The Measurement Of Financial Extreme Risk Based On SV And EVT Models

With tighter relationship of international financial market and serious volatility offinancial assets, the local, regional and international financial crisis seem to behappened more frequent and with short time intervals. National financial regulators,market participants and investors are particularly sensitive to the volatility of financialasset value. Thus, it becomes urgent to put forward and build more effective andpractical financial risk management methods and techniques. However, the pasttheoretical assumptions or applicable assumptions in financial risk managementmethods have larger deviation compared with the actual, such as the classic financialassets return sequence of normal distribution assumption has serious discrepancy to thehigh peak and thick tail characteristics of the facts. Therefore, how to describe anddepict the actual tail characteristics of financial assets, how to search for a moreappropriate sample distribution form, and how to build a more accurate financial riskmeasure model, these problems are vital to improve the financial risk measure method.There exist many risk management thinking and methods, such as riskdiversification, risk hedging, risk transfer, risk avoidance, risk compensation and so on,but in the face of current high effective information transmission and the coming bigdata era of financial economics, some qualitative financial risk management methodslike risk aversion which seems passive and even a bit subjective. It is not suitable for thebusiness status of financial institutions; even serve for the real economy timely andeffectively. Therefore, the core of risk management should be transformed into themeasurement of value at risk. But VaR（value at risk）, as a kind of simple understandingand quantitative risk management technique, has become the current internationalstandards in the field of financial risk management, especially in the Basel Accord, ithas already been applied as an important regulatory inspection index in bankingmanagement. At the same time, during the related literature of financial riskmanagement, the way to improve VaR measure method or the thinking of VaRtransformation also has been prompted. So, building accurate VaR calculation model isthe major aim of our study. The kernel of VaR measurement is to correctly judge thevolatility shape of sample data, based on this, the article combines the SV （stochasticvolatility） model with EVT （extreme value theory） model as the main line, viadifferent theory recombination, for example volatility conditions distribution, volatilitystructure transformation etc., trying to build more accurate financial risk measurement model. The main research contents and innovations embody as follows:First of all, analyze model theory and modeling idea. Based on the analysis of SVmodel, EVT model, and classical VaR measure method, the paper builds dynamic VaRmodel based on the SV-EVT model. This part mainly focuses on the analysis of basicfeatures of financial assets time series data, the volatility study of SV model, and thestandardized treatment of mean equation’s volatility residual, then it discusses theextreme value theory to fit and forecast the extreme value quantile of sample taildistribution. It systematically illustrates the SV-EVT combined model in VaR measuremethod for the first time.Secondly, conduct empirical test and practical application. In view of the SV-EVTmodel has been build, we introduce different sample distribution functions or volatilitystate transformation into the portfolio model to conduct empirical test through highfrequency financial data. Among them, by applying GHSKt （generalized hyperbolicskew students t） distribution to reflect the characteristics of volatility asymmetric thicktail, putting forward to combine this distribution with SV model to build the SV-GHSKtmodel for the first time. After that, we apply the model to simulate sample data andmake the residual sequence standardized. Finally we form the SV-GHSKt-EVTportfolio model with extreme value theory. Empirical analysis verifies that theSV-GHSKt-EVT model has more advantages than other heteroscedastic model orsimple SV model. It provides a more practical method for high peak and thick tailfeatures, stochastic volatility and leverage effect of financial assets risk measurement. Inaddition, for the problem of volatility state transition, in the past study of volatilitymodel, the volatility state transition is a single continuous state, so we consider puttingmarkov structure transformation model into the SV model which will be able to reflectthe characteristics of structure transition, and combined with thick tail t distribution, wepropose the MSSV-t model for the first time. We also apply the traditional extremetheory to analyze the residual sequence of MSSV-t model, finally conclude the VaRmodel based on MSSV-t-EVT model. The empirical analysis also shows thatMSSV-t-EVT model can effectively identify volatility transformation characteristics ofsample data, and also can reasonably measure risk transition, especially in the highconfidence level.Lastly, analyze multivariate extreme value and model expanding. The formerSV-EVT model only consider the risk condition of one dimension sample variable, so itseems not applicable for the multi-dimensional market or financial asset portfolio’s overall risk. Therefore, considering the risk correlation of dual market structure, weintroduce Copulas theory into SV-EVT model, and utilize this model to describemarginal distribution of the binary variables assets volatility. At the same time, in orderto highlight related timeliness between different financial market risks, we also deformCopulas function to time-varying Copulas function for the purpose of establishingCopulas-SV-EVT model. The empirical analysis shows that compared with static riskmeasure methods, the time-varying Copulas-SV-EVT model can more effectively reflectthe relationship between different financial assets. Although the time-varying Copulas-SV-EVT model does not solve the problem of overall portfolios risk measure, but itestablishes a good foundation for the follow-up study.The paper describes sample variables’ random characteristics and transformablecharacteristics through the combination of SV model and EVT model. It conforms to thebasic conditions and practical requirements of VaR measurement and expands extremevalue theory’s application in financial risk management field. It also providestheoretical methods of preventing extreme financial risk for financial marketparticipants, especially the regulators and investment institutions.