This paper develops a gradient-based optimization method to estimate stochastic dynamic models in economics and finance and assess identification globally. By extending the state space to include all model parameters and approximating the mapping between parameters and moments, we only need to solve the model once to structurally estimate parameters. We approximate the mapping between parameters and moments by training a neural network on model-simulated data and then use this mapping to find the set of parameters that minimizes a function of the distance between model and data moments. We show how the mapping between parameters and moments can also be used to assess identification globally, detecting issues that a local diagnostic would miss. We illustrate the algorithm by solving and estimating a dynamic corporate finance model with endogenous investment, costly equity issuance, and capital adjustment costs. In this application, our method reduces the estimation time from many hours to a few minutes.

In many areas of economics, relatively simple models developed for insight are used as quantitative guides. We study the accuracy of such simple quantitative guidance in an area where it has been widely adopted — lifecycle portfolio choice among stocks, bonds, and liquid accounts — by developing a machine-learning algorithm to solve for optimal portfolio choice in a calibrated lifecycle model that includes many features of reality modeled only separately in previous work. Both for optimizing households and for households that under-save, the average fully optimal portfolio at each age conforms well to current simple age-dependent prescriptive rules until shortly before retirement, validating existing analyses. We further show that the consumption-equivalent losses from conditioning portfolio shares on age alone are substantial, around 2 to 3 percent of consumption. Fully optimal equity shares have substantial heterogeneity, particularly by wealth level, state of the business cycle, and dividend-price ratio, implying substantial gains to further customization in these dimensions.

Media: Wall Street Journal, Morningstar

Mortgage borrowers are "locked in": forgoing moves to keep low mortgage rates. We study the general equilibrium effects of mortgage lock-in on housing markets. We provide causal evidence that lock-in increases prices, particularly in expensive areas, because locked-in borrowers would otherwise demand less housing. We design a spatial housing ladder model with long-term mortgages, generating a distribution of locked-in rates and equilibrium effects on mobility and prices consistent with the data. A temporary rate hike causes lock-in, increasing housing demand and prices, especially in expensive areas. A $10k tax credit to starter-home sellers modestly unlocks mobility while increasing trade-up home prices.

​

Media: NPR Planet Money (The Indicator), NPR Planet Money, NPR Morning Edition

One in seven Americans carry medical debt, with $88 billion reported on consumer credit reports. In April 2023, the three major credit bureaus stopped reporting medical debts below $500. We study the effects of this information deletion on consumer credit scores, credit limits and utilization, repayment behavior, and payday borrowing. Using a machine learning model, we show that small medical debts are not meaningfully predictive of defaults, suggesting their deletion should have minimal effect on lending decisions. We test this prediction using two complementary research designs. First, a regression discontinuity analysis comparing individuals just above and below the $500 threshold finds no direct benefits from the information deletion, ruling out small changes in credit access and financial health. Second, to assess potential indirect effects, we classify consumers based on whether their predicted probability of default increases or decreases when medical debts are deleted. A difference-in-differences analysis comparing these groups before and after the 2023 policy change reveals no evidence of negative spillover effects. Finally, we show that larger medical debts ( greater than $500) are also not meaningfully predictive of default, suggesting that eliminating medical debts entirely from credit reports, as planned under a January 2025 decision by the Consumer Financial Protection Bureau, is unlikely to affect credit outcomes.