Grouped Model Averaging for Finite Sample Size

This paper studies grouped model averaging methods for finite sample size situation. Sufficient conditions under which the grouped model averaging estimator dominates the ordinary least squares estimator are provided. A class of grouped model averaging estimators, g-class, is introduced, and its dominance condition over the ordinary least squares is established. All theoretical findings are verified by simulation examples. We also apply the methods to the analysis of the grain output data of China.