지은이: Hugo A. Keuzenkamp (1995)
Keynes disliked econometrics. Moreover, he did not understand much of it. This, at least, is the view of many economists and econometricians who recall their vague, and usually indirect, knowledge of the Keynes-Tinbergen controversy. ( ... ) The controversy with Tinbergen is frequently regarded as a deplorable clash between an old and a new era in economics. Occasionally, Keynes is credited with raising the problem of misspecification, without having an established vocabulary in which to formulate the problem. However, in this paper I will argue that Keynes' critique is not primarily one of mis-specification. It is neither based on an objection to econometrics and probabilistic inference in general, nor does it follow from an outdated misunderstanding of the crucial issues at stake. Keynes understoot them all to well.
The debate between Keynes and Tinbergen has its roots in one of the first and most ambitious efforts to test economic theories statistically. On request of the League of Nations, Haberler( 1958) had provided an overview of business cycle theories. As a follow-up, Tinbergen was asked to confront those rival theories with data. The resulting work(Tinbergen, 1939a, b) roused strong sentiments. The debate concentrated on the question whether statistical methods are proper tools for testing economic theories.
The basic issue at stake was: is the multiple regression model,
a valid or fruitful tool for testing economic hypotheses? This question has to be answered taking into consideration the possibility of serious specification uncertainly and a limited set of (unique, non-experimental) data, where k (the potential number of regressors) may well exceed n (the given number of observations). Is this a practical, or is it a logical problem? Once started, the debate on those questions never concluded.
 See Leamer (1978) for a recent perspespective.
Unlike Tinbergen, who was a very pragmatic research worker, Keynes was preoccupied with the ^logical^ conditions for probabilistic inference, as may be clear from his earlier work, Keynes( 1973a). Keynes argued that the application of statistical methods to the analysis of investment behaviour (the example presented by Tinbergen, 1939a, to clarify his method) was was the least promising starting point as this is a case where those logical conditions were not even remotely met. In his paper I aim at clarifying why not. I will focus on the logical issue raised by Keynes, which does not seem to be generally understood or appreciated. This logical point may have been phrased obscurely, but it is worth further investigation for a better understanding of the foundations of econometric inferenceㅡeven today.