High Frequency Data VaR Research Based On Realized Range

With the continuous development of Chinese financial market, the identificationand measurement of financial market risk not only gradually become the focus center ofthe financial institutions and regulatory authorities, but also the research center of thescholars of the financial market research. The traditional risk measurement is mostlybased on the GARCH models and SV models built by choosing the low frequency datain the day, although these models achieved many achievements in depicting marketvolatility characteristics and applying in the risk management, the trading in financialmarket is frequent and only using the low frequency data will lose a lot of important dayinformation, which leads to the errors in volatility estimating and volatility forecasting.The high frequency data can well preserve the day market information and it is helpfulto better study the financial market risk. Besides, the traditional GARCH models cannot be directly used in the high frequency data modeling analysis, then searchingeffective high frequency data risk measurement model is not only good for providingtheoretical reference and policy suggestions to financial institutions and regulatoryauthorities, but also providing a new research direction for the scholars of studying thefinancial market risk.As a kind of effective method of measuring risks, VaR have got favour by the riskmanagers and many researchers for its simple calculation, the intuitive method and theadvantages in accurately measuring the risk. With the development of the VaR theory,its application in the risk management is becoming more and more wide, using differentmethods to calculate more accurate VaR has been the center problem in quantifying therisk analysis. Since the high frequency data research have been proposed in the1990s, ithas attracted the widely attentions of the scholars at home and abroad. Scholars, such asAndersen, put forward the method of using high frequency data to estimate volatility,which called realized volatility, has led many scholars to research the realized volatilityand related areas of high frequency volatility. In view of this, this paper introduced therealized range, one of the research method of the high frequency, into calculatingVaR,and compared the result with the GARCH model based on low frequency data todetermine the advantages and disadvantages in forecasting VaR, etc, at the same time,itprovides the reference for applying the realized range in volatility estimation and risk management.This paper combine the high-frequency data with VaR theory, removing themicrostructure error and calendar effect by weighting the realized range and choosingproper sampling frequency, then applying the weight realized range to the volatilityforecast and VaR. By backing test, we compared the VaR calculated by weight realizedragne and by low frequency data under different distribution, which is the main researchdirection and innovation, it also contribute to the research and application of realizedrange theory.In the empirical analysis part, this paper first define the data and give the result ofselecting the data frequency. Secondly, the paper shows the descriptive statistics of therealized sequence and daytime sequence, followed by unit root test, correlation test, etc.through the autocorrelation function diagram, this paper illustrates the existence ofcalendar effect and the role of the weight method to the realized range in eliminating thecalendar effect. Then, by R/S method, this paper test the long memory characteristic ofthe realized volatility, realized range and weight realized range, the result shows all therealized sequence exist long memory characteristic. Based on the calculated Hurstexponents, the long memory model(ARFIMA model) has been built and made acomparing analysis with GARCH model in volatility forecast. Finally, this paper bringthe ARFIMA based on weight realized range and the GARCH model in the VaRcalculation. Taking the time sequence have the leptokurtosis and fat-tail characteristicinto account, the paper used the normal distribution, T distribution and generalized errordistribution to find the optimal distribution form, then tested the calculation results ofVaR through the LR likelihood ratio.