AlgoEvolve: LLM-driven Meta-evolution of Algorithmic Trading Programs
AlgoEvolve is a new framework that uses Large Language Models (LLMs) to create and improve trading strategies through a two-level evolutionary process. It adapts to changing markets and outperforms traditional models, achieving an annualized Sharpe ratio of 5.60. The system evolves both trading logic and the prompts guiding LLMs, introducing terms like Prompt Genome and semantic mutation. Despite strong results, the research notes a complexity ceiling and questions about scalability and real-world use.
What it examines
This paper introduces AlgoEvolve, a framework using Large Language Models (LLMs) to evolve and improve algorithmic trading strategies. It combines an inner loop for strategy generation and an outer loop for evolving search prompts, aiming to adapt to noisy, changing financial markets and discover robust trading logic.
What it concludes
AlgoEvolve shows LLMs can autonomously create and refine trading strategies, achieving high risk-adjusted returns. Its approach can be used for continual program synthesis in finance and other complex fields. Future work may focus on modular program synthesis to overcome complexity limits and expand practical applications.
Evidence objects
AlgoEvolve leverages Large Language Models in a novel two-level evolutionary process, autonomously adapting trading strategies to changing markets and achieving a remarkable annualized Sharpe ratio of $5.60$, far surpassing traditional methods.
key_findings bullet 1 · key_findings · validation V0
The framework introduces a meta-evolutionary architecture: LLMs act as 'semantic mutation operators' refining Python strategies, while evolving promptstermed 'Prompt Genome'enables reasoning-driven, regime-adaptive strategy discovery beyond human-designed rules.
key_findings bullet 2 · key_findings · validation V0
Rigorous walk-forward testing on real market data confirms AlgoEvolve's effectiveness, but the system faces a 'complexity ceiling' in synthesizing advanced logic, raising questions about scalability and real-world deployment for future research.
key_findings bullet 3 · key_findings · validation V0
AlgoEvolve presents a novel hierarchical LLM-driven evolutionary framework for algorithmic trading, uniquely using LLMs as semantic mutation operators in both inner and meta-evolutionary loops. This enables autonomous synthesis and prompt optimization, offering interpretability, regime adaptation, and zero inference-time latency. Its emergent logic outperforms human-designed strategies, demonstrating significant originality and impact.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
Recent work shows that Large Language Models (LLMs) can act as semantic mutation operators for the evolutionary discovery of programs and proofs. Most current applications focus on static coding benchmarks. We extend this paradigm to algorithmic trading. This domain is uniquely challenging because it is noisy, non-stationary, and highly discontinuous. We present AlgoEvolve, an LLM-driven evolutionary framework that generates, evaluates, and iteratively improves executable trading strategies. These strategies are expressed as Python code and evaluated through a rigorous testing protocol. Across multiple experiments, the system exhibits emergent regime-adaptive strategy logic, including autonomous shifts in trading rules. We further introduce a meta-evolutionary outer loop that evolves the prompts guiding program synthesis in the inner loop. This outer loop discovers improved search heuristics. These heuristics balance exploration and exploitation while reducing zero-trade failures. They consistently outperform initial human-designed instructions. The results demonstrate that LLM-based semantic evolution provides a viable approach for continual program synthesis in complex environments.
Source row: 150 · abstract type: unknown