Estimating the latent effect of unemployment benefits on unemployment duration

We estimate the effect of a shortening of unemployment benefit entitlements on unemployment duration. Previous studies on the same or related problems have not taken into account that the competing risks duration model is not identified and we shed first light on the question whether the non identification problem may preclude informative results. It turns out that the identification bounds for the parameters of interest are very wide in the absence of strong assumptions. We suggest an assumption on the dependence structure between risks which is milder than what conventional duration models assume. Under this assumption, the identification bounds are tighter and become informative for the direction of the treatment effect. We find evidence that the unemployed with higher pre-unemployment earnings are more likely to enter full-time employment and, in particular, subsidized self-employment. -- dependent censoring ; partial identification ; difference-in-differences ; unemployment duration ; unemployment benefits