Model Selection and Model Averaging in Nonparametric Instrumental Variables Models

This paper considers the problem of choosing the regularization parameter and the smoothing parameter in nonparametric instrumental variables estimation. We propose a simple Mallows’ Cp-type criterion to select these two parameters simultaneously. We show that the proposed selection criterion is optimal in the sense that the selected estimate asymptotically achieves the lowest possible mean squared error among all candidates. To account for model uncertainty, we introduce a new model averaging estimator for nonparametric instrumental variables regres-sions. We propose a Mallows criterion for the weight selection and demonstrate its asymptotic optimality. Monte Carlo simulations show that both selection and averaging methods generally achieve lower root mean squared error than other existing methods. The proposed methods are applied to two empirical examples, the class size question and Engel curve