AlgoEvolve: LLM-Driven Meta-Evolution for Algorithmic Trading.

The domain of algorithmic trading presents unique challenges due to its noisy, non-stationary, and discontinuous nature. This research introduces AlgoEvolve, an innovative framework that leverages large language models (LLMs) for the evolutionary discovery and refinement of trading programs. AlgoEvolve generates executable Python-based trading strategies, which are then rigorously evaluated. The system demonstrates an ability to develop emergent, regime-adaptive strategy logic, including autonomous adjustments to trading rules in response to market conditions. A key innovation is a meta-evolutionary outer loop that evolves the prompts used to guide the LLM's program synthesis in the inner loop. This outer loop discovers superior search heuristics, effectively balancing exploration and exploitation while minimizing failures. These LLM-generated heuristics consistently outperform initial human-designed instructions, showcasing the potential of semantic evolution for continuous program synthesis in complex, real-world environments like financial markets.