Testing instrument validity for LATE identification based on inequality moment constraints

This paper proposes bootstrap tests for the validity of instrumental variables (IV) in just identified treatment effect models with endogeneity. We demonstrate that the IV assumptions required for the identification of the local average treatment effect (LATE) allow us to both point identify and bound the mean potential outcomes (i) of the always takers (those treated irrespective of the instrument) under treatment and (ii) of the never takers (never treated irrespective of the instrument) under non-treatment. The point identified means must lie within their respective bounds, which provides four testable inequality moment constraints for IV validity. Furthermore, we show that a similar logic applies to testing the assumptions needed to identify distributional features (e.g., local quantile treatment effects). Finally, we discuss how testing power can be increased by imposing dominance/equality assumptions on the potential outcome distributions of different subpopulations.