ABIDES-Economist: Agent-Based Simulator of Economic Systems with Learning Agents

Abstract

We present ABIDES-Economist, an agent-based simulator for economic systems that includes heterogeneous households, firms, a central bank, and a government. Agent behavior can be defined using domain-specific behavioral rules or learned through reinforcement learning by specifying their objectives. We integrate reinforcement learning capabilities for all agents using the OpenAI Gym environment framework for the multi-agent system. To enhance the realism of our model, we base agent parameters and action spaces on economic literature and real U.S. economic data. To tackle the challenges of calibrating heterogeneous agent-based economic models, we conduct a comprehensive survey of stylized facts related to both microeconomic and macroeconomic time series data. We then validate ABIDES-Economist by demonstrating its ability to generate simulated data that aligns with the relevant stylized facts for the economic scenario under consideration, following the learning of all agent behaviors via reinforcement learning. Specifically, we train our economic agents' policies under two broad configurations. The first configuration demonstrates that the learned economic agents produce system data consistent with macroeconomic and microeconomic stylized facts. The second configuration illustrates the utility of the validated simulation platform in designing regulatory policies for the central bank and government. These policies outperform standard rule-based approaches from the literature, which often overlook agent heterogeneity, shocks, and agent adaptability.