SSRN-大语言模型与中国股票市场收益率可预测性研究

Large Language Models and Return Prediction in China First Version: November 7th, 2023 This Version: November 7th, 2024 Lin Tan, Huihang Wu and Xiaoyan Zhang Abstract We examine whether large language models (LLMs) can extract contextualized representation of Chinese public news articles to predict stock returns. Based on representativeness and influences, we consider seven LLMs: BERT, RoBERTa, FinBERT, Baichuan, ChatGLM, InternLM, and their ensemble model. We show that news tones and return forecasts extracted by LLMs from Chinese news significantly predict future returns. The value-weighted long-minus-short portfolios yield annualized returns between 35% and 67%, depending on the model. Building on the return predictive power of LLM signals, we further investigate its implications for information efficiency. The LLM signals contain firm fundamental information, and it takes two days for LLM signals to be incorporated into stock prices. The predictive power of the LLM signals is stronger for firms with more information frictions, more retail holdings and for more complex news. Interestingly, many investors trade in opposite directions of LLM signals upon news releases, and can benefit from the LLM signals. These findings suggest LLMs can be helpful in processing public news, and thus contribute to overall market efficiency. Keywords: return prediction, news articles, large language models, information efficiency, Chinese stock market. JEL Codes: C52, C5, G1, G14.  Lin Tan (tanl.19@pbcsf.tsinghua.edu.cn), Huihang Wu (wuhh@pbcsf.tsinghua.edu.cn), and Xiaoyan Zhang (zhangxiaoyan@pbcsf.tsinghua.edu.cn) are all at the PBC School of Finance at Tsinghua University. We thank Kaiji Chen, Lei Chen, Lin William Cong, Byoung-Hyoun Hwang, Fuwei Jiang, Raymond Kan, Xing Liu, Xiumin Martin, Jun Tu, Xiaolu Wang, Dacheng Xiu, Nianhang Xu, Bernard Yeung, Xintong Zhan, Xingjian Zheng, Guofu Zhou, conference participants at 2024 ABFER-JFDS Conference on AI and Fintech, 2024 China Fintech Research Conference (CFTRC), 2024 Summer Institute in Finance (SIF) Annual Conference, and seminar participants at Sun Yat-Sen University, Tsinghua University and 2024 Summer Institute in Digital Finance (SIDF) for their helpful comments. We appreciate Zicheng Wang’s research assistance. We gratefully acknowledge financial support from the National Natural Science Foundation of China [Grant No. 72350710220]. All remaining errors are our own. Corresponding author: Xiaoyan Zhang, PBC School of Finance, 43 Chengfu Road, Beijing, China, 100083, zhangxiaoyan@pbcsf.tsinghua.edu.cn. Large Language Models and Return Prediction in China Abstract We examine whether large language models (LLMs) can extract contextualized representation of Chinese public news articles to predict stock returns. Based on representativeness and influences, we consider seven LLMs: BERT, RoBERTa, FinBERT, Baichuan, ChatGLM, InternLM, and their ensemble model. We show that news tones and