Consider the canonical form of the Stein problem of estimating a vector #x3B8; on the basis of the observed vector T which isN(#x3B8;,I)-distributed. Suppose also that some of the components of #x3B8; may be zero and that we wish to simultaneously select and estimate the non-zero components. Although existing Stein estimators perform very well in terms of total squared error loss risk, they are not attuned well to this selection requirement. We introduce two new estimators which effectively address the selection requirement while also having risk performance reasonably comparable to that of Stein estimators.
展开▼
