Working Papers

Breaking Bad Opioid Sorting, 2026.
with Han Ng, and Jia Xiang

• Abstract: In many markets, consumers rely on experts for access to goods and services. Consumer and expert preferences over outcomes are linked through consumers’ choice of experts, or sorting. We investigate how consumer sorting contributes to observed differences in outcomes and its implications for expert-targeted policies. Using rich employer-sponsored health insurance claims data on opioid prescriptions for chronic pain, we first show that prescription intensity is highly dispersed across physicians. We decompose the variance of physicians’ prescription decisions and find that patient sorting is three times as important as physicians’ inherent prescription propensity for prescription intensity dispersion. Most of this sorting cannot be justified on medical grounds. We develop and estimate an equilibrium model of patient choice of physicians and physicians’ prescription decisions. Patients optimally choose their physicians based on both their opioid preferences and their expectations of physicians’ prescription decisions. Our counterfactual analysis shows that expert-targeted policies to curb non-medically grounded opioid prescribing will be severely attenuated by patient resorting. We propose an alternative policy that eliminates sorting based on non-medically grounded preferences, while preserving appropriate care for patients with medical needs for prescription opioids.

Optimal Estimation of Discrete Choice Demand Models with Consumer and Product Data, 2025.
with Paul Grieco, Charles Murry, and Joris Pinkse
Revisions requested at Econometrica

• We propose a likelihood-based estimator with exogeneity restrictions for BLP-type demand models combining microdata, product shares, and prices.
• Our estimator is implemented in the easy-to-use companion Julia package Grumps.jl. Install it by typing using Pkg; Pkg.add("Grumps") in the Julia REPL.
• previously circulated as Conformant and Efficient Estimation of Discrete Choice Demand Models

Dispersion, Discrimination, and the Price of Your Pickup, 2024.

• Abstract: Using repeat purchase data on pickup trucks, I establish that the same consumers pay persistently high or persistently low prices across vehicle purchases. Less than 1% of this persistence can be explained by demographics. This result suggests that dealers use consumer information beyond coarse demographics to personalize prices. Using a novel discrete choice model with personalized pricing, I study the role of consumer information firms use for pricing in the welfare effects of price discrimination. To do so, I overcome a common problem in settings with transaction data: personalized prices of non-chosen alternatives are unobservable. I solve this problem by recovering unobserved personalized prices and consumer-specific price sensitivity from the observed transaction price via firms’ first-order conditions. I simulate two counterfactuals: uniform pricing and price discrimination based on coarse demographic groups. Compared to uniform pricing, personalized pricing increases profits and total welfare but, on average, harms consumers. On the other hand, compared to uniform pricing, price discrimination based only on demographics is not profitable. This highlights the importance of the amount of consumer information firms can use for pricing in the welfare effects of price discrimination.