WIP Seminar: Vito Stefano Bramante

Title: "Statistical Profiling Back to Work: A Machine Learning Algorithm for Assignment to Active Labor Market Programs" (with Giulio Zanella)

  • Data: 30 giugno 2022 dalle 17:30 alle 18:30

  • Luogo: Seminar Room - Piazza Scaravilli, 2 + Microsoft Teams Meeting

Abstract

A key task of the Public Employment Service is to assign job seekers to Active Labor Market Programs, so as to speed up re-employment, reduce unemployment benefits outlay, and increase tax revenue. Selecting the best program for a specific job seeker requires to figure out treatment effects conditional on a specific profile, an inferential problem that purely predictive algorithms are not equipped to solve. In this work we use administrative data on the universe of benefit claimants in Trento province during 2019-2021 to develop an algorithm that is able to select the best program for each worker at the outset of an unemployment spell, based on observable characteristics. The algorithm is based on instrumental forest, a recent innovation in the statistical literature that combines causal inference and machine learning. Drawing from the judge fixed-effects literature, we exploit the quasi-random nature of the assignment of job seekers to caseworkers to remove selection bias.