From Rational Agents to Learning Agents: A Structured Review and Conceptual Framework for MARL in Economic Modeling

This paper provides a systematic review of the emerging literature on multiagent reinforcement learning (MARL) in economics. The review is motivated by the rapid expansion of MARL methods in computational modeling and the lack of a clear framework that connects this literature to established economic approaches. The paper first introduces the shared dynamic structure underlying MARL and economic decision problems, including Markov decision processes and Bellman equations, and clarifies the distinction between adaptive agents and rule-based agents. It then organizes existing applications of MARL in four domains: macroeconomics, finance, market design, and behavioral economics. Across these domains, the review identifies the main research themes, common modeling choices, and current limitations. A central argument of the paper is that progress in this field depends not only on stronger algorithms but also on clearer standards for economic interpretation, model design, and empirical credibility. To address this issue, the paper develops a trustworthiness-centered evaluation perspective and outlines an operational architecture for applying MARL in economic research. The contribution is conceptual rather than algorithmic: The paper offers a shared analytical vocabulary, a structured map of the literature, and a roadmap for future work at the intersection of economics and computer science.