Research On Stock Market Return And Volatility Using Internet Information

According to the behavioral finance theory,besides fundamental macro-economic factors,stock market price is also influenced by individual investor or noise trader to a large extent.In order to reveal the influence mechanism,network data is employed to investigate the relationship between online data and the stock market and forecast stock market return and volatility from the perspective of individual investor sentiment and individual investor attention respectively.In this dissertation,we conduct empirical analysis in the following two aspects.(1)Base on the individual investor's sentimental information,we investigate the relationship between investor sentiment and stock market return,and assess the value of Internet sentiment information.We first apply methods of Chinese text sentiment analysis to extract network emotion time series from Sina Weibo texts,and then separately use the mean Granger causality and quantile Granger causality test to explore whether there is a causal relationship between online emotions and stock market returns.An empirical study is conducted on the CSI 300 Index.The empirical result show that there exists significant Granger causal relationships under the framework of quantile not in the mean.These Granger causal relationships are found at upper or lower quantiles through the quantile Granger causality test.In addition,their magnitude and manner are heterogeneously affected by Internet sentiment at different market stages.The results imply that Internet sentiment has a significant impact on stock market returns and can predict the tail(top or bottom)behavior of stock market returns.(2)From the perspective of individual investors' attention,we examine the ability of Internet search data in predicting stock market and explore the contributions and complements of Internet search and macro financial data.Based on the basic GARCH-MIDAS model,we establish univariate and multivariable GARCH-MIDAS models,and apply them to predict stock market volatility using the Internet search data,macroeconomic data,and mixed data.Through quantitively measuring the influence of Internet search on volatility prediction,we further discuss the natures of Internet search data.The empirical results show that Internet search data can be used to predict stock market volatility,and is complementary to macroeconomic variables.Including the combination of Internet search data and macroeconomic variables in the GARCH-MIDAS model is helpful to reduce prediction error.In addition,we find that the ability of Internet search to interpret the stock market is heterogenous in different market environments,which is consistent with the conclusion from before Internet sentiment research.We can argue that the individual investors' attention and individual investment have similar characteristics.This dissertation demonstrates the feasibility and effectiveness of using network data for stock market analysis,which provides a new perspective for the theoretical research of stock market.The high correlation between network information and the stock market also confirms the efficacy of behavioral finance.Our conclusions provide new research ideas for the study of stock market pricing,returns forecasting and volatility estimating,and provide empirical evidence to extract information from massive network data effectively and use network information correctly.