SMALL SAMPLE PROPERTIES OF THE TWO-STAGE AITKEN PREDICTOR UNDER SPECIFICATION ERROR

Abstract

In this paper, we examine the small sample properties of the two-stage Aitken estimator under the specification error (omitted variables). It is shown analytically and numerically that the predictor incorporating the two-stage Aitken estimator becomes less efficient than the predictor incorporating the ordinary least squares estimator as the magnitude of the specification error gets larger, even if heteroscedasticity is severe. In addition, an empirical example is given in the final section.