Portal de Periódicos da CAPES

Comment on: Reflections on the Probability Space Induced by Moment Conditions with Implications for Bayesian Inference

2016; Oxford University Press; Volume: 14; Issue: 2 Linguagem: Inglês

10.1093/jjfinec/nbv011

1479-8417

John Geweke,

Complex Systems and Time Series Analysis

This contribution reflects the author's on-going study of issues that arise for Bayesian inference using observed functions of the observables that the model addresses directly. In this paper, "model" becomes moment conditions, and much of the same technology can be applied. The questions that arise, at least in the absence of a proper prior distribution, are deep and difficult. The paper states that they are much alleviated given a proper prior distribution. I agree. To the extent that investigators continue to provide Bayesian interpretation of moment condition estimates in the absence of an explicit prior distribution, this is important work. Without it, things can go off the rails, as the paper points out. There is a rather long list of things that can, and do, go wrong when prior distributions are not stated explicitly. Even if the impact of the current paper were only to firmly establish the case studied as having these characteristics, that is a significant contribution. More important, such trainwrecks illustrate well-established pitfalls of non-Bayesian reasoning—I learned the hard way, by falling into these pits more than once and then looking carefully at what happened. If this paper is responsible for moving some investigators to this point more quickly, then from my perspective that would be an even more important contribution to the profession.