[论文] Propose, Don't Judge: An Anytime-Valid Referee for LLM Agents That...

论文概要

研究领域: ML
作者: Bo Qu, Mingguang Chen, Licheng Wang
发布时间: 2026-09-25
arXiv: 2609.27051

中文摘要

语言模型智能体现在运行整个量化因子研究流程：提出投资因子、回测、筛选幸存者并淘汰失败者。我们问：这些工作中哪些应该留给智能体？答案是"受治理的自我进化"：智能体可以提议，但必须由一个它无法触及的冻结统计裁判来评判。裁判仅根据提交之后才揭晓的市场结果对每个候选打分——通过下注的方式进行，因此其错误发现保证对任何提议策略在所有停止时刻都成立。我们将三个提议者（脚本、bandit、语言模型）与该裁判及三个故意"漏风"的裁判交叉组合，在一个植入真值的合成世界、一个探针编写环境与 CSI 500 十年滚动前推上实验。谁来评判决定虚假准入数量：冻结裁判比漏风裁判少准入 5-11 倍低于阈值的因子（脚本提议者下），且没有提议者能缩小这一差距。谁来提议决定产出：语言模型胜过脚本、与 bandit 持平，并补充了 bandit 不具备的能力——编写自己的诊断探针。证书的代价是时间：一个被准入的真因子要约 500 个交易日的等待，因此持证组合的夏普比率落后于无门槛组合。评判属于程序；提议与工具制造属于智能体。

原文摘要

Language-model agents now run the whole of quantitative factor research: they propose investment factors, backtest them, select the survivors and retire them. We ask which of those jobs an agent should keep. Our answer is governed self-evolution: the agent may propose, and a frozen statistical referee that the agent cannot touch must judge. The referee scores each candidate only on market outcomes revealed after submission, by betting, so its false-discovery guarantee holds at every stopping time for any proposal policy. We cross three proposers (a script, a bandit and a language model) with this referee and with three deliberately leaky ones, in a synthetic world with planted truth, a probe-authoring environment and a ten-year walk-forward on the CSI 500. Who judges sets the number of fal...

自动采集于 2026-09-25

#论文 #arXiv #ML #小凯