TIME-SERIES

FORECASTING
c 2000 by Chapman & Hall/CRC
CHAPMAN & HALL/CRC
Boca Raton London New York Washington, D.C.
Chris Chatfield
Reader in Statistics
Department of Mathematical Sciences
University of Bath, UK
TIME-SERIES
FORECASTING
Library of Congress Cataloging-in-Publication Data
Chatfield, Christopher.
Time-series forecasting / Chris Chatfield.
p. cm.
Includes bibliographical references and index.
ISBN 1-58488-063-5 (alk. paper)
1. Time-series analysis. 2. Forecasting-Statistical methods. I.
Title.
QA280 .C42 2000
519.55dc21 00-050843
CIP
This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use.
Apart from any fair dealing for the purpose of research or private study, or criticism or review, as permitted
under the UK Copyright Designs and Patents Act, 1988, this publication may not be reproduced, stored
or transmitted, in any form or by any means, electronic or mechanical, including photocopying, micro-
filming, and recording, or by any information storage or retrieval system, without the prior permission
in writing of the publishers, or in the case of reprographic reproduction only in accordance with the
terms of the licenses issued by the Copyright Licensing Agency in the UK, or in accordance with the
terms of the license issued by the appropriate Reproduction Rights Organization outside the UK.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. Specific permission must be obtained in writing from CRC Press LLC
for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.
Trademark Notice: Product or corporate names may be trademarks or
used only for identification and explanation, without intent to infringe.
registered trademarks, and are
Visit the CRC Press Web site at www.crcpress.com
2000 by Chapman &C Hall/CRC
No claim to original U.S. Government works
International Standard Book Number 1-58488-063-5
Library of Congress Card Number 00-050843
Printed in the United States of America 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
To
Steve and Sarah, Rob and Lali
wishing you fulllment and contentment
c 2000 by Chapman & Hall/CRC
Contents
Preface
Abbreviations and Notation
1 Introduction
1.1 Types of forecasting method
1.2 Some preliminary questions
1.3 The dangers of extrapolation
1.4 Are forecasts genuinely out-of-sample?
1.5 Brief overview of relevant literature
2 Basics of Time-Series Analysis
2.1 Dierent types of time series
2.2 Objectives of time-series analysis
2.3 Simple descriptive techniques
2.4 Stationary stochastic processes
2.5 Some classes of univariate time-series model
2.6 The correlogram
3 Univariate Time-Series Modelling
3.1 ARIMA models and related topics
3.2 State space models
3.3 Growth curve models
3.4 Non-linear models
3.5 Time-series model building
4 Univariate Forecasting Methods
4.1 The prediction problem
4.2 Model-based forecasting
4.3 Ad hoc forecasting methods
4.4 Some interrelationships and combinations
5 Multivariate Forecasting Methods
5.1 Introduction
5.2 Single-equation models
5.3 Vector AR and ARMA models
c 2000 by Chapman & Hall/CRC
5.4 Cointegration
5.5 Econometric models
5.6 Other approaches
5.7 Some relationships between models
6 A Comparative Assessment of Forecasting Methods
6.1 Introduction
6.2 Criteria for choosing a forecasting method
6.3 Measuring forecast accuracy
6.4 Forecasting competitions and case studies
6.5 Choosing an appropriate forecasting method
6.6 Summary
7 Calculating Interval Forecasts
7.1 Introduction
7.2 Notation
7.3 The need for dierent approaches
7.4 Expected mean square prediction error
7.5 Procedures for calculating P.I.s
7.6 A comparative assessment
7.7 Why are P.I.s too narrow?
7.8 An example
7.9 Summary and recommendations
8 Model Uncertainty and Forecast Accuracy
8.1 Introduction to model uncertainty
8.2 Model building and data dredging
8.3 Examples
8.4 Inference after model selection: Some ndings
8.5 Coping with model uncertainty
8.6 Summary and discussion
References
c 2000 by Chapman & Hall/CRC
Preface
Forecasting is an important activity in economics, commerce, marketing
and various branches of science. This book is concerned with forecasting
methods based on the use of time-series analysis. It is primarily intended
as a reference source for practitioners and researchers in forecasting,
who could, for example, be statisticians, econometricians, operational
researchers, management scientists or decision scientists. The book could
also be used as a text for a graduate-level course in forecasting. Some
application areas, such as meteorology, involve specialist methods which are
not considered in this book. The specialist area of judgemental forecasting
is also not covered. Rather we concentrate on time-series forecasting
methods which are applicable more generally, but especially in economics,
government, industry and commerce.
The scope of the subject is wide and the topics I have chosen to cover
reect my particular interests and concerns. There are now several books,
both at introductory and intermediate levels, which aim to cover the basic
range of time-series forecasting methods. This book does not attempt to
duplicate this material and is not meant to be a how-to-forecast manual. It
would need to be several times longer in order to provide all the necessary
information about theory and applications for all the methods reviewed
in this book. Rather this book was conceived as providing a summary of
the basics of univariate time-series analysis and modelling (Chapters 2 and
3), including recent arrivals such as GRAPH models and neural networks,
followed by a fairly brief catalogue of the many time-series forecasting
methods, both univariate (Chapter 4) and multivariate (Chapter 5). The
latter chapter, together with Chapter 6, also attempt to compare the more
important method, both in terms of their theoretical relationships (if any)
and their practical merits including their empirical accuracy. While the
search for a best method continues, it is now well established that no
single method will outperform all other methods in all situations, and, in
any case, it depends on what is meant by best ! The context is crucial.
This book also covers two other general forecasting topics, namely the
computation of prediction intervals (Chapter 7) and the eect of model
uncertainty on forecast accuracy (Chapter 8). These important aspects of
forecasting have hitherto received rather little attention in the time-series
literature. Point forecasts are still presented much more frequently than
interval forecasts, even though the latter are arguably much more useful.
There are various reasons for the overemphasis on point forecasts and it is
c 2000 by Chapman & Hall/CRC
hoped that the presentation in Chapter 7 will help to change the position.
The model uncertainty problem arises when the analyst formulates and ts
a model to the same set of data as is the usual procedure in time-series
analysis. The analyst typically behaves as if the selected model is the correct
model and ignores the biases which arise when the same data are used to
choose, t and use a model. Standard least-squares theory no longer applies
but is typically used anyway. Prediction intervals are typically calculated
conditional on the tted model and are often found to be too narrow. This
provides a link with the material in Chapter 7.
While self-contained in principle for a reader with a reasonably thorough
general statistical background, this book will be more accessible for
someone who knows something about the basics of time-series analysis and
has, for example, some understanding of trend, seasonality, autoregressive
(AR) and moving average (MA) models.
I would like to acknowledge helpful comments on draft versions of the
book from various people including Keith Ord, Sandy Balkin and two
anonymous referees. I thank Ken Wallis for providing Figure 7.1. The work
for Example 8.4 was carried out jointly with Julian Faraway (University of
Michigan, USA). As always, any errors or obscurities that remain are my
responsibility, however much I may wish to avoid this!
The time-series data used as examples in this book are available via my
website http://www.bath.ac.uk/mascc/ at the University of Bath. They
include an extended S&P500 series of 9329 observations from 3/07/1962
to 22/07/1999. I would like to have included more examples, but the
manuscript is already long enough and time is pressing. My forecasted
date of delivery has proved remarkedly inaccurate! I hope that you, the
reader, nd the book clear and helpful. Please feel free to tell me what you
think as constructive criticism is always welcome.
Chris Chateld
Department of Mathematical Sciences
University of Bath
Bath, U.K., BA2 7AY
email: cc@maths.bath.ac.uk
October 2000
c 2000 by Chapman & Hall/CRC
Abbreviations and Notation
AR Autoregressive
MA Moving Average
ARMA Autoregressive Moving Average
ARIMA Autoregressive Integrated Moving Average
SARIMA Seasonal ARIMA
ARFIMA Fractionally Integrated ARIMA
TAR Threshold Autoregressive
GARCH Generalized Autoregressive Conditionally Hereoscedastic
SWN Strict White Noise
UWN Uncorrelated White Noise
NN Neural Network
MSE Mean Square Error
PMSE Prediction Mean Square Error
MAPE Mean Absolute Prediction Error
P.I. Prediction Interval
ac.f. Autocorrelation function
acv.f. Autocovariance function
x
N
(h) Forecast of x
N+h
made at time N
B Backward shift operator
The operator (1 B)
E Expectation or Expected value
I The identity matrix a square matrix with ones on the diagonal
and zeroes otherwise
N(,
2
) Normal distribution, mean and variance
2

Chi-square distribution with degrees of freedom

Z
t
or
t
Uncorrelated white noise with constant mean and variance
Vectors are indicated by boldface type, but not scalars or matrices.
c 2000 by Chapman & Hall/CRC