Agent-Based Modeling for Optimal Economic Policy with Exogenous Shocks

Abstract

My thesis explores the application of reinforcement learning and agent-based computational economics to the problem of optimal policy in an environment with exogenous shocks. I develop an agent-based general equilibrium economic model that is considerably simpler and more extensible than existing agent-based models. For this model, I introduce a price adjustment process and prove that under suitable assumptions, the process converges in a single market with variable supply and demand, as well as in an exchange economy. I then present simulation results where agents in economy learn via deep reinforcement learning methods, and show that agents learn to behave in accordance with theoretical predictions. Finally, I introduce exogenous unemployment shocks to the economy, and show that a social planner trained with reinforcement learning is able to improve welfare for the workers in the economy.