Job Market Paper

Learning on the Job -- PDF, Slides
with Fil Babalievsky, William Jungerman

What are the sources of worker learning within the firm? How much of a worker’s human capital growth comes from firm specific factors, such as the learning environment, as opposed to their own ability, and the composition of their coworkers? In this paper, we introduce a novel labor search model with multi-worker firms, learning from coworkers, and heterogeneity in learning-by-doing rates that can vary at the worker and firm level. Despite its complexity, we show that it is possible to solve such a model by leveraging recent advances from the machine learning literature. We use French administrative data to discipline the parameters of the model, specifically by targeting how wage growth varies across workers, firms, and the distribution of coworker wages. With the calibrated model in hand we perform a series of structural and statistical decompositions to test how much of the variance of human capital is driven by learning from coworkers and by heterogeneous learning-by-doing, and find that learning from coworkers is the dominant source of learning in the economy. Switching off learning from coworkers lowers human capital and wages by more than 25%, and differences in the composition of coworkers accounts for more than 50% of the variance of human capital growth rates.

Working Papers

Endogenous Firm Structure and Worker Specialization -- PDF, Slides

What tasks must be performed to produce a good? Which occupations are well suited to do those tasks? And what are the gains to worker specialization within the firm? I use Brazilian administrative data to document new facts about how firms vary the types of workers that they choose to hire as they grow larger. Bigger firms hire more distinct occupations. They also hire a set of workers whose cognitive, manual, and interpersonal skills are more dispersed than at small firms. I then develop a structural model of how firms choose which types of workers to hire, and how they assign tasks to these workers. I propose a novel identification strategy for how to indirectly infer the (multi-dimensional) distribution of skill requirements for tasks that firms face and using only cross-sectional data on which occupations firms choose to hire, and in what proportion, across the firm size distribution. I estimate my model using Brazilian manufacturing firms, and show that more than 1/3 of the variance in firm-level TFP is due to firms’ endogenous choices of which types of workers to hire (and how specialized those workers should be). I find that gains from increasing firm specialization are about 1.3% of output, and that the costs of shutting down worker specialization within firms are large, leading to a 9.6% decrease in total output. I find similar gains in more narrowly defined industry codes such as leather goods.

Deep Reinforcement Learning for Economics -- PDF
with Fil Babalievsky, William Jungerman

This paper provides a self-contained guide to deep reinforcement learning methods for economists. These tools allow agents to find the policy function that maximizes their expected discounted stream of rewards under very few assumptions about the structure of the model, and are potentially applicable to a wide class of economic models. We begin by translating the language of reinforcement learning to the language of economics. Next, we introduce neural networks, a class of function approximators that have proven useful for reinforcement learning. We use a standard Bewley (1977)-style consumption- savings problem as our test case since we can easily check for correctness. We conclude by offering a practical guide with implementation details and a discussion of what kinds of problems are more amenable to reinforcement learning than conventional techniques.

In Progress

Credit Access and the Earnings Mobility of Workers and Entrepreneurs
with Carter Braxton, Kyle Herkenhoff, Gordon Phillips

Does greater access to credit increase the earnings mobility of workers and entrepreneurs? Has the expansion of consumer credit contributed to the increase in earnings inequality? We answer the first question by linking individual credits reports to administrative earnings data for workers as well as entrepreneurs. We answer the second question by developing a tractable labor sorting model with human capital accumulation. We link TransUnion credit reports to the LEHD on scrambled social security numbers. We stratify individuals based on credit scores (the marginal cost of credit), and credit limits (the stock of credit), and we document their lifecycle earnings mobility patterns from 1998 to 2008. We instrument access to credit using house price variation and credit account ages in 1998. We find that credit access has an insignificant effect on earnings mobility among initially low earning households. We find that credit access has significant positive significant effect on the earnings mobility of high earning households. We find similar results for entrepreneurial income, with those who have initially high entrepreneurial earnings benefiting the most from credit access. We estimate our model to match these facts, and then we counterfactually shut down credit markets. We find that credit access, while welfare improving, significantly increases measured wage and entrepreneurial income inequality.

Publications

Do Long-Haul Truckers Undervalue Future Fuel Savings? -- Journal Link
with Adam Copeland, John J. Stevens
Energy Economics (June 2019) vol. 81, pp. 1148-1166

The U.S. federal government enacted fuel efficiency standards for medium and heavy trucks for the first time in September 2011. Rationales for using this policy tool typically depend upon frictions existing in the marketplace or consumers being myopic, such that vehicle purchasers undervalue the future fuel savings from increased fuel efficiency. We measure by how much long-haul truck owners undervalue future fuel savings by employing recent advances to the classic hedonic approach to estimate the distribution of willingness-to-pay for fuel efficiency. We find significant heterogeneity in truck owners’ willingness to pay for fuel efficiency, with the elasticity of fuel efficiency to price ranging from 0.51 at the 10th percentile to 1.33 at the 90th percentile, and an average of 0.91. Combining these results with estimates of future fuel savings from increases in fuel efficiency, we find that long-haul truck owners’ willingness-to-pay for a 1 percent increase in fuel efficiency is, on average, just 29.8% of the expected future fuel savings. These results suggest that introducing fuel efficiency standards for heavy trucks might be an effective policy tool to raise medium and heavy trucks’ fuel economy.