Abstract

This paper is concerned with estimation and inference on structural parameters in instrumental variable regression models that involve not only many weak instruments but also many included exogenous regressors. This setting is practically important to accommodate saturated specifications for proper interpretation as the local average treatment effect and fixed effect dummies, for example. We propose the cross-fitted instrumental variable estimator, establish its consistency and asymptotic normality under mild assumptions, and derive a consistent variance estimator for valid inference on the structural parameters. Simulation results illustrate desirable finite sample performances of the proposed method.