CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1764
Title: Linear empirical Bayes for hierarchical normal models with double shrinkage Authors:  Woncheol Jang - Seoul National University (Korea, South)
Hoyoung Park - Seoul National University (Korea, South) [presenting]
Abstract: Empirical Bayes shrinkage estimators are proposed for hierarchical normal models with unknown heteroscedastic errors. The main advantages are twofold. First, the proposed method consider the hierarchical normal model with unknown and unequal variance. Second, our method does not require a specific form of priors, so it is robust to misspecification of hierarchical models. We show theoretical optimality of the proposed estimator under regularity conditions. The performance of the proposed methods are illustrated via numerical studies and a case study in baseball.