One-step L-estimators for the linear model (Q1109452)

A linear model of the form \(Y_ j=x_ j'\theta_ 0+e_ j\), \(j=1,...,n\), where \(Y_ 1,...,Y_ n\) are observations on a response variable, \(x_ j'=(1,x_{j2},...,x_{jp})\) \(j=1,...,n\) is a sequence of p-vectors \((p>1)\), \(\theta_ 0\) is an unknown parameter vector to be estimated and \(e_ j\), \(j=1,...,n\) is a sequence of independent and identically distributed random variables, is considered. Given a preliminary estimate of \(\theta_ 0\), a class of estimators based on linear combinations of order statistics is proposed and investigated with respect to asymptotic properties. In particular, under a set of regularity conditions, the estimators are shown to be consistent and asymptotically normally distributed.