Can Investor Sentiment Improve The Accuracy Of Realized Volatility Prediction?

Under the background of economic globalization,especially the outbreak of global epidemic in recent years,the stock market has experienced unprecedented fluctuations.This fluctuation increases the uncertainty and risk of the stock market and affects the normal operation of the stock market.Volatility is a hot issue in economic and financial research.It is directly related to market uncertainty and affects the behavior of investment enterprises and individuals.In order to reduce this uncertainty,it is very important to accurately measure the volatility of stock index earnings.On the other hand,due to the low degree of marketization of China’s stock market,the trend of the stock market is easily affected by various news,so this paper considers the volatility of China’s stock market through the linear introduction of investor sentiment factors,which has theoretical guidance and practical significance for Chinese enterprises and individuals to conduct asset investment and risk management.At the beginning,the article combed the various methods of measuring investor sentiment indicators and the relationship between stock market volatility and investor sentiment studied by scholars from various countries over the years.It can be seen that although there are many methods of measuring investor sentiment indicators,there is actually no unified standard.Based on the reality of China,this paper will use the method of Text-CNN machine learning model to label the news data of all constituent stocks of the CSI 300 Index from January 1,2016 to December31,2021,and then construct the investor sentiment proxy variable of the CSI 300 Index according to the market value of each constituent stock(by daily weight)through the sentiment index construction method to study its relationship with the volatility of the CSI 300 Index.Secondly,this paper then describes the relevant theories of efficient market hypothesis and behavioral finance,in which the behavioral finance theory mainly introduces the prospect theory and behavioral portfolio theory.Then in terms of investor sentiment,this paper believes that investors’ irrational emotional behavior mainly affects the stock market from cognitive bias and preference bias,which provides a theoretical basis for the rationality of investor sentiment variables.Finally,this paper selects the 5-minute high-frequency data of the Shanghai and Shenzhen 300 Index from January 1,2016 to December 31,2021 as the empirical sample,and then introduces the constructed investor sentiment index into various heterogeneous autoregressive time series(HAR)models in a linear way.In the empirical study,sample period 3:2 is divided into estimated periods within the sample and estimated periods outside the sample.Then,the forward rolling prediction method is used to forecast daily,weekly and monthly out-of-sample.The empirical research finds that:(1)When forecasting daily,weekly or monthly volatility,all HAR-IS family models in the sample show that the coefficient of investor sentiment variables is very significant,and the goodness of fit of HAR-IS family models is higher than that of HAR family models,indicating that investor sentiment index is a good factor to predict the actual volatility of the CSI 300 index,and this conclusion is robust.In addition,for the daily,weekly and monthly volatility prediction results,the best fit is the weekly volatility,followed by the monthly volatility,and finally the daily volatility.(2)The DM and MCS tests outside the sample show that both the daily,weekly and monthly volatility prediction results show that the HAR-IS family model has better prediction effect,among which the HAR-CJ-IS model has the best effect in predicting daily volatility;Among the models for forecasting weekly and monthly volatility,the HAR-SJ-IS model has the best forecasting effect.