Research projects

During the second year, I continued to work on two projects that investigate the ideas outlined inthe initial proposal. While the drafts are not ready yet, there is substantial development in both of them, and I expect to produce drafts by the Spring.

IThe first project is now called “ On Policy Evaluation under Sequential Exogeneity” and is joint with a Ph.D. student at Berkeley, Yahu Cong. It develops the ideas described in the previous report. In particular, we demonstrate how the pretreatment data can be used to quantify the worst case bias caused by deviations from strict exogeneity. To do so, we consider an asymptotic regime where such deviations are small and show how to emply the observed data on the adoption process to calculate the bias in this regime. Our results allow researchers to use algorithms proposed by Rambachan and Roth (2022) in models with sequential exogeneity and provide a design-based foundation for their approach. Currently, we are working on an empirical application, and once it is finalized, we will be able to produce the draft. The intermediate results of this project were presented twice during the past year. The first time they were presented at the Economics Department at Stanford University in July 2022, and the second time at CEMFI Fall Econometrics Conference in November, 2022.

The second project, joint with David Hirshberg and called “Randomization-based inference for Synthetic Control estimators,” focuses on the properties of the Synthetic Control Method (SCM) in models with sequential exogeneity. During the last year, we were able to obtain a tighter characterization of the asymptotic behavior of this estimator and demonstrate that without further regularization, it always exhibits asymptotic bias, which dominates the variance and thus cannot be used to conduct inference. Currently, we are developing a proposal that addresses this problem by appropriate regularization, essentially assuming that recent periods are more relevant for policy adoption than the more distant ones. We expect that the improved estimator will have better properties and currently work on showing that formally. The intermediate results of this project were presented twice during the past year. The first time they were presented at the Synthetic Control conference at Princeton, in June 2022, and the second time at the ICSDS conference in Florence in December 2022.