A0742
Title: Bayes extended estimators with shrinkage priors for multivariate normal models
Authors: Michiko Okudo - The University of Tokyo (Japan) [presenting]
Fumiyasu Komaki - RIKEN CBS (Japan)
Abstract: The focus is on constructing predictive densities for multivariate normal models with unknown mean vectors. Bayesian predictive densities based on shrinkage priors often have complex representations and its computation requires approximation by taking the average of plugin densities. We approximate Bayesian predictive densities to reduce computational time and space by projecting them onto normal models with unknown mean and unknown covariance matrices, which include the original model as a subspace. We evaluate the Kullback-Leibler risk performance of the proposed methods, and compare them with those of Bayesian predictive densities with the uniform prior. Pythagorean relation of Bayesian predictive densities and its projection is also shown.
