基于LSTM网络和投资者情绪的A股大盘指数可预测性研究

Predictability of A-Share Market Index Based on LSTM Network and Investor Sentiment

摘要: 资本市场的股票价格走势深刻反映并影响着国家的宏观经济政策。随着国家对深度学习和机器学习在经济学问题中应用的推动，将这些技术与经济学结合已成为当前的研究热点。近年来，提取文本中的有效信息并研究其对股票价格波动的影响，已引起学术界广泛关注。本文利用2023年12月4日至2024年12月2日的A股大盘指数数据，基于财经新闻情感分析构建的投资者情绪指数，提出了一种融合情感特征的股指预测模型。研究表明，加入投资者情绪指标的预测模型在预测精度和预测值与真实值的接近度等方面显著优于其他机器学习模型，投资者情绪指数呈现出比其他特征变量更强的特征重要性。将预测的股指进行替换后，结论仍保持一致。

Abstract: The stock price trends in capital markets profoundly reflect and influence national macroeconomic policies. With the growing emphasis on the application of deep learning and machine learning in economics, integrating these technologies with economic research has become a current focus. In recent years, extracting valuable information from texts and studying its impact on stock price fluctuations have attracted widespread academic attention. This paper utilizes A-share market index data from December 4, 2023, to December 2, 2024, and an investor sentiment index constructed based on financial news sentiment analysis. It proposes a stock index prediction model that incorporates sentiment features. The study shows that the model with the investor sentiment index significantly outperforms other machine learning models in terms of prediction accuracy and the closeness between predicted and actual values. Additionally, the investor sentiment index demonstrates stronger feature importance than other variables. The results remain consistent when the predicted stock index is replaced.