The Influence Of Co-jump Among Index Constituent Stocks On Volatility Forecasting

Based on the classical jumping diffusion theory and under the framework of HAR model,the model which involves in the jump intensity in this paper is different from the traditional volatility forecast model which only considers the jump amplitude.We find that the model including the jumping intensity improves the volatility prediction ability in some extent,especially the cojump components.First,we use the nonparametric method to decompose the realized volatility into a continuous path sample variance representing the continuous change of the price and a jump variation representing the drastic change of the price.The BN-S univariate jump test is used to identify the jump occurence,and then we use the BLT method to identify whether there is a cojump among the index constituent stocks.On this basis,we use the Hawkes process to model the index jump intensity and the cojump intensity among the constituent stocks,respectively.Finally,The HAR-RV-CJ-CJI model is constructed by adding the independent variables to the classical HAR framework.The results show that the HAR-RV-CJ-CJI model has obvious improvement effect for the volatility forecasting compared to the HAR-RV model.In determining the frequency of high frequency data acquisition,we use the volatility signature plots(VSP)to determine the optimal sampling frequency,in order to reach a balance between information richness and noise minimum contained in the data.We found that different from foreign literature commonly used five minutes sampling frequency,the domestic market is more suitable for ten minutes sampling frequency.In order to improve the recognition accuracy of the jump,we use the WSD correction factor to correct the calendar effect of the realized volatility when the BLT test is used to identify the cojump among the constituent stocks.This paper finds that the calendar effect is closer to the "W-shape" rather then the "U-shape".