Can Algorithms Reliably Predict Long-Term Unemployment in Times of Crisis? – Evidence from the COVID-19 Pandemic

"In this paper, we compare two popular statistical learning techniques, logistic regression and random forest, with respect to their ability to classify jobseekers by their likelihood to become long-term unemployed. We study the performance of the two methods before the COVID-19 pandemic as wel...

Full description

Saved in:
Bibliographic Details
Published inIDEAS Working Paper Series from RePEc
Main Authors Kunaschk, Max, Lang, Julia
Format Paper
LanguageEnglish
Published St. Louis Federal Reserve Bank of St. Louis 01.01.2022
Subjects
Coronaviruses
COVID-19
Pandemics
Regression analysis
Unemployment
Online AccessGet full text

Cover

More Information
Summary:"In this paper, we compare two popular statistical learning techniques, logistic regression and random forest, with respect to their ability to classify jobseekers by their likelihood to become long-term unemployed. We study the performance of the two methods before the COVID-19 pandemic as well as the impact of the pandemic and its associated containment measures on their prediction performance. Our results show that random forest consistently out-performs logistic regression in terms of prediction performance, both, before and after the beginning of the pandemic. During the lockdowns of the first wave, the number of unemployment entries and the fraction of individuals that become long-term unemployed strongly increases, and the prediction performance of both methods declines. Finally, while the composition of the (long-term) unemployed changed at the beginning of the COVID-19 pandemic, we do not find systematic patterns across groups with different levels of labor market attachment or across different sectors of previous employment in terms of declines in prediction performance." (Author's abstract, IAB-Doku) ((en))
DOI:10.48720/IAB.DP.2208