Awarded the Royal Economic Society Junior Symposium Best Paper Award, April 2019
This paper analyzes the returns to training-occupation combinations. I use administrative panel data on apprenticeships and employment for German workers, and identify the returns using data on occupation-specific vacancies. For the estimation, I set up a Roy model and extend existing control function approaches to deal with selection in a two-stage, high-dimensional setting. I find sizable returns to training in one’s occupation, and substantial bias when not controlling for selection. Returns are decreasing in the task distance between training and occupation. I argue that imperfect information leads to ex-post suboptimal training choices, and that retraining could address this ex-ante friction.
Temporal-Difference Estimation of Dynamic Discrete Choice Models (with K. Adusumilli)
We study the use of Temporal-Difference learning, a popular Reinforcement Learning technique, for estimating the structural parameters in dynamic discrete choice models. Our proposed algorithms are based on the conditional choice probability approach, but use functional approximations to estimate various value terms in the pseudo-likelihood function. We suggest two approaches: The first approach, the linear semi-gradient method, provides functional approximations to the recursive terms using basis functions. Computationally, it involves solving a low dimensional linear equation. The second approach, Approximate Value Iteration, builds a sequence of approximations to the recursive terms by solving a non-parametric estimation problem in each step. Almost any machine learning method devised for prediction can be used for approximation; we particularly focus on neural networks. Both approaches are fast and have the advantage of naturally allowing for continuous and high dimensional state spaces. Furthermore, they do not require specification of transition probabilities. For the estimation of dynamic games, they do not require integrating over the actions of other players, which further heightens the computational advantage. Our estimators are consistent, and efficient under discrete state spaces. For continuous states, we propose locally robust corrections in order to achieve parametric rates of convergence. Monte Carlo simulations for a dynamic firm entry problem, a dynamic firm entry game and a version of the famous Rust (1987) engine replacement problem confirm the properties of our algorithms in practice.
Research in Progress
Early Career Shocks, Skill Match, and the Nature of Human Capital
Young workers’ careers are importantly affected by the economic conditions around their time of labor market entry. This paper studies skill match as a mechanism behind these effects. Using a large administrative panel of German workers who completed training in a specific occupation, I construct an objective measure of skill match which records whether a worker is employed in the occupation they got trained in. I document that this measure is strongly procyclical for young cohorts, and use variation in local unemployment rates to show that high entry unemployment persistently reduces a worker’s skill match. These results display considerable heterogeneity across trainings. To shed light on the underlying mechanisms, I set up a simple model of selection into occupations. Working in an occupation different from one’s training comes at a cost, but may be optimal in response to occupation-specific shocks. Novel data on unemployment in target occupations confirms that skill-specific labor demand at entry importantly affects both skill match and wages. Across trainings, the cyclicality of the training occupation interacts with measures of human capital specificity to explain heterogeneity in the effects of entry conditions. These results have important implications for policies aimed at mitigating the effects of entry conditions for young workers.