Computation of solutions to dynamic models with occasionally binding constraints

We construct the first algorithm for the perfect foresight solution of otherwise linear models with occasionally binding constraints, with fixed terminal conditions, that is guaranteed to return a solution in finite time, if one exists. We also provide a proof of the inescapability of the “curse of dimensionality” for this problem when nothing is known a priori about the model. We go on to extend our algorithm to deal with stochastic simulation, other non-linearities, and future uncertainty. We show that the resulting algorithm produces fast and accurate simulations of a range of models with occasionally binding constraints.