Asymptotic risk comparison of improved estimators for normal convariance matrix
Asymptotic risk comparison of improved estimators for normal convariance matrix
复制标题
正态协方差矩阵改进估计量的渐近风险比较
DOI：
10.21099/tkbjm/1496159454
复制
发表时间：
1982
影响因子：
0.7
通讯作者：
N. Sugiura
中科院分区：
文献类型：
--
作者：
N. Sugiura
本文计算了n维正态分布协方差阵I的经验Bayes估计2H（Haff [5]）的渐近风险，并与James和Stein的极小极大估计IJS的渐近风险作了比较。对于p^6，证明了SJS总是渐近地优于IH，尽管前导项是相同的。对任意p中的某个I，本文提出了新的渐近占优SJS的估计。给出了一些数值比较。普通estimators10和minimax estimators2JS的确切风险也计算和比较渐近的近似是优秀的。
Asymptotic risks of the empirical Bayes estimators 2H by Haff [5] for a covariance matrix I in a />-dimensionalnormal distributionare computed and compared with that of James and Stein'sminimax estimatorsIJS. For p^6, it is shown that SJS are always betterthan IH asymptotically,though the leading terms are the same. New estimatorswhich dominate SJS for some I in any p asymptoticallyare proposed. Some numerical comparisons are given. Exact risks for ordinary estimatorsl0 and minimax estimators2JS are also computed and compared with asymptotic ones for which the approximations are shown to be excellent.