FinCon: A Synthesized LLM Multi-Agent System with Conceptual Verbal Reinforcement for Enhanced Financial Decision Making

• DOI: 10.48550/arxiv.2407.06567
• Publication Date: 2024-07-09
• Authors: Yangyang Yu; Zhiyuan Yao; Haohang Li; Zhiyang Deng; Yupeng Cao; Zhi Chen; Jordan W. Suchow; Rong Liu; Zhenyu Cui; Denghui Zhang; K. P. Subbalakshmi; Guojun Xiong; Yueru He; Jimin Huang; Li Dong; Qianqian Xie
• Keywords: Conceptual framework

Large language models (LLMs) have demonstrated notable potential in conducting complex tasks and are increasingly utilized various financial applications. However, high-quality sequential investment decision-making remains challenging. These require multiple interactions with a volatile environment for every decision, demanding sufficient intelligence to maximize returns manage risks. Although LLMs been used develop agent systems that surpass human teams yield impressive returns, opportunities enhance multi-sourced information synthesis optimize outcomes through timely experience refinement remain unexplored. Here, we introduce the FinCon, an LLM-based multi-agent framework CONceptual verbal reinforcement tailored diverse FINancial tasks. Inspired by effective real-world firm organizational structures, FinCon utilizes manager-analyst communication hierarchy. This structure allows synchronized cross-functional collaboration towards unified goals natural equips each greater memory capacity than humans. Additionally, risk-control component enhances decision quality episodically initiating self-critiquing mechanism update systematic beliefs. The conceptualized beliefs serve as future agent's behavior can be selectively propagated appropriate node requires knowledge updates. feature significantly improves performance while reducing unnecessary peer-to-peer costs. Moreover, demonstrates strong generalization capabilities tasks, including single stock trading portfolio management.