Although the title of this book suggests a focus on forecasting, it should be compulsory reading for all engaged in applying econometrics in a time-series context and especially for those who are responsible for teaching others. The text continually provides insights from the authors' wealth of practical experience linking empirical modelling to forecasting. Professional forecasters will be interested in this book, and it should feature on reading lists for (advanced) undergraduate and postgraduate courses in econometrics and applied econometrics.
A previous volume by the same authors was concerned with forecasting stationary time series and with integrated variables that could be modelled in a stationary system. This book moves the setting on to recognise the importance of deterministic non-stationarities (structural breaks). It consists of 12 chapters and a preface that should not be skipped.
Chapter one usefully reviews the main results from the earlier book and so sets the scene for the value added in the present volume. Chapters two, three and four consider the sources of forecast failure, the role of deterministic shifts in these failures and other possible sources of forecast failure.
Useful tables in chapter two list the taxonomic possibilities of sources of forecast error, with structural change affecting the equilibrium mean particularly important for forecast failure. Practically, modellers have adopted the strategies of adjusting intercepts to "correct" their forecasts or of working with differenced data. Chapters five and six consider how these strategies work. Chapters seven, eight and eleven consider three empirical applications, first to consumers' expenditure and then to a monetary model and a wage-price-unemployment system. Chapter seven illustrates the essence of results from the earlier chapters and is accessible to readers who are not comfortable with a multivariate approach. Chapters eight and ten are concerned with different aspects of the multivariate examples. Chapter nine introduces the concept of co-breaking. Chapter ten is an interesting departure from the usual framework (vector auto-regression with equilibrium correction) and considers some non-linear regime-switching models. Chapter 12 is a postscript, and the book concludes with some useful exercises, several of which could be the basis of research proposals. The book is well presented, with good use of tables and figures.
The empirical examples indicate what in practice are the "deterministic non-constancies (that) seem responsible for many major episodes of empirical forecast failure". For example, in the context of a four-equation monetary model, financial innovation in the form of interest payments on retail sight deposits can be interpreted as a deterministic shift.
Readers familiar with previous work by the authors will correctly forecast that this volume will be tightly written, with many subtle and important distinctions, and will fit into a progressive research strategy. The book is likely to be referred to again and again as the reader gradually appreciates the full importance of its concepts (for example, co-breaking). Anyone interested in what time-series econometrics has to offer should buy this book or recommend it to their library.
Kerry Patterson is professor of econometrics, University of Reading.
Forecasting Non-Stationary Economic Time Series
Author - Michael P. Clements and David F. Hendry
ISBN - 0 262 032 4
Publisher - MIT Press
Price - £21.95
Pages - 362