Research On Predictive Power Of HAR-RV Type Model Based On Investor Sentiment Index Expansion

With the rapid development of computer technology and the enhancement of the availability of high-frequency data,the realized volatility constructed by the intraday return information of financial assets has become a hot research topic in the academic circle.It's main research direction includes the characteristics of realized volatility of financial assets mining and prediction research.Among them,introducing feasible exogenous variables on the basis of the classical model is the main research idea of forecasting volatility in the academic community.Based on this,taking the Shanghai composite index as the research object,this paper constructs two subjective emotion proxy indexes by using the textual emotion analysis method,and combines the objective emotion proxy index using the traditional principal component analysis method to construct the composite investor sentiment index.Second,we add the composite investor sentiment index into basic model linearly,including HAR-RV,HAR-CJ and HAR-RSV model.in order to further research on asymmetric effect of investor sentiment,the sentiment index is decomposed into optimistic and pessimistic sentiment indicators to expand the basic model,and then a total of nine models are constructed.After that,in-sample parameter estimation for all models,as well as prediction out of the sample are performed;Finally,the SPA method is used to test the predictive power of all the models,and also we test the robustness of the research results.We have obtained some results as follows:(1)The two subjective proxy indexes of investor sentiment constructed by the method of textual sentiment analysis have good correlation with the closing price of Shanghai stock index respectively,and the correlation between the composite index of investor sentiment and the closing price of Shanghai stock index is about 0.92.(2)The in-sample parameter estimation results show that investor sentiment contains useful information for predicting future realized volatility.All the expanded models have better explanatory ability to the realized volatility than the original models,and on the fitting ability,model expanded by the decomposed emotional indicators is higher than that expanded by the undecomposed ones.Among them,the fitting ability of HAR-RSV-CISIPN model is the strongest.(3)Investor sentiment index has asymmetric influence on stock market volatility.Both optimism and pessimism can positively repair volatility.This is consistent with the investor behavior of chasing the rise and killing the fall in the stock market;In addition,the impact intensity of optimistic investor sentiment is greater.This is caused by the greedy nature of people and short-selling restrictions in the stock market.(4)the out-of-sample prediction results show that the prediction ability of the extended model is enhanced compared with the corresponding basic model,and the prediction accuracy of the model with the decomposed investor sentiment index is improved compared with that with the undecomposed investor sentiment index.Among them,the predictive power of HAR-RSV-CISIPN model is the best.(5)The results of robustness test show that the prediction of the future realized volatility with a longer period is still consistent with the prediction of the future oneday realized volatility.It is also found that the influence of investor sentiment on volatility does not weaken with time forward,and the influence intensity reaches the maximum when the forecast cycle is month.This is probably a sign that investor sentiment has a long memory.To sum up,it can be seen that investor sentiment plays an important role in predicting asset volatility,so the future risk control and asset allocation in the stock market should pay more attention to the investor sentiment,and how to construct more accurate investor sentiment indicator is also an important part at present.