基于分位数Granger因果的网络情绪与股市收益关系研究

 行为金融学理论认为，股票市场的价格变动除受宏观基本因素影响外，还在很大程度上受众多个体投资者或噪音交易者行为左右。中国股票市场拥有庞大的个人投资者群体，且股民群体与网民群体之间具有高度耦合性，使用网络情绪等信息能够探索中国股市收益变动基本规律。       为揭示个体投资者行为对股市收益的影响，以个体投资者情绪为视角，以网络环境中个体投资者的情绪信息为切入点，检验投资者情绪与股市收益的关联关系，评估网络情绪信息价值。使用中文文本情感分析方法，从新浪微博文本中提取出网络情绪时间序列；分别运用均值Granger因果和分位数Granger因果检验方法，探讨网络情绪波动与股市收益之间是否存在因果关系；将股票市场发展阶段进行细致划分，研究不同市场阶段下网络情绪波动与股市收益之间的因果关系。       对沪深300指数收益进行实证研究，结果表明，尽管在均值框架下网络情绪波动与股市收益之间因果关系并不明显，但基于分位数Granger因果分析却发现两者在极端分位点区间处存在广泛且显著的因果关系。数据显示，在40个因果关系检验中，分位数Granger因果检验的因果关系发现了23个显著的因果关系，发现率为57.5%，远高于均值Granger因果检验的7.5%。此外，股市收益受到网络情绪波动影响的程度和方式在不同市场阶段下有所不同。       研究结果具有一定的理论意义和应用价值。在一些特定分位点区间网络情绪波动对股市收益存在显著因果关系影响，这为在特定条件下股市收益的可预测性提供了佐证。网络情绪能够预测股市收益的尾部(上尾或下尾)行为特征，可以为金融风险防范提供决策参考。研究结果为股票市场的定价、收益预测和波动率估计等相关研究提供了新的研究思路，也为网络情绪信息使用提供了新的方向。

Exploring the Relationship between Internet Sentiment and Stock Market Returns Based on Quantile Granger Causality Analysis

 According to the behavioral finance theory, besides fundamental macro-economic factors, the change of stock market price is also influenced by individual investors or noise traders to a large extent. There are a large amount of individual investors in Chinese stock market. Most of them are also Internet users. Due to the high degree of coincidence between the group of individual investors and Internet users, it is possible to explore the fundamental rule of stock returns in China using internet sentiment information.       To investigate the impacts of individual investor behavior, we examine the relationship between investor sentiment and stock returns. In particular, we focus on and evaluate the impacts of internet sentiment information on stock returns from the perspective of individual investor sentiment. First, we extract several Internet Sentiment time series from Sina Weibo texts using Chinese text sentiment analysis approach. Second, we explore the causal relationship between internet sentiment changes and stock market returns using both the mean and quantile Granger causality tests. Third, Chinese stock market is divided into three stages based on its trends and we study the Granger causal relationship between internet sentiment changes and stock returns in different stages.       An empirical study is conducted on the CSI 300 index. The empirical results show that there is a widespread and significant causal relationship between the two variables at extreme quantile intervals via the quantile Granger causality test, instead of the naive mean Granger causality test. Specifically, in 40 causal relationship tests, the quantile Granger causality test confirms that there are 23 significant causal relationships. The discovery ratio is 57.5%, which is obviously larger than that of 7.5% of the naive mean Granger causality test. In addition, the effects of internet sentiment changes on stock returns are heterogeneous over the different market stages.       The empirical findings have great significance both in theory and practice. A significant causal impact of internet sentiment on the stock market returns has proved to be for some specific quantile intervals. It provides evidences for the predictability of stock market returns under certain conditions. Therefore, internet sentiment can be used to predict the tail(for instance upper or lower tail) of stock return, which offers a decision-making scheme to avoid financial risk. These results eventually provide a new idea for stock market research involving asset pricing, return prediction, volatility estimation, and so on. They have also shed some lights on a new direction for the application of internet sentiment information.