An empirical Bayes estimator for thek-dimensional regression parameter vector that incorporates prior information in the form ofmlinear restrictions is proposed. Ifmis larger than 2 and smaller thank-2, the proposed estimator dominates the ordinary least-squares estimator as well as the Bock-Judge variant Stein-type estimator with respect to quadratic risk. The gain in risk is explicitly given in the paper.
展开▼