Primary Submission Category: Difference in Differences
On policy evaluation with sequential exogeneity
Authors: Dmitry Arkhangelskiy, Yahu Cong,
Presenting Author: Dmitry Arkhangelsky*
We develop a systematic approach to causal inference with panel data under the assumption of sequential exogeneity. First, we propose a causal model that incorporates strict and sequential exogeneity. Focusing on staggered adoption design, we show that no linear estimator can generally identify a convex combination of treatment effects. This result contrasts sharply with recent literature on strictly exogenous models and older panel data literature on sequential exogeneity. To mitigate this negative result, we propose two approaches. First, we show how to quantify the worst-case bias caused by sequential exogeneity by connecting it to the deviation from the parallel trends. Second, we show that a convex combination of treatment effects can be identified if we have access to an ex-ante control group.