Predicting re-employment: machine learning versus assessments by unemployed workers and by their caseworkers
-
Series
-
SpeakerGerard van den Berg (University of Groningen)
-
FieldEmpirical Microeconomics
-
LocationTinbergen Institute Amsterdam, room 1.01
Amsterdam -
Date and time
November 05, 2024
15:30 - 16:30
Abstract
We analyze three sources of information on the individual probability of re-employment within 6 months (RE6), among individuals sampled from the inflow into unemployment. First, they are asked for their perceived probability of RE6 (sample N=1200). Second, their caseworkers reveal whether they expect RE6. Third, random-forest machine learning methods are trained on big administrative data on the full inflow, to predict individual RE6. We compare the predictive performance of these measures and consider the gains of combining them. Correcting the machine learning algorithm if the unemployed themselves predict long-term unemployment leads to a "super-predictor" with a superior performance. This gain is concentrated among risk averse unemployed who may have collected more information on future idiosyncratic events.