Primary Submission Category: Synthetic Control Method
Augmented Balancing for Difference-in-Differences: A Synthesis
Authors: Apoorva Lal, Yiqing Xu, Ziyi Liu,
Presenting Author: Apoorva Lal*
We synthesize recent estimators in synthetic control literature and show that these estimators are special cases of augmented balancing in a (potentially staggered) difference-in-differences setup, differing in their choice of outcome model and the balancing scheme generating the weights. Through a series of Monte Carlo exercises, we demonstrate that, unlike other estimators that are typically fragile in certain scenarios, the synthetic difference-in-differences (SDID) estimator is remarkably robust to different data-generating processes, primarily due to improved overlap (both cross-sectional and temporal) generated by dual weights and its ability to accommodate weak signals in low-rank structures. Moreover, we show that proper data pre-processing, coupled with a flexible outcome model such as a factor model incorporating covariates, can improve the performance of SDID. We offer practical recommendations and a software implementation for practitioners.