Approximate and Almost-Exact Aggregation in Dynamic Stochastic Heterogeneous-Agent Models

The paper presents a new method to solve DSGE models with a great number of heterogeneous agents. Using tools from systems and control theory, it is shown how to reduce the dimension of the state and the policy vector so that the reduced model approximates the original model with high precision. The method is illustrated with a stochastic growth model with incomplete markets similar to Krusell and Smith (1998), and with a model of heterogeneous firms with state-dependent pricing. For versions of those models that are nonlinear in individual variables, but linearized in aggregate variables, approximations with 50 to 200 state variables deliver solutions that are precise up to machine precision. The paper also shows how to reduce the state vector even further, with a very small reduction in precision.