arXiv cs.CL· Jiacheng Guo, Suozhi Huang, Shuzhen Li, Yunlong Gao, Zerui Cheng, Jason Ge, Shushu Liang, Zihao Li, Hao Lu, Ming Yin, Shilong Liu, Jiashuo Liu, Xu Kuang, Mengdi Wang·· 9 小时前AI 评分38
语言模型能否学会预测股价?研究者构建 BETA 沙盒并用智能体强化学习训练出 AURA
Can Language Models Learn to Forecast Stock Prices?
AI 导读
研究者构建 BETA 沙盒,让模型调用工具检索决策日前可得的价格、指标与新闻,再预测标普 500 个股未来收益,发现前沿模型存在测试期越近分数越低、过度预测上涨、预测幅度偏小三个弱点。
正文 · 原文
Authors:Jiacheng Guo, Suozhi Huang, Shuzhen Li, Yunlong Gao, Zerui Cheng, Jason Ge, Shushu Liang, Zihao Li, Hao Lu, Ming Yin, Shilong Liu, Jiashuo Liu, Xu Kuang, Mengdi Wang
Abstract:We study whether language models can learn to forecast stock returns through training. We build BETA, a chronological sandbox over S&P 500 stocks in which a model calls tools to retrieve prices, indicators, and news available up to a decision date, and then forecasts the stock's future return. Evaluating frontier models in this sandbox reveals three weaknesses: their scores decline as the test period approaches the present while a simple momentum baseline stays stable, they predict gains far more often than gains occur, and they predict moves much smaller than those that are realized. We first fine-tune on filtered, hindsight-generated tool-use trajectories for historical tasks judged predictable from the evidence available at the time. We then apply reinforcement learning with rewards based on realized returns. During RL, long investigations frequently exceed the context window. We therefore introduce a selective compaction tool that allows the model to compress chosen portions of its interaction history. Selective compaction changes the context available to subsequent decisions, so policy updates must preserve the context in which each action was generated. Meanwhile, rollouts vary substantially in the number and length of their trainable turns, making a shared group-level baseline less effective. We address these challenges with Context-Aligned Turn-Level PPO, which uses a turn-level critic and computes policy updates using each turn's actual context. The resulting models, which we call AURA, substantially reduce all three weaknesses. Training improves the 8B model's score from 24.0 to 47.0 and the 30B model's from 27.4 to 54.7, with the latter surpassing all evaluated frontier models on our benchmark and achieving its largest margin on the most recent test period. These results suggest that agentic RL can improve performance on tasks with noisy, delayed real-world outcomes.
| Comments: | 21 pages, 10 figures |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2609.36914 [cs.CL] |
| (or arXiv:2609.36914v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2609.36914 arXiv-issued DOI via DataCite |
Submission history
From: Jiacheng Guo [view email]
[v1]
Tue, 29 Sep 2026 07:34:39 UTC (352 KB)
[v2]
Tue, 17 Nov 2026 11:47:09 UTC (526 KB)
来源:arXiv cs.CL · arxiv.org