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

“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.