Monopsony at Work? The Short- and Long-run Effects of Labor Restrictions on Refugees' Economic Integration
Wth Andreas Beerli, Selina Kurer, Dominik Hangartner, Michael Siegenthaler.
How do policies that restrict refugees' labor market access affect their economic integration? This paper analyzes the employment, wage, and job mobility effects of labor market policies regulating whether, where, and for whom refugees are allowed to work. The empirical design exploits the exogenous assignment of refugees to Swiss cantons upon arrival and the rich spatio-temporal variation in cantonal labor market policies. We merge newly collected data on cantonal policies 1999-2016 to high-quality administrative data on refugees' asylum processes and social security earnings records. Using a range of panel data models, we find negative short- to medium-term employment and earnings effects of banning refugees from working in the first months after arrival, prioritizing residents over refugees, and restricting refugees' labor markets geographically and sectorally. These effects are substantial in size: moving from the least to the most restrictive policy mix reduces the average employment rate of refugees in the first five years after arrival from 23% to 16%. The priority requirement and sectoral and geographical restrictions also lower refugees' hourly wages. Consistent with dynamic models of monopsony, the wage effects arise because the two policies reduce refugees' chances to switch to more demanding and better-paid jobs. Restrictive policies do not affect emigration, not even among refugees that are only temporally admitted. This enables us to trace the longer-term scars of labor market policies: we find that priority and blocking policies reduce refugees' labor market earnings up to five and six years, respectively, after they cease applying. Together, these results suggest that labor restrictions for refugees burden both refugees and host communities with significant costs.
pystacked: Stacking generalization and machine learning in Stata
With Christian B Hansen, Mark E Schaffer
pystacked implements stacked generalization (Wolpert, 1992) via Python's scikit-learn. Stacking combines multiple supervised machine learners--the "base" or "level-0" learners--into a final prediction to improve performance. The currently supported base learners include regularized regression, random forest, gradient boosting and neural nets (multi-layer perceptron). \pystacked can also be used with a single base learner and, thus, provides an easy-to-use API for scikit-learn's machine learning algorithms.
ddml: Double-debiased machine learning in Stata
With Christian B Hansen, Mark E Schaffer, Thomas Wiemann
The Effects of Cash-Based Interventions on the Integration of Venezuelan Immigrants in Peru
With Marine Casalis, Dominik Hangartner, Rodrigo Sánchez
Incentive or Impediment? The Short- and Long-Term Impact of Low Welfare Support on Refugee Integration
With Dominik Hangartner, Selina Kurer