Primary Submission Category: Difference in Differences, Synthetic Control, Methods for Panel and Longitudinal Data
Granular Synthetic Control
Authors: Yuhang Zhang, Haitian Xie,
Presenting Author: Yuhang Zhang*
Unlike cross-sectional or short-panel causal inference methods such as unconfoundedness, instrumental variables, and difference-in-differences, which delineates nonparametric identification and semiparametric estimation stages, the classic synthetic control method estimates the counterfactual outcome directly through weighted combinations of control units, suppressing a distinct nonparametric identification stage. In this paper, we introduce a refined synthetic control method that, with access to granular-level data, develops nonparametric identification and then proceeds to semiparametric estimation. Leveraging micro-level data allows for more flexible, covariate-specific weighting, and requires only two time periods for identification. We explore this methodology in both panel and repeated cross-sectional settings, developing doubly robust identification and proposing semiparametric estimation based on double/debiased machine learning methods. The effectiveness of our approach is demonstrated through an empirical application.