This paper presents an algorithm for solving nonlinear dynamic stochastic models that computes value function by simulations. We argue that the proposed algorithm can be a useful alternative to the existing methods in some applications.