While a broad theoretical literature examines the optimal timing of unemployment insurance benefits, little is known about the empirical importance of underlying mechanisms like duration dependence and dynamic selection. Using administrative unemployment records from Germany, we estimate a structural job search model with savings to empirically disentangle various forms of duration dependence and dynamic selection. Duration dependence and dynamic selection are identified separately using multiple unemployment spells of individuals. This allows us to assess their contribution to an increasing or decreasing profile of benefits over time. We also analyze the local consumption smoothing gains and moral hazard costs to characterize the impact of duration dependence and dynamic selection on the optimal insurance problem. Our results show that they push towards increasing schedules, primarily by lowering the moral hazard costs of providing benefits later in the spell.