Book
reviews
391
There is a brief section on Bayesian methods, mentioning work on AR(p) models, updating inferences about the mean (or mean vector) for linear series, and a connection with intervention analysis. The chapter ends with a striking example showing a 600 observation realization of a non-linear process. Viewed from the perspective of a linear, stationary model, it appears to contain many outliers; however, relative to the given model, there are no outliers. The remaining twelve chapters include general material on the accommodation of outliers, discordancy tests, directional data, slippage tests, multivariate data, regression, designed experiments, and Bayesian approaches. The book misses some key research relating to outliers in time series data. For example, Mandelbrot (1982) discusses outliers in certain non-stationary series, but is not mentioned in the second edition. His ideas may prove fundamental in effectively removing outlier problems in many real-world time series. Also not mentioned is Martin, Samarov and Vandaele’s major paper (1983). In sum, the second edition provides a great service by updating and summarizing the general body of literature in this growing field. Hopefully, before long, researchers will concentrate enough on the field of outliers in time series to remove the epithet ‘a little-explored area’, and accomplish enough to be of additional benefit to the forecaster. David H. Nash Drexel University, Philadelphia, ?A, USA
References Mandelbrot, B., 1982, The many faces of scaling: Fractals, geometry of nature, and economics, in: W.C. Schieve and P.M. Allen, eds., Self-organization and Dissipative structures (University of Texas Press, Austin, TX). Martin, Samarov, and Vandaele, 1983, Robust methods for ARIMA models, in: A. Zellner, ed., Proceedings of ASA - Census - NBER Conference on Applied Time Series Analysis of Economic Data (Bureau of the Census, Washington, DJC).
Charles S. Gross and Robin T. Peterson, Business Forecasting, 2nd ed. (Houghton Boston 1983) pp. 400.
Mifflin
Company,
Gross and Peterson have described their book as ‘an introduction to business forecasting for current and future managers. Thus, it seems that their intended audience in&ides the practitioner, as well as the upper division undergraduate college student. The text is extremely readable and the recommended prerequisites of introductory statistics, computer science, and high school algebra should be sufficient background for most of the material. This reviewer has used this book twice in his senior marketing forecasting class and has had no problems with text readability. The stated emphasis of the book is on sales forecasting even though the authors indicate that many of the methods described could be used for other purposes. Additionally, the authors’ emphasis is placed on micro-firm and industry forecasting and not macro-aggregate economic forecasting. Their reasoning is that macroeconomic forecasts are provided by many organizations and consequently companies can use these sources. A brief description of each chapter il be given. This will be followed by an assessment of the book’s strengths and weaknesses. Last, the extent to which the text fits the needs of the practitioner and the academic community will be discussed. Chapter 1 gives a good overview of how forecasting is organized and how it fits into the organizational planning process. The forecasting process is described briefly. Chapter 2 presents a
392
Book
reviews
cursory treatment of the major judgmental forecasting methods. Chapters 3 through 5 discuss quantitative methods. Informal and formal smoothing methods. regression, and decomposition methods are covered. Chapters 6 through 8 are devoted to miscellaneous forecasting methods such as surveys and test markets, models based on learned behavior, and model building/simulation. Chapter 9 contains a brief survey of forecasting using economic indicators, the replacement cycle, and input/output analysis. Chapter 10 addresses the assessment and implementation of forecasting models. Chapter 11 provides closing comments about the importance and future of forecasting. There are several positive aspects of this text. As mentioned earlier, it is extremely readable; examples illustrate the calculations for most of the quantitative methods discussed. There is a summary at the end of each chapter. Discussion questions, recommended readings, and a short case study are also provided for each chapter. Fortran computer codes are included in the text for some of the time-series methods. A 149 page instructor’s manual is also available. Some useful, simple forecasting methods usually overlooked by other forecasting texts have been included (such as the moving percentage change method). The book also has certain weaknesses. (1) Some topics are given more space than needed, while some more important topics are underemphasized. For example, budgets are overemphasized in Chapter 1. Decision analysis covered in Chapter 2 is given almost as much space as decomposition methods discussed in Chapter 5. It would have been more appropriate to give less space to decision analysis and more space to major forecasting methods. Methods such as the sales force composite and Delphi Technique are covered too briefly even for an introductory book on forecasting. (2) Sometimes the authors depart from their stated focus on sales forecasting, For example, mostof their discussion on decision analysis deals with identifying the value of research, not with sales forecasting. Another illustration of this is the discussion on Monte Carlo simulation in Chapter 8. The example used to illustrate simulation doesn’t show the value of using simulation for sales forecasting but, rather,its use in determining from which supplier to buy machines. (3) The overall organization of the book is somewhat confusing. For example, regression analysis is discussed from the causal modeling viewpoint in Chapter 4. However, this chapter is placed between two chapters that solely describe time-series methods (smoothing methods in Chapter 3 and decomposition methods in Chapter 5). Another example is that forecasting by use of customer or user expectations is briefly discussed in Chapter 2. However, Chapter 6, in describing the use of surveys, is devoted to the same topic. One last example is that brief discussions of Box-Jenkins and adaptive filtering are given in the decomposition chapter rather than in the time-series smoothing chapter, as might be more appropriate. (4) Some important topics are notably missing or given only passing attention. Since the authors claim that improved initialization schemes and optimal weighting routines eliminate the need for different approaches, adaptive response rate single exponential smoothing (ARRSES), and Holt’s two parameter exponential smoothing are not illustrated. Another example is the absence of discussion on the use of the Durbin-Watson statistic in measuring autocorrelation, though the topic of autocorrelation is covered. The examination of residuals in regression analysis is given only passing treatment. Additionally, no discussion is given for the preparation of data before they are used for forecasting. In summary, even with the weaknesses mentioned, this bpok adequately gives an introduction to business forecasting. Managers will get a quick overview of the major methods used in forecasting
Book reviews
393
although they probably will need other sources to implement some of the methods discussed. As mentioned earlier, this reviewer has used the text for an introductory course in forecasting. This text has worked satisfactorily but most instructors will probably want to supplement the text with outside readings or additional notes, as this reviewer has. James E. Cox, Jr. Illinois State University, Normal, IL 61761, USA
Zvi Griliches and Michael D. Intriligator, Amsterdam, 1983) pp. 771.
eds., Handbook
of Econometrics,
vol. 1 (North Holland,
The first of the three volumes of the Handbook of Econometrics is in three sections. The first of which, entitled ‘Mathematical and Statistical Methods in Econometrics’ contains two chapters: ‘Linear Algebra and Matrix Methods in Econometrics’ by Henri Theil and ‘Statistical Theory and Econometrics’ by Arnold Zellner. The next section entitled ‘Econometric Models’ has articles on ‘Economic and Econometric Models’ by Michael Intriligator, ‘Identification’ by Cheng Hsiao and ‘Model Choice and Specification Analysis’ by Edward Learner. The final section is concerned with ‘Estimation and Computation’ and contains articles on ‘Nonlinear Regression Models’ by Takeshi Amemiya, ‘Specification and Estimation of Simultaneous Equation Models’ by Jerry Hausman, ‘Exact Small Sample Theory in the Simultaneous Equation Model’ by Peter Phillips, ‘Bayesian Analysis of Simultaneous Equation Systems’ by Jacques Dreze and Jean Francois Richard, ‘Biased Estimation’ by George Judge and M.E. Bock, ‘Estimation for Dirty Data and Flawed Models’ by William Krasker, Edwin Kuh and Roy Welsch and finally ‘Computational Problems and Methods’ by Richard Quandt. As one would expect, given the well known and distinguished contributors, this is an extremely impressive volume that assembles several really excellent surveys on the current state of knowledge in different fields of econometrics. The editors are also to be congratulated for producing a volume where there is a natural progression from one authors chapter to another and a volume that contains remarkably few misprints. The aim of this set of volumes is to produce a ‘definitive source, reference and teaching supplement for use by professional researchers and advanced graduate students’. In their different ways, all the chapters are of very high quality; while some are naturally more immediately accessible than others. For example, no graduate student is likely to encounter undue difficulties with Intriligator’s third chapter which motivates the formulation of different models. The preliminary chapter by Theil contains a lot of useful results on matrix algebra and shows how they can be applied to specific econometric problems and to specific economic theory in consumer demand analysis etc. While Theil’s chapter is also of high quality, his review of the literature is quite selective and researchers wanting a glossary of results, e.g., the various results on matrix differentiation, will be disappointed. One of my favorite chapters in the book ‘is Arnold Zellner’s survey of statistical methodology. Like his well known survey articles on causality (1979a) and the use of statistics in econometrics (1979b), this chapter is a beautifully written account; where necessary containing terse details of statistical results mixed with insights into the foundations of probability theory, both classical and Bayesian. Zellner surveys the main univeriate and multivariate distributions, asymptotic theory, estimation theory and Bayesian approaches. In a few cases, e.g., the Cramer-Rao inequality, a proof is given; but in general the treatment mainly consists of results and commentary. Zellner manages to compress some of the main results on asymptotic theory, pp. 110-116 into seven pages and I think succeeds in