AlphaDiverse: Post-Training Local Quantitative Research Agents for Diverse Exploration in Alpha Factor Mining

(Work done during an internship at Shanghai Non-convex Intelligent Technology.)

Qingzhuo Wang, Zikun Wei, Zhihua Wei, Wen Shen

AlphaDiverse combines multi-agent alpha factor research, diverse research path collection, and post-training of local agents. Complementary research plans and varied research environments produce diverse traces for supervised fine-tuning of local Planner and Realizer agents. A joint GRPO method then optimizes both agents using rewards for predictive quality and complementary factor contributions.

Research feedback is restricted to an inner data period, with the frozen final model evaluated on a later outer period. Experiments across four Chinese stock universes show competitive predictive performance alongside broader exploration of economic mechanisms.

Recommended citation: Qingzhuo Wang, Zikun Wei, Zhihua Wei, Wen Shen. (2026). "AlphaDiverse: Post-Training Local Quantitative Research Agents for Diverse Exploration in Alpha Factor Mining." arXiv 2026.
Download Paper