Paper Type
Short
Paper Number
PACIS2026-2149
Description
Digital capital market interactions face strategic disclosure risks, yet static empirical methods fail to capture dynamic investor-company-regulator tripartite games. This paper develops a localized LLM multi-agent simulation framework (LangGraph + LoRA-tuned Qwen2.5-7B) for capital market regulation, simulating tripartite dynamic interactions with 3 regulatory intensity levels (0.25/0.50/0.80) and generating 2400 experimental records. Results confirm a significant negative correlation between regulatory intensity and stock price performance, a transparency saturation effect in long-term games, and linear amplification of penalty intensity with regulatory stringency. The framework addresses cloud LLM inefficiencies and inaccurate regulatory intensity quantification, enabling low-cost, large-scale policy simulation. It provides a data-driven experimental platform for regulatory policy optimization and enriches LLM applications in financial dynamic game research.
Recommended Citation
Huang, Jinshui, "A Localized LLM Multi-Agent Framework for Tripartite Game Simulation in Capital Market Regulation" (2026). PACIS 2026 Proceedings. 11.
https://aisel.aisnet.org/pacis2026/practioner/practioner/11
A Localized LLM Multi-Agent Framework for Tripartite Game Simulation in Capital Market Regulation
Digital capital market interactions face strategic disclosure risks, yet static empirical methods fail to capture dynamic investor-company-regulator tripartite games. This paper develops a localized LLM multi-agent simulation framework (LangGraph + LoRA-tuned Qwen2.5-7B) for capital market regulation, simulating tripartite dynamic interactions with 3 regulatory intensity levels (0.25/0.50/0.80) and generating 2400 experimental records. Results confirm a significant negative correlation between regulatory intensity and stock price performance, a transparency saturation effect in long-term games, and linear amplification of penalty intensity with regulatory stringency. The framework addresses cloud LLM inefficiencies and inaccurate regulatory intensity quantification, enabling low-cost, large-scale policy simulation. It provides a data-driven experimental platform for regulatory policy optimization and enriches LLM applications in financial dynamic game research.
Comments
16-Practitioner