Minimax multiple shrinkage estimation (Q1082010)

A Stein estimator of the form \[ d_ v(Y)=Y-[(p-2)/\| Y-v\|^ 2](Y-v) \] shrinks Y towards a target \(v\in R^ p\). The paper proposes multiple shrinkage estimators (msest's) for cases where prior information suggests several different choices for the target. A special form is given by \(d_*(Y)=\sum_{k}r_ k(Y)d_ k(Y)\) where \(r_ k\) are adaptive functions of Y which place increasingly weight on the \(d_ k\) which are shrinking most. The paper analyses several situations and resulting msest's, especially minimax multiple shrinkage Stein estimators. The main results of the paper are shown to generalize easily for more realistic normal-theoretic settings.