Sebastian Ernst

This paper evaluates the use of algorithmic risk profiling to direct job search support toward individuals most at risk of long-term unemployment. Using three regression discontinuity designs and rich administrative data from the Flemish Public Employment Service (VDAB) in Belgium, we address two key dimensions: who and when to target. First, job search counseling raises employment on average, but treatment effects decline sharply with predicted risk. This reveals a fundamental trade-off between targeting those most at risk and those most responsive to intervention. Second, we find no evidence of depreciation in treatment effects at the individual level, but strong dynamic selection: as spells lengthen, surviving job seekers have both higher baseline risk and lower treatment responsiveness, generating a trade-off between the timing of treatment and the composition of the remaining pool. We develop and calibrate a conceptual framework for optimal targeting whose key sufficient statistic is the ratio of predicted treatment effects to predicted job-finding probabilities. Targeting rules combining both objects substantially outperform the status quo, yielding welfare gains nearly three times larger than random assignment, while targeting on risk scores alone performs worse than random targeting.

This paper analyses how risk profiling is used to assign unemployed job seekers to job search counseling in Flanders, Belgium. We compare algorithmic selection to self-selection and selection by job search counselors. We discuss practical challenges for the implementation of risk profiling and highlight avenues for further research. We find that algorithmic assignment is used for only a small fraction of the sample and that job search counselors appear to have valuable private information on job seekers' reemployment prospects beyond what is captured by the algorithmic risk score.