Forecasting Economic Time Series
Swanson, Norman R., Journal of the American Statistical Association
Michael P. CLEMENTS and David F. HENDRY. Cambridge, U.K.: Cambridge University Press, 1998. xxii + 368 pp. ISBN 0-521-63242-0. $69.95 (H) $24.95 (P).
The oft subjective and sometimes maligned area of economics that is macroeconomic time series forecasting receives a carefully critical and seriously detailed treatment in this book. Perhaps one of the most appealing features of the book is the systematic way in which it outlines and uncovers problems in forecasting, lays out possible solutions, and uses Monte Carlo, theoretical, and empirical evidence to assess the potential solutions. This approach leads to a well-balanced and useful collection of diagnoses and prescriptions that are in many ways indicative of the current state of the art. Another appealing feature is that beginning researchers who are generally interested in serious (empirical) scientific investigation can learn much from noting how Clements and Hendry uncover, assess, and examine important issues in the area of economic forecasting. A third feature worth noting is the plethora of insightful and detailed empirical and Monte Carlo evidence, much of it drawn from published articles in the a rea of forecasting written by Clements, Hendry, and various co-authors.
This book is actually the first of a two-volume set on macroeconomic forecasting based on the Marshall Lectures at Cambridge University. The second volume is titled Forecasting Non-Stationary Economic Time Series (Clements and Hendry 1999). Although it might be thus assumed that this book is written at an introductory level, that is not the case. The book is aimed at advanced undergraduate students, graduate students, professionals, and academics, particularly those with an interest in empirical macroeconomics and forecasting. Readers who have had a first course in time series econometrics, have received serious undergraduate training in econometrics, or who have skimmed relevant chapters in a book like that of Hamilton (1994), stand to gain the most from reading this book. In summary, the book is somewhat more advanced than Elements of Forecasting (Diebold 1998), Business Forecasting (Hanke and Reitsch 1998), and Forecasting in Business and Economics (Granger 1989), for example, and is perhaps closest to Ti me Series Models for Business and Economic Forecasting (Franses 1998) and Forecasting Economic Time Series (Granger and Newbold 1986).
The book essentially starts at the beginning with a discussion of the development of forecasting since the 1920s, and very quickly moves to comprehensive discussions of methods for achieving robustness against certain forms of structural break and model misspecification, methods for comparing forecasts and assessing forecast accuracy, Monte Carlo techniques for forecasting, and methods for multistep dynamic estimation. The authors even go so far as to develop what they call a "formal taxonomy of forecasting errors," which they propose as an aid to the systematic evaluation of economic forecasts. Perhaps one of the most appealing features of the book is that although it covers advanced topics less technically oriented readers can benefit greatly from reading it, as it begins with a serious survey of forecasting first principles, forecast evaluation, and forecasting in univariate processes and proceeds gradually to a discussion of forecasting in cointegrated systems and other more advanced topics. However, hav ing said this, I also should point out that the artificial data example constructed in Chapter 1 is sufficiently advanced as to involve the use of I(1) and I(0) variables, vector error-correction models, and Johansen (1988, 1991) trace test statistics, for example. This is because Clements and Hendry aim to illustrate at the outset many of the problems and issues that they discuss later.
Another appealing feature is that although the authors suggest various routes to successful forecasting, they also summarize the advantages and disadvantages of following alternative routes. …