scientific article; zbMATH DE number 3874403
Let \(X_ 1,...,X_ n\) be a sample from a normal distribution \(N_ p(\mu,\Sigma)\), where \(\mu\in R^ p\) and \(\Sigma\) is a positiv definite matrix, both \(\mu\) and \(\Sigma\) being unknown. It is well known that, under the loss function \(tr(\Sigma^{-1}d-I)^ 2\), the classical estimators of covariance matrix such as MLE and UMVUE are not admissible and that the best affine invariant estimator is an improved one. But the improved estimator is not explicitly obtainable and the admissibility was not discussed previously. In this paper, it is shown that the best affine invariant estimator is not admissible and an improved estimator is explicitly constructed, which is a scale invariant estimator. The technic used in this paper is a generalization of that used in improving the best invariant estimators of variance or generalized variance, i.e. generalizing a relevant proposition on quadratic functions to quadratic forms.
- Improved estimation of the covariance matrix and the generalized variance of a multivariate normal distribution: some unifying results
- Improvement on the best invariant estimators of the normal covariance and precision matrices via a lower triangular subgroup
- On the best equivariant estimator of covariance matrix of a multivariate normal population
- Improving on the best affine equivariant estimator of the ratio of generalized variances
- Estimation of a covariance matrix under Stein's loss
- Improving on the best affine equivariant estimator of the ratio of generalized variances
- Improved minimax estimation of the bivariate normal precision matrix under the squared loss
- Improvement on the best invariant estimators of the normal covariance and precision matrices via a lower triangular subgroup
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3339931)