Website CV Paper
Fields
Econometrics, Applied Micro
Job market paper
Sharp tests of instrument validity for general outcome distributions
In this paper I study nonparametric potential outcome instrumental variable models with multi-valued treatments and shape restrictions. I develop novel identification and inference results to obtain the first specification test that incorporates all observable implications in this class of models when outcomes are continuous. The main identification result shows that, even if the outcome has infinite support, the sharp testable implications of the exclusion restriction are equivalent to a finite system of linear inequalities. The resulting linear system involves parameters defined by nondifferentiable functions of the data for which existing inference procedures are not applicable. I prove that these parameters are equivalent to the expected loss of nonparametric classifiers. Leveraging this equivalence, I develop a joint inference procedure and obtain a formal test of instrument validity. I apply the proposed test to the Oregon Health Experiment (OHE). While previous research on the OHE demonstrates an impact on insurance take-up, I reject the hypothesis that the intervention affected insurance coverage only on the extensive margin. Leveraging a shape constraint that models this empirical finding, I obtain estimates of the effect of insurance on health and hospital utilization that are qualitatively consistent with previous findings but reflect the intensive margin effect of the intervention.