LLM-Powered Multi-Agent System for Automated Crypto Portfolio Management

The article presents a multi-agent system (MAS) for automated cryptocurrency portfolio management, integrating three specialized agents: a Crypto Agent for market dynamics, a News Agent for sentiment analysis, and a Trading Agent for executing strategies. Evaluated over a 52-week backtest, the best configuration achieved a cumulative return of 133.52% with a Sharpe ratio of 1.502, significantly outperforming single-agent models and traditional benchmarks. This approach enhances interpretability by ensuring that all portfolio decisions are traceable to the agents' reasoning, addressing the opacity issues of deep learning models in high-volatility environments.