OPTIMAL MODEL AVERAGING BASED ON GENERALIZED METHOD OF MOMENTS

We propose a model averaging method that combines estimators from the generalized method of moments (GMM). Unlike other GMM-based model averaging procedures, this method allows all candidate models to be misspecified (not locally misspecified). We prove that when all candidate models are misspecified, the proposed method is optimal in the sense of minimizing the estimation loss; when there exists at least one correctly specified model, the method can achieve the common root-n convergence rate. Simulation experiments and an application to a housing market show the superiority of our method over other methods.