Pré-publication, Document de travail: Most unemployment insurance (UI) schemes mandate a single benefit schedule,while little empirical findings support this mandate. In this paper, I exploit aFrench program where workers are given a choice between two different UI sched-ules, providing an ideal setup to evaluate both moral hazard and selection intoUI. Using high-quality administrative data, I measure significant adverse selectionby relating the entitlement choice with the characteristics of the insured. Moralhazard is even larger, as shown by a fuzzy regression discontinuity design using aneligibility criterion: choosing a short schedule with higher average benefits increasesunemployment duration by six months.