Predictive Counterfactuals for Treatment Effect Heterogeneity in Event Studies with Staggered Adoption

Abstract

I propose a machine learning-based approach for estimating treatment effect heterogeneity in event studies with staggered adoption. The first step is to use machine learning algorithms to predict counterfactual outcomes in the absence of treatment. A distribution of effects can then be obtained by taking the difference between realized and counterfactual outcomes. From this distribution, it is possible to estimate the average treatment effect on the treated (ATT) or conditional ATT, depending on the researcher's objective. With simulations, I show that the ML estimates are unbiased and more efficient than estimates from conventional approaches. My proposal serves as an alternative to standard two-way fixed effects regressions that are potentially biased in the presence of heterogeneous treatment effects. I conclude with an application to real data from a residential energy efficiency program, revealing substantial heterogeneity of energy savings depending on the types and levels of upgrades performed.

Keywords: Causal Inference, Machine Learning, Event Studies, Energy Efficiency, Air Polution

JEL Classification: C18, C55, Q49