References & Citations
Mathematics > Statistics Theory
Title: Mixing Least-Squares Estimators when the Variance is Unknown
(Submitted on 2 Nov 2007)
Abstract: We propose a procedure to handle the problem of Gaussian regression when the variance is unknown. We mix least-squares estimators from various models according to a procedure inspired by that of Leung and Barron (2007). We show that in some cases the resulting estimator is a simple shrinkage estimator. We then apply this procedure in various statistical settings such as linear regression or adaptive estimation in Besov spaces. Our results provide non-asymptotic risk bounds for the Euclidean risk of the estimator.
Submission history
From: Christophe Giraud [view email] [via CCSD proxy][v1] Fri, 2 Nov 2007 18:36:35 GMT (78kb)
Link back to: arXiv, form interface, contact.