Business Cycles, Indicators, and Forecasting.

In the past ten years, advances in macroeconomic theory, statistical methods and computational power have led to the development of new forecasting methods that are slowly beginning to replace the traditional structural econometric models. In Business Cycles, Indicators, and Forecasting, the editors, James H. Stock and Mark W. Watson, have compiled a collection of eight time-series forecasting papers that were originally presented at a conference of the National Bureau of Economic Research in May of 1991. The individual papers in this volume can be grouped into three broad areas. The first four papers provide an exposition of recently developed statistical techniques. Specifically, these papers employ time-series and Bayesian vector autoregressive (BVAR) methodologies which can be used to generate stochastic simulations to produce probability forecasts of discrete recession/expansion events. The second group consists of a pair of empirical papers which explore the historical relations among specific time series. The final group contains two methodological papers that address dynamic factor models in high dimensional systems and nonlinear time-series modeling. Following each paper the text offers a comment which contains a detailed criticism of each paper's analytical content. In this review, I will offer a more descriptive discussion of each author's contribution.

The first group of papers begins with Victor Zarnowitz and Phillip Braun's encyclopedic paper entitled "Twenty-two Years of the NBER-ASA Quarterly Economic Outlook Surveys: Aspects and Comparisons of Forecasting Performance." The authors examine the general problems and the history of forecasted evaluations and surveys. They also detail the NBER-ASA data collection methods and discuss the analytical implication of the data set which contains twenty variables and runs for eighty-six consecutive quarters. In the second paper, "A Procedure for Predicting Recessions with Leading Indicators: Econometric Issues and Recent Experience" James H. Stock and Mark W. Watson set the analytical tone for the remaining three papers in this group. The Stock and Watson paper offers a detailed account of the construction and empirical performance of a recession index. The model generates "three indexes of overall economic activity on a monthly basis: an experimental coincident index (the XCI); and experimental leading index (the XLI), which is a forecast of the growth in the XCI over the subsequent six months; and an experimental recession index (the XRI), which estimates the probability that the economy will be in recession six months hence" [p. 96]. In the third paper "Estimating Event Probabilities from Macroeconomic Models Using Stochastic Simulation" Ray C. Fair explains that probability questions can be directly addressed within the context of macroeconomic modeling by using stochastic simulation. Fair uses a model that is dynamic, nonlinear, simultaneous and can be designed to have autoregressive errors of any order. He uses the model to run a number of simulations on several economic events and then compares the models performance with real-time data. Fair estimates event probabilities for two consecutive quarters of negative real GNP growth as well as two quarters of inflation over seven percent. The final paper in this group is Christopher A. Sims's "A Nine-Variable Probabilistic Macroeconomic Forecasting Model." This paper is an extension of the Bayesian vector autoregressive six variable model developed by Robert Litterman in the early 1980s. This current model extends the previous work by augmenting the original model with the ability to account for nonnormality of forecast errors and by allowing for time-varying variances as well as time-varying autoregressive coefficients. The new model also adds trade-weighted value of the dollar, Standard and Poors 500 stock index and the commodity price index to the original variable list of T-bill, M1, GNP deflator, Real GNP, business investment and unemployment.

Benjamin M. Friedman and Kenneth N. Kuttner begin the empirical section with a paper entitled "Why Does the Paper-Bill Spread Predict Real Economic Activity?" The paper explores the reasons why the difference between the interest rates on commercial paper and Treasury bills is such a good indicator of real economic activity. Friedman and Kuttner examine default risk, monetary policy effects, differential tax treatments and changing cash requirements of borrowers for possible indications of the variables predicative power. The second paper in this section is "Further Evidence on Business-Cycle Duration Dependence" by Francis X. Diebold, Glenn D. Rudebusch and Daniel E. Sichel. The paper attempts to find out if expansions, contractions or whole cycles are more or less likely to end as they grow older. The authors employ an exponential-quadratic hazard model to examine duration dependence and use data from the U.S., Great Britain, France and Germany. They conclude that all four countries exhibit evidence of positive duration dependence in prewar expansions but not for prewar contractions.

The final two papers focus on the methodological aspects of economic time series and forecasting. In "A Dynamic Index Model for Large Cross Sections" Danny Quah and Thomas J. Sargent present a framework for analyzing commonalities in dynamic models and data structures where the cross-sectional dimension is potentially as large as the time series dimension. The paper develops index structures and shows how standard econometric techniques can be adapted to handle random fields. The final paper is "Modeling Nonlinearity Over the Business Cycle" by Clive W. J. Granger, Timo Terasverta and Heather M. Anderson. In this paper the authors compare a wide variety of linear models with those of nonlinear structure. They conclude that nonlinear models are superior in sample but not out of sample. The paper also addresses the dangers of overfitting due to data mining.

This book is not intended for a general audience. It is a selection of highly technical statistical techniques that are directed toward the forecasting professional. To gain the full benefit from this work, one must approach the book with a good understanding of the mechanics of time-series forecasting. However, those already working in the field will find in the book a very convenient cross section of the current state of the art in time-series economic forecasting.

Michael C. Carroll Colorado State University