A Guide to Econometrics, 5th ed

Business Economics, April, 2004 by Jan Kmenta

A Guide to Econometrics, 5th ed

By Peter Kennedy, 2003. Cambridge, MA: The MIT Press. Pp. 623. $75.00 cloth, $37.95 paper.

The fact that there is a new fifth edition of Peter Kennedy's A Guide to Econometrics is eloquent testimony to the popularity of the preceding editions. The book is not a textbook for an econometric course but a nonmathematical complement to a standard econometrics textbook at any level. Its aim is to provide an overview of econometric topics and to give an intuitive interpretation of econometric concepts and methods without the usual clutter of technical detail. Business economists will appreciate this approach. The virtue of the newest edition lies not only in the additions and extensions introduced but also in the updating of the huge stock of references cited throughout the book.

As in the earlier editions, the author follows a uniform structure of the chapters. Each chapter consists of the main text pertaining to the subject of the chapter, followed by "general notes" and concluded by "technical notes." Mathematical exposition is reserved to the concluding part, and even there it is used very sparingly and in an uncomplicated way.

The chapters cover virtually all topics discussed in standard econometric courses, whether undergraduate or graduate. In general, the coverage is comprehensive, although different readers might wish for more attention to their favorite topics. My wish list includes more coverage of some of the most exciting and relatively recent innovations in econometrics now finding their way into business and finance, such as Robert Engle's ARCH model and Halbert White's heteroskedasticity-consistent standard errors. But these are unavoidable aspects of a comprehensive exposition of any subject.

Apart from the uniqueness of the style of Kennedy's text, the book is also unique in a way that distinguishes it from other econometric texts, namely that for anybody interested in the subject it is fun to read. The text is full of clever insights, witty comments and humorous citations. The author's excellent understanding of all things econometric is remarkable, and his command of the econometric literature is most impressive. Considering the amount of work that must have gone into writing this book, it is clear that for the author it was a labor of love. In his comments on each topic the author thinks of practically everything. Again and again, I thought of some omitted aspect of the topic under discussion only to find it mentioned in the next paragraph or on the next page. Furthermore, the models and methods discussed are presented in a simple and easily understandable way. Notable examples of this would include generalized method of moments, duration models, bootstrapping, and especially neural networks. Worth mentioning is the new chapter on panel data, and the business economist will appreciate the new chapter on "applied econometrics" with its most useful discussion of common mistakes.

With all these enthusiastic accolades, the obvious question has to do with the possible existence of any weakness of the book. In this regard the only significant weakness, at least in my view, is the treatment of "time series econometrics." The author notes that the introduction of the time-series analysis approach to modeling dynamic economic relations in the 1980s has radically changed the way of dealing with time series observations in econometrics. The most effective acknowledgement of this fact--not yet known to the author at the time of writing this book--was the 2003 Nobel Prize award to the two architects of the "time series revolution," Robert Engle and Clive Granger. Nevertheless, the "time series econometrics" has weaknesses that are not mentioned by the author and should interest business economists.

One weakness of the time series approach in modern econometrics is methodological. The traditional and common approach to econometric problems of dealing with economic relationships usually starts with the regression model that applies regardless of the type of observations used for estimation and testing. The author himself states this very clearly by noting that "(m)ost econometric problems can be characterized as situations in which one (or more) of these five assumptions (of a classical regression model) is violated in a particular way ... (The researcher) then turns ... to see whether the OLS estimator retains its desirable properties and if not what alternative estimator should be used" (pages 47-48). The violated assumption in the case of variables observed over time is presumably that of a finite variance. The standard methodology of econometric research would call for the determination of the detrimental consequences concerning the properties of OLS estimators and for their remedy. In the parlance of time series analysts, inference using OLS is "invalid" or "spurious," but the author makes no reference to the standard and precisely defined properties of consistency, efficiency, and normality in presenting time-series analysis. Certainly business students with an eye to using this material will notice the disconnect.