A0407
Title: Fixed-effect estimators under strict exogeneity for averages of unknown heterogeneous effects
Authors: Myoung-jae Lee - Korea University (Korea, South) [presenting]
Abstract: "Fixed-effect estimator (FIX)" with a constant treatment effect specification is popular with panel data, but treatment effects are unknowingly heterogeneous in reality, not constant; also, the panel model can be misspecified in other ways. For these problems, the following points are made using nonparametric "causal reduced forms". Firstly, under a strict exogeneity, FIX is consistent with a possibly negative weighted average of heterogeneous treatment effects despite model misspecifications, if a set of restrictive conditions holds; however, negative weights make the estimand non-causal. Secondly, a "modified FIX" is consistent with the same estimand without those restrictive conditions; this is an improvement, but the estimand is still non-causal. Thirdly, the "modified FIX using only the treatment-varying subsample" is consistent with a nonnegatively weighted average effect, which is thus causal. Simulation and empirical studies demonstrate these points.
