On predictive density estimation for location families under integrated absolute error loss (Q2405154)

The paper considers the problem of estimating a \(d\)-dimensional unimodal spherically symmetric density and establishes the existence of a clear relationship between a predictive density estimation problem and a point estimation. It is proved that for the class of log-concave densities a scale expansion, which is induced by a predictive density estimator \(q(x,y)=c^{-1}q(e c^{-1})\), \(e=xy-x\), performs better, in terms of the integrated absolute error loss than a plug-in rule under certain conditions on \( c\). A set of 9 theorems, corollaries and lemmata are proved for sustaining the claims. They are illustrated by discussing some examples.