基于BERT-BILSTM模型的投资者情绪对股价短期预测  

The Study of Investor Sentiment of Short-term Stock Price Prediction Basedon BERT-BILSTM Model

作　　者：陈志芳 李佳员 CHEN Zhifang;LI Jiayuan(College of Statistics and Mathematics,Inner Mongolia University of Finance and Economics,Hohhot,China,010070)

机构地区：[1]内蒙古财经大学统计与数学学院,内蒙古呼和浩特010070

出　　处：《财经理论研究》2024年第6期28-38,共11页Journal of Finance and Economics Theory

基　　金：国家社会科学基金重大项目(18ZDA127);内蒙古自然科学联合基金项目(2023LHMS07009);内蒙古自治区研究生科研创新项目(KCLX2023-120);内蒙古经济数据分析与挖掘重点实验室研究课题(SX23005)。

摘　　要：行为金融学认为股票价格受到投资者情绪的影响。本研究旨在探讨投资者情绪对股票短期收益的影响及其预测作用。通过网络爬虫技术采集东方财富股吧中沪深300指数成分股的金融文本数据,并根据文本内容是否包含对未来几个交易日股票收益的预期,将近1万条样本数据分为“看涨”“看跌”和“中性”三种情感。基于这些分类数据,训练了BERT-BILSTM模型以实现海量文本的自动分类。这是一种将BERT和BILSTM结合的模型,通过预训练和深度学习的方式,提高了情感分类的准确性和效率。随后,本文构建了投资者情绪和情绪分歧度指标,用于分析投资者情绪对股票收益率和交易量的预测作用。实证结果显示,投资者情绪对当日股票收益率有显著正向影响,并对未来1~5个交易日的股票收益率有明显的预测作用;情绪分歧度对当日股票交易量有显著正向影响,且能预测未来1~3个交易日的股票交易量。本文还进行了稳健性检验,更改了投资者情绪指标的计算方式,在不改变控制变量的前提下,稳健性检验的结果依然显示情绪指标能够显著影响当日及未来数日的股票收益率和交易量,只是系数的显著性有所降低。这进一步验证了本文模型的可靠性和适用性。在本文的实证分析中,还发现股吧发帖量对股票收益率显著负相关,这可能反映了市场中高讨论热度并未转化为正向的市场表现,反而可能引发市场的过度反应。综合上述发现,本文的实证结果支持假设H1和H2的推断,即投资者情绪和情绪分歧度对股票市场有显著影响。Behavioral finance suggests that stock prices are influenced by investor sentiment.This paper investigates the impact of investor sentiment on short-term stock returns and its predictive power.Financial text data for the constituent stocks of the CSI 300 index were collected from the stock forums of Oriental Fortune using web scraping technology.Based on whether the text expressed expectations about stock returns in the coming trading days,approximately 10,000 samples were categorized as"bullish,""bearish,"or"neutral."Using this labeled data,the BERT-BILSTM model was trained to automatically classify large volumes of text.The model combines BERT(Bidirectional Encoder Representations from Transformers)and BILSTM(Bidirectional Long Short-Term Memory)to improve the accuracy and efficiency of sentiment classification through pre-training and deep learning techniques.This paper constructs metrics for investor sentiment and sentiment divergence to analyze their predictive role on stock returns and trading volume.The empirical results show that investor sentiment has a significant positive effect on same-day stock returns and can significantly predict stock returns over the next one to five trading days.Sentiment divergence has a significant positive effect on same-day trading volume and can predict trading volume for the next one to three trading days.A robustness test was also conducted,altering the calculation of investor sentiment indicators while keeping control variables constant.The test results still show that sentiment indicators significantly affect same-day stock returns and trading volume in the following days,though the significance of the coefficients decreases slightly.This further validates the reliability and applicability of the model used in this paper.Additionally,the empirical analysis reveals that the volume of forum posts is negatively correlated with stock returns.This may suggest that high levels of market discussion do not lead to positive performance;instead,they may trigger market overreactio

关 键 词：投资者情绪 情绪分歧度 股票收益率 文本挖掘 

分 类 号：F830.91[经济管理—金融学]