【作者】 刘海；

【导师】 陶志富；

【作者基本信息】 安徽大学 ， 应用统计（专业学位）， 2022， 硕士

【摘要】 自证券市场诞生以来,股票市场的高收益性吸引着大量投资者进行投资,然而其背后的高风险性使得投资者急切需要一种能够对股票市场进行有效预测的方法。因此对于股票市场的预测历来是一个热点问题。传统的股票市场预测研究使用将股票交易是产生的历史数据为变量进行建模分析,而行为金融学理论的发展使得研究者们通过分析投资者情绪也可以预测股票市场,本文将两者进行结合,展开研究。首先,本文将通过网络爬虫技术获得的投资者评论文本数据进行数据清洗,采用word2vec模型进行文本数据向量化表示,并对现今主要的几种情感分析方法进行比较分析,建立基于Stacking算法的情感分析模型,将投资者评论数据进行分析构建投资者情绪特征。其次选取股票交易数据指标,例如股价、交易量等,在对投资者情绪特征与股价之间的因果分析进行Granger检验分析,验证投资者情绪的确能够影响股价的走势。最后本文建立融合投资者情绪特征的SA-SVR模型、SA-BP模型、SA-LSTM模型,以及建立未融合投资者情绪特征的SVR模型、BP模型与LSTM模型,分别从短期、中期、长期三种时间跨度去检验模型预测结果的精确性以及稳定性。本文的实证对象是上证综合指数,实验结果表明,在对投资者评论文本数据进行情感分类中,基于Stacking算法的情感分析模型的性能较好,其召回率为0.81,精确率为0.79,表明基于Stacking算法的情感分析模型能够准确的判断文本数据的情感倾向,并提取投资者情绪特征。同时在对股指进行预测的实验中,融合投资者情绪特征的三类模型性能优于未融合投资者情绪特征的三类模型,验证投资者情绪特征有利于对股票市场的预测研究。同时SA-LSTM模型的预测性能在6中模型中最强,短期预测的MAPE值为0.0092,中期预测的MAPE值为0.0129,长期预测的MAPE值为0.0087。研究发现在对上证综合指数收盘价的预测中,SA-LSTM提取了投资者情绪特征和指数交易指标中的信息量对股指收盘价进行有效预测。更多还原

【Abstract】 Since the birth of the stock market,the high yield of the stock market has attracted a large number of investors to invest.However,the high risk behind it makes investors urgently need a method that can effectively predict the stock market.Therefore,the prediction of the stock market has always been a hot issue.The traditional stock market prediction research uses the historical data generated by stock trading as variables for modeling and analysis,while the development of behavioral finance theory makes researchers can also predict the stock market by analyzing investor sentiment.This paper combines the two and carries out research.Firstly,this paper cleans the text data of investor comments obtained by web crawler technology,uses word2 vec model to quantify the text data,compares and analyzes several main emotion analysis methods,establishes a emotion analysis model based on stacking algorithm,analyzes the investor comment data,and constructs the emotional characteristics of investors.Secondly,select stock trading data indicators,such as stock price and trading volume,and conduct Granger test analysis on the causal analysis between investor sentiment characteristics and stock price to verify that investor sentiment can indeed affect the trend of stock price.Finally,this paper establishes SA-SVR model,SA-BP model and SA-LSTM model that integrate the characteristics of investor sentiment,as well as SVR model,BP model and LSTM model that do not integrate the characteristics of investor sentiment.It tests the accuracy and stability of the prediction results of the model from three time spans: short-term,medium-term and long-term.The empirical object of this paper is the Shanghai Composite Index.The experimental results show that in the emotional classification of investor comment text data,the emotional analysis model based on stacking algorithm has better performance,with a recall rate of 0.81 and an accuracy rate of 0.79.It shows that the emotional analysis model based on stacking algorithm can accurately judge the emotional tendency of text data and extract the emotional characteristics of investors.At the same time,in the experiment of predicting the stock index,the performance of the three types of models integrating the characteristics of investor sentiment is better than that of the three types of models not integrating the characteristics of investor sentiment.Verifying the characteristics of investor sentiment is conducive to the prediction of the stock market.At the same time,the prediction performance of SA-LSTM model is the strongest among the six models.The MAPE value of short-term prediction is0.0092,the MAPE value of medium-term prediction is 0.0129,and the MAPE value of longterm prediction is 0.0087.It is found that in the prediction of the closing price of the Shanghai Composite Index,SA-LSTM extracts the characteristics of investor sentiment and the amount of information in the index trading indicators to effectively predict the closing price of the stock index.更多还原