A comparison of hazard rates of duration outcomes before and after policy changes is hampered by non-identification if there is unobserved heteogeneity in the effects and no model structure is imposed. We develop a discontinuity approach that overcomes this by exploiting variation in the moment at which different cohorts are exposed to the policy change, i.e. by considering spells crossing the policy change. We prove identification of average treatment effect on hazard rates without model structure. We estimate these effects by local linear kernel hazard regression. We use the introduction of the NDYP programme for young unemployed individuals to estimate average programme participation effects on the exit rate to work.