Research
Job Market Paper
Young workers quickly transition to higher-paying and less routine occupations within the first decade of their careers. This paper studies how on-the-job learning shapes the impacts of automation shocks within and across career paths. I develop a dynamic general equilibrium model of the labor market in which young workers sort into occupations that differ in their learning opportunities. Combining the structure of the occupational choice model with large-scale résumé data, I recover a flexible on-the-job learning technology that rationalizes observed patterns of early-career occupational mobility. The calibrated model shows how an automation shock to one occupation can induce greater automation further up the occupational career ladder. For a broad-based automation shock, adjustments in workers’ skill accumulation decisions provide an important margin of long-run labor-supply flexibility, substantially lowering the pass-through of automation exposure to relative wages.
Working Papers
with John Grigsby, Jonathon Hazell, and Abdoulaye Ndiaye
Forthcoming, American Economic Review: Insights
We introduce dynamic incentive contracts into a model of unemployment fluctuations. Our main result is that wage cyclicality from incentives does not affect the response of unemployment to productivity shocks. The response of unemployment is the same, to a first-order, in two economies: one with flexible incentive pay, and another with exogenously fixed wages. This equivalence is due to movements in effort. Under the optimal incentive contract, firms’ profits do not change when wages fall, because the effort of the worker falls too.
with Alicia Zhang
Who drives the diffusion of new technologies? We provide novel evidence that more experienced software developers use newer programming languages than their less experienced counterparts. Leveraging panel data on individual-level programming activity, we find that developers continue to adopt newer languages as they gain more coding experience. The fact that less experienced developers use older languages is surprising, as we also show that newer languages offer a salary premium. We interpret these facts through the lens of a directed job search model featuring an individual technology adoption decision. In the model, workers must pay costs to switch technology vintages. Newer technologies offer higher wages and a lower cost of switching to the technological frontier, which incentivizes continuous upgrading. Shocks that reduce aggregate productivity, increase worker turnover, and reduce labor market matching efficiency slow the diffusion of new technologies.
Work in Progress
with Xiang Zhang
How do the labor market dynamics of inventors affect firm innovation? We provide new empirical evidence on how research productivity changes following inventor job moves, both for the moving inventors and for their former co-inventors who remain at the firms they leave. We link patent records to inventors' employment histories and use a matched event-study design to compare similar inventors moving to young firms and large incumbents. We find that inventors moving from startups to large incumbents experience a 47% increase in patent output. However, their departures are followed by a 17% decline in patenting among former co-inventors who remain at startups. We find no comparable decline among co-inventors following inventor departures from large incumbents. These findings reveal an asymmetry in inventor mobility: moving to large incumbents is associated with higher productivity for inventors but lower patenting among their former collaborators at startups. To quantify the aggregate implications of these effects, we develop a model of creative destruction with inventor mobility and examine how the allocation of inventors across firms shapes innovation and firm dynamics.
Using data from the 1984-2020 Displaced Worker Surveys, I show that average earnings losses following job displacement have been decreasing over time and that this trend is driven by workers displaced from jobs in lower skill occupations. The average decline for this group of workers reflects lower rates of occupational switching following displacement, as well as smaller earnings losses conditional on switching occupations, and is present within broad sectors and occupational categories. Among occupation switchers, I find that losses have declined the most for workers with higher occupational tenure. I find no such trend among workers displaced from higher skill occupations. These findings are consistent with declining skill sensitivity in the production technology employed in lower skill occupations, consistent with recent work by Edmond and Mongey (2026). A straightforward implication of this theory—declining earnings inequality within lower skill occupations—finds preliminary support in the data. Cross-sectional earnings profiles in lower skill occupations have also flattened over this time period, suggesting a role for dynamic interactions between declining skill sensitivity and human capital accumulation.
with Agustín Barboza
How do firing costs shape which automation technologies firms adopt, and what happens to incumbent workers working in automated tasks following adoption? We develop a dynamic partial equilibrium model in which firms jointly choose their production technology, layoffs, hiring, and occupational retraining. Firing costs operate through two margins. Conditional on adoption, they encourage firms to retain and retrain incumbent workers for non-automated tasks rather than rely on external hiring, increasing within-firm occupational reallocation and reducing layoffs. At the adoption margin, firing costs tax low-productivity technologies that are more likely to lead to worker displacement. Quantitative experiments suggest that firing costs primarily change how adopters reorganize their workforce, while having more selective effects on which technologies firms adopt. With uncertainty and attrition, firing costs can also incentivize firms to delay their adoption. We test the model's empirical predictions for occupational reallocation using establishment-level data on works council status and industrial robot adoption from Germany, complementing recent work by Findeisen et al (2026).