Interests: Industrial Organization, Applied Micro, Experiments, Market Design, Machine Learning, Health Economics, and Agentic AI
This project evaluates affordable housing allocation as a one-sided matching market in which households face significant search and application costs before entering decentralized lotteries. Using a structural model and counterfactual simulations, I compare the status quo uncoordinated waitlists to centralized mechanisms (Deferred Acceptance and Top Trading Cycles) under varying friction regimes. The main result is that participation frictions are the primary driver of inequity: eliminating search and application burdens can remove most justified envy with only modest changes in gross match welfare. In contrast, in highly congested markets like Massachusetts, centralizing the allocation rule alone (e.g., Deferred Acceptance) is close to neutral, while Top Trading Cycles increases welfare but at substantial fairness costs.
This project introduces a practical method for measuring preferences in settings where revealed-preference data are scarce. I construct renter “personas” from the American Community Survey and use a state-of-the-art large language model to evaluate affordable housing opportunities on a 0–10 suitability scale, interpreting these scores as cardinal utilities. The synthetic utilities show strong within-house alignment and meaningful heterogeneity across households, and they respond intuitively to key attributes (rent, school quality, safety, commute time). The approach provides a scalable tool for market design and policy analysis when direct surveys or administrative preference data are unavailable.
Integrated into my JMP
Medicare’s major coverage programs share providers, creating scope for spillovers across segments. We study whether incentives in the Medicare Shared Savings Program (MSSP) spill over from Traditional Medicare to Medicare Advantage (MA). MSSP can affect MA through two channels. First, a benchmark channel: because MA benchmarks are tied to local fee-for-service spending, MSSP-induced spending reductions mechanically lower MA plan payments. Second, a provider channel: because providers treat both Traditional Medicare and MA beneficiaries, MSSP-induced changes in care delivery may spill over directly to MA enrollees. We focus on the provider channel. To isolate it, we condition on plan-level MA benchmarks and exploit quasi-experimental variation generated by the 2019 Pathways to Success reform, which changed the timing of non-rebasing years across MSSP cohorts and thereby altered the strength of savings incentives. Combining this policy variation with pre-reform county ACO shares and baseline plan enrollment weights, we construct a Bartik-style instrument for plan-level exposure to MSSP activity. We find that greater MSSP exposure lowers MA plans’ expected cost of coverage and increases plan generosity. A 1% increase in exposure reduces expected cost per beneficiary by 0.17%, increases rebates by 2.1%, lowers out-of-pocket maxima by 0.37%, and raises the probability of offering a zero-premium plan by 0.57 percentage points. The results imply that MSSP affects MA not only mechanically through benchmarks, but also through provider behavior that spills over across Medicare programs.