Research On The Impact Of Investor Sentiment On Stock Price Forecasting Based On Neurual Network Model

Since the birth of the A-share market,A-shares have experienced several sharp rises and falls.These irrational sharp rises and falls are difficult to be fully explained with traditional financial theories.Based on investor sentiment,behavioral finance theory exactly explains various financial anomalies and irrational behaviors.Investor sentiment has undoubtedly contributed to the irrational prosperity and panic decline in the stock market.Taking these phenomenon as an entry point,we study the effect of market investor sentiment on stock market returns and the heterogeneous listed company's stock returns using data mining models.First,we select the number of new account openings(NIA),turnover rate(TURN),consumer confidence index(CCI),trading volume(VOL),China Securities Investor Confidence Index-Buy Index(BII)as proxy indicators of investor sentiment.Use the first principal component analysis eliminates the effects of advance and lag effects,and the second principal component analysis to eliminate the correlation of macroeconomic variables,and then remove the effect of macroeconomic changes to investors through regression.And then the partial least squares regression method is used to extract the part related to the real investor sentiment to obtain the final investor sentiment comprehensive index.Next,using the random forest algorithm,a high-importance technical index is selected from a series of representative candidate technical indexes as the input variables of the data mining model,namely BBI,BBIBOLL,BIAS,BOLL and RSI.Then,the comprehensive index of investor sentiment and the screened technical index are used as input variables of the model to train and establish a BP neural network model.Finally,through the analysis of pure technical analysis indicators,and the difference in the accuracy of predicting stock prices by the neural network model established by adding the comprehensive indicator of investor sentiment after processing,the influence of investor sentiment index in stock forecasting modeling is studied,and according to The company's characteristic variables,such as scale indicators and growth indicators,cross-classify companies and study what types of stocks are more susceptible to investor sentiment.After the above research steps,this article mainly draws the following conclusions:The average absolute error between the predicted value of the neural network model modeled with pure technical indicators and the actual value of the stock price is MAD = 19.6854.The average absolute error between the predicted value and the actual value of the stock price of the BP neural network established by using the investor sentiment index and the filtered technical index as input variables is MAD =19.6854.The conclusion shows that compared with the BP neural network modeled solely by technical indicators,the data mining model that incorporates investor sentiment variables performs better in predicting the accuracy of stock prices,and has a greater Advantage.In the study of cross-classification of companies by characteristic variables such as scale indicators and growth indicators,the type of stocks with the smallest degree of deviation between the predicted value of the neural network model and the actual value of the stock price is a small market value stock portfolio(MAD = 7.173),which fit the models beat,and the stock type with the largest deviation between the predicted value of the neural network model and the actual value of the stock price is the large market value growth stock portfolio(MAD = 23.703),which performs worst,and the small market value growth stocks The fitting effect between the portfolio and the large-cap value stock portfolio is somewhere between the above two characteristic portfolio stock types.