Asymptotic variance estimation in multivariate distributions (Q1175675)

The point estimation problem of the parameter \(\sigma^ 2\) under a general loss function \(W\) in the case of a multivariate location-scale family \[ P_{\mu\sigma}(A)=P(\sigma^{-1}(A-\mu)), \] involving an unknown location vector \(\mu\) and unknown scale parameter \(\sigma^ 2\), is considered. The asymptotic admissibility of the estimators of \(\sigma^ 2\) of the form \[ \delta(\bar Y,S)=c_ 0S^ 2(1-n^{- 1}g(\bar Y n^{1/2}/S)),\;\;where \] \[ S^ 2=\sum_ 1^ n(Y_ j-\bar Y)^ T(Y_ j-\bar Y),\;\;\bar Y=n^{-1}\sum_ 1^ n Y_ j,\;Y_ 1,\dots,Y_ n\in P_{\mu,\sigma} \] and \(c_ 0\) is such that \[ \min_ C E_{01}W(cS^ 2)=E_{01}W(c_ 0S^ 2), \] is considered. It is demonstrated that this problem is related with the admissibility of the estimator of random quadratic polynomials based on normal random vectors.