Research
Working Papers
-
Distributional impacts of centralized school choice (with Suk Joon Son)
Abstract: Informational frictions in centralized school choice can significantly influence its distributional consequences. Recognition of such frictions is also necessary to accurately measure welfare. We build a model of school applications, allowing applicants to consider only a limited set of schools and to have mistaken beliefs about their admission chances. Quasi-experimental variation and rich information in students’ rank-ordered lists enable identification. Utilizing this model, we evaluate the impacts of centralized school choice in New York City on racial segregation and equity in welfare, decomposing the contributions of the frictions and the preferences of students and schools. We also quantify matching stability and deviations from truthful reporting. Our results show that while school choice improves welfare across races, limited consideration substantially compromises these gains, particularly for Black and Hispanic students. A counterfactual policy involving personalized school recommendations designed using our model is projected to recover 20–36% of the welfare losses.
-
Computationally feasible identification-robust inference on discrete choice demand Job Market Paper
Abstract: The possibility of instruments being weak is a central debate in empirical industrial organization, particularly in the context of demand estimation. This paper applies an identification-robust confidence set proposed by Andrews (2018) in the context of the discrete choice demand model by Berry, Levinsohn, and Pakes (1995), offering a technique to enhance its computational feasibility. Given that the computational cost stems from grid searches over a large dimensional parameter space, I provide a condition under which the dimensionality of the required grid can be significantly reduced. Monte Carlo simulations show that the robust confidence set obtained by utilizing my dimension reduction technique achieves desired properties when the condition is met. Even when the potentially restrictive condition, namely homoscedasticity, is violated, this technique is able to achieve a close approximation of the confidence set that would have been obtained through a full grid search. This suggests that the technique can also guide the formation of a full grid, especially when a researcher lacks prior information.
-
Standard errors in demand estimation with Hausman instruments
Abstract: This paper shows how the common practice of calculating the standard errors of demand estimators can be inaccurate when using Hausman instruments. The typical method ignores a correlation pattern arising from essential endogeneity of Hausman instruments, usually underestimating the true variance of the estimators as a consequence. I explore methods to robustly estimate the variance of the demand estimator for some popular classes of Hausman instruments, including region-based and adjacency-based instruments. Monte Carlo simulations are conducted to evaluate their performances. The results suggest using moderately scoped Hausman instruments and correlation-robust variance estimators, to reduce the true variance of the demand estimator and to accurately estimate the variance.
Publications
- Long-term changes in old-age incomes: Results from cohort analyses (2015) with Chulhee Lee, Journal of Korean Economic Studies, 33(3), 5–34, in Korean.