Time-Weighted Difference-in-Differences: Accounting for Common Factors in Short T Panels

This paper proposes a time-weighted difference-in-differences (TWDID) estimation approach that is robust against interactive fixed effects in short T panels. Time weighting substantially reduces both bias and variance compared to conventional DID estimation through balancing the pre-treatment and post-treatment unobserved common factors. To conduct valid inference on the average treatment effect, I develop a correction term that adjusts conventional standard errors for weight estimation uncertainty. Revisiting a study on the effect of a cap-and-trade program on NOx emissions, TWDID estimation reduces the standard errors of the estimated treatment effect by 10% compared to a conventional DID approach. In a second application I illustrate how to implement TWDID in settings with staggered adoption of the treatment.