Learning Macroeconomic Policies based on Microfoundations: A Stackelberg Mean Field Game Approach

The Lucas critique emphasizes the importance of considering microfoundations, how micro-agents (i.e., households) respond to policy changes, in macroeconomic policymaking. However, due to the vast scale and complex dynamics among micro-agents, predicting microfoundations is challenging. Consequently, this paper introduces a Stackelberg Mean Field Game (SMFG) approach that models macroeconomic policymaking based on microfoundations, with the government as the leader and micro-agents as dynamic followers. This approach treats large-scale micro-agents as a population, to optimize macroeconomic policies by learning the dynamic response of this micro-population. Our experimental results indicate that the SMFG approach outperforms real-world macroeconomic policies, existing AI-based and economic methods, enabling the learned macroeconomic policy to achieve the highest performance while guiding large-scale micro-agents toward maximal social welfare. Additionally, when extended to real-world scenarios, households that do not adopt the SMFG policy experience lower utility and wealth than adopters, thereby increasing the attractiveness of our policy. In summary, this paper contributes to the field of AI for economics by offering an effective tool for modeling and solving macroeconomic policymaking issues.