统计与数据科学学院知识库

  自我国股票市场成立以来，其发展速度一直处于快速增长的状态，迅速成为全球最大的新兴市场之一。作为中国资本市场重要的组成部分，创业板市场的迅猛发展在为投资者和企业带来机遇的同时，也伴随着一定的风险。创业板的价格波动性较大，这不仅对投资者的决策带来极大的不确定性，还可能对企业的融资计划、政府的政策制定以及其他市场主体的行为产生深远影响。同时，随着互联网的普及与发展，投资者可以在互联网上实时发送和收集丰富的信息，而这些信息也在一定程度上影响着投资者的判断，进而影响其决策。并且大多数投资者缺乏足够的专业知识，容易受到市场情绪的波动影响，进而做出跟风或非理性的决策。这类情绪化的投资行为，常常导致股票市场收益率的剧烈波动。

  本文基于目前的文献研究，分别以创业板指数和创业板股票为研究对象进行具体分析，主要研究工作如下：（1）采用文本数据对创业板指数进行分析，为了分析投资者的情绪对指数收益率的影响，提升模型的预测性能，本文收集创业板指数相关的投资者评论文本数据，构造情感指数，再利用ARMA-GARCH模型分析情感指数对收益率的解释作用。同时利用KPCA将与创业板股票相关的宏观经济数据、历史价格数据进行特征提取。最后将构造的投资者情感指与主成分作为输入变量对创业板股票收益率进行预测。（2）上述方法中，在影响因子中加入投资者情感指数的方法获得了较好的预测性能。因此继续采用该方法，构造针对创业板股票的投资者情感指数对股票收益率进行预测，然后通过对比验证构造的投资者情感指数的有效性。（3）由于创业板股票来自创业板板块，除了受到股票自身影响，还可能会受创业板指数影响，因此，考虑将针对创业板指数的情感指数纳入创业板板块内具体股票的收益率预测。利用网格搜索算法寻求构造的“创业板指数投资者情感指数”与“个股投资者股票资者情感指数”的最佳权重并进行加权处理构造投资者综合情感指数。接着，利用VMD对每只股票的投资者综合情感指数进行分解，再利用样本熵重构为高中低频数据进行分析。最后，将投资者综合情感指数作为一个输入变量，采用第二步中预测效果较好的SVR和RF模型对每只股票的收益率进行分析，并采用评价指标来评估提该方法的预测性能。

  实证结果表明：（1）本研究提出的结合创业板投资者情感指数的创业板指数收益率预测模型与其他预测模型相比，在水平和方向预测精度上均获得了更好的预测性能。（2）针对创业板股票收益率预测，分别从数据和方法层面分析：在数据层面，结合创业板股票投资者情感指数的数据集在水平和方向预测精度上均显著优于经济数据集。在方法层面，基于ISI-KPCA-SVR的预测方法和ISI-KPCA-RF获得了更好的预测性能。（3）研究发现，将针对指数的投资者情感指数和针对股票的投资者情感指数进行加权组合后的投资者综合情感指数与经过KPCA特征提取的宏观指标数据集同时作为预测因子预测创业板股票收益率时，可以有效提高预测精度。

Since the establishment of China's stock market, its development speed has been in a state of rapid growth, and it has rapidly become one of the largest emerging markets in the world. As an important part of China's capital market, the rapid development of gem brings opportunities for investors and enterprises, but also accompanied by certain risks. The price volatility of gem is large, which not only brings great uncertainty to investors'decision-making, but also may have a far-reaching impact on the financing plans of enterprises, government policy making and the behavior of other market players. At the same time, with the popularization and development of the Internet, investors can send and collect rich information in real time on the Internet, and these information also affect investors' judgment to a certain extent, and then affect their decision-making. And most investors lack sufficient expertise and are vulnerable to fluctuations in market sentiment, thus making follow-up or irrational decisions. Such emotional investment behavior often leads to sharp fluctuations in stock market yields.

Based on the current literature research, this paper takes the gem index and GEM stocks as the research objects for specific analysis. The main research work is as follows: (1) using text data to analyze the gem index, in order to analyze the impact of investor sentiment on the index yield and improve the prediction performance of the model, this paper collects the text data of investor comments related to the gem index, constructs the sentiment index, and then uses ARMA GARCH for impact analysis. At the same time, KPCA is used to extract the characteristics of macroeconomic data and historical price data related to GEM stocks. Then, the constructed investor sentiment index and principal component are used as input variables to predict the return of GEM stocks. (2) Among the above methods, the method of adding investor sentiment index to the influencing factors has achieved better prediction performance. Therefore, we continue to use this method to construct an investor sentiment index for the gem index to predict stock returns, and then verify the effectiveness of the constructed investor sentiment index by comparison. (3) Because the research object is from the gem sector and may be affected by the gem index, it is considered to include the emotional indicators for the gem index in the yield forecast of specific stocks in the gem sector. The grid search algorithm is used to find the best weight of the constructed "gem index investor sentiment index" and "stock investor sentiment index" and weighted. Then, the comprehensive emotional index of each stock is decomposed by VMD, and then the sample entropy is reconstructed into high, medium and low frequency data for analysis. Finally, taking the comprehensive emotional index as an input variable, the SVR and RF models with better prediction effect in the second step are used to analyze the yield of each stock, and the evaluation indicators are used to evaluate the prediction performance of this method.

The empirical results show that (1) compared with other prediction models, the gem index yield prediction model combined with the gem investor sentiment index proposed in this study has better prediction performance in terms of horizontal and directional prediction accuracy. (2) For the forecast of gem stock returns, it is analyzed from the data and method levels respectively: at the data level, the data set combined with gem stock investor sentiment index is significantly better than the economic data set in terms of horizontal and directional forecast accuracy. At the method level, the prediction method based on isi-kpca-svr and isi-kpca-rf achieve better prediction performance. (3) It is found that when the comprehensive sentiment index weighted by the investor sentiment index for the index and the investor sentiment index for the stock and the macro index data set extracted by KPCA features are used as predictors to predict the return of GEM stocks, the prediction accuracy can be effectively improved.