CSI 300 Index Realized Volatility Modeling And Its Application Based On High-frequency Data

Volatility modeling is the key element of risk management of financial market,as well as the key to asset allocation decision-made and asset pricing.As a part of financial asset price movement process,jump comprises the important part of financial asset volatility.In-depth analysis of jump behavior is of great significance to estimate the financial asset volatility accurately.The traditional volatility calculation has trouble in complex parameters estimation and low accuracy,failing in combining with high frequency data effectively.Based on the framework of realized volatility theory and high-frequency data,we make CSI 300 index as research object and construct its realized volatility.Furthermore,we separate its continuous and jump part and work on measuring three different jump-related varies which include jump variance,jump direction and jump intensity.As we did so,we manage to get most out of jump information in order to improve the accuracy of volatility forecasting.And finally,we apply the estimated volatility to the risk measurement of financial market.Main work methods and conclusions of this paper are listed as follows:Firstly,based on nonparametric method,we separate volatility’s continuous and jump part according to BN-S method and threshold method respectively and further compare the capability of these two methods.Results show that the realized volatility and jump variance both exhibit the character of leptokurtic,heavy tail and cluster,and the realized volatility also show significant slow decaying.The adjusted return(normalized by true,not estimated realized volatility)is nearly normal distribution.Threshold Bipower Variance calculated by threshold method is a better integral volatility estimator than the Bi-power Variance calculated by traditional BN-S methods.C-Tz jump test statistic of threshold method captures jump more accurately.Secondly,by measuring jump variance and separating it into positive part and negative part,we can make further analysis of jump direction information.And the application of exponential weighted moving method helps to exchange the original jump occurrence frequency data into new time series that can be further fitted by the Hawkes model and return jump intensity data that we needed.Before using HAR-class model for volatility modeling,we apply a so-called "average through window moving" method to relevant model varies(such as the Realized Volatility and the Threshold Bi-power Variance etc.).In the same time,we exploit jump information by considering its three-dimensional components to the volatility forecast model which has been proved as the optimal one.Empirical studies show that both positive and negative jump variance increase volatility while the impact is asymmetric,since the negative part’s influence is more significant.Jump intensity is a self-exciting process and has a negative correlation with volatility.However,the event of "jump happening" in t day will exacerbate the volatility of day t + 1.Applying "average through window moving" method to relevant model varies can significantly improve the prediction ability and this ability can be further improved by exploiting multi-dimensional jump information.Finally,the return is normalized by volatility that estimated by different HAR-class model and called the adjusted return.Let its distribution under different assumption,that is,normal distribution and skewed-t distribution,we calculate the corresponding VaR(value at risk)and compare the results.Conclusions show that risk measurement of the market is the most accurate one under the optimal volatility forecast model and the skewed-t distribution assumption.