The History of Econometric Ideas (Historical Perspectives on Modern Economics)

Rating:
Summary: Morgan is,like Tinbergen and Frisch,addicted to the N(0,1)
Review: Mary Morgan's book is an interesting summary of the historical development of the use of statistical analysis in economics, which culminated in the birth of econometrics in the mid-1930's ,due to the work of Tinbergen and Frisch and to the work of Haavelmo in the 1940's through the mid 1950's.She traces the development of the application of statistics to economic data ,starting with discussions of laws of error through the development of least squares by Legendre,Laplace,and Gauss to applied work done by Edgeworth and Mitchell.The heart of the book involves the work of Tinbergen,Frisch,and Haavelmo in their development of multiple regression and correlation analysis and their use of these techniques to supposedly test different economic theories based on the analysis of time series data.Morgan does a very good job as long as she sticks to historical summary.Unfortunately, she,far to often,errs grievously when she attempts to assess the relative merits of different competing theoretical arguments.This is most obvious in her coverage of the 1939-40 debate between John Maynard Keynes and Jan Tinbergen over the logical foundations of econometrics that took place in the Economic Journal(EJ).It also shows up in her strange claim that econometrics is an experimental method,which it certainly is not and will never be as long as the basic assumptions of the multiple regression and correlation approach are based on the mere assumption of a normal probability distribution. Morgan covers the Tinbergen-Keynes debate on pages 121-134.There is no evidence in this book that Morgan ever read any part of Keynes's A Treatise on Probability(1921;TP),much less the crucial parts of the TP that would allow a reader of the 1939-1940 exchange to understand the full force of Keynes's critique(chapter 17,pp.205-214,chapters 29 and 30,chapter 32,pp.391-393, pp.397-398,especially Keynes's discussion on p.398 of using the Lexis Q-test to test for the stability of, and justification for assuming the use of, a normal probability distribution,chapter 33,pp.408-409,415-416,and 420-421,especially ft.1 on pp.420-421,which summarizes the conclusions one can expect to derive from an application of the Lexis-Q test).The logical core of Keynes's critique is contained on page 568 of the Sept.,1939 EJ.Keynes states:"...the most important condition is that the environment in all relevant respects,other than the fluctuations in those factors of which we take particular account,shuold be uniform and homogeneous over a period of time."Keynes then suggests the formal procedure required to implement the application of the Lexis Q-test-breaking up the series(time series) into a set of subseries and then testing to see if each of the subseries' coefficients are generally the same(uniform) as the coefficients of the series.The results would be normal,subnormal,and supernormal(see Keynes's ft.1 on pp.420-421 of the TP).One criticism of Keynes(minor)is that he did not explicitly mention the Lexis Q-test.Keynes probably did not want to preclude other tests for normality,such as the Chi-Square.No where in Tinbergen's two volume work is there any such test.Keynes,however,could have told Tinbergen the results of such a test in advance.Due to constantly changing expectations of future expected profits and constant new investment in a capital stock changing due to constant technological change,advance and innovation over the long run period of time that would constitute the relevant time series,the necessary uniformity and homogeneity needed to apply a normal probability assumption would fail the test. The Keynes of the TP also had a suggested solution-use Cheybshev's inequality(see chapter 29 of the TP) and other measures of central tendency(law of errors) such as the median,harmonic mean and geometric mean(see chapter 17 of the TP).Morgan discusses the assumption of normality(see pp.8-9,60,113,117,240)that underlies the entire edifice of Tinbergen's econometrics without mentioning a single piece of empirical,experimental or historical evidence that would supply some amount of justification for such an assumption. In many respects,Keynes's argument against the misuse and abuse of the assumption of a normal probability distribution to test different theories of the business cycle ,using time series data, can be viewed as an earlier underdeveloped version of Benoit Mandelbrot's overwhelming critique of the misuse and abuse of the normal distribution in finance and portfolio theory.It is now obvious,due primarily to Mandelbrot's 50 years of scholarly work,that predictions about future price movements in the stock,money,bond,commodity, and currency exchange markets are all wrong.It should not be surprising that attempts to predict changes in the business cycle over the 65 years since the Tinbergen-Keynes debate have also turned out to be badly off the mark.I recommend that a reader of Morgan's book also purchase Hugo Keuzenkamp's 2001 book,titled Probability,Econometrics and Truth and Benoit Mandelbrot's 2004 book,The (Mis) Behavior of Markets in order to fully grasp the case against present day econometric practice.Econometrics is presently rsduced to the short run estimation of coefficients based on the use of cross-sectional data.