Reproduce “CosmosAlpha: Multi-Relational Graph Search with Pareto-Guided Exploration for LLM-Driven Alpha Mining” 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

Zhan-He Wang, Zi-Yang Lang, Yi-Yang Li, Qing-Yu Lan, Yun-Song Piao

CosmosAlpha-Graph searches for factors by connecting formulas through parentage, shared expression structure, and stock exposures, then using graph neighborhoods to guide an LLM’s proposals. Against the authors’ MCTS-based proxy, it ranks better on seven predictive and portfolio measures in both Chinese A-share universes. Its mean rank is 1.43 versus 7.14 on CSI 300, on held-out 2021–2025 data, not live-trading evidence.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.