Error de especificación en mínimos cuadrados generalizado

Abstract

When a linear regression model is misspecified by a left out variable the Generalized Least Squares (GLS) estimates are biased, and the bias depends crucially on how the excluded and the included variables are generated and not on how the true error terms are gcnerated. The bias in GLS estimates is larger than that of ordinary least squares estimates except when the left out variable has higher autocorrelation than the included variablft. The relative effidency of GLS with respect to ordinary Least Squares does not depend on the left out variable.