Automatic Debiased Machine Learning for Instrumental Variable Models of Complier Treatment Effects

We propose debiased machine learning estimators for complier parameters, such as local average treatment effect, with high dimensional covariates. To do so, we characterize the doubly robust moment function for the entire class of complier parameters as the combination of Wald and $\kappa$ weight formulations. We directly estimate the $\kappa$ weights, rather than their components, in order to eliminate the numerically unstable step of inverting propensity scores of high dimensional covariates. We prove our estimator is balanced, consistent, asymptotically normal, and semiparametrically efficient, and use it to estimate the effect of 401(k) participation on the distribution of net financial assets.