UC San Diego

Chapter 1 studies the instrument validity for local average treatment effects. we provide a testable implication for instrument validity in the local average treatment effect (LATE) framework with multivalued treatments. Based on this testable implication, we construct a nonparametric test of instrument validity in the multivalued treatment LATE framework. The test is asymptotically consistent. The size of the test can be promoted to the nominal significance level over much of the null, indicating a good power property. Simulation evidence is provided to show the good performance of the test in finite samples. Chapter 2 constructs improved nonparametric bootstrap tests of Lorenz dominance based on preliminary estimation of a contact set. Our tests achieve the nominal rejection rate asymptotically on the boundary of the null; that is, when Lorenz dominance is satisfied, and the Lorenz curves coincide on some interval. Numerical simulations indicate that our tests enjoy substantially improved power compared to existing procedures at relevant sample sizes. Chapter 3 proposes a sieve focused GMM (SFGMM) estimator for general high-dimensional semiparametric conditional moment models in the presence of endogeneity. Under certain conditions, the SFGMM estimator has oracle consistency properties and converges at a desirable rate. We then establish the asymptotic normality of the plug-in SFGMM estimator for possibly irregular functionals. Simulation evidence illustrates the performance of the proposed estimator.