Title: Inference for high-dimensional instrumental variables regression

Abstract: This paper concerns statistical inference for the components of a high-dimensional regression parameter despite possible endogeneity of each regressor. Given a first-stage linear model for the endogenous regressors and a second-stage linear model for the dependent variable, we develop a novel adaptation of the parametric one-step update to a generic second-stage estimator. We provide conditions under which the scaled update is asymptotically normal. We then introduce a two-stage Lasso procedure and show that the second-stage Lasso estimator satisfies the aforementioned conditions. Using these results, we construct asymptotically valid confidence intervals for the components of the second-stage regression coefficients. We complement our asymptotic theory with simulation studies, which demonstrate the performance of our method in finite samples.