Restricted Linear Models: Which Estimator Performs Better?

• Activity Number: 130
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economic Statistics Section
• Author(s): Luis Frank*+
• Companies: University of Buenos Aires
• Keywords: Restricted estimatores ; Small Samples ; Simulation ; Regression

Abstract

Given y=Xß+e - e distributed N(0,V) - subject to Rß=r, the paper evaluates the efficiency of different "feasible" restricted heteroskedastic estimators with small samples. The findings suggest that estimating V by a two-step procedure or by the Oberhofer-Kmenta procedure and introducing it either in the generalized least squares (RGLS) solution or the restricted least squares (RLS) is advisable when working with small samples (n<50) and well conditioned Xs, compared with other estimators. If samples are very small and X is well conditioned, both RGLS and RLS may not outperform the LS solution. However, this last result does not hold for ill-conditioned Xs. The MINQUE estimator is definitely not recommended when n<50.