Website
Fields
Econometrics
Job market paper
Learning Tail Features by Random Forest: with Applications to Health Economics Datasets
This paper develops Pareto-GRF, a nonparametric estimator of conditional tail exponents using a Pareto approximation and honest generalized random forests. I establish consistency, pointwise asymptotic normality, and feasible variance estimation, separating forest localization bias from finite-threshold approximation bias. An application to the top one percent of medical expenditures in the 2016–2023 Medical Expenditure Panel Survey documents heterogeneity in tail severity associated with age and cancer diagnoses, conditional on exceeding a high spending threshold.