RIEB, Kobe University

A Pragmatic Method for Model-Selection Based on the Widely Applicable Bayesian Information Criterion

In this study, we discuss the use of the Widely Applicable Bayesian Information Criterion (WBIC) when prior information is unknown. We provide a numerical example whereby if the prior is arbitrarily set to be tight or weak, the marginal likelihood can fail to find the best econometric model, even though its likelihood function is consistent with true data-generating process. Given this fact, we propose combining WBIC and a noninformative prior. This procedure objectively selects econometric models in a Bayesian context, and yields a reasonable result in such a situation.