Recursive computation of heterogeneous agent models

The paper presents a method for the recursive computation of models with a continuum of heterogeneous agents. Following the literature, it is assumed that the cross-sectional distribution of wealth in the economy can be represented, to a sufficient degree of accuracy, by a finite number of summary statistics. The main innovation of the method is a very flexible representation of the cross-sectional distribution that can be adapted in the process of the solution. This helps in achieving high accuracy using only a small number of statistics. The solution is computed by backward induction on a discretization of the state space. The paper discusses approximation techniques to make the algorithm stable and fast. Theoretical considerations and practical experience indicate that the method has very good convergence properties. It is illustrated by an application to a standard model of heterogeneous agents that face uninsurable income risk. Address of the author: Michael Reiter Universitat Pompeu Fabra Department of Economics and Business Ramon Trias Fargas, 25–27 08005 Barcelona Spain e-mail: [email protected]