The Intraday VaR Measure Of Chinese Index Future Based On Ultra-high Frequency Data

With the development of China's securities market,financial derivatives have been more abundant,which cause a bigger volatility in the market.Therefore,it is particularly important to measure and control the risk.Traditional risk measurement is usually based on low frequency daily trading data,easy to overlook the information in a day,which may lead to considerable errors in risk measurement.Ultra-high frequency time series contain all the accurate information of each tick-by-tick data,which contain more plenty of market volatility information,just like accurate fluctuation amplitude in a day.Therefore,to seek appropriate effective risk measurement model with ultra-high frequency data,will help investors hnow the risk,which also help regulatory authorities to control the risk better.It's a custom to use regularly spaced daily data for traditional GARCH class models.However,high frequency data are generally irregularly spaced transaction data which is recorded when the event occurred.Therefore,this article will use the applicable models for irregularly spaced transaction data.They are Logarithmic Autoregressive Conditional Duration Model(Log-ACD)and Asymmetry Autoregressive Conditional Muration Model(AACD)which considers the direction of price movements.After that,to estimate intraday VaR and test the ability among the two ACD models and AR-GARCH model.This paper selects the main contract of CSI 300 stock index future as a representative of China's securities market,to value and research the intraday risk in China.The data contains each tick data from 2010/4/19 to 2015/12/31.First determining the threshold of price volatility,which makes every average duration of price volatility remains stable,then taking account the time-of-day effect,then calculate the VaR of the main contract of CSI 300 stock index future in 30 minutes with three models and conduct a comprehensive analysis with different examinations to test their effect,correlation,independence and precision.They are Kupiec test,Dynamic Quantile Test,GMM Duration-Based Test,MSE and MAE test.In order to test the effect of the comparison above,we evaluate stability the Log-ACD model and AACD model.In the end,we compare the practical application effect of the models in the bear market in 2015.The empirical results show that:(I)in the event of an average price duration of approximately 5 minutes,the confidence level of 95%,and 5% significance level for the inspection results,AR-GARCH model obtains the highest percentage of Kupiec test,showing that AR-GARCH model has a highest percentage which the estimating VaR is equal to the expected failure rate,the percentage is 88.64%,and the AACD model is 6.82% lower than ARGARCH model,the Log-ACD behaves the worst.However,AACD model gets the best performance in the independence test,correlation test and precision estimate inspection.The AR-GARCH follows the second,and Log-ACD model is the worst.Overall,compared with AR-GARCH model and Log-ACD model,AACD model is the best to estimate VaR.As a new method to control model,it has a strong application significance.(II)After changing the average price duration and confidence level,the behavior of three stay stable.On the one hand,it shows the feasibility under the initial conditions of an average price duration of 5 minutes and the 95% confidence level,on the other hand also proves the AACD model has the advantages of stability.(III)Taking the recent bear market as example(From June 15,2015 to August 26,2015),all models could not pass Kupeic test but AACD model still performs the best in all the other tests,which shows that the intraday volatility risk is pretty big,the ability of risk control remains to be improved.However,compared with the other two models,AACD model has a larger improvement in independence,correlation,and precision.AACD model can be well used in the intraday risk measurement and management of the securities market in China.