Improvement on the Best Invariant Estimators of the Normal Covariance and Precision Matrices via a Lower Triangular Subgroup

Abstract

This paper addresses the problems of estimating the normal covariance and precision matrices. A commutator subgroup of lower triangular matrices is considered for deriving a class of invariant estimators. The class shows inadmissibility of the best invariant and minimax estimator of the covariance matrix relative to quadratic loss. Also, in estimation of the precision matrix, a dominance result is given for improvement on a minimax estimator relative to the Stein loss.