Dynamic Synthetic Controls: Accounting for Varying Speeds in Comparative Studies

• In Event: Methodological Advances in Causal Inference

Abstract

Synthetic controls are widely used to estimate the causal effect of a treatment. However, synthetic controls do not account for the different speeds at which units may react to changes. Reactions may be inelastic or `sticky’, and hence slower, because of different regulatory, institutional, or political environments. We show that these different speeds can lead to biased estimates of causal effects and that it is not possible to address the problem by controlling for speed. We address this problem by developing a dynamic synthetic control method that accounts for varying speeds in time series and thereby improves synthetic control estimates. We apply the new method to re-estimate the effects of terrorism on income (Abadie 2003), of tobacco laws on consumption (Abadie 2010), and of the German reunification on GDP (Abadie 2015). We also evaluate the performance of the method on a Monte-Carlo simulation of synthetic data sets. We find that our method reduces the errors in the estimates of the true treatment effects by up to 70% compared to traditional synthetic controls. This gain in efficiency improves our ability to make robust inferences.

Authors

• Jian Cao, Trinity College, Dublin

• Thomas Chadefaux, Trinity College Dublin