Properties of ordinary least squares estimators in regression models with nonspherical disturbances

Abstract

A general discussion is presented of the properties of the OLS estimator in regression models where the disturbances do not have a scalar identity covariance matrix. Several new and interesting characterizations are provided together with a synthesis of existing results. A distinguishing feature of the paper is the unified treatment of the issues of efficiency and inference. The theoretical material is illustrated by examples covering a wide range of models of interest to applied econometricians. © 1992.