基于异构自回归模型的股市波动率研究

The Study on Stock Market Volatility Based on Heterogeneous Autoregressive Model

摘要: 本文以传统的异构自回归模型(HAR)为基础，考虑到新闻信息文本对于股票市场波动的影响，构建了新的情感词典，将情绪指标(AI)与基础的HAR模型相结合，并将深度学习中具有学习长短期依赖状态特点的门控循环单元(GRU)模型引入进来，形成了GRU-AI模型。然后介绍了波动率预测模型中的评估方法——损失函数法以及模型信度集(MCS)检验法。最后进行实证分析表明将情绪指标加入到波动率模型中可以提高模型样本外的预测精度。

Abstract: Based on the traditional heterogeneous autoregressive model (HAR), considering the impact of news information text on stock market fluctuations, this paper constructs a new emotion dictionary, combines emotion indicator (AI) with the basic HAR model, and introduces the Gated Recycle Unit (GRU) model, which has the characteristics of long and short term dependent state learning in deep learning, to form the GRU-AI model. And then the evaluation methods of volatility prediction model—loss function method and model reliability set (MCS) test method are introduced. Finally, the empirical analysis shows that adding sentiment index to the volatility model can improve the prediction accuracy of the model out of sample.