English

Covariance matrix estimation under data-based loss

Statistics Theory 2020-12-23 v1 Applications Statistics Theory

Abstract

In this paper, we consider the problem of estimating the p×pp\times p scale matrix Σ\Sigma of a multivariate linear regression model Y=Xβ+EY=X\,\beta + \mathcal{E}\, when the distribution of the observed matrix YY belongs to a large class of elliptically symmetric distributions. After deriving the canonical form (ZTUT)T(Z^T U^T)^T of this model, any estimator Σ^\hat{ \Sigma} of Σ\Sigma is assessed through the data-based loss tr(S+Σ(Σ1Σ^Ip)2)(S^{+}\Sigma\, (\Sigma^{-1}\hat{\Sigma} - I_p)^2 )\, where S=UTUS=U^T U is the sample covariance matrix and S+S^{+} is its Moore-Penrose inverse. We provide alternative estimators to the usual estimators aSa\,S, where aa is a positive constant, which present smaller associated risk. Compared to the usual quadratic loss tr(Σ1Σ^Ip)2(\Sigma^{-1}\hat{\Sigma} - I_p)^2, we obtain a larger class of estimators and a wider class of elliptical distributions for which such an improvement occurs. A numerical study illustrates the theory.

Keywords

Cite

@article{arxiv.2012.11920,
  title  = {Covariance matrix estimation under data-based loss},
  author = {Anis M. Haddouche and Dominique Fourdrinier and Fatiha Mezoued},
  journal= {arXiv preprint arXiv:2012.11920},
  year   = {2020}
}