Reproduce “MadEvolve: Evolutionary Optimization of Trading Systems with Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yurii Kvasiuk, Tianyi Li, Owen Colegrove, Moritz Münchmeyer
An LLM-driven evolutionary optimization framework, MadEvolve, successfully developed and improved quantitative trading systems by generating novel strategies and features. The framework achieved out-of-sample Sharpe ratio increases of up to 1.83 points and PnL gains exceeding 25x, demonstrating algorithmic discovery in noisy financial environments.
Key insights
- LLM-driven agentic evolutionary search can effectively optimize complex problems in noisy financial environments, extending its applicability beyond deterministic domains and showing resilience to inherent market noise.
- Observed performance improvements in trading systems are attributable to genuine algorithmic innovations and structural changes, rather than merely increased position sizing or artifacts of statistical overfitting.
- An ensemble of diverse LLMs contributes complementary mutation strategies, leveraging different model strengths to accelerate the discovery of high-performing and robust solutions in the evolutionary process.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.