A new class of blased estimate in linear regression

Abstract

Consider the linear regression model , Eε=0 and . Motivated by an interpretation of ridge estimate , we propose a new class of biased estimate to combat multicollinearity, where 0<d<l is a parameter and is the least squares estimate. combines the advantages of and Stein estimate . Theory and simulation results show that has the similar good property as . The advantage of over is that is a linear function of d. So the selection of d is simple.