Structured Approximate Dynamic Programming for Simulating Heterogeneous Agents in Incomplete Markets

Abstract

The increasing use of dynamic stochastic general equilibrium models by central banks to conduct policy analysis necessitates the development of new algorithmic techniques capable of solving these problems. This thesis uses structured approximate dynamic programming to solve the Krusell-Smith model, a prototypical dynamic stochastic general equilibrium model that simulates heterogeneous agents in incomplete markets. We formulate the Krusell-Smith model as a stochastic programming problem and solve it using two di erent structure enforcing algorithms. In the process, we also develop a new structured approximate dynamic programming algorithm we call ADPLP. We compare performance of our approximate solutions relative to a benchmark solution and show that both our algorithms have good accuracy and convergence for small time horizons. We close by conducting sensitivity analysis policy studies using the Krusell-Smith model to simulate the e ects of the European Central Bank's recent monetary policy.