Research On The Influence Of Investor Sentiment On The Stock Market

The participants in financial market are not always “rational”,which is different from the hypothesis of effective market theory of traditional finance,which results in many financial anomalies that traditional financial models cannot explain.Subsequently,behavioral finance emerged,which explained these problems better than the traditional model.China's stock market has achieved remarkable progress in recent thirty years,but compared with the developed western market,the development of policy system,supervision of regulatory departments,degree of government intervention and information asymmetry are not mature enough.It is of theoretical and practical significance to analyze China's stock market with investor sentiment in the perspective of behavioral finance.In view of the literature at home and abroad,many scholars have considered the stock market analysis in the perspective of investor sentiment.But on the quantification of investor sentiment,most studies used a single index as sentiment proxies or constructed sentiment index by multiple indirect indicators,but using text mining methods to construct sentiment index is less and is not mature.This paper used principal component analysis and text mining method to construct sentiment index respectively,and compared the advantages and disadvantages.In addition,on the prediction of investor sentiment on the stock market,the majority of scholars used the traditional GARCH model,but for the perplexing stock market,GARCH model has some limitations,this paper used BP neural network prediction model to predict the closing price,and compared GARCH prediction model and BP neural network model.First of all,two kinds of investor sentiment index are constructed by principal component analysis and text mining.This paper selected 7 stock market sentiment proxies by learning from the classic construction method of BW index,performed principal component analysis afer excluded the macro factors,so as to construct the sentiment index;then took up all the articles from the web crawler from the Sina Finance website market review section from January 2010 to August 2017 as the original text corpus,built Web text sentiment index by the word segmentation,POS tagging,building the dictionary.Secondly,the GARCH model is built to analyze the impact of investor sentiment on the closing price of stock market,and predict theclosing price.Considering the limitation of GARCH prediction,a prediction model based on BP neural network is established to predict the closing price and calculated prediction error.To predict the closing price,a prediction model of closing price and two prediction models of two kinds of sentiment index are established,to analyze whether the sentiment index has a significant effect on the closing price.Finally,by comparing the prediction errors of the built models,the advantages and disadvantages of the two kinds of sentiment index and the advantages and disadvantages of the two forecasting models are compared.Through the above research,this paper got the following conclusions: investor sentiment and stock market closing price were Granger causes each other;the influence of emotion on the closing price was significant and they were positive correlation,positive mood will make the closing price rises,depression will make closing price reduction;the prediction errors of sentiment index were smaller than the prediction errors of the closing price;two kinds of sentiment index had predictive ability,but the sentiment index built by text mining had stronger predictive ability of closing price;BP neural network model was better than the GARCH model,and it can be extended to predict the price of other financial markets.