The econometric analysis of time series

The econometric analysis of time series

Book forecasting makes no mention of the many innovations limiting demand, including time-of-day electricity pricing, interruptible service and self-...

105KB Sizes 74 Downloads 1250 Views

Book

forecasting makes no mention of the many innovations limiting demand, including time-of-day electricity pricing, interruptible service and self-generation that are likely to prove so important. The final part of the book. Managing the Forecasting Function, asks questions like “what needs to be forecast”-a little late to be asked in chapter 29, “how to integrate forecasting and decision making” and “what is the future of forecasting”. The discussion of the last question makes no mention of the wide variety of existing new methods of forecasting that are being discussed in the econometric and statistical literature. A manager with little experience in forecasting may well find this book a useful introduction to the field but he or she should be strongly encouraged not to stop here, but to read other accounts, as there is a lot more to be learned. C. W.J. GRA NGER University of Calijorniu San Diego, U.S.A.

A.C. HARVEY The Econometric Analysis of Time Series Philip Allan, Oxford, 1981, xi + 384 pages, &l&95 One of the causes of the increasing popularity of time series methods is the fact that they produce forecasts that, even if based on simple analysis, show an unexpectedly high quality. Many empirical comparisons confirmed the idea that simple (i.e. univariate) models are a match for the huge econometric models used in practice. Naive methods were rebaptized to semi-naive (Christ), quasi-naive (Cooper), not-quite-so-naive (Nelson) and even super-sophisticated naive (Nerlove). One of the explanations of this remarkable forecasting performance concerns the possible inadequacy of the underlying theories. In “Econometric Analysis of Time Series” A.C. Harvey uses a somewhat different argument. Most of the available economic theory is of a
Reviews

319

tion and estimation of dynamic models (chapters 7, 8 and 9). The first five chapters present the necessary statistical and methodological basis. Chapter six, on static regression models with autoregressive-moving average disturbances, is a logical link between the two parts. A second objective, mentioned by the author, is the presentation of an ‘integrated approach’ to the maximum likelihood estimation/testing problem. First, the basic idea of maximum likelihood is introduced and illustrated by some simple examples. Next, the Cramer-Rao Lower Bound is derived and the properties of the ML-estimator are presented. These sections form an introduction to a more general treatment of ML-estimation of different forms of regression models. Computational difficulties arising in ML-applications are discussed in a fine chapter on numerical optimisation. Also, likelihood ratio methods are treated as tests of misspecification. In the final chapter the approach is used in simultaneous equation estimation. Model selection strategy is the third theme of ’ the book. In particular if dynamic models are involved, the question whether transfer functions or stochastic difference equations should be preferred is discussed. The text is clearly written in a good layout. Knowledge of elementary matrix algebra is assumed throughout the book as well as acquaintance with regression principles. Exercises (some of them with answers) are included in each chapter. Harvey’s book is a fine textbook on the use of time series in econometric model building and can be recommended to both students and practitioners. E.G. F. van WINKEL University of Technology Eindhoven, Netherlands

H. TOUTENBURG Prior Information

in Linear Models

Wilev. Chichester. 1982. gl6.50 This book looks at the problem of estimation for the general linear model where the error terms are allowed to be correlated. The usual unbiased estimate of the regression parameter vector j3 in this model is the generalized least squares estima-