Frequentist and Subjectivist Perspectives on the Problems of Model Building in Economics

Abstract

"It is difficult for me to tone down the missionary zeal acquired in youth, but perhaps the good battle is justified since there are still many heathens."I. J. Good (1976, p. 126) T he "heathens" to whom I. J. Good refers are the non-Bayesians of the world and they are my targeted audience as well. Specifically, I plan to discuss, in as simple and nontechnical a fashion as possible, the subjectivist-Bayesian attitude toward model building in econometrics and to contrast it with the standard frequentist attitude. Most of the issues I raise are familiar to statisticians but not to economists. Rather than give the suspicious reader a menu of Bayesian techniques, I hope to create an interest in acquiring a taste for the Bayesian cuisine by recommending five pragmatic principles.While neither the Bayesian nor the non-Bayesian flocks are homogeneous with respect to the beliefs of their own members, I believe the "within-group" differences

Unawareness, priors and posteriors

“…[...] The only problem is the difficulty in realistically specifying a "prior" distribution for parameters that one hadn't explicitly recognized before seeing the data." And Poirier [17], who calls these 'post-data priors', writes: "I believe such priors exist before the data, but that the difficulties of formulating priors cover them up. The data encounter that provokes the window [i.e.…”

Asset Returns and State-Dependent Risk Preferences

“…Much has been written (Zellner 1971, Leamer 1978, Poirier 1988, Poirier 1995 on the theoretical justifications for doing so, and the recent development of Markov chain Monte Carlo techniques has greatly increased the feasibility of Bayesian methods of inference.…”

How Likelihood and Identification Went Bayesian

“…Against the enthusiasm of Savage and Lindley, of the econometricians Leamer (1978) and Poirier (1988), is the caution of Gelman, Carlin, Stern and Rubin (1995, pp. 9, 190) for whom the principle is of limited interest and its suggestion that analysis be done \without further regard to how the experiment was actually planned or performed" potentially misleading.…”