Utilizing RBC system for taxation policy evaluation: An adaptive interaction framework based on deep reinforcement learning

Shuang Luo, Shunyu Liu, Tianrun Cai, Chao Wu

Abstract: The economic system serves as the foundation for the organization and coordination of economic activities within a society, involving various interacting agents, such as workers, firms, and the government. Simulating the behavior of different agents within an economic system can aid policymakers in evaluating the potential consequences of various economic policies, thereby facilitating the development of more effective policies. However, conventional agent-based modeling (ABM) is limited by hand-crafted interaction rules, resulting in a heavy reliance on simulated outcomes in formulating static individual behaviors. In this work, we propose an adaptive interaction framework to dynamically model the effects of various taxation policies within the Real-Business-Cycle (RBC) simulation system. The RBC system simulates the interaction between workers, firms, and the government to explain economic fluctuations and conduct economic policy evaluations in the business cycle. In contrast to previous works, our framework allows mutual adaptation among all interacting members to attain a dynamic equilibrium, yielding a more accurate representation of real-world economic operation. Specifically, the decision-making process of the government is driven by a deep reinforcement learning agent, enabling it to update taxation policies based on different objective signals from the RBC system. Additionally, workers and firms can also adjust their behaviors through imitation learning, aiming to optimize their utility under various taxation policies. Simulation results indicate that the proposed framework can capture the complex dynamics of various interacting agents and effectively evaluate the effects of different taxation policies.