We propose a method for solving continuous-state and action-stochastic dynamic programs that is a hybrid between the continuous space projection methods introduced by Judd and the discrete space methods introduced by Bellman. Our hybrid approach yields a smooth representation of the value...