SupplyNet: Large Language Model-Based Multi-Agent Simulation of Credit Risk Contagion in Supply Chain Networks

Modern supply chains have evolved into digitally enabled value networks that coordinate production, financing, and information flows across globally connected firms. Although this interdependence improves operational efficiency, it also amplifies systemic vulnerability. However, existing simulation models of supply chain risk contagion often rely on mechanistic assumptions: they either assume that risk transmission follows a fixed infection rate or represent supply chain firms as homogeneous nodes that make decisions according to preset rules. These approaches overlook the influence of cognitive factors on supply chain risk propagation in the real world. Large language models (LLMs) can incorporate cognitive factors into supply chain simulations by enabling agents to interact through natural-language dialogue. Accordingly, we propose SupplyNet, an LLM-driven framework for simulating liquidity risk contagion in supply chains. To the best of our knowledge, it is the first LLM-driven simulation framework for liquidity risk contagion in supply chains. We validate the framework on a supply chain topology constructed from real-world data. Building on this foundation, we systematically analyze the propagation pathways of liquidity risk across the supply chain network and evaluate the extent to which various factors mitigate its spread.