A Multi-agent System Based On LLM For Trading Financial Assets

In this paper, we propose an agent-based system including Large Language Models (LLM).It depicts a multi-agent system architecture for cryptocurrency trading, consisting of five distinct agents: an Analyst, a Data Scientist, a Strategy Developer, a Trading Advisor, and a Risk Manager.Each agent performs a specific task using search and scraping tools to gather necessary information.These agents then collaborate to manage all aspects of the trading process.The OpenAI API is utilized to generate outputs from the agents.Three trading scenarios are proposed: moderate risk tolerance with position trading, high risk tolerance with scalping, and low risk tolerance with swing trading.Experimental results show that each agent effectively contributes to the trading strategy.For instance, position trading benefits from comprehensive risk analysis and mitigation strategies, scalping relies on detailed execution plans and risk management protocols, and swing trading focuses on market trends and regulatory impacts.