关于极值理论（EVT）的风险价值估计.pdfVIP

实现的，文中第三章对这一点用了较大的篇幅进行论述。对于这两种不同的算法， 木文都给出了理论推导和相应的实证分析，结果均表明，对于较高的置信水平，两 类计算方法的表现都优于正态分布假设下的各种计算方法。第四章将以极值理论为 基础的算法和传统的计算方法进行了综合，给出了在任意概率水平下的分位数的估 计，从而将风险价值的估计从仅仅局限于高概率水平的尾部扩展到整个数轴，弥补 了极值理论算法只适宜高概率水平的不足。第五章是本文的总结部分，在这一章中， 详细指出了极值理论方法的各种不足并对金融风险管理方法的未来发展做了一些展 望。 关键词:风险价值 POT模型 广义帕雷托分布 二次子样 渐近均方误差 ABSTRACT ValueatRisk(VaR)isnowoneofthemostpopularmethodswhichareusedto managefinancialriskintheworld.Manycorporationsandscholarstakegreateffortsto studythisproject.Tothisdayalotofcomputingarithmetichasbeenbroughtforwardsuch asJ.P.MorgansRiskMetrics,MonteCarlosimulationandsoon.However,inthemost methodsofVaRcalculation,normaldistributionofreturnsisunquestionablychoseto serveastheelementaryhypothesis.Manyempiricalstudiesindicatethatthereal distributionofthepercentagepricechangeisnotnormaldistributionforithaspalpable fattertailsandathinnerwaist.SotheVaRonthebasisofnormaldistributionoftenleads tounderestimationoftherealrisk.Thereasonliesinthehypothesiswhichcannotexhibit thetailscharacteroftherealloss.InthispapertwonewarithmeticsbasedontheExtreme ValueTheoryareappliedtoconquertheshortcomingofnormaldistributionhypothesisin ordertoincreasetheaccuracyofVaRestimation. OneofthearithmeticevolvesaroundthePeakoverThresholdmodel(POT),its postulateistoinvestigatetheexcessdataofreturnsamples.AccordingtotheExtreme ValueTheory,wecanobtaintheparticulardistributionofthemostleftsamplesinvirtueof GeneralizedParetoDistributionfunction(GPD),andthenValueatRiskcanbedirectly calculatedintermofitsdefinition.Inaddition,weputforwardtheconceptofExpected Shortfall(ES)andsimultaneouslygivethedetailedcalculationformulas.Positive analyseshowsthatwecanreceiveexcellenteffectwiththeintegrationofVaRandES.In thefinalityofthischapter,wepresenttwomethodsoftestingthemodelsvalidity.The otheriscalledtwostepsubsamplebootstrap.Itspivotalpointliesinconfirmingthe estimatedvalueofthetailsparametersbytwicesampling.Afterthisstep,wecanob