Research
Research
Gaussian Rank Correlation and Regression [Paper]
with Dante Amengual and Enrique SentanaIn A. Chudik, C. Hsiao and A. Timmermann (eds.) Essays in honor of M. Hashem Pesaran: Panel Modeling, Micro Applications and Econometric Methodology, Advances in Econometrics 43B, pp. 269-306, Emerald 2022Policy Learning with Selected Quasi-experimental Data (JMP) [Draft Coming Soon]
Abstract: Many empirical research efforts aim to provide policymakers with optimal treatment assignment rules using quasi-experimental methods, such as the Difference-in-Differences (DiD) approach. When observational data is used, researchers may be concerned that the data is not randomly selected from the target population. In this paper, I discuss how to learn the optimal treatment assignment rules robustly when the bias of the observed sample is upper bounded in the selection sense — that is, for observations with different characteristics in the target population, the ratio of being observed or not observed in the sample is upper bounded by some constant. Different welfare criteria, including utilitarian maximization, maximizing gains, and minimizing regrets, are considered, and the worst-case welfare-maximizing rules are derived. For estimation and inference, I propose estimators based on Conditional Value-at-Risk (CVaR) estimators and provide results for both panel and repeated cross-sectional data structures. Finally, I illustrate the practical application by revisiting a quasi-experiment on tariff reductions between South Africa and Mozambique, demonstrating how to obtain the optimal policy under various welfare criteria while accounting for sample bias.
"Sensitivity Analysis for Functional Parallel Trends Assumption"