Research On Realized Volatility Forecast Modeling Of CSI 300 Index Introducing Investor Sentiment

The volatility of financial assets penetrates the capital market,and good forecasts of volatility are conducive to asset pricing,hedging,and risk management.With the development of the era of big data on the Internet,the increasing availability of intraday high-frequency transaction data has provided new means for the study of financial volatility.On the other hand,with the rise of behavioral finance,research on investor sentiment and stock market returns and its fluctuations has become a hot topic.In addition,institutional and individual investors in China's securities market are mostly short-term speculators.Complex emotional factors affect Larger,so that the price of financial assets may deviate from the real price.Therefore,this article considers research and introduction of a volatility forecasting model of emotional factors,which has theoretical guidance and practical significance for Chinese investors to make reasonable use of business data,reasonably obtain more accurate volatility forecasts,and reasonably conduct investment portfolio and risk management.This paper first constructs investor sentiment indicators that are weighted by heat based on text sentiment analysis,and combines the daily news enthusiasm and news sentiment of the constituent stocks of the Shanghai and Shenzhen 300 Index to construct IS index data for Chinese investor sentiment;then based on the residual obey Under the four distribution assumptions of Gaussian,student t,GED and Skew-t distribution,the IS index volatility value is constructed using the AR-(E)GARCH model;finally,the IS index and volatility variables representing investor sentiment are simultaneously linearly introduced into the HAR family model to construct New models of the HAR-IS-GARCH family and the HAR-IS-EGARCH family are used to explore the improvement of the prediction accuracy of the original HAR family benchmark models that predict daily,weekly,and monthly volatility models.The empirical data uses the five-minute high-frequency trading data of the Shanghai and Shenzhen 300 Index of the Chinese stock market from January 1,2015 to December 31,2019 to construct a 5-minute realized RV volatility model.The sample period is 3: 2 In-sample estimation period and out-of-sample prediction period,and rolling forward prediction method is used for out-of-sample prediction.In the model parameter estimation of the sample,it was found that:(1)Whether it is predicting daily,weekly or monthly volatility,investor sentiment and its fluctuation variables are very significant,indicating that investor sentiment and its fluctuations can effectively Acting on stock market volatility,this conclusion is robust in three forecast length models.Among them,the best fitting degree is the weekly volatility,and the fitting ability to predict the monthly volatility has improved the most.(2)Whether predicting daily,weekly,or monthly volatility,the new model of the HAR-IS-EGARCH family of models that introduces investor sentiment and its volatility under a Skew-t distribution always has the highest in-sample goodness,and It is found that using EGARCH model to describe the fluctuation can improve the fitting degree of the model compared to GARCH model.The results of the DM test and MCS test of the predictive ability of the out-of-sample model show that:(1)No matter which forecast length model introduces investor sentiment and its volatility,changing the residual to obey the Skew-t distribution and selecting the EGARCH model to characterize the volatility can effectively improve the out-of-sample prediction performance.(2)The new HAR-IS-EGARCH family of model based on the Skew-t distribution,which introduces investor sentiment and its volatility,has the best out-of-sample prediction performance in predicting daily,weekly and monthly volatility.Among them,when predicting the monthly volatility,the performance of out-of-sample prediction is improved more significantly.In addition,it is further proved that the volatility model of the investor sentiment indicator with heat weighting has better in-sample fitting ability and out-of-sample forecasting performance than the volatility model of the investor sentiment indicator without heat weighting.