hozdenor_Tepper_2020 (2).pdf (1.23 MB)
Download file

Occupational Networks and Automation

Download (1.23 MB)
posted on 18.09.2020, 19:07 by Hakki OzdenorenHakki Ozdenoren
This dissertation investigates the labor market consequences of technological change. Chapter I builds an occupational network based on the flows of workers
between occupations and shows that the network has a core/periphery structure. Core occupations employ most of the workforce, require fewer skills, and pay lower
wages. At the same time, they act as bridges between other occupations and provide insurance value to the workers in other occupations in case they lose their jobs. A key result in this chapter is to show that the core occupations become more likely to be automated thanks to the advances in technologies like machine learning and cloud computing. Consequently, automation is expected to have far more
significant consequences than what would be implied by its direct impact. If the occupations with the highest probability of being automated disappear, 7% of the workforce would be displaced from their jobs. Moreover, almost 10% of the edges between occupations would dissolve, further aggravating the impact of automation. Chapter II develops a structural model of occupational choice that endogenizes
worker flows between occupations. It extends the dynamic discrete choice model of occupational choice to include search frictions and transition costs and embeds it into a general equilibrium search environment. Using the Survey of Income and Program Participation and O*NET datasets, search frictions and transition costs are structurally estimated. Results show that transition costs that workers face in automatable jobs are particularly high, and search frictions significantly curtail workers’ abilities to transition away from jobs vulnerable to labor substituting
technology. Furthermore, low-cost transitions for these workers are towards other highly automatable occupations. Consequently, if such occupations would undergo automation in a similar timeline, the impact of new technologies would be significantly amplified. Finally, a counter-factual is performed where automation decreases revenues of manual firms in “Transportation and material moving” occupations by twenty-five percent. The new steady-state features 150,000 more unemployed workers. Analyzing transition dynamics reveals that unemployment is
considerably higher during the transition, and that it takes about seven years for unemployment rates to reach their steady-state values—a significant portion of a worker’s career. Chapter III uses the framework developed in Chapter II to evaluate two strands of labor market programs that aim to help unemployed workers: a Trade Adjustment
Assistance inspired Automation Adjustment Assistance (AAA) program and Unemployment Insurance (UI). The AAA program that provides relief conditional on being unemployed from the automated occupation introduces adverse incentives and induces workers to stay in the automated occupation. UI policies do no carry this risk as workers need not be unemployed from a specific occupation to be eligible for benefits. This chapter considers three alternative UI policies. The first policy is the current implementation of UI in the US economy. The second policy is a UI policy optimized for the pre-automation economy in the absence of automation, which we call SS-Optimal. SS-Optimal policy increases replacement ratio from 30%
to 71% and increases the welfare by 0.01%. However, when automation begins, both the current and SS-Optimal UI lead to a massive budget shortfall. The final policy considered is a dynamically-optimal UI policy that takes the transition induced by the automation into account. Dynamically-optimal unemployment insurance provides almost full insurance and increases welfare by 0.01% while keeping a balanced budget. Therefore, UI programs must anticipate and adjust accordingly with technological developments.




Degree Type



Tepper School of Business

Degree Name

  • Doctor of Philosophy (PhD)


Laurence Ales Christopher Sleet

Usage metrics