We give conditions under which the smoothness properties of the value and policy functions of a dynamic program at discount factor zero extend to small positive discount factors. We apply this to a model of Bayesian learning by a decision maker who does not know which of several parameters is...