Penalized Generalized Method of Moments with Many Weak Instrumental Variables
时间：2016-04-28 14:00 ~ 15:00
This paper addresses the issue of statistical inference for instrumental variable models that are weakly identified. We propose a nuclear (Ky Fan) norm regularized two step GMM method with growing number of weak instrumental variables. In choosing the regularization parameter we adopt the Donald and Newey (1990) criterion which aims at minimization of the MSE. The main innovation of our paper is that we rely on the ‘blessing of dimensionality’, specifically, the many weak IVs are welcoming in the efficient estimation of structural parameters. And we use the penalized GMM method to control for the bias of many weak IV. We show the asymptotic property of our estimator which is consistent and asymptotic normal with optimal variance. We provide simulation results which demonstrate that our method performs favorably compared to other existing methods in finite sample. We also apply our method in the classic returns to education studies.
About the Speaker: