Occupational Choice, Human Capital and Learning: A Multi-Armed Bandit Approach

This paper introduces a model of worker matching at the occupation level. In our setup, young workers, while employed in an occupation, accumulate human capital and also learn about their underlying productivity in that occupation. Human capital is partially transferable to other occupations and similarly, the information acquired in one occupation is useful for the worker’s productivity elsewhere. Workers with low tenure levels, as well as low-paid workers, are the ones most likely to switch occupations, consistent with our empirical findings. Though the model is quite general, we show that Gittins indices can be used in this setup to preserve tractability. We discuss potential applications ranging from assessing the impact of AI and automation to the evaluation of policies such as unemployment benefits, sector-specific subsidies, or minimum wages.