Abstract:
In this paper we investigate the finite sample risk performance of feasible generalised least squares estimators applied in models with serially correlated error terms. The risk functions of the ordinary least squares, generalised least squares and feasible generalised least squares estimators are derived under the asymmetric Linear-Exponential loss function. A numerical evaluation using simulation is used to compare the risk functions. Our numerical results show that the relative risk gains of the feasible generalised least squares estimators over the ordinary least squares estimator increases with higher loss asymmetry, particularly for larger serial correlation coefficients.
