Simulating Market Equilibrium with Large Language Models

Large Language Models (LLMs) have the potential to simulate complex human decision-making and economic behavior, making them well-suited for training simulations. This study explores LLMs' ability to simulate a market equilibrium game commonly used in MBA classrooms to train future business leaders. We test three simulation architectures: prompt-based, Retrieval Augmented Generation-based, and controller-based. The prompt-based approach struggled with limited context, while the RAG-based system improved information retrieval but occasionally veered off course. To address these challenges, we developed a controller-based simulation integrating multiple LLM agents and custom tools, which enhanced both control and accuracy. Our results show that this approach not only improves the fidelity of economic simulation, but also enriches the learning experience for students by providing more accurate, engaging, and realistic business case environments.