Distribution Regression in Duration Analysis: an Application to Unemployment Spells

This article proposes estimation and inference procedures for distribution regression models with randomly right-censored data. The proposal generalizes classical duration models to a situation where slope coefficients can vary with the elapsed duration, and is suitable for discrete, continuous or mixed outcomes. Given that in general distribution regression coefficients do not have clear economic interpretation, we also propose consistent and asymptotically normal estimators for the average distribution marginal effects. Finite sample properties of the proposed method are studied by means of Monte Carlo experiments. Finally, we apply our proposed tools to study the effect of unemployment benefits on unemployment duration. Our results suggest that, on average, an increase in unemployment benefits is associated with a nonlinear, non-monotone effect on the unemployment duration distribution and that such an effect is more pronounced for workers subjected to liquidity constraints.