Researchers presented the EvolveTrade framework for self-learning AI traders

Researchers presented EvolveTrade — a framework for automatic optimization of trading AI-agents' strategies. The system allows models to dynamically adjust system prompts based on accumulated experience and transaction results without changing the weights of the underlying LLM.

In contrast to traditional agents with fixed instructions, EvolveTrade uses a separate "Policy Agent". This component analyzes the history of decisions and portfolio performance metrics, then updates the rules for data collection, signal verification, and risk management. This approach ensures the agent's adaptation to changing market regimes without the need to retrain the main model.

Experiments showed that using self-developing policies leads to an increase in the Sharpe Ratio and cumulative return (Cumulative Return) compared to agents working on static prompts. Analysis of the system's behavior confirmed that updated policies more often activate code analysis and calculation tools specific to the current market situation.

The authors note that adapting the procedures for using tools is a critical factor in increasing the reliability of autonomous financial systems. The work is published under the CC BY 4.0 license, which allows the use of the proposed methodology for developing more flexible trading algorithms.