Eco3S Framework Simulates Socio-Economic Systems with LLM-Based Agents

The emergence of large language models (LLMs) has revitalized interest in agent-based modeling (ABM) for simulating complex systems. However, current LLM-based ABM approaches struggle with dynamic agent-environment interactions, flexible counterfactual analysis, and automating scientific simulation processes. To overcome these limitations, researchers introduce Eco3S, a novel framework designed for socio-economic system simulation, particularly useful for economic research and policy analysis. Eco3S incorporates three core mechanisms: a Co-evolving Environment Design that creates a bidirectional feedback loop between agents and their environment for realistic emergent behaviors; Structural Causal Simulation, inspired by structural causal models, enabling flexible interventions for diverse causal inference tasks; and a Simulation-Analysis-Refinement Paradigm, which iteratively improves experimental designs based on previous simulation outcomes. The framework has been validated through experiments replicating established economic studies and phenomena, demonstrating its effectiveness, scalability, and generalizability for rigorous economic research and policy-making.