Multivariate Versus Multinomial Probit: When are Binary Decisions Made Separately also Jointly Optimal?
Dale J. Poirier*, Deven Kapadia
Last modified: %2012-%06-%22
Abstract
We provide an analysis of the question in the title in terms of a bivariate probit framework representing two (possibly correlated) separate decisions, and a multinomial probit framework representing the four possible outcomes viewed as one joint decision. We offer a Bayesian treatment that builds on Weeks and Orme (1998) and Di Tommaso and Weeks (2000) who showed that the bivariate probit corresponds to a singular four-dimensional multinomial probit under testable restrictions. We also discuss extensions to trivariate and quadrivariate probit.