TradingAgents: Multi-agent LLM financial trading framework## Summary

TradingAgents: Multi-agent LLM financial trading framework Summary In recent years, multi-agent collaboration driven by large language models (LLM) has made significant progress in automated decision-making and problem solving. However, in the financial field, most research is still limited to a single agent processing specific trading tasks, or using a multi-agent framework to independently collect data, and has not yet fully simulated the "division-collaboration" team dynamics within a real trading company. The "TradingAgents" framework proposed in this article draws on the organizational structure of professional trading companies and creates a multi-agent team based on LLM, including multiple roles engaged in fundamental research, sentiment analysis, technical analysis, and trading with different risk preferences; the framework also sets up long-short view researchers (Bull and Bear researchers) to balance market views, and equips a risk management team to monitor exposures. Under this system, trading agents make decisions by integrating team debates and historical data, simulating a real-world collaborative trading environment. Experimental results show that compared with the baseline model, this framework has significant improvements in indicators such as cumulative returns, Sharpe ratio, and maximum drawdown, proving the potential of multi-agent LLM in financial trading applications. Introduction The combination of large language models (LLMs) and…