Research On Stock Index Trend Prediction Based On Investor Sentiment

As China’s market economy develops,the stock market assumes an increasingly significant role in the national economy.As an excellent investment method and a barometer of macroeconomic changes,the trend prediction of stock indexes has garnered the interest of an increasing number of researchers.However,the stock market is irrational,and the stock index data has non-stationary,nonlinear,and highly noisy characteristics.Consequently,it is challenging to accurately predict the trend of stock index changes using only conventional time series models.In addition,the unique structure of Chinese shareholders causes the psychological state and emotions of shareholders to influence the trend of the stock index.With the advancement of machine learning,the neural network can capture complex sequence properties and nonlinear structures,resulting in an improvement in the accuracy of stock market trend forecasting.This paper establishes a stock index change trend prediction model based on investor sentiment that combines quantitative investor sentiment with stock index historical data information and LSTM neural network for predictive analysis.This paper uses the dictionary-based text sentiment analysis method to quantify the crawled stock bar comments into sentiment scores and combines it with the search index and information index in the Baidu index to create an investor sentiment index in order to quantify the investor psychology.Due to the severe multicollinearity of the stock index historical data information,the functional principal component analysis method is chosen for dimensionality reduction in this paper.The B-spline basis function method was used to fit the data,after which the features of the 14 curves were extracted,yielding three principal components with a cumulative contribution rate of 95.8%: the investor sentiment factor,the factors affecting the stock’s own value,and the Transaction Data Influence Factors.The paper concludes by incorporating the principal components of the stock index historical data and the principal components combined with investor sentiment into the LSTM model for predictive analysis.Using the Shanghai Stock Exchange 50 Index as an example,the aforementioned models are constructed.In order to evaluate the generalizability of the model,different-sized stock indices(the CSI 300 Index and the CSI 500 Index)are chosen for input prediction.The evaluation index system is evaluated in light of the constructed prediction results.By analyzing the changes in prediction accuracy before and after the addition of investor sentiment and by comparing the prediction accuracy of the model applied to stock indices of various sizes,the following conclusions are drawn: 1.The model proposed in this paper is effective for stock index forecasting;2.The introduction of investor sentiment will improve the prediction accuracy of the model,and the prediction accuracy of different stock indexes will be improved to varying degrees,with the impact on the CSI 300 index being the most pronounced,followed by the CSI 500 index,and the impact on the Shanghai Stock Exchange 50 Index being the weakest;3.The model is more suitable for indices with a large proportion of the financial sector,and the universality is not good.In accordance with the research methodology,this paper offers the following recommendations from the perspectives of three distinct financial market entities: 1.For investors,when making stock investment decisions,attention should be paid to the accuracy of the source of information;2.For publicly traded companies,periodic investor exchange meetings should be held to accurately disclose relevant information so that investors can make rational investments;3.For government regulatory authorities,on the one hand,it strengthens the guidance of small and medium investors and improves their ability to identify various comment information based on online public opinion.On the other hand,it monitors the stock market through data analysis and prohibits individual and institutional investors from publishing comment information in bulk for the purpose of manipulating stock prices.