High-frequency Data Volatility Modeling And Risk Measurement

In 2015,China's A-share experienced a process of sharp fluctuations.The thrilling market makes investors unforgettable.This year is bound to go down in China's financial history.The cruelty market once again reminded investors of the importance of risk management,and volatility measurement occupies a pivotal position in risk management,with the increasing acquirability of financial high-frequency data,the"realized measure",which based on the non-parametric method of high-frequency data becomes more and more important in volatility research.The "realized measure" change the financial volatility from a hidden variable to direct a variable,and provides a reliable measure for market volatility.But it lacks the ability of prediction.In view of the advantages of the traditional GARCH models,it is a hot topic to combine it with realized measure in volatility modeling.The Realized GARCH model proposed by Hansen et al(2012)associates the conditional volatility of GARCH models with the realized measure,and thus obtained a complete GARCH analysis framework.However,it also has some shortcomings.For the universal existence of sharp peak,thick tail and asymmetrical distribution in financial data,this paper extends Realized GARCH to thick tail distribution,and make the power of leverage function realxed to estimated parameters,in order to further deal with the leverage effect,sets the form of leverage function as asymmetric relaxed powers.At the same time,in the consideration of the effects of different realized measure to model,this paper not only uses RV measure,but also introduces other three realized measure,which constitutes the contrast models of tail risk measurement.In addition to comparing the results of VaR measurement,and also introduce two loss functions in the view of financial risk management to make up the deficiency of VaR.We also use SPA test which based on bootstrap method to further improve the robustness of model comparison results.And in this paper the Monte Carlo simulation method is used to compare the prediction results.The empirical results based on the 5-minute high-frequency data of Shanghai Stock Exchange show that the application of Realized GARCH to risk measurement has a good result,and the processing of the power in leverage function significantly improves the accuracy of the tail risk measurement.The asymmetric relaxed power to positive and negative information improves the measurement precision of tail risk,and it shows a similar result in the one-step prediction of the realized measure.The selection of different measures has a great influence on the model's risk measurement precision,at the different risk levels,the performance of each realized measure is different,but the treatment of the leverage function has improvement on the tail risk measurement precision in all models,and the improvement is more significant when there is an asymmetric relaxed power.