A0331
Title: Binary response model with many weak instrumental variables
Authors: Dakyung Seong - University of Sydney (Australia) [presenting]
Abstract: An endogenous binary response model with many weak instruments is considered. In contrast with linear simultaneous equations models, binary response models with endogenous regressors and many weak instruments have received minimal attention from researchers despite their practical importance. Two consistent and asymptotically normally distributed estimators are proposed: a regularized conditional maximum likelihood estimator (RCMLE) and a regularized nonlinear least square estimator (RNLSE), using regularization in the first stage. Standard regularization schemes such as Tikhonov regularization and Spectral cut-off can be employed for the proposed estimators, and consistent estimators of their asymptotic variances are also provided. Monte Carlo simulations show that both the RCMLE and the RNLSE outperform existing estimators when many weak instruments are present. We apply our estimators to two empirical examples to demonstrate their empirical relevance.
