Spmv Approaches to Dynamic Discrete Choice Models with Limited Transition
Dynamic optimization problems often involve continuous state variables. Casting such problems into dynamic discrete choice models usually requires variable discretization. When there are multiple state variables, many discretized future states will be visited with only very small probability conditional on current states. We investigate pruning these small transition probabilities and applying the sparse matrix-vector multiplication method in value function iterations. We assess our method in a numerical example inspired by Rust (1987) and Barwick and Pathak (2015). Our method substantially improves computational performance and reduces memory requirements with little loss in accuracy