Scholion.ai

Monte Carlo Approximations in Bayesian Decision Theory

Jun Shao

Journal of the American Statistical Association · 1989 · Volume 84 · Issue 407

DOI: 10.1080/01621459.1989.10478826

Abstract

In decision-making problems, the Bayesian action and its posterior expected loss usually cannot be obtained analytically. In this article we study a Monte Carlo method for approximating the Bayesian action and its posterior expected loss. The Monte Carlo approximation to the Bayesian action is obtained through (a) approximating the posterior expected loss function by using the Monte Carlo integration method and (b) searching the minimum of the approximated posterior expected loss function. As the Monte Carlo sample size diverges to infinity, the Monte Carlo approximations are shown to be convergent in general situations. The rates of the convergence are also obtained under some regularity conditions on the loss function. Two accuracy measures of the Monte Carlo approximations are proposed. Some examples are presented for illustration.

Enable JavaScript for citation tools and additional paper interactions.