A fast and low computational memory algorithm for non-stochastic simulations in heterogeneous agent models

Abstract

Heterogeneous agent models in macroeconomics generally require numerical computation of the cross-sectional distribution of agents. The standard textbook approach is to fully approximate the Markov kernel that iterates the distribution forward in time as a Markov transition matrix, which can be costly in terms of computational time and memory when the state space is large. This note provides an alternative algorithm that is simple, requires much less computational memory, and is substantially faster than the standard algorithm.