Jirnyi, A., and V. Lepetyuk
WP-AD 2011-21

Keywords: reinforcement learning,incomplete market,aggregate uncertainty,Krusell-Smith,knn estimator

Abstract: We develop a method of solving heterogeneous agent models in which individual decisions depend on the entire cross-sectional distribution of individual state variables, such as incomplete market models with liquidity constraints. Our method is based on the principle of reinforcement learning, and does not require parametric assumptions on either the agents’ information set, or on the functional form of the aggregate dynamics.