Reproduce “Automate Strategy Finding with LLM in Quant Investment” on OpenResearch

About this paper

Zhizhuo Kou, Holam Yu, Junyu Luo, Jingshu Peng, Xujia Li, Chengzhong Liu, Juntao Dai, Lei Chen and 2 more

Researchers from The Hong Kong University of Science and Technology and Peking University developed a three-stage framework leveraging Large Language Models (LLMs) and a multi-agent system to automate quantitative investment strategy discovery and portfolio management. The system achieved a 53.17% cumulative return on the SSE50 index, dramatically outperforming the benchmark's -11.73% over the same period, and demonstrated robust, adaptive performance across diverse market conditions.

Key insights

• LLMs can effectively extract and formulate executable alpha factors from diverse financial literature, enabling a more adaptive and comprehensive alpha discovery process than traditional methods.
• A multi-agent evaluation system, integrating Confidence Score and Risk Preference agents, is critical for achieving robust and adaptive alpha selection, particularly in volatile market conditions and for downside protection.
• The synergistic combination of LLM-generated alphas, even those with individually lower Information Coefficients, contributes significantly to the overall strong predictive power and superior portfolio performance.

Summary by alphaXiv.