AN ADAPTIVE EMPIRICAL BAYES ESTIMATOR OF THE MULTIVARIATE NORMAL-MEAN UNDER QUADRATIC LOSS

For shrinkage estimators to achieve significant risk improvement over their traditional competitors, one must identify the region or subspace where the unknown location vector lies or is thought likely to lie a priori. When vague or conflicting priors suggest that a broad class of estimators may be effective under a squared-error-loss measure, new minimax or near-minimax empirical Bayes estimators are proposed that make use of Stein's unbiased estimator of the risk to identify the optimum risk-effective shrinkage estimator. © 1990.