This paper studies the predictability of long-term unemployment (LTU) using rich administrative data from Sweden. We establish substantial heterogeneity in LTU risk across individuals, accounting for both observed and unobserved heterogeneity using a wide range of observable predictors and multiple-spell outcomes, respectively. We apply our prediction algorithm to study the dynamics of job finding over the unemployment spell and the business cycle. Selection effects can explain most of the decline in average job finding over the unemployment spell but little of its cyclicality. We also find sizable heterogeneity in the profiles of job finding over the unemployment spell but not over the business cycle.
Müller et al. (2025) studied this question.