BIASED CROSS-VALIDATION IN A KERNEL REGRESSION ESTIMATION

Abstract

This article is concerned with the problem of choosing a bandwidth for nonparametric regression. We consider a method based on an biased estimate of mean average squared error. It is seen that the bandwidth chosen by biased cross-validation method, is asymptotically optimal and has small sample variability. In a simulation study, we show that this bandwidth is closer to optimum bandwidth than other bandwidths when the underlying regression function is sufficiently smooth.