Primary Submission Category: Difference in Differences
Sequential Synthetic Difference in Differences
Authors: Aleksei Samkov, Dmitry Arkhangelsky,
Presenting Author: Aleksei Samkov*
We study the estimation of treatment effects of a binary policy in environments where the rollout of the treatment is staggered. We show that the identification problem for a particular factor model can be solved using the Synthetic Difference in Difference (SDiD) method. We analyze the class of the corresponding estimators and connect their asymptotic behavior to a class of oracle estimators. We derive the optimal oracle estimator in this class and use this connection to propose a new procedure — sequential SDiD — and show that it is asymptotically unbiased, normal, and efficient. The method developed in this paper presents a natural alternative to the conventional DiD strategies in staggered adoption designs.