大语言模型、文本情绪与金融市场  

Large Language Model and Textual Sentiment Analysis in Chinese Stock Markets

作　　者：姜富伟 刘雨旻 孟令超[4] Jiang Fuwei;Liu Yumin;Meng Lingchao(School of Economics,Xiamen University;The Wang Yanan Institute for Studies in Economics,Xiamen University;School of Finance,Central University of Finance and Economics;China School of Banking and Finance,University of Interational Business and Economics)

机构地区：[1]厦门大学经济学院 [2]厦门大学王亚南经济研究院 [3]中央财经大学金融学院 [4]对外经济贸易大学中国金融学院

出　　处：《管理世界》2024年第8期42-59,共18页Journal of Management World

基　　金：国家社科基金重大项目“三重压力下双支柱调控的政策效应评估与优化研究”(22&ZD063);国家自然科学基金面上项目(72072193,71872195,72342019);中央财经大学青年科研创新团队和科教融合研究生学术新星孵化计划(202303)的资助。

摘　　要：“人工智能+”行动是发展新质生产力的重要途径,其在金融领域的应用有助于金融强国建设。本文创新性地融合结构化金融市场数据和非结构化金融文本大数据,并结合中国特色金融市场的独特特征,训练了一个更适用于我国金融领域的中文金融大语言模型,并开展金融市场情绪测度和资产价格风险预测。研究发现,与传统字典法相比,使用中文金融大语言模型构建的大模型情绪在金融市场回报预测方面表现显著更佳。大模型情绪对很多宏观经济变量也有显著预测能力,能够捕捉非理性情绪冲击对宏观经济基本面的影响。大模型情绪在经济下行和极端风险事件期间的预测效果更强,契合了金融理论中非理性情绪对金融市场和宏观经济会产生非对称与非线性影响的结果。综上,本研究展现了“人工智能+”行动在我国金融领域应用落地的潜在技术路径和理论逻辑。The AI Plus initiative is an important way to develop new quality productive forces,and its implementation in the financial industry contributes to building a financial powerhouse.This paper trains a Chinese financial large language model more suitable for Chinese stock markets by innovatively combining structured financial data and unstructured financial textual big data and taking into account the unique characteristics of the Chinese financial market.This paper then measures the financial market sentiment and predicts the financial market risks using the trained large language model.Empirical results show that the constructed large model sentiment variable using the Chinese financial large language model outperforms traditional dictionary methods in predicting asset prices.The large model sentiment variable can significantly predict macroeconomic variables across dfferent economic conditions,capturing the impact of irrational shocks on macroeconomic fundamentals.The predictive power of the large model sentiment variable on market returns is stronger during economic downturns and extreme risk events,aligning with the theory that irrational sentiment has asymmetric and nonlinear effects on financial markets and the macroeconomy.Overall,our paper illustrates the potential pathways and the theoretical logic for integrating the Al Plus initiative into the financial industry.

关 键 词：文本情绪 深度学习 大语言模型 资产定价 

分 类 号：F832.5[经济管理—金融学] TP18[自动化与计算机技术—控制理论与控制工程]