Publications
with Miguel Espinosa and Arthur Seibold. Econometrica (2022)
We study firm responses to a large-scale change in apprenticeship regulation in Colombia. The reform requires firms to train, setting apprentice quotas that vary discontinuously in firm size. We document strong heterogeneity in responses across sectors, where firms in sectors with high skill requirements tend to avoid training apprentices, while firms in low-skill sectors seek apprentices. Guided by these reduced-form findings, we structurally estimate firms' training costs. Especially in high-skill sectors, many firms face large training costs, limiting their willingness to train apprentices. Yet, we find substantial overall benefits of expanding apprenticeship training, in particular when the supply of trained workers increases in general equilibrium. Finally, we show that counterfactual policies that take into account heterogeneity across sectors can deliver similar benefits from training while inducing fewer distortions in the firm-size distribution and in the allocation of resources across sectors.
with Robert E. Lucas Jr. and Esteban Rossi-Hansberg. AEJ: Macroeconomics (2019)
We develop a theory of career paths and earnings where agents organize in production hierarchies. Agents climb these hierarchies as they learn stochastically from others. Earnings grow as agents acquire knowledge and occupy positions with more subordinates. We contrast these and other implications with U.S. census data for the period 1990 to 2010, matching the Lorenz curve of earnings and the observed mean experience-earnings profiles. We show the increase in wage inequality over this period can be rationalized with a shift in the level of the complexity and profitability of technologies relative to the distribution of knowledge in the population.
Economica (2019)
In this note, I propose two alternative frameworks to study idea diffusion models with cohort structures. Both frameworks fix Lucas (2009) aggregation mistake while keeping the analytical tractability of the model and its insights. The frameworks differ in their assumptions on the meeting process. First I study a continuous arrival process where agents meet others at each point in time, and then a more commonly used Poisson process where meeting opportunities arrive stochastically at some given Poisson rate. I generalize the growth formula in Lucas (2009) and show both models yield the same growth rate on a balanced growth path. Moreover, I show the continuous arrival process can be viewed as the limit of Poisson processes where the meeting rate increases but the quality of meetings decreases.
Working Papers
Universities, Spatial Misallocation and Growth
with Xiao Ma and Yiran Zhang
We study how spatial spillovers from universities shape aggregate economic growth. Using Chinese data on manufacturing firms, patents, and land transactions, we show that proximity to universities raises innovation efficiency, primarily through knowledge diffusion and access to university-trained talent. New firms benefit disproportionately from these spillovers yet are less likely to locate nearby, where incumbents occupy scarce land. We build a quality-ladder model in which firms choose both innovation effort and location. Creative destruction reallocates product lines but not physical space: an innovating entrant inherits the incumbent's product but not its land, generating spatial misallocation absent from standard growth models. The social planner concentrates more R&D effort near universities than the market does, and this spatial distortion compounds the familiar under-investment in innovation. Calibrating to Chinese data, preliminary counterfactual exercises suggest that university spillovers are a sizable source of growth. Eliminating the proximity advantage in innovation productivity would cut the balanced growth rate by more than half, while land reallocation policies yield additional but more modest gains.
Population Aging, Firms and Worker Composition
with Cheng Chen, Aspen Gorry and Takahiro Hattori
Population aging changes the composition of workers in the labor force. We develop a framework to understand how worker composition affects productivity in an economy with heterogeneous firms and apply it to Japan. Reallocating workers toward the optimal composition of experience and skill would increase output by 3.8\%. Including worker aging and reallocation costs, we find strong complementarity between workers' skill and firm productivity, which generates differential hiring strategies across firms and heterogeneous effects of population aging on wages and market concentration. Using Japanese establishment data, we validate key model predictions: larger firms disproportionately hire young workers, and regions with slower working-age population growth exhibit lower concentration of employment among top establishments.
with Jeremy Pearce
This paper studies how the speed-quality tradeoff in innovation interacts with firm dynamics, concentration, and economic growth. Using microdata and a change in policy on patent duration, we document the existence of this tradeoff both in the aggregate and at the firm level. We show long-run trends in the increasing speed of innovation alongside declining quality at large firms, with the allocation of inventors playing an essential role. We develop a theoretical framework incorporating the speed-quality tradeoff and show that allocating less labor towards speed increases growth, particularly in the presence of private benefits to innovation. We estimate a quantitative endogenous growth model to study how firms' substitution across speed and quality interacts with aggregate outcomes. Quantitatively, the transition to faster, lower-quality innovations has a significant impact on growth, mainly through a shift in innovation production technology. We argue that even when the reallocation of inventors across speed and quality has modest effects on growth, endogenizing both the speed and quality decisions of firms is crucial for studying innovation.
with Ufuk Akcigit, Ernest Miguelez, Stefanie Stantcheva and Valerio Sterzi (R&R Econometrica)
An inventor’s own knowledge is a key input in the innovation process. This knowledge can be built by interacting with and learning from others. This paper uses a new large-scale panel dataset on European inventors matched to their employers and patents. We document key empirical facts on inventors’ productivity over the life cycle, inventors’ research teams, and interactions with other inventors. Among others, most patents are the result of collaborative work. Interactions with better inventors are very strongly correlated with higher subsequent productivity. These facts motivate the main ingredients of our new innovation-led endogenous growth model, in which innovations are produced by heterogeneous research teams of inventors using inventor knowledge. The evolution of an inventor’s knowledge is explained through the lens of a diffusion model in which inventors can learn in two ways: By interacting with others at an endogenously chosen rate; and from an external, age-dependent source that captures alternative learning channels, such as learning-by-doing. Thus, our knowledge diffusion model nests inside the innovation-based endogenous growth model. We estimate the model, which fits the data very closely, and use it to perform several policy exercises, such as quantifying the large importance of interactions for growth, studying the effects of reducing interaction costs (e.g., through IT or infrastructure), and comparing the learning and innovation processes of different countries.
This paper studies the convergence of wage distributions between the South and Non-South U.S. metropolitan areas from 1940 to 2010. Initially, the Southern cities exhibited lower average wages and a more dispersed log wage distribution. After 1940, there was a remarkable convergence of the entire wage distribution between the two regions, resulting in nearly identical distributions by 2010. I propose a spatial economics model with a continuum of heterogeneous agents, integrating regional technological differences that made low-skilled workers in the Non-South more productive, driving higher average wages and reduced dispersion. The calibrated model indicates that disparities in low-skill services and manufacturing technologies primarily contribute to initial wage distribution gaps, highlighting the dynamic role of regional technological factors in shaping U.S. wage distributions during the studied period.