Reproduce “FINCOSMOS: An Open-World Evaluation Paradigm for Agents in Finance” 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
Yupeng Cao, Xueqing Peng, Lingfei Qian, Haohang Li, Yangyang Yu, Anke Xu, Wenbo Cao, Yan Wang and 9 more
FINCOSMOS evaluates financial agents by asking them to elicit hidden strategy details, source their own market data, and honor a data cutoff. It grades delivered files against the requested strategy and independently checks data dates and backtest calculations. Across 150 sessions on ten frozen, long-only U.S. equity tasks, none met acceptance; all 35 user-signed-off sessions failed at least one veto.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.