Research On High Frequency Volatility Model Based On Overnight Information Shock

In recent years,with the development of big data and cloud storage technology,5 minute,1 minute or higher frequency transaction data of financial assets can be recorded,stored and analyzed.Volatility measurement,fitting and prediction based on high-frequency trading data has become one of the research hotspots in the field of asset pricing and risk management.Specially,realized volatility(RV)and realized range-based volatility(RRV)are the two types of high-frequency volatilities that have attracted the most attention.However,the current mainstream modeling methods for RV and RRV only focus on the autocorrelation and endogenous effects of volatility,but ignore the external information shocks and its asymmetric effects.So there may be errors in the prediction of high frequency volatility.In addition,are HAR-RV models better than HAR-RRV models?It is also worth exploring further.Based on this,This paper first reviews the mainstream modeling methods of RV and RRV.Then this paper uses overnight information as the source of the external information shocks,and uses the GARCH family models to estimate the overnight volatility,finally constructs HAR-RV-CJ-GARCH/TGARCH-based models and HAR-RRV-CJ-GARCH/TGARCH based on the external information shocks.These two types of models not only consider the autocorrelation and jump characteristics of high-frequency volatility,but also take into account the agglomeration and asymmetry of the external information shocks.Subsequently,this paper selects the high-frequency trading data of the CSI 300 Index as a research sample,constructs the two-scale Extreme Range Volatility(TSRV)as the true volatility,and uses the MCS test to evaluate the prediction capabilities of these two types of models and the existing mainstream models.At the same time,the paper also compares and analyzes the fitting and prediction capabilities of HAR-RV and HAR-RRV models.The main research conclusions are as follows:(1)Compared with the current mainstream models,HAR-RV-CJ-GARCH/TGARCH models and HAR-RRV-CJ-GARCH/TGARCH models that contains overnight information shocks and consider the agglomeration of volatility always have better fitting ability and prediction effect.(2)In all high-frequency volatility models that consider the impact of overnight information,HAR-S-RV-CJ-GARCH/TGARCH models and HAR-S-RRV-CJ-GARCH/TGARCH models which distinct the different impact between positive and negative semivariance have better prediction results.Further,in the non-stationary period,HAR-S-RV-CJ-TGARCH model and HAR-S-RRV-CJ-TGARCH model have the best prediction effect;but in the stationary period,HAR-S-RV-CJ-TGARCH model and HAR-S-RRV-CJ-TGARCH model are only optimal in short-term prediction,while the HAR-S-RV-CJ-GARCH model and HAR-S-RRV-CJ-GARCH model have better long-term prediction capabilities.(3)The comparison of the prediction capabilities of HAR-RRV models and the HAR-RV models shows that the forecasting effect of HAR-RRV models are better than HAR-RV models during the non-stationary period;And during the stationary period,HAR-RRV models are only better in short-term prediction,while HAR-RV models have better long-term prediction performance.