-Income Countries. Brookings Papers on Economic Activity, No. 1
(2002), pp. 257-286, Vol. 2002.

208 Amundi Investment Strategy Collected Research Papers


WP-048

The Asset- and Mortgage-Backed


Securities Market in Europe
MARIELLE DE JONG,
Head of Fixed-Income Quantitative Research, Amundi
VINH CAM ANH NGUYEN,
Consultant of Fixed-Income Quantitative Research, Amundi
HUBERT VANNIER,
Senior Portfolio Manager, Amundi

March 2015

How do European asset- and mortgage-backed securities fare


today five years after the 2008 crisis they have been incriminated
in? We make an assessment of the asset class in today’s
renormalized market conditions.
We explore its return-to-risk profile in a standard mean-
variance framework, taking the view of a long-term Euro Area
bond investor. We make evident that the securities significantly
reduce investment risk and in the same time improve the outlook
for return, when combined with other European bonds.

Amundi Investment Strategy Collected Research Papers 209


I. Introduction

What was discredited as a plague or even identified as the very symbol for the intoxication of
the financial markets in 2008 and 2009, has recovered remarkably since. The asset- and
mortgage-backed securities in Europe, abbreviated by ABS usually and worth one-and-a-half
trillion euro in all, are back at more reasonable credit spread levels, under 3% on average,
down from 9% in the heat of the financial crisis. Confidence seems restored and excessive
liquidity shortfalls have ceased.
Two major structural reforms are debit to this, as Jeanniard (2011) points out. Firstly the
stakeholders operating on the European ABS markets have changed since the crisis. The bulk
of investors who were refinancing long-term assets with short-term positions and got caught
out by the rupture in market liquidity, have given way to longer-term investors. And secondly
the structure of certain instruments has been simplified making the asset class more
transparent as a result. Certain safety nets embedded in the instruments have been tested for
real, which has sparked the market confidence. With that the sting has been taken out of the
asset class in Europe. This is not necessarily the case in the United States where unresolved
issues remain, in particular in connection with the insolvency position of the nationalized loan
corporations Fannie Mae and Freddie Mac. There is no equivalent for this in Europe, instead
mortgage-backed securities bear the credit risk themselves through a pooling-and-tranching
system very much like the American asset-backed instruments..
In this study we look at the European asset- and mortgage-backed securities market and
compare their investment profile with that of Euro Area sovereign bonds. We do this in the
standard mean-variance framework, both in absolute return terms and in a relative
benchmark-enhancement setup. We measure by how much the investment opportunity of a
bond investor is set to expand by including these assets.

II. Data and test methodology

II.1. Expected returns

Among the European ABS markets we have selected those who are best suited for a mean-
variance analysis. For that matter we have retained the most senior tranches only with an

210 Amundi Investment Strategy Collected Research Papers


AAA rating at inception, so as to play down default risk (tail events) and bring about the more
mainstream market risk. It results in a set of nine indices representing the high-quality
European ABS market. Six contain residential mortgage-backed securities (RMBS), one
contains commercial mortgage-backed securities (CMBS), one auto loans and one small-to-
medium size enterprise collateralized loan obligations (SME CLO).
We have retrieved total return series for the indices as calculated by Markit on a weekly basis
over a seven-year period from January 2007 to February 2014, and we have retrieved returns
for four Barclays Euro Treasury indices. The returns include coupon payments, price variation
and -in the case of ABS data- the pre-empted payments of principal as well. Markit
establishes ABS market prices on the basis of surveys among a set of broker houses who
participates to give regular price quotes. Further documentation on the price calculations and
index construction can be found on the respective web sites of the data providers.
Key data features are given in Table 1. In the first column are the number of securities in the
index, in (a) the weighted average life (WAL) for the ABS which compares with the modified
duration for the sovereigns, in (b) the average spreads for the ABS as calculated by JP
Morgan for February 2014, which together with the euro swap rate of corresponding WAL,
given in (c), add up to the yields-to-maturity (YTM), in (d). For the treasury indices the YTM
are as calculated by Barclays for February 2014.

Table 1 Test bed: nine European ABS indices and four Euro Treasury indices
Index # WAL/duration spread vs swap euro swap yield to maturity
issues in years (a) in bp (b) in bp (c) in bp (d)
EU Auto loans 26 1.0 34 38 72
EU CMBS 18 2.0 198 45 243
EU RMBS 283 5.2 59 102 161
Spanish RMBS 135 6.7 208 143 351
Spanish SME CLO 13 1.7 173 45 218
Italian RMBS 41 3.5 178 81 259
Portuguese RMBS 22 8.2 268 161 429
Dutch RMBS 74 4.0 59 81 140
UK PRMBS 31 1.3 54 38 92

French Treasuries 42 6.7 142


Spanish Treasuries 34 5.8 244
Italian Treasuries 58 6.1 255
German Treasuries 54 6.7 99
* Data source: Markit iBoxx for ABS data, JP Morgan for spreads, Bloomberg for the euro swap rates
and Barclays for the euro treasury indices.

Amundi Investment Strategy Collected Research Papers 211


In the portfolio optimizations we carry out in this study, we consider the YTM as given in
Table 1, to represent the expected asset returns. This makes sense for a long-term investor. If
the intention is to hold the assets all along until maturity, the YTM will be exactly the
investment return, that is the carry seized over the holding period. If the assets are to be held
for the medium-to-long term, the yields-to-maturity are the unbiased estimates for future
returns, in the sense that they give market-neutral expectations.

II.2. Expected volatility

The historical volatility levels of the ABS returns are compared with those of sovereigns in
Exhibit 2. The ones given in the Table have been measured over the entire observation period,
which comprises two crises: the ABS liquidity crisis in 2008-2009 and the euro sovereign
debt crisis that peaked in 2010-2011. The volatility levels were thus higher than they are
nowadays now that both crises have calmed down.
In Exhibit 2 the volatility levels have been measured over a one-year trailing time-window
and are compared over time for three indices: the Spanish RMBS, Spanish Treasuries and
German Treasuries. It can be seen that the Spanish RMBS became twice as volatile over the
first crisis years and that the volatility of Spanish Treasuries sparked over the second crisis
years. The other ABS indices have gone through very much the same orbit as Spanish RMBS,
while among the sovereigns the divide was general between core and peripheral countries.
In the portfolio optimizations we have carried out, we have taken the prudent stance to retain
the relatively high volatility levels measured over the entire observation period as given in
Exhibit 2, as the expected volatilities. By that we incorporate the possibility of a new ABS
crisis into the risk forecasts. Such scenario is conservative compared to the more realistic
situation that the ABS liquidity crisis has faded and is not likely to reoccur.
ote that the asset-backed securities are on the whole less volatile than sovereigns
notwithstanding the two crises, which leads to higher return-to-risk ratios.

212 Amundi Investment Strategy Collected Research Papers


Exhibit 2 Return volatilities and Sharpe ratios
Index volatility Sharpe
EU Auto loans 1.2 0.38
EU CMBS 4.3 0.51
EU RMBS 2.4 0.56 14%
Spanish RMBS 4.5 0.72 Spanish RMBS volatility
German Treasuries volatility
Spanish SME CLO 3.2 0.60 12%
Spanish Treasuries volatility

Italian RMBS 3.7 0.63 10%

Portuguese RMBS 6.6 0.62 8%

Dutch RMBS 1.7 0.68 6%


UK PRMBS 2.8 0.24
4%

French Treasuries 4.9 0.24 2%

Spanish Treasuries 7.7 0.28


0%
Italian Treasuries 6.4 0.36 04/01/2008 04/01/2009 04/01/2010 04/01/2011 04/01/2012 04/01/2013

German Treasuries 4.7 0.16


*annualized return volatilities

II. . o elatio

The return correlations measured over the entire observation period are given in Table 3. ote
that the correlation between the two asset classes, in the off-diagonal blocks, is close to zero.
It means that there is little price influence between ABS and sovereigns, giving much scope
for risk diversification between them. We have verified that the inter-class correlation remains
nil during the two crisis periods. This can be seen in the Appendix where correlation tables
have been measured over two sub-periods, from 2007 to 2009 and from 2010 to 2013.

Table 3 Correlation between the assets


BS
LO

RM
EC
BS
s
an

BS

BS
RM

SM

se

BS
lo

RM

RM
BS

BS

y
e

ĐŽƌƌĞůĂƚŝŽŶ
M
to

an
gu
ish

sh
CM

RM

ce
n

PR
Au

h
rt u

rm
i

ain
lia
an

an

tc

an

ly

ϮϬϬϳͲϮϬϭϯ
Du

UK

Ge
EU

EU

EU

It a

It a
Po
Sp

Sp

Sp
Fr

EU Auto loans ϭ Ϭ͕ϯϭ Ϭ͕ϰϭ Ϭ͕ϯϲ Ϭ͕ϰϬ Ϭ͕Ϭϲ Ϭ͕Ϯϭ Ϭ͕ϯϱ Ϭ͕ϯϱ Ϭ͕Ϭϭ ͲϬ͕Ϭϭ Ϭ͕ϬϮ Ϭ͕ϬϮ
EU CMBS Ϭ͕ϯϭ ϭ Ϭ͕ϰϯ Ϭ͕ϯϳ Ϭ͕Ϯϰ Ϭ͕ϭϵ Ϭ͕Ϯϯ Ϭ͕Ϯϵ Ϭ͕ϯϭ Ϭ͕Ϭϭ ͲϬ͕Ϭϭ Ϭ͕Ϭϰ ͲϬ͕ϬϮ
EU RMBS Ϭ͕ϰϭ Ϭ͕ϰϯ ϭ Ϭ͕ϴϳ Ϭ͕ϰϵ Ϭ͕ϱϴ Ϭ͕ϱϰ Ϭ͕ϲϮ Ϭ͕ϯϳ Ϭ͕Ϭϰ ͲϬ͕Ϭϭ Ϭ͕ϭϬ Ϭ͕Ϭϭ
Spanish RMBS Ϭ͕ϯϲ Ϭ͕ϯϳ Ϭ͕ϴϳ ϭ Ϭ͕ϰϰ Ϭ͕ϯϯ Ϭ͕ϯϭ Ϭ͕ϯϲ Ϭ͕Ϯϰ Ϭ͕ϭϯ Ϭ͕Ϭϯ Ϭ͕ϭϲ Ϭ͕Ϭϵ
Spanish SME CLO Ϭ͕ϰϬ Ϭ͕Ϯϰ Ϭ͕ϰϵ Ϭ͕ϰϰ ϭ Ϭ͕ϭϱ Ϭ͕Ϯϱ Ϭ͕ϯϭ Ϭ͕Ϯϴ Ϭ͕Ϭϯ ͲϬ͕Ϭϱ Ϭ͕Ϭϰ Ϭ͕ϬϮ
Italian RMBS Ϭ͕Ϭϲ Ϭ͕ϭϵ Ϭ͕ϱϴ Ϭ͕ϯϯ Ϭ͕ϭϱ ϭ Ϭ͕ϰϬ Ϭ͕Ϯϴ Ϭ͕ϭϴ ͲϬ͕Ϭϯ ͲϬ͕Ϭϯ Ϭ͕Ϭϳ ͲϬ͕Ϭϵ
Portuguese RMBS Ϭ͕Ϯϭ Ϭ͕Ϯϯ Ϭ͕ϱϰ Ϭ͕ϯϭ Ϭ͕Ϯϱ Ϭ͕ϰϬ ϭ Ϭ͕Ϯϰ Ϭ͕ϭϵ ͲϬ͕ϭϴ ͲϬ͕ϭϲ ͲϬ͕ϭϭ ͲϬ͕Ϭϴ
Dutch RMBS Ϭ͕ϯϱ Ϭ͕Ϯϵ Ϭ͕ϲϮ Ϭ͕ϯϲ Ϭ͕ϯϭ Ϭ͕Ϯϴ Ϭ͕Ϯϰ ϭ Ϭ͕ϰϳ Ϭ͕ϬϬ ͲϬ͕Ϭϭ Ϭ͕Ϭϰ Ϭ͕ϬϬ
UK PRMBS Ϭ͕ϯϱ Ϭ͕ϯϭ Ϭ͕ϯϳ Ϭ͕Ϯϰ Ϭ͕Ϯϴ Ϭ͕ϭϴ Ϭ͕ϭϵ Ϭ͕ϰϳ ϭ Ϭ͕ϬϬ Ϭ͕Ϭϭ Ϭ͕Ϭϳ ͲϬ͕Ϭϭ
France Ϭ͕Ϭϭ Ϭ͕Ϭϭ Ϭ͕Ϭϰ Ϭ͕ϭϯ Ϭ͕Ϭϯ ͲϬ͕Ϭϯ ͲϬ͕ϭϴ Ϭ͕ϬϬ Ϭ͕ϬϬ ϭ Ϭ͕ϯϲ Ϭ͕ϯϴ Ϭ͕ϴϱ
Spain ͲϬ͕Ϭϭ ͲϬ͕Ϭϭ ͲϬ͕Ϭϭ Ϭ͕Ϭϯ ͲϬ͕Ϭϱ ͲϬ͕Ϭϯ ͲϬ͕ϭϲ ͲϬ͕Ϭϭ Ϭ͕Ϭϭ Ϭ͕ϯϲ ϭ Ϭ͕ϴϬ Ϭ͕ϭϰ
Italy Ϭ͕ϬϮ Ϭ͕Ϭϰ Ϭ͕ϭϬ Ϭ͕ϭϲ Ϭ͕Ϭϰ Ϭ͕Ϭϳ ͲϬ͕ϭϭ Ϭ͕Ϭϰ Ϭ͕Ϭϳ Ϭ͕ϯϴ Ϭ͕ϴϬ ϭ Ϭ͕ϭϬ
Germany Ϭ͕ϬϮ ͲϬ͕ϬϮ Ϭ͕Ϭϭ Ϭ͕Ϭϵ Ϭ͕ϬϮ ͲϬ͕Ϭϵ ͲϬ͕Ϭϴ Ϭ͕ϬϬ ͲϬ͕Ϭϭ Ϭ͕ϴϱ Ϭ͕ϭϰ Ϭ͕ϭϬ ϭ

Data source : Markit and Barclays. Calculations made by the authors.

Amundi Investment Strategy Collected Research Papers 213


On the basis of the observed correlation we have built a risk model to estimate the structural
covariance between the assets and discard spurious relations. We do this by means of
principle component analysis (PCA), see Jolliffe (2002) for a general reference.
The first four components that result from running a PCA over the observation period are
displayed in Figure 4. The eigenvalues corresponding to these components indicate how much
of the total variance they explain when taken in proportion to the sum of eigenvalues. We
report that they explain 30 , 18 , 11 and 9 of the variance respectively, which is
significant. In Figure 4 the individual sensitivities of the assets to the four components are
given.

Figure 4 Principal components of the correlation matrix

Ϭ͕ϲ

Ϭ͕ϰ

Ϭ͕Ϯ

ͲϬ͕Ϯ (1) ABS risk vs non-ABS


ͲϬ͕ϰ (3) Core / peripherals - ABS

ͲϬ͕ϲ
EU Auto EU CMBS EU RMBS Spanish Spanish Italian Portuguese Dutch UK PRMBS France Spain Italy Germany
loans RMBS SME CLO RMBS RMBS RMBS

Ϭ͕ϴ

Ϭ͕ϲ

Ϭ͕ϰ

Ϭ͕Ϯ

ͲϬ͕Ϯ

ͲϬ͕ϰ (2) Interest rate risk


ͲϬ͕ϲ (4) Core / peripherals - sovereigns
ͲϬ͕ϴ
EU Auto EU CMBS EU RMBS Spanish Spanish Italian Portuguese Dutch UK PRMBS France Spain Italy Germany
loans RMBS SME CLO RMBS RMBS RMBS

We give an interpretation of the PCA results.


1) The first component, to which all asset-backed securities are sensitive and sovereigns
are almost insensitive, represents a binary ABS versus non-ABS risk factor.
2) The second, to which only the sovereigns are sensitive, represents the interest rate risk
factor.

10

214 Amundi Investment Strategy Collected Research Papers


3) The third component makes a distinction within the ABS class it opposes the
peripheral countries to the core countries within the Eurozone, with the exception of
Spanish small-sized collateral loans.
4) The fourth factor does the same for the sovereign bonds it opposes Spain and Italy to
Germany and France. This factor has emerged since the sovereign debt crisis. It is
interesting to note that this factor seems to have had a knock-on effect on the asset-
backed securities.

Given that the four components are statistically significant and have an intuitive interpretation
we retain them as factors, denoted , in a linear-factor model. The other PCA components are
neither significant nor intuitive. We have retained the residual variances of the assets as well,
denoted V i2 , that remain after subtracting the common factor returns. Formally we specify the

return of asset i over time t as

Di  ¦
4
it 1
Ei ˜ t  H it (1)

so that the covariance between two assets is specified by

°¦
­ 4
Ei ˜V 2 ˜ E i iz
®
1
cov(i, ) (2)
°̄¦
4
1
E i ˜ V 2 ˜ E i  V i2 i i

We obtain the modelled correlations as given in Table 5. We make the assumption that this
correlation structure is persistent going forward. The model fits the data well as can be seen
by the resemblance with the observed correlation given in Table 3.

Table 5 Modelled correlation between the assets


BS
LO

RM
EC
BS
s
an

BS

BS
RM

SM

se

BS
lo

RM

RM
BS

BS

y
e

ŵŽĚĞůůĞĚ
M
to

an
gu
ish

sh
CM

RM

ce
n

PR
Au

h
rt u

rm
i

ain
lia
an

an

tc

an

ly

ĐŽƌƌĞůĂƚŝŽŶ
Du

UK

Ge
EU

EU

EU

It a

It a
Po
Sp

Sp

Sp
Fr

EU Auto loans ϭ Ϭ͕ϰϬ Ϭ͕ϯϲ Ϭ͕ϯϭ Ϭ͕ϱϳ ͲϬ͕ϭϯ Ϭ͕Ϭϱ Ϭ͕ϲϮ Ϭ͕ϲϯ Ϭ͕Ϭϴ ͲϬ͕ϭϮ ͲϬ͕Ϭϵ Ϭ͕ϭϮ
EU CMBS Ϭ͕ϰϬ ϭ Ϭ͕ϯϲ Ϭ͕Ϯϳ Ϭ͕ϯϰ Ϭ͕ϭϭ Ϭ͕ϭϮ Ϭ͕ϰϮ Ϭ͕ϯϲ ͲϬ͕ϬϮ ͲϬ͕ϬϮ Ϭ͕Ϭϭ ͲϬ͕ϬϮ
EU RMBS Ϭ͕ϯϲ Ϭ͕ϯϲ ϭ Ϭ͕ϲϬ Ϭ͕ϰϱ Ϭ͕ϱϰ Ϭ͕ϯϴ Ϭ͕ϲϮ Ϭ͕ϯϲ Ϭ͕ϬϮ Ϭ͕Ϭϭ Ϭ͕Ϭϴ Ϭ͕ϬϬ
Spanish RMBS Ϭ͕ϯϭ Ϭ͕Ϯϳ Ϭ͕ϲϬ ϭ Ϭ͕ϯϲ Ϭ͕ϯϰ Ϭ͕Ϯϭ Ϭ͕ϰϳ Ϭ͕Ϯϴ Ϭ͕ϮϮ Ϭ͕ϭϭ Ϭ͕ϭϲ Ϭ͕ϭϵ
Spanish SME CLO Ϭ͕ϱϳ Ϭ͕ϯϰ Ϭ͕ϰϱ Ϭ͕ϯϲ ϭ Ϭ͕Ϭϳ Ϭ͕ϭϰ Ϭ͕ϱϰ Ϭ͕ϰϲ Ϭ͕Ϭϵ ͲϬ͕ϭϭ ͲϬ͕Ϭϴ Ϭ͕ϭϯ
Italian RMBS ͲϬ͕ϭϯ Ϭ͕ϭϭ Ϭ͕ϱϰ Ϭ͕ϯϰ Ϭ͕Ϭϳ ϭ Ϭ͕ϯϳ Ϭ͕ϮϮ ͲϬ͕Ϭϯ ͲϬ͕ϮϬ Ϭ͕Ϭϰ Ϭ͕Ϭϵ ͲϬ͕Ϯϱ
Portuguese RMBS Ϭ͕Ϭϱ Ϭ͕ϭϮ Ϭ͕ϯϴ Ϭ͕Ϯϭ Ϭ͕ϭϰ Ϭ͕ϯϳ ϭ Ϭ͕ϮϮ Ϭ͕Ϭϱ ͲϬ͕Ϯϳ ͲϬ͕Ϯϱ ͲϬ͕ϮϮ ͲϬ͕Ϯϭ
Dutch RMBS Ϭ͕ϲϮ Ϭ͕ϰϮ Ϭ͕ϲϮ Ϭ͕ϰϳ Ϭ͕ϱϰ Ϭ͕ϮϮ Ϭ͕ϮϮ ϭ Ϭ͕ϱϳ ͲϬ͕ϬϮ ͲϬ͕Ϭϭ Ϭ͕Ϭϯ ͲϬ͕Ϭϯ
UK PRMBS Ϭ͕ϲϯ Ϭ͕ϯϲ Ϭ͕ϯϲ Ϭ͕Ϯϴ Ϭ͕ϰϲ ͲϬ͕Ϭϯ Ϭ͕Ϭϱ Ϭ͕ϱϳ ϭ ͲϬ͕Ϭϯ Ϭ͕Ϭϰ Ϭ͕Ϭϳ ͲϬ͕Ϭϲ
France Ϭ͕Ϭϴ ͲϬ͕ϬϮ Ϭ͕ϬϮ Ϭ͕ϮϮ Ϭ͕Ϭϵ ͲϬ͕ϮϬ ͲϬ͕Ϯϳ ͲϬ͕ϬϮ ͲϬ͕Ϭϯ ϭ Ϭ͕ϯϵ Ϭ͕ϯϴ Ϭ͕ϵϭ
Spain ͲϬ͕ϭϮ ͲϬ͕ϬϮ Ϭ͕Ϭϭ Ϭ͕ϭϭ ͲϬ͕ϭϭ Ϭ͕Ϭϰ ͲϬ͕Ϯϱ ͲϬ͕Ϭϭ Ϭ͕Ϭϰ Ϭ͕ϯϵ ϭ Ϭ͕ϴϱ Ϭ͕ϭϮ
Italy ͲϬ͕Ϭϵ Ϭ͕Ϭϭ Ϭ͕Ϭϴ Ϭ͕ϭϲ ͲϬ͕Ϭϴ Ϭ͕Ϭϵ ͲϬ͕ϮϮ Ϭ͕Ϭϯ Ϭ͕Ϭϳ Ϭ͕ϯϴ Ϭ͕ϴϱ ϭ Ϭ͕ϭϬ
Germany Ϭ͕ϭϮ ͲϬ͕ϬϮ Ϭ͕ϬϬ Ϭ͕ϭϵ Ϭ͕ϭϯ ͲϬ͕Ϯϱ ͲϬ͕Ϯϭ ͲϬ͕Ϭϯ ͲϬ͕Ϭϲ Ϭ͕ϵϭ Ϭ͕ϭϮ Ϭ͕ϭϬ ϭ

11

Amundi Investment Strategy Collected Research Papers 215


II. . a o it opti i atio

In the next section we carry out Markowitz (1952) optimization analyses in the traditional
Capital Asset Pricing Model framework, see Sharpe (1964). This established analysis
technique has its known limitations, which we discuss briefly for the case of asset-backed
securities.
All variables are based on estimations which may be erroneous. The precautions we make to
avoid this are stipulated in this section. For the expected returns in particular we make note
that they are based on the current yields-to-maturity and as such represent expected carry
performance only. A possible tightening or loosening of the credit spreads is not considered.
The asset returns are assumed to be normally distributed. Based on the stable price behaviour
of the ABS over the last five years we make the pro ection that this will continue going
forward. The selection of senior, Triple A tranches has been made deliberately to favour this
situation.
Practical issues in particular market liquidity is not taken into account. There are the two sides
to consider. The sell side is since the easing of the crisis in 2009 no longer an obstacle. The
ma ority of ABS sales take place through bids-wanted-in-competition vehicles (BWIC) which
are fluid. The buy side for ABS has become slow, since the securities are in ma ority held by
long-term investors. The risk related to this situation is to miss investment opportunity, which
is not the same severity of risk five years ago when investors got caught out by the sudden
market drought.

III. Portfolio optimisation

III. . ol te i a d et opti i atio

In order to evaluate the absolute benefit of mixing ABS and sovereigns in an investment
portfolio, we carry out absolute risk/return optimizations given the return potential of the
assets (in Table 1), their volatility levels (in Table 2) and the covariance structure (in Table 5).
We build long-only and fully-invested portfolios while varying the aversion to risk and by
that trace the efficient frontier. The resulting portfolios are given in Figure 6.

12

216 Amundi Investment Strategy Collected Research Papers


The portfolios on the efficient frontier, among which the minimum-risk portfolio A, the
maximum-Sharpe portfolio B, and the maximum-return portfolio C, are heavily invested in
ABS, as can be seen in the Table. This is not surprising given the favourable features of asset-
backed securities: their relatively low return volatilities, their significant return potential
combined with the low correlation levels with sovereigns.
In the Figure the efficient frontier is compared with a basis portfolio that is 100 invested in
sovereigns, denoted D. The Figure shows by how much ABS could hypothetically add value
to such a portfolio in absolute terms, that is without taking into account any investment
constraints. Of course these portfolios are not realistic options for a large investor who faces
implementation constraints.

Figure 6 Added value of ABS to a portfolio invested in sovereigns – in absolute term

min. max. max.


risk Sharpe return sov
A B C
ABS 80 82 100 0
sovereigns 20 18 0 100

return 183 307 429 158


risk 2.0 2.5 5.0 4.7

III.2. e c a e a ced i ve t e t opti i atio

Considering a more realistic situation where the investments are being compared with a given
benchmark, we take the case of an investor whose performance is compared with the Euro
Treasuries bond index. ence, the portfolio risk, defined as the tracking error (TE), is nil
when holding the index positions and increases as more active positions are being added to
the portfolio. In the same way the portfolio return is measured to the extent that it can beat the
benchmark (add alpha). In this setting we do the same exercise as is done above, we vary the
risk aversion while optimizing the alpha with respect to the tracking error and by that trace the
efficient frontier of optimal portfolios. The results are in Figure 7.

13

Amundi Investment Strategy Collected Research Papers 217


Figure 7 Added value of ABS to a portfolio invested in sovereigns – in relative terms

min. risk max. IR max. alpha


D E F
ABS 0 14 100
sovereigns 100 86 0

alpha 0 44 257
TE 0 1.0 7,0

The minimum-risk portfolio D is the benchmark, and the maximum-return portfolio F is the
same as portfolio C. Portfolio E maximizes the relative return-to risk profile, the Information
Ratio (IR). It is plotted onto the absolute return and risk axes in Figure 6 and is displayed in
Figure 8. The portfolio beats the benchmark both in terms of risk and return. By adding 14
of ABS to a sovereign-invested portfolio the overall risk reduces from 4.7 to 4.2 , while
the return potential increases from 158 to 216 basis points. As a result, under the hypothesis
of a risk-free return at 0.15 (the 1-year German sovereign yield), the Sharpe ratio increases
from 0.30 to 0.48.

Figure 8 The maximum-IR portfolio (E)


Ϭй ϰй
Ϯй
ϭϬй
ϭϴй
EU CMBS
Spanish RMBS
Portuguese RMBS
France
ϭϲй Spain
Italy
Germany
ϰϵй

14

218 Amundi Investment Strategy Collected Research Papers


III. . t e te t

In a last experiment we run a stress test, by which we measure what would happen to the
optimized portfolio E in case a new ABS crisis would reoccur. If the crisis would manifest in
the same magnitude as in 2007-2009, the correlation within the ABS class would double, their
volatility levels would double as well yet there would be little to no contagion to the
sovereign bonds (see Appendix). In that situation the volatility of the optimized portfolio
would rise. The risk with respect to the benchmark, the tracking error, would increase from
0.9 to 1.1 . The increase is limited due to the fact that the ABS pocket is relatively small
and that there is no contagion.

IV. Conclusion

We make the observation that five years after the financial crisis, the asset- and mortgage-
backed securities in Europe compared to euro sovereigns (i) yield higher, (ii) are less volatile
and (iii) are less correlated, both between themselves and with respect to sovereigns. This is
the interest of our paper. It comes to no surprise that with those favourable features the ABS
take a predominant position in the return-to-risk optimal portfolio.

References
Jeanniard, R. (2011) “How have asset-backed securities gone from one crisis to another, from
a plague to a safe haven security”, Amundi Cross asset investment strategy n 11.

Jolliffe, I. (2002) “Principal Component Analysis”, Springer Series in Statistics.

Markowitz, H. (1952) “Portfolio selection”, o al o i a ce 7 (1).

Sharpe, W. (1964) “Capital asset prices: a theory of market equilibrium under conditions of
risk”, o al o i a ce 19 (3).

15

Amundi Investment Strategy Collected Research Papers 219


Appendix

In the Table below are the correlations that have been measured over two sub-periods: one
from 2007 to 2009 over the ABS crisis and one from 2010 to 2013 which includes the
sovereign debt crisis.
It is relevant to note that
i. the correlation structure remains stable over the entire period, in particular the
correlations between the two asset classes remain close to zero in both sub-periods.
ii. the correlation within the ABS class doubles in the ABS crisis period,
iii. the average correlation between the sovereign bonds falls sharply during the sovereign
debt crisis (between core and peripheral countries),

Ad Table 3 Correlation measured over two sub-periods

During the ABS liquidity crisis


BS
LO

RM
EC
BS
s
an

BS

BS
RM

SM

se

BS
lo

RM

RM
BS

BS

y
e

ĐŽƌƌĞůĂƚŝŽŶ
M
to

an
gu
sh

sh
CM

RM

ce
n

PR
Au

h
rt u

rm
i

ain
lia
an

an

tc

an

ly
ϮϬϬϳͲϮϬϬϵ
Du

UK

Ge
EU

EU

EU

It a

It a
Po
Sp

Sp

Sp
Fr
EU Auto loans ϭ Ϭ͕ϯϳ Ϭ͕ϱϱ Ϭ͕ϰϴ Ϭ͕ϱϬ Ϭ͕ϭϰ Ϭ͕ϰϴ Ϭ͕ϯϵ Ϭ͕ϯϳ Ϭ͕Ϭϯ ͲϬ͕Ϭϱ Ϭ͕Ϭϯ Ϭ͕Ϭϰ
EU CMBS Ϭ͕ϯϳ ϭ Ϭ͕ϰϲ Ϭ͕ϰϭ Ϭ͕Ϯϵ Ϭ͕ϭϵ Ϭ͕ϯϲ Ϭ͕ϯϯ Ϭ͕ϯϯ ͲϬ͕Ϭϰ ͲϬ͕Ϭϴ ͲϬ͕ϬϮ ͲϬ͕ϬϮ
EU RMBS Ϭ͕ϱϱ Ϭ͕ϰϲ ϭ Ϭ͕ϴϴ Ϭ͕ϰϴ Ϭ͕ϱϱ Ϭ͕ϲϭ Ϭ͕ϳϰ Ϭ͕ϰϮ Ϭ͕Ϭϱ Ϭ͕Ϭϱ Ϭ͕ϭϮ Ϭ͕Ϭϯ
Spanish RMBS Ϭ͕ϰϴ Ϭ͕ϰϭ Ϭ͕ϴϴ ϭ Ϭ͕ϰϯ Ϭ͕Ϯϴ Ϭ͕ϯϮ Ϭ͕ϰϱ Ϭ͕Ϯϴ Ϭ͕ϭϱ Ϭ͕ϭϮ Ϭ͕ϮϬ Ϭ͕ϭϯ
Spanish SME CLO Ϭ͕ϱϬ Ϭ͕Ϯϵ Ϭ͕ϰϴ Ϭ͕ϰϯ ϭ Ϭ͕ϭϬ Ϭ͕ϯϲ Ϭ͕ϯϰ Ϭ͕ϯϱ Ϭ͕Ϭϰ Ϭ͕ϬϬ Ϭ͕ϭϯ Ϭ͕Ϭϰ
Italian RMBS Ϭ͕ϭϰ Ϭ͕ϭϵ Ϭ͕ϱϱ Ϭ͕Ϯϴ Ϭ͕ϭϬ ϭ Ϭ͕ϱϳ Ϭ͕ϯϳ Ϭ͕Ϯϱ ͲϬ͕ϭϮ ͲϬ͕Ϭϴ ͲϬ͕Ϭϭ ͲϬ͕ϭϲ
Portuguese RMBS Ϭ͕ϰϴ Ϭ͕ϯϲ Ϭ͕ϲϭ Ϭ͕ϯϮ Ϭ͕ϯϲ Ϭ͕ϱϳ ϭ Ϭ͕ϰϵ Ϭ͕ϯϰ ͲϬ͕ϮϬ ͲϬ͕ϭϴ ͲϬ͕ϮϬ ͲϬ͕ϭϳ
Dutch RMBS Ϭ͕ϯϵ Ϭ͕ϯϯ Ϭ͕ϳϰ Ϭ͕ϰϱ Ϭ͕ϯϰ Ϭ͕ϯϳ Ϭ͕ϰϵ ϭ Ϭ͕ϱϬ ͲϬ͕Ϭϭ Ϭ͕Ϭϰ Ϭ͕Ϭϳ ͲϬ͕Ϭϯ
UK PRMBS Ϭ͕ϯϳ Ϭ͕ϯϯ Ϭ͕ϰϮ Ϭ͕Ϯϴ Ϭ͕ϯϱ Ϭ͕Ϯϱ Ϭ͕ϯϰ Ϭ͕ϱϬ ϭ ͲϬ͕Ϭϭ ͲϬ͕Ϭϭ Ϭ͕ϭϭ ͲϬ͕ϬϮ
France Ϭ͕Ϭϯ ͲϬ͕Ϭϰ Ϭ͕Ϭϱ Ϭ͕ϭϱ Ϭ͕Ϭϰ ͲϬ͕ϭϮ ͲϬ͕ϮϬ ͲϬ͕Ϭϭ ͲϬ͕Ϭϭ ϭ Ϭ͕ϵϰ Ϭ͕ϴϱ Ϭ͕ϵϴ
Spain ͲϬ͕Ϭϱ ͲϬ͕Ϭϴ Ϭ͕Ϭϱ Ϭ͕ϭϮ Ϭ͕ϬϬ ͲϬ͕Ϭϴ ͲϬ͕ϭϴ Ϭ͕Ϭϰ ͲϬ͕Ϭϭ Ϭ͕ϵϰ ϭ Ϭ͕ϴϳ Ϭ͕ϵϭ
Italy Ϭ͕Ϭϯ ͲϬ͕ϬϮ Ϭ͕ϭϮ Ϭ͕ϮϬ Ϭ͕ϭϯ ͲϬ͕Ϭϭ ͲϬ͕ϮϬ Ϭ͕Ϭϳ Ϭ͕ϭϭ Ϭ͕ϴϱ Ϭ͕ϴϳ ϭ Ϭ͕ϳϳ
Germany Ϭ͕Ϭϰ ͲϬ͕ϬϮ Ϭ͕Ϭϯ Ϭ͕ϭϯ Ϭ͕Ϭϰ ͲϬ͕ϭϲ ͲϬ͕ϭϳ ͲϬ͕Ϭϯ ͲϬ͕ϬϮ Ϭ͕ϵϴ Ϭ͕ϵϭ Ϭ͕ϳϳ ϭ

16

220 Amundi Investment Strategy Collected Research Papers


During the sovereign debt crisis

BS
LO

RM
EC
BS
s
an

BS

BS
RM

SM

se

BS
lo

RM

RM
BS

BS

y
e
ĐŽƌƌĞůĂƚŝŽŶ

M
to

an
gu
sh

sh
CM

RM

ce
n

PR
Au

h
rt u

rm
i

ain
lia
an

an

tc

an

ly
ϮϬϭϬͲϮϬϭϯ

Du

UK

Ge
EU

EU

EU

It a

It a
Po
Sp

Sp

Sp
Fr
EU Auto loans ϭ Ϭ͕Ϯϱ Ϭ͕Ϭϳ Ϭ͕Ϭϯ Ϭ͕Ϭϳ ͲϬ͕Ϭϵ Ϭ͕Ϭϲ Ϭ͕ϭϳ Ϭ͕Ϯϰ ͲϬ͕Ϭϯ Ϭ͕ϬϮ Ϭ͕ϬϮ Ϭ͕ϬϬ
EU CMBS Ϭ͕Ϯϱ ϭ Ϭ͕ϯϯ Ϭ͕Ϯϱ Ϭ͕ϭϮ Ϭ͕ϮϮ Ϭ͕ϮϬ Ϭ͕Ϯϭ Ϭ͕ϯϳ Ϭ͕Ϭϴ Ϭ͕Ϭϯ Ϭ͕Ϭϵ ͲϬ͕Ϭϭ
EU RMBS Ϭ͕Ϭϳ Ϭ͕ϯϯ ϭ Ϭ͕ϴϲ Ϭ͕ϱϬ Ϭ͕ϲϮ Ϭ͕ϱϴ Ϭ͕ϯϰ Ϭ͕Ϯϵ Ϭ͕Ϭϰ ͲϬ͕Ϭϱ Ϭ͕ϭϬ ͲϬ͕Ϭϭ
Spanish RMBS Ϭ͕Ϭϯ Ϭ͕Ϯϱ Ϭ͕ϴϲ ϭ Ϭ͕ϰϲ Ϭ͕ϰϭ Ϭ͕ϯϲ Ϭ͕ϭϲ Ϭ͕ϭϴ Ϭ͕ϭϮ ͲϬ͕Ϭϭ Ϭ͕ϭϱ Ϭ͕Ϭϰ
Spanish SME CLO Ϭ͕Ϭϳ Ϭ͕ϭϮ Ϭ͕ϱϬ Ϭ͕ϰϲ ϭ Ϭ͕ϮϮ Ϭ͕Ϯϯ Ϭ͕Ϯϯ ͲϬ͕Ϭϱ Ϭ͕ϬϮ ͲϬ͕Ϭϵ ͲϬ͕ϬϮ Ϭ͕ϬϬ
Italian RMBS ͲϬ͕Ϭϵ Ϭ͕ϮϮ Ϭ͕ϲϮ Ϭ͕ϰϭ Ϭ͕ϮϮ ϭ Ϭ͕ϯϰ Ϭ͕ϭϲ Ϭ͕ϭϭ Ϭ͕Ϭϯ ͲϬ͕Ϭϭ Ϭ͕ϭϭ ͲϬ͕Ϭϯ
Portuguese RMBS Ϭ͕Ϭϲ Ϭ͕ϮϬ Ϭ͕ϱϴ Ϭ͕ϯϲ Ϭ͕Ϯϯ Ϭ͕ϯϰ ϭ Ϭ͕ϭϮ Ϭ͕ϭϴ ͲϬ͕ϭϴ ͲϬ͕ϭϱ ͲϬ͕Ϭϴ ͲϬ͕Ϭϰ
Dutch RMBS Ϭ͕ϭϳ Ϭ͕Ϯϭ Ϭ͕ϯϰ Ϭ͕ϭϲ Ϭ͕Ϯϯ Ϭ͕ϭϲ Ϭ͕ϭϮ ϭ Ϭ͕Ϯϳ Ϭ͕Ϭϰ ͲϬ͕Ϭϲ Ϭ͕Ϭϯ Ϭ͕Ϭϱ
UK PRMBS Ϭ͕Ϯϰ Ϭ͕ϯϳ Ϭ͕Ϯϵ Ϭ͕ϭϴ ͲϬ͕Ϭϱ Ϭ͕ϭϭ Ϭ͕ϭϴ Ϭ͕Ϯϳ ϭ Ϭ͕Ϭϰ Ϭ͕Ϭϳ Ϭ͕ϭϭ Ϭ͕ϬϮ
France ͲϬ͕Ϭϯ Ϭ͕Ϭϴ Ϭ͕Ϭϰ Ϭ͕ϭϮ Ϭ͕ϬϮ Ϭ͕Ϭϯ ͲϬ͕ϭϴ Ϭ͕Ϭϰ Ϭ͕Ϭϰ ϭ Ϭ͕ϭϴ Ϭ͕ϭϴ Ϭ͕ϳϲ
Spain Ϭ͕ϬϮ Ϭ͕Ϭϯ ͲϬ͕Ϭϱ ͲϬ͕Ϭϭ ͲϬ͕Ϭϵ ͲϬ͕Ϭϭ ͲϬ͕ϭϱ ͲϬ͕Ϭϲ Ϭ͕Ϭϳ Ϭ͕ϭϴ ϭ Ϭ͕ϳϵ ͲϬ͕ϭϯ
Italy Ϭ͕ϬϮ Ϭ͕Ϭϵ Ϭ͕ϭϬ Ϭ͕ϭϱ ͲϬ͕ϬϮ Ϭ͕ϭϭ ͲϬ͕Ϭϴ Ϭ͕Ϭϯ Ϭ͕ϭϭ Ϭ͕ϭϴ Ϭ͕ϳϵ ϭ ͲϬ͕ϮϬ
Germany Ϭ͕ϬϬ ͲϬ͕Ϭϭ ͲϬ͕Ϭϭ Ϭ͕Ϭϰ Ϭ͕ϬϬ ͲϬ͕Ϭϯ ͲϬ͕Ϭϰ Ϭ͕Ϭϱ Ϭ͕ϬϮ Ϭ͕ϳϲ ͲϬ͕ϭϯ ͲϬ͕ϮϬ ϭ

1. .

17

Amundi Investment Strategy Collected Research Papers 221


222 Amundi Investment Strategy Collected Research Papers
WP-046

Option Pricing under Skewness


and Kurtosis using
a Cornish Fisher Expansion

SOFIANE ABOURA,
Associate Professor – Paris-Dauphine University
DIDIER MAILLARD,
Professor - Cnam, Senior Advisor, Amundi

January 2015

This paper revisits the pricing of options, in a context of financial


stress, when the underlying asset’s returns displays skewness
and excess kurtosis. For that purpose, we use a Cornish- Fisher
transformation for valuing option contracts with an exact
formula allowing for heavytails.
An application to the FTSE 100 stock index option contracts
during October 2008 provides evidence about the capability of
the Cornish-Fisher model to improve calibration and pricing
performance during a period of stress.

Amundi Investment Strategy Collected Research Papers 223


1 Introduction
Financial markets have been subject to stress periods throughout their history. This raises
the question of whether option pricing models have the ability to derive fair contract prices
and risk measures in taking such potential stress conditions into account. Although the fi-
nancial literature on option theory has documented many well-known pervasive features that
affect pricing, these are not taken into account in the classic Black-Scholes-Merton frame-
work.

In order to remedy the assumption of a Gaussian marginal distribution for the underlying
asset returns in the classical Black-Scholes-Merton model, three principal approaches are
proposed in the literature: stochastic volatility models1 , jump-diffusion process for the price
dynamics2 , and stochastic volatility with a jump-diffusion process3 . There are also more
general non-Gaussian alternative classes for the underlying asset4 .

Alternatively, others have considered semi-parametric option pricing formulae, particularly


when it is not always possible to present the exact distribution in a tractable form; therefore,
much effort has been made to approximate the exact distribution. Jarrow and Rudd (1982)
model the distribution of stock price with a Edgeworth series expansion. Corrado and Su
(1996a) model the distribution of stock log prices with a Gram-Charlier series expansion,
while Corrado and Su (1996b) performed the same type of study with an Edgeworth expan-
sion5 . This method focuses on the skewness and kurtosis deviation from normality for stock
1
Scott (1987), Wiggins (1987), Johnson and Shanno (1987), Hull and White (1987, 1988), Stein and Stein
(1991), Heston (1993), Nandi (1998), Duan (1995), Heston and Nandi (2000), Hagan, Kumar, Lesniewski
and Woodward (2000).
2
Merton (1976), Bates (1988), Camara and Heston (2008).
3
Bates (1996), Bakshi, Cao and Chen (1997), Das and Sundaram (1999), Bates (2000), Duffie, Pan and
Singleton (2000), Pan (2002), Eraker, Johannes and Polson (2003), Eraker (2004), Duan, Ritchken and Sun
(2007).
4
Madan and Seneta (1990), Madan and Milne (1991), Madan Carr and Chang (1998), Bouchaud, Sornette
and Potters (1997), Barndorff-Nielsen (1998), Eberlein, Keller and Prause (1998), Aparicio and Hodges
(1998), De Jong and Huisman (2000), Matacz (2000), Corrado (2001), Savickas (2002), Borland (2002),
Carr, Geman, Madan and Yor (2002, 2003), Rockinger and Abadir (2003), Kleinert (2004), Dupoyet (2004),
Pochard and Bouchaud (2004), Borland and Bouchaud (2004), Carr and Wu (2004), Dutta and Babbel
(2005), Sherrick, Garcia and Tirupattur (1996), Albota and Tunaru (2005), Markose and Alentorn (2005),
Bakshi, Madan and Panayotov (2008), Aboura, Valeyre and Wagner (2014).
5
See Barton and Dennis (1952) or Stuart and Ord (1987) for discussion on the distinction between

224 Amundi Investment Strategy Collected Research Papers


returns. For clarity, it should be noted that, to correct the bias of the Black-Scholes (1973)
model, Corrado and Su (1996a) sum up the Black-Scholes formula with the adjustment terms
accounting for non-normal skewness and kurtosis by truncating the expansion after the fourth
moment. Under risk-neutral probability, they apply the Gram-Charlier density function to
derive European call price formula. The main advantage of Gram-Charlier and Edgeworth
expansions is that they allow for additional flexibility over a normal density because they
introduce a skewness and kurtosis parameter in the distribution. However, these approaches
have noticeable drawbacks. Rockinger and Jondeau (2001) note that, since Gram-Charlier
expansions are polynomial approximations, they have the important drawback of yielding
negative values for a probability. Actually, it is not guaranteed to be positive, and therefore
may violate the domain of validity of the probability distribution. This arises from the fact
that the expansions are usually truncated after the fourth power, which may imply negative
densities over some interval of their domain of variation (Leon, Mencia and Sentana (2007)),
thereby probabilities can be negative for such expansions. This is an undesirable outcome
because this situation might occur when the financial markets are in distress, which means
that these nearly Gaussian distributions may fail when they are needed most. Therefore,
imposing positivity constraints will require the Gram-Charlier approach to be used only for
weak departures from normality. Moreover, Blinnikov and Moessner (1998) observe that for
strongly non-Gaussian cases, Edgeworth expansion has a small domain of applicability in
practical cases, since it diverges like the Gram-Charlier series. Note that Gram-Charlier is
itself a particular density expansion of the Edgeworth expansions class.

Despite the large number of published works on the issue of pricing derivatives during stress
periods, the idea to consider a weakly non-Gaussian distributions with an exact formula that
allows for skewness and excess kurtosis, to our best knowledge, has been so far ignored. Here
we aim to address this issue by deriving a weakly-non Gaussian distribution to price option
contracts. We believe that pricing option contracts during financial stress periods induces
a trade-off between sophistication (to account for marketplace stylized facts) and simplicity
(to avoid any type of risk arising from the model). Most of the aforementioned papers have
probably underestimated this trade-off, either because the model is too complex and risky
Edgeworth and Gram-Charlier expansions.

Amundi Investment Strategy Collected Research Papers 225


for the business purpose or it includes approximations that might lead to errors. We argue
that the Cornish-Fisher expansion addresses this trade-off by considering risk neutral non-
normal third and fourth moments without any approximation. Actually, this expansion is
a way to transform a standard Gaussian random variable into a non-Gaussian random vari-
able. The transformation needs to be bijective so that the quantiles of the distribution are
conserved. However, computing the moments of the distribution resulting from the Cornish-
Fisher transformation, although simple in theory, is difficult in practice.

Fortunately, the use of Cornish-Fisher expansion helps with avoiding two main typical pit-
falls in these series expansions. First and most important, the area of the domain of validity,
expressed in actual skewness and kurtosis, seems to give sufficient room for manoeuvre in
most circumstances, which is not necessarily the case for other expansions. This is a major
point for avoiding the case of negative probabilities, typical for higher order expansions,
when pricing far-from the money option contracts. What can be shown is that the domain
of validity of the Cornish-Fisher is much wider than in the Gram-Charlier case. Second, the
transformation gives a simple relation between the skewness and kurtosis parameters and
the Value-at-Risk or Expected Shortfall measures.
Finally, our formula provides simple and flexible ways to fit a large variety of skewness and
kurtosis data. Indeed, it is written in a simple form, following Corrado and Su (1996a,b),
where the third and fourth moments are additional parameters that correct the Black-
Scholes-Merton formula. This simplifies the costly numerical analysis, as the model is easy
to estimate but also simple to interpret.

This research is relevant for three principal reasons. First, it addresses the issue of option
pricing with a weakly non-Gaussian model that accounts for explicit skewness and kurtosis
having a large domain of validity. Second, it addresses the issue of option pricing in a context
of financial stress. To our best knowledge, no existing studies compare their model’s per-
formance during stress periods using intra-day data and short-term out-the-money options,
which are the major sources of mispricing. In contrast, our database is mainly composed by
short-term out-the-money contracts. Third, our model compares the model’s performance
to a well-known trading rule, that is used as a benchmark, while the vast majority of the
existing studies do comparisons with the Black-Scholes (1973) model or its extensions. This

226 Amundi Investment Strategy Collected Research Papers


well known trading rule, called ’sticky strike’, was popularized by Derman (1999) to manage
the smile dynamics. The sticky strike rule models the volatility as remaining constant to
a given strike, whatever the underlying asset moves up or down instantaneously. Ciliberti,
Bouchaud and Potters (2008) show that the sticky strike rule is exact for small maturities.
Surprisingly, there are few studies using this rule for option pricing (Hagan, et al (2002),
Daglish, Hull and Suo (2002) and Ciliberti, Bouchaud and Potters (2008)).

The main contribution of this paper is to derive a weakly non-Gaussian European-style op-
tion pricing model, allowing for explicit third and fourth moments estimated implicitly from
the derivatives market. To our knowledge, no research paper has addressed the issue of
pricing option with a Cornish-Fisher transformation.

The paper is organized as follows: Section 2 displays the characteristics of the novel model;
Section 3 exposes the empirical results; and Section 4 summarizes and concludes.

2 The Cornish-Fisher option pricing model


2.1 The Cornish-Fisher transformation
The Cornish-Fisher expansion, if properly used (Maillard (2012)), allows the generation of
distributions with the desired volatility, skewness and kurtosis. The Cornish-Fisher expan-
sion relies on the polynomial transformation of a normal standard distribution z into a
distribution Z:

s k s
Z = z + (z 2 − 1) × + (z 3 − 3z) × − (2z 3 − 5z) × (1)
6 24 36
s and k are parameters which determine skewness and kurtosis, but except for very low
values, they do not coincide with effective skewness s∗ and kurtosis k∗ . The parameters will
be computed to achieve the effective skewness and kurtosis:

M3
s∗ = (2)
M21.5

Amundi Investment Strategy Collected Research Papers 227


M4
k∗ = −3 (3)
M22

We compute the moments of the Cornish-Fisher distribution:

M1 = 0
M2 = 1 + 961 k2 + 1296
25 4
s − 361 ks2
76 3 85 5
M3 = s − 216 s + 1296 s + 14 ks − 144
13
ks3 + 321 k2 s (4)
M4 = 3 + k + 167 k2 + 323 k3 + 3072
31 4
k − 2167 4
s − 48625 6 21665 8
s + 559872 s
7 2 113 4 5155 6 7 2 2 2455 2 4 65 3 2
− 12 ks + 452 ks − 46656 ks − 24 k s + 20736 k s − 1152 k s

As Z is non-standard (zero mean but variance slightly different from one), we will use the
transformation leading to a new definition of Z by retaining:

z + (z − 1) × 6s + (z 3 − 3z) × 24k − (2z 3 − 5z) × 36s


Z=  (5)
1 + 961 k2 + 1296
25 4
s − 361 ks2

Obviously, this new definition has the same higher moments than the initial transformation.

Recall that both the Cornish-Fisher and Gram-Charlier expansions are means of transform-
ing a Gaussian distribution into a non-Gaussian distribution, the skewness and the kurtosis of
which can be controlled if the transformations are properly implemented. Their expansions
differ in that the Cornish-Fisher is a transformation of quantiles, whereas the Gram-Charlier
is a transformation of a probability density. Both transformations must be implemented
with care, as their domain of validity does not cover the whole range of possible skewness
and kurtosis coefficients. Figure 1 exhibits the domain of validity of both expansions in the
skewness-kurtosis plane. What is most appealing, in the Cornish-Fisher expansion, is that
its domain of validity is much wider than in the Gram-Charlier case. This characteristic
reveals, unambiguously, how much more realistic the Cornish-Fisher expansion is, as com-
pared to the Gram-Charlier expansion; judging by the magnitude of the skewness-kurtosis
domain, which potentially encapsulates most of the tradable assets.

228 Amundi Investment Strategy Collected Research Papers


Figure 1: Domain of validity of the Cornish-Fisher and Gram-Charlier models

Amundi Investment Strategy Collected Research Papers 229


2.2 The option pricing model
The Cornish-Fisher expansion is a simple transformation of quantiles, but the corresponding
probability density is rather complicated. There is currently no way to compute the integrals
analytically, despite numerous approaches. We therefore decided to proceed with necessary
numerical computations to compute the quantiles. Our computations give the Cornish-Fisher
transformation’s density of probability:

z2
1 e− 2
Φ(Z) = √  k s2  2 (6)
2π z 2 8 − 6 + z 3s + 1 − k8 + 5s36

with:
 √  √
3 −q+ Z + (q− Z )2 + 4 p3 3 −q+ Z − (q− Z )2 + 4 p3
z = a3 + b
2
b 27
+ b
2
b 27

a = k −ss2
8
−3
k s2
b= 24
− 18
2 s2
1− k8 + 5s
p= k 2
36
− 13 36
2
−s
24 18
( 24 − s18 )2
k

k 5s2 s3
−s s(1− 8 + 36 )
q= k s2
− 181 k 2 − 272 216
2
8
−3 ( 24 − s18 )2 k
( 24 − s18 )3

We use the weakly non-Gaussian probability density Φ(.), derived from the Cornish-Fisher
expansion, to control for skewness and excess kurtosis. Φ(Z) is the density evaluated on the
random variable Z. Assuming risk-neutrality, we derive theoretical European call option for-
mula as the value of the expected payoff, discounted by the risk-free rate rf at the contract’s
time to maturity (T − t). For a call option CCF with strike K and underlying asset price ST
at maturity T , the value will be given by:
 +∞
CCF = e−rf (T −t) (ST − K)+ Φ(Z)d(Z) (7)
−∞

If the skewness and the excess kurtosis are zero, this model reduces to the Black-Scholes
(1973) model with CCF = CBS when s∗ = 0 and k∗ = 0.

To illustrate the model application, we attempt to reproduce smile curves as a function of

230 Amundi Investment Strategy Collected Research Papers


kurtosis and skewness. The model allows to compute an option price depending on maturity,
risk-free interest rate, volatility and skewness and kurtosis. To do so, we infer the implied
volatility for various levels of strike prices. The computations are made for a call option with
one-month maturity, a 2% risk-free rate and a 20% volatility.

Figure 2 reproduces volatility smiles according to various levels of kurtosis and skewness.
First, the deeper is the kurtosis, the higher is the volatility curvature. Second, the impact
of skewness leads to a rotation in the smile where a negative skewness induces a negative
slope.

Amundi Investment Strategy Collected Research Papers 231


Figure 2: Volatility smiles according to kurtosis and skewness

232 Amundi Investment Strategy Collected Research Papers


3 The empirical results
3.1 The dataset
The data set6 consists of European option contracts on the FTSE 100 stock index during
October 2008. The data set contains 364 intra-daily observations from October 1 to October
31. Bid-ask spreads are used in the study for quality measure. This month is critical since
it includes the highest volatility peaks of the 2008 financial crisis. It therefore allows for
testing our model on one of the most financially distressed periods, contrary to the approach
of the vast majority of research papers. Three types of option contracts are discarded. First,
only the two shortest maturities are used due to computational time consideration, with a
maximum time horizon of three months. The risk-free interest rates are the two week Libor
rates. Second, only near-the-money and out-the-money prices are used, since out-the-money
contracts are more liquid than in-the-money contracts. Third, in order to have only signifi-
cant option prices, we retain at each strike, the lowest spread identified from 16:00 to 16:04.
Figure 3 plots the volatility smile as a function of strike price K during October 2008.

6
We would like to thank Sebastien Valeyre for having provided us with this data set that comes from
Liffe-Nyse-Euronext.

Amundi Investment Strategy Collected Research Papers 233


Figure 3: Volatility smiles in October 2008

3.2 The model estimation


We consider a minimizing procedure for the model estimation. The first step corresponds to
the in-the-sample model calibration, while the second step corresponds to the out-the-sample
valuation. For the in-the-sample procedure, the model is calibrated on the intra-daily data
by minimizing the following sum of squared errors:
N 
t
min e2K,t (8)
(σ,s∗ ,k∗ )
K=1

With eK,t representing the normalized difference ( CMarket


St
−CCF
) between the market call op-
tion prices CMarket and the theoretical Cornish-Fisher call option prices CCF for the K th
strike price with K = 1, ..., NT and the underlying stock index price St at date t. The
estimation procedure is implemented on the set of parameters (σ, s∗ , k∗ ) that is estimated
implicitly by the quadratic minimization procedure, yielding the implied volatility, the im-
plied skewness, and the implied kurtosis, respectively.

234 Amundi Investment Strategy Collected Research Papers


To assess the differences between market prices and model prices for an out-the-sample fit,
we compute the mean price forecast error to quantify the error magnitude.


Nt
e2K,t (9)
K=1

With eK,t again representing the normalized difference between the market call option prices
and the call option prices computed by the Cornish-Fisher model for the K th strike price
with K = 1, ..., NT and the underlying stock index price St at date t. The set of esti-
mated parameters (σ, s∗ , k∗ ) are computed from the in-the-sample calibration and remain
constant for the one-day ahead out-the-sample pricing; this assumption is reasonable since
these parameters are relatively stable for short horizons. Hence, during the out-the-sample
procedure, we re-compute option prices of the current day using the previous day’s implied
volatility, implied skewness, and implied kurtosis; the interest rate r is set constant for the
period. Therefore, only the underlying stock index price St and the time to maturity (T − t)
change.

3.3 The empirical findings


Table 1 summarizes the model calibration and pricing on FTSE 100 call options for the
month of October 2008. The choice of this month is motivated by the highest concentra-
tion of volatility peaks during the 2008 financial crisis. The option mispricing overall affects
short-term contracts and out-the-money contracts. For that reason, it is relevant to test the
Cornish-Fisher model on the October 2008 month options, which consist of short-term (less
then 3 months) and generally deep-out-the money contracts (average moneyness defined by
(S/K) is 0.90). Therefore, we posit that the Cornish-Fisher model is able to fairly price the
option contracts characterized by underlying asset prices that are weakly non-Gaussian.

Amundi Investment Strategy Collected Research Papers 235


Table 1: Cornish-Fischer Model Calibration and Pricing
The in-the-sample fit (see column "Calibration") shows that the sum of squared errors made
by the Cornish-Fisher model (49.42%) is only half of the errors made by the sticky strike
model. This signifies that the in-the-sample fit brought by the weakly non-Gaussian model
adheres fairly well to the market data during this turbulent period. The empirical results
reveal that the Cornish-Fisher average implied volatility is 55.14%, which is 20% higher than
the sticky strike implied volatility for the same period. The Cornish-Fisher implied volatil-
ity ranges from 32.75% to 71.89%, which represents a variation of as much as 100% in only
eleven trading days. The volatility peak of 71.89% occurred on October 17, after a stock
index decline from 4,327.30 (October 14) to 3,824.33 (October 16), followed by an increase
to 4,204.29 (October 20).

The Cornish-Fisher average implied skewness s∗ is -1.72 (ranging from -2.42 to -0.37), while
the average implied kurtosis k∗ is 7.80 (ranging from 1.23 to 16.64). Under these circum-
stances, a larger domain of validity is paramount since it allows the model to capture strong
market swings that affect these three structural parameters (σ, s, k) within a short interval
of time.

236 Amundi Investment Strategy Collected Research Papers


Figure 4: Cornish-Fischer model calibration

Figure 4 illustrates the ranges of values for the three parameters estimated during October
2008. Given the volatile nature of this month, the implied skewness and kurtosis may appear
relatively stable in general.

To assess the out-of-sample forecasting (see column "Pricing") performance of the Cornish-
Fisher model, we set constant the implied volatility σt−1 , implied skewness st−1 , and implied
kurtosis kt−1 calibrated from the previous trading day t−1, in order to price option contracts
for the day t. It appears that the sum of squared errors made by the Cornish-Fisher model
represent two thirds (76.59%) of the error made by the sticky strike model.

Overall, we conclude that the Cornish-Fisher model has improved calibration and pricing
accuracy during the most volatile month of the 2008 financial crisis. This can be explained
that this weakly non-Gaussian model has a larger domain of validity than comparable models
(Gram-Charlier, Edgeworth etc.) and is easier to estimate and more stable than the stochas-
tic volatility models. The day of October 17 represents the biggest stock index price swing of
the period, characterized by the highest volatility (71.89%), the highest kurtosis (16.64), the

Amundi Investment Strategy Collected Research Papers 237


second highest negative skewness (-2.40), and the highest in-the-sample fit error (5.0e-04);
the highest out-the-sample fit error (9.3e-03) occurred immediately prior, on October 15.
The model’s capability to adhere to the data is fairly good, although it has been disturbed
by the sudden regime change of the underlying stock index from October 15 to October 20.
Note that a possible improvement would have been to extend the empirical test to a larger
set of data.

4 Conclusion
This paper derives a new option pricing model based on a weakly non-Gaussian risk-neutral
probability density. This density relies on a Cornish-Fisher transformation with an exact
formula allowing for heavy-tails in the presence of non-normal skewness and kurtosis. An
in and out-the-sample analysis is carried out on intra-day data from the FTSE 100 stock
index options during October 2008, which was the most volatile month of the 2008 financial
crisis. We conclude that the Cornish-Fisher model has improved calibration and pricing
performance, as comparison to the sticky strike model. This improvement is due to its larger
domain of validity. It is worth emphasizing that this weakly non-Gaussian model is eas-
ier to implement than the stochastic volatility models or other alternative classes of stable
non-Gaussian models. This work can be applied to large data sets or extended to risk man-
agement measures, such as VaR and Expected Shortfall.

References
[1] Bakshi, G., C., Cao, and Z., Chen, 1997, Empirical performance of alternative option pricing models,
Journal of Finance, 52, 2003-2049.

[2] Barton, D.E., and K.E., Dennis, 1952, The conditions under which Gram-Charlier and Edgeworth
curves are positive definite and unimodal, Biometrika, 39, 425-427.

238 Amundi Investment Strategy Collected Research Papers


[3] Black F., and M., Scholes, 1973, The pricing of options and corporate liabilities, Journal of Political
Economy, 81, 637-659.

[4] Blinnikov, S., and R., Moessner, 1998, Expansions for nearly Gaussian distributions, Astronomy and
Astrophysics, 130, 193-205.

[5] Borland, L., 2002, A theory of non-Gaussian option pricing, Quantitative Finance, 2, 415-431.

[6] Borland, L., and J.P., Bouchaud, 2004, A non-Gaussian option pricing model with skew, Quantitative
Finance, 4, 499-514.

[7] Bouchaud, J-P., D., Sornette, and M., Potters, 1997, Option pricing in the presence of extreme fluctu-
ations, Mathematics of Derivative, Cambridge University Press, 112-125.

[8] Camara, A., and S., Heston, 2008, Closed-Form option pricing formulas with extreme events, Journal
of Futures Markets, 28, 213-230.

[9] Carr, P., H., Geman, D., Madan, and M., Yor, 2003, Stochastic volatility for Levy processes, Mathe-
matical Finance, 13, 345-382.

[10] Ciliberti, S., J-P. Bouchaud, and M., Potters, 2009, Smile dynamics, a theory of the implied leverage
effect, Wilmott Journal, 1, 87-94.

[11] Cornish, E., and R., Fisher, 1937, Moments and cumulants in the specification of distributions, Revue
de l’Institut International de Statistiques 5, 307-320

[12] Corrado, C., and T., Su, 1996a, Skewness and kurtosis in S&P 500 index returns implied by option
prices, Journal of Financial Research, 19, 175-92.

[13] Corrado, C., and T., Su, 1996b, S&P 500 index option tests of Jarrow and Rudd’s approximate option
valuation formula, Journal of Futures Markets, 16, 611-30.

[14] Daglish, T., J., Hull, and W., Suo., 2007, Volatility surfaces: theory, rules of thumb and empirical
evidence, Quantitative Finance, 7, 507-524.

[15] Derman, E., and I., Kani, 1994, Riding on a smile, Risk, 7, 32-39.

[16] Derman, E., 1999, Regimes of volatility, Risk, 55-59.

Amundi Investment Strategy Collected Research Papers 239


[17] Duan, J.C., 1995, The GARCH option pricing model, Mathematical Finance, 5, 13-32.

[18] Duffie, D., J., Pan, and K., Singleton, 2000, Transform analysis and asset pricing for affine jump
diffusions, Econometrica, 68, 1343-1376.

[19] Dumas,B., J., Fleming, and R., Whaley, 1998, Implied volatility functions : empirical tests, The Journal
of Finance, 53, 2059-2105.

[20] Dupire, B., 1994, Pricing with a smile, Risk, 7, 18-20.

[21] Hagan, P., D., Kumar, A.S., Lesniewski, and D.E., Woodward, 2002, Managing smile risk, Wilmott
Magazine, 1, July, 84-108.

[22] Heston, S.L., 1993, A closed-form solution for options with stochastic volatility with applications to
bond and currency options, Review of Financial Studies, 6, 327-343.

[23] Hull, J., and A., White, 1987, The pricing of options on assets with stochastic volatility, Journal of
Finance, 42, 281-300.

[24] Kleinert, H., 2004, Option pricing for non-Gaussian price fluctuations, Physica A, 338, 151-269.

[25] Leon A., J., Mencia, and E., Sentana, 2007, Parametric properties of semi-nonparametric distributions
with applications to option valuation, working paper, University of Alicante

[26] Maillard, D., 2012, A User’s guide to the Cornish-Fisher expansion, SSRN Working Papers Series.

[27] Maillard, D., 2013, More on Cornish-Fisher: Distribution density and boundary conditions, SSRN
Working Papers Series.

[28] Markose, S., and A., Alentorn, 2005, Option pricing and the implied tail index with the Generalized
Extreme Value (GEV) distribution, working paper, University of ESSEX.

[29] Merton, R.C, 1973, Theory of rational option pricing, Bell Journal of Economics and Management
Science 4, 141-183.

[30] Merton, R.C, 1976, Option pricing when underlying stock returns are discontinuous, Journal of Finan-
cial Economics, 3, 125-144.

[31] Rockinger, M., and E., Jondeau, 2001, Gram-Charlier densities, Journal of Economic Dynamics and
Control, 25, 1457-1483.

240 Amundi Investment Strategy Collected Research Papers


WP-045

Modelling Tail Risk in a


Continuous Space
DIDIER MAILLARD,
Professor - Cnam, Senior Advisor – Amundi

December 2014

Non normal distributions are a fact of life. In the financial world, many
distributions display tail risk, i.e. (negative) skewness and excess kurtosis.
Being able to model such risk is useful in various and important fields: risk
measurement, fund management performance evaluation, asset pricing…
One way to model tail risk is to introduce discontinuities, such as jumps,
to describe the distribution of values or returns. It is however possible, and
often convenient, to model tail risk in a continuous space. Both Cornish-
Fisher and Gramm-Charlier expansions (which is the simple form of a
family of Edgeworth expansions) are means to transforming a Gaussian
distribution into a non-Gaussian distribution, the skewness and the kurtosis
of which can be controlled if the transformations are properly implemented.
This may be useful for modelling distributions for a wide range of issues,
especially in risk assessment and asset pricing. The expansions differ in
their nature: Cornish-Fisher is a transformation of a random variable, or of
quantiles, meanwhile Gramm-Charlier is a transformation of a probability
density. Both transformations must be implemented with care, as their
domain of validity does not cover the whole range of possible skewnesses
and kurtosis. It appears that the domain of validity of Cornish-Fisher is
much wider that the domain of validity of Gramm-Charlier. This, and the
fact that Cornish-Fisher provides easily the quantiles of the distribution,
gives it an advantage over Gramm-Charlier in several configurations.

Amundi Investment Strategy Collected Research Papers 241


1 – Introduction

The Cornish Fisher expansion is a way of transforming a Gaussian distribution into a non-
Gaussian distribution, the skewness and the kurtosis of which can be controlled if the
transformation is properly implemented (Maillard, 2012). The Gramm Charlier expansion is
also a way of transforming a Gaussian distribution into a non-Gaussian one, with the desired
skewness and kurtosis.

Those expansions may prove very useful to model uncertain variables or events which
obviously are not normally distributed. In the field of finance, one observes that return or
changes in asset prices distributions display (generally negative) skewness and (generally
positive) excess kurtosis. These moments should be taken into account when risk is assessed,
and in asset pricing.

Gramm-Charlier has been used in option pricing, for instance in a seminal paper by Corrado
& Sue (1996). Cornish-Fisher has been used in several papers considering the risk of asset
returns: see for example Cao & alii (2010), or Fabozzi & alii (2012). Maillard (2013b) uses
Cornish-Fisher to estimate the cost of tail risk in a managed portfolio according to a
manipulation-proof performance measure. Aboura & Maillard (2014) use Cornish-Fisher for
the purpose of option pricing.

This paper attempts to assess the respective merits of both transformations.

2 – Nature of the expansions

Cornish-Fisher and Gramm-Charlier expansions are not about the same object. The Cornish-
Fisher expansion is a transformation of a standard Gaussian random variable z into a non-
Gaussian variable Z, such that:

S K S2
(CF ) Z z  ( z 2  1)  ( z 3  3z )  (2 z 3  5 z )
6 24 36

Where S and K are parameters tied to skewness and kurtosis respectively.

242 Amundi Investment Strategy Collected Research Papers


The Gramm-Charlier expansion transforms a standard Gaussian probability density ij into a
non-Gaussian probability density ĭ, such that:

ª S K º ª S K º
(GC ) ) ( z ) M ( z ) «1  H 3 ( z )  H 4 ( z )» M ( z ) «1  ( z 3  3 z )  ( z 4  6 z 2  3)»
¬ 6 24 ¼ ¬ 6 24 ¼
z2
1 
M ( z) e 2

2S

Hn is the Hermite polynomial of order n

d n  x2 / 2
(1) n e x
2
/2
H n ( z) e
dx n

Note that the term of « expansion » is related to the fact that the probability laws derived from
the Cornish-Fisher or Gramm-Charlier expansions are « approximations » of any probability
law displaying the same four moments. But they are not in themselves approximate
probability laws, provided they lie within a certain domain of validity.

3 – Moments

Moments are easy to compute in the case of Gramm-Charlier (see for example Jondeau and
Rockinger (2001), or Appendix).

It ensues that volatility is unitary and that skewness and excess kurtosis 1 are equal to the S
and K parameters respectively:

(GC )
V 1
Sˆ S
Kˆ K

1
We mean by excess kurtosis the difference between kurtosis and 3, which is the kurtosis of a Gaussian
distribution.

Amundi Investment Strategy Collected Research Papers 243


The computation of moments for Cornish-Fisher is rather more complex, and has been done
by Maillard (2012).

(CF )
1 2 25 4 1
V 1 K  S  KS 2
96 1296 36
76 3 85 5 1 13 1
S S  S  KS  KS 3  K 2 S
Sˆ 216 1296 4 144 32
1.5
§ 1 2 25 4 1 ·
¨1  K  S  KS 2 ¸
© 96 1296 36 ¹
ª 7 2 3 3 31 7 4 25 6 21665 8 7 2º
«3  K  16 K  32 K  3072 K  216 S  486 S  559872 S  12 KS »
4

« »
« 113 KS 4  5155 KS 6  7 K 2 S 2  2455 K 2 S 4  65 K 3 S 2 »
«¬ 452 46656 24 20736 1152 »¼
Kˆ 2
3
§ 1 2 25 4 1 2·
¨ 1  K  S  KS ¸
© 96 1296 36 ¹

Except for very small (absolute) values for S and K, volatility differs (slightly) from 1, and
skewness and kurtosis differ (sometimes hugely) from the skewness and kurtosis parameters.

It is thus possible to build distributions with the desired four first moments with both
expansions. Choosing the parameters is straightforward in the Gramm-Charlier case. It is
somewhat more arduous in the Cornish-Fisher case: one has to compute the S and K
parameters by reversing the two expressions giving the actual skewness and kurtosis (which
must be done numerically), and then correct the Cornish-Fisher expansion by dividing by the
value of volatility, to obtain a random variable with unitary variance and the desired skewness
and volatility.

4 – Probability densities

By definition, the probability density corresponding to Gramm-Charlier is given by the


expansion:

244 Amundi Investment Strategy Collected Research Papers


ª S K º ª S K º
(GC ) ) ( z ) M ( z ) «1  H 3 ( z )  H 4 ( z )» M ( z ) «1  ( z 3  3 z )  ( z 4  6 z 2  3)»
¬ 6 24 ¼ ¬ 6 24 ¼

The probability density corresponding to Cornish-Fisher may be computed from the definition
of the random variable (see Maillard (2013a)), but its expression is somewhat more complex:

M ( z)
(CF ) ) ( Z )
§K S · 2
S K 5S 2
z 2 ¨¨  ¸¸  z  1  
©8 6 ¹ 3 8 36

with:

 q ' Z / a 3  q'Z / a 3 2  4 3
p  q ' Z / a 3  q'Z / a 3 2 4 3
p
3 27 3 27
z ] SK ( Z ) a ' 0 / 3  
2 2
s K S
a 3 k  2s a' 0
2
k s
k  2s 2 24 6
1  3k  5s 2 1 s2 s 1 s (1  3k  5s 2 ) 2 s3
p  q'  
k  2s 2

3 k  2s 2
2
k  2s 2

3 k  2s 2 2
27 k  2s 2
3

Though complex, this formula may be encapsulated into a single spreadsheet cell.

It is possible to plot the density function and its deformation according to kurtosis (see
Maillard (2013a) for more details), for the whole distribution.

Amundi Investment Strategy Collected Research Papers 245


and for the tail of the distribution (here the right-hand tail),

The impact of skewness (here positive skewness) is less visible on the whole distribution,

246 Amundi Investment Strategy Collected Research Papers


It appears however clearly on the tail (here the right hand tail):

Those probability densities are useful for computing numerically integrals, for instance in
issues of option pricing, and this seems to confer Gramm-Charlier an advantage of simplicity.
However, such integrals may also be computed using the quantiles of the distribution (by

Amundi Investment Strategy Collected Research Papers 247


and for the tail of the distribution (here the right-hand tail),

The impact of skewness (here positive skewness) is less visible on the whole distribution,

248 Amundi Investment Strategy Collected Research Papers


It appears however clearly on the tail (here the right hand tail):

Those probability densities are useful for computing numerically integrals, for instance in
issues of option pricing, and this seems to confer Gramm-Charlier an advantage of simplicity.
However, such integrals may also be computed using the quantiles of the distribution (by

Amundi Investment Strategy Collected Research Papers 249


equipondering their values). In that case, the advantage goes to Cornish-Fisher, which gives
an immediate value of the quantiles. In Gramm-Charlier case, there is no simple expression of
the quantiles.

5 – Domain of validity

This is a very important point. Any system of probability should present two features:
- Non-negativity: any possible event should have a probability equal or superior to zero,
- Unitary sum: the probabilities of all possible events should add to one.

For the Gramm-Charlier expansion, which is expressed in terms of a probability density, it


means that:

ª S K º ª S K º
(GC ) ) ( z ) M ( z ) «1  H 3 ( z )  H 4 ( z )» M ( z ) «1  ( z 3  3 z )  ( z 4  6 z 2  3)» t 0 z
¬ 6 24 ¼ ¬ 6 24 ¼
f

³ )( z )dz
f
1

It is easy to show (see Appendix, moment M0) that the second condition is fulfilled. As for the
first one, a 4th-order polynomial has to be always positive. The condition therefore has been
studied in particular by Jondeau & Rockinger (2001).

They prove that the boundary of the domain of validity has a parametric definition given by:
H 3 ( z)
Sˆ ( z ) 24
z 6  3z 4  9 z 2  9
H 2 ( z)
Kˆ ( z ) 72 6
z  3z 4  9 z 2  9

It is possible to plot this boundary in the skewness/kurtosis plane. The domain of validity is
the inner part of the boundary.

250 Amundi Investment Strategy Collected Research Papers


Chart 1

It appears that kurtosis cannot exceed 4, and skewness 1.05 in absolute value. The domain of
validity is thus quite limited. It is not rare to observe in returns distribution kurtosis in excess
of 4 and skewnesses in excess of 1 in absolute terms 2.

For Cornish-Fisher, the probability density adds to 1 by definition. The non-negativity


condition is equivalent to the monotonicity of the transformation of the quantiles, i.e. that (the
positive sign resulting from the fact that Z is positive for large positive values of z):

2
Gramm-Charlier is the simplest (lowest polynomial order) of a family of density expansions known as the
Edgeworth expansions. The higher order expansions are also subject to nonnegativity problems. In addition, their
sum does not necessarily equal 1.

Amundi Investment Strategy Collected Research Papers 251


dZ
! 0 z
dz

The condition for that to hold is that S and K are subject to the following inequality:

S2 § K S 2 ·§ K 5S 2 ·
 4¨¨  ¸¨1   ¸d0
9 ©8 6 ¸¹¨© 8 36 ¸¹

This result is known for a long time. It is interesting to rewrite it (see Maillard (2012), with a
presentation of the derivation) as:

K S
9k 2  (3  33s 2 )k  30 s 4  7 s 2 d 0 k s
24 6
1  11s 2  s 4  6 s 2  1 1  11s 2  s 4  6 s 2  1
dkd
6 6

It implies that S cannot exceed 2.485 (for the square root to be real), and it gives an equation
of the boundary in the (S,K) plane.

252 Amundi Investment Strategy Collected Research Papers


Chart 2

Domain of validity of the CF expansion

14

12

10
Kurtosis parameter K

8
K'
K"
6

0
-2,5 -2 -1,5 -1 -0,5 0 0,5 1 1,5 2 2,5
Skew ness param eter S

However, one should remember at this stage that the skewness and kurtosis parameters are not
the actual skewness and kurtosis. The equation of the boundary in the ( Ŝ , K̂ ) plane could be
obtained by reversing the relationship but it is not easily tractable.

However, one can obtain a parametric representation of the boundary using S as a parameter.
S leads to Ŝ , and to K̂ through K.

This gives in Chart 3 the boundary of the domain of validity of the Cornish-Fisher
transformation.

Amundi Investment Strategy Collected Research Papers 253


Chart 3

The domain of validity is much wider in the Cornish-Fisher case, as may be seen in chart 4.

254 Amundi Investment Strategy Collected Research Papers


Chart 4

There have been proposals to extend the domain of validity of the transformation by
“rectifying” them: see in the case of Gramm-Charlier Jondeau & Rockinger (2001), or more
generally Chernozhukov & alii (2007). Their aim is to correct the breaches of non-negativity.
Those rectifications lead to new distribution laws, which may not be as parsimonious in their
implementation as Cornish-Fisher or Gramm-Charlier.

6 – Links with VaR and CvaR

Contrarily to Gramm-Charlier, Cornish-Fisher provides a simple expression of the quantiles


of the distribution. It is therefore convenient to compute easily values at risk (VaR), which are
tied to a quantile in the unfavourable part of the distribution (VaR is volatility times minus the

Amundi Investment Strategy Collected Research Papers 255


quantile times expected value minus present value). Omitting those constants, VaR at
threshold 1-a may be written as:

VaR1D ZD

It may be shown easily (see Maillard (2012)), using a Cornish-Fisher expansion, that:

S S2 K
VaR1D vD  (1  vD2 )  (5vD  2vD3 )  (vD3  3vD )
6 36 24

where vD  zD N 1 (D ) is value-at-risk in the Gaussian case.

It is also easy to obtain another, more consistent, measure of risk, the conditional value at risk
(CVaR)

ª S S2 Kº
CVaR1D yD «1  vD  (1  2vD2 )  (1  vD2 ) »
¬ 6 36 24 ¼
zD2
1 1 
where yD e 2 is the conditional value-at-risk in the Gaussian case.
D 2S

Cornish-Fisher provides thus a simple method for correcting risk measures that would prevail
in a Gaussian situation for skewness and kurtosis.

A caveat: The expressions of VaR and CVaR depend on the skewness and kurtosis
parameters. Those should be obtained by reversing the two expressions giving the actual
skewness and kurtosis as a function of the skewness and kurtosis parameters.

256 Amundi Investment Strategy Collected Research Papers


7 – Conclusions

Due to its much wider domain of validity, Cornish-Fisher should be preferred in most cases. It
has also the advantage of giving a simple expression of the quantiles, which may be quite
useful in numerical simulations.

Amundi Investment Strategy Collected Research Papers 257


References

Aboura, Sofiane and Didier Maillard, 2014, “Option Pricing under Skewness and Kurtosis
using a Cornish Fisher Expansion”, Working Paper

Cao, Zhiguang, Richard D.F. Harris and Jian Shen, 2010, “Hedging and Value at Risk: A
Semi-Parametric Approach”, Journal of Future Markets 30(8), 780-794

Chernozhukov, Victor, Ivan Fernandez-Val and Alfred Galichon, 2007, “Rearranging


Edgeworth-Cornish-Fisher Expansions, Working Paper 07-20, Working Paper Series
Massachussets Institute of Technology, Department of Economics, August 2007

Cornish, E., and R. Fisher, 1937, “Moments and Cumulants in the Specification of
Distributions”, Revue de l’Institut International de Statistiques 5, 307-320

Corrado, Charles J. and Tie Su, 1996, “Skewness and Kurtosis in S&P 500 Index Returns
implied by option prices”, The Journal of Financial Research, Vol. XIX, N° 2 pages 175-192

Fabozzi, Frank. J, Svetlovar T. Rachev and Stoyan V. Stoyanov, “Sensitivity of portfolio VaR
and Cvar to portfolio return characteristics”, Working Paper, Edhec Risk Institute, January
2012

Jondeau, Eric, and Michael Rockinger, 2001, “Gramm-Charlier densities”, Journal of


Economic Dynamics & Control, 25 (2001) pages 1457-1483

Leon, Angel, Javier Mencia and Enrique Santana, “Parametric Properties of Semi-
Nonparametric Distributions, with Applications to Option Valuation”, Journal of Business &
Economic Statistics, April 2009, Vol. 27, No. 2

Maillard, Didier, 2012, “A User’s Guide to the Cornish Fisher Expansion”, Working Papers
Series, SSRN n° 1997178, February 2012

258 Amundi Investment Strategy Collected Research Papers


Maillard, Didier, 2013a, “More on Cornish-Fisher: Distribution Density and Boundary
Conditions”, SSRN Working Paper n°2236338

Maillard, Didier, 2013b, “Manipulation-Proof Performance Measure and the Cost of Tail
Risk”, SSRN Working Paper n°2276050

Spiring, Fred, “The Refined Positive Definite and Unimodal Regions for the Gram-Charlier
and Edgeworth Series Expansion”, 2011, Advances in Decision Sciences, Research Paper
No 463097

Amundi Investment Strategy Collected Research Papers 259


Appendix

Gramm-Charlier moments

f f
ª S i 3 K º
³ z )( z)dz ³ «¬