分类号:F830密级:公开UDC:单位代码:10424学位论文VaRVaRVaRVaR与与与与CVaRCVaRCVaRCVaR的对比研究及实证分析的对比研究及实证分析的对比研究及实证分析的对比研究及实证分析郭花申请学位级别:硕士学位专业名称:运筹学与控制论指导教师姓名:李述山职称:教授山东科技大学二零零七年五月论文题目:论文题目:论文题目:论文题目:VaRVaRVaRVaR与与与与CVaRCVaRCVaRCVaR的对比研究及实证分析的对比研究及实证分析的对比研究及实证分析的对比研究及实证分析作者姓名:作者姓名:作者姓名:作者姓名:郭花入学时间:入学时间:入学时间:入学时间:2004200420042004年9999月专业名称:专业名称:专业名称:专业名称:运筹学与控制论研究方向:研究方向:研究方向:研究方向:预决策理论及应用指导教师:指导教师:指导教师:指导教师:李述山职职职职称:称:称:称:教授论文提交日期:论文提交日期:论文提交日期:论文提交日期:2007200720072007年5555月论文答辩日期:论文答辩日期:论文答辩日期:论文答辩日期:2007200720072007年6666月授予学位日期:授予学位日期:授予学位日期:授予学位日期:CONTRASTCONTRASTCONTRASTCONTRASTSTUDYSTUDYSTUDYSTUDYANDANDANDANDEMPIRICALEMPIRICALEMPIRICALEMPIRICALANALYSISANALYSISANALYSISANALYSISOFOFOFOFVARVARVARVARANDANDANDANDCVARCVARCVARCVARAAAADissertationDissertationDissertationDissertationsubmittedsubmittedsubmittedsubmittedininininfulfillmentfulfillmentfulfillmentfulfillmentofofofofthethethetherequirementsrequirementsrequirementsrequirementsofofofofthethethethedegreedegreedegreedegreeofofofofMASTERMASTERMASTERMASTEROFOFOFOFSCIENCESCIENCESCIENCESCIENCEfromfromfromfromShandongShandongShandongShandongUniversityUniversityUniversityUniversityofofofofScienceScienceScienceScienceandandandandTechnologyTechnologyTechnologyTechnologybbbbyyyyGuoGuoGuoGuoHuaHuaHuaHuaSupervisor:Supervisor:Supervisor:Supervisor:ProfessorProfessorProfessorProfessorLiLiLiLiShushanShushanShushanShushanCollegeCollegeCollegeCollegeofofofofinformationinformationinformationinformationsciencesciencesciencescienceandandandandEngineeringEngineeringEngineeringEngineeringMayMayMayMay2007200720072007声明本人呈交给山东科技大学的这篇硕士学位论文,除了所列参考文献和世所公认的文献外,全部是本人在导师指导下的研究成果。该论文资料尚没有呈交于其它任何学术机关作鉴定。硕士生签名:日期:AFFIRMATIONAFFIRMATIONAFFIRMATIONAFFIRMATIONIIIIdeclaredeclaredeclaredeclarethatthatthatthatthisthisthisthisdissertation,dissertation,dissertation,dissertation,submittedsubmittedsubmittedsubmittedininininfulfillmentfulfillmentfulfillmentfulfillmentofofofofthethethetherequirementsrequirementsrequirementsrequirementsforforforforthethethetheawardawardawardawardofofofofDoctorDoctorDoctorDoctorofofofofPhilosophyPhilosophyPhilosophyPhilosophyininininShandongShandongShandongShandongUniversityUniversityUniversityUniversityofofofofScienceScienceScienceScienceandandandandTechnology,Technology,Technology,Technology,isisisiswhollywhollywhollywhollymymymymyownownownownworkworkworkworkunlessunlessunlessunlessreferencedreferencedreferencedreferencedofofofofacknowledge.acknowledge.acknowledge.acknowledge.TheTheTheThedocumentdocumentdocumentdocumenthashashashasnotnotnotnotbeenbeenbeenbeensubmitsubmitsubmitsubmittedtedtedtedforforforforqualificationqualificationqualificationqualificationatatatatanyanyanyanyotherotherotherotheracademicacademicacademicacademicinstitute.institute.institute.institute.Signature:Signature:Signature:Signature:Date:Date:Date:Date:山东科技大学硕士论文摘要1摘要从中国成功入世以来,金融服务业不可避免的同国际标准接轨,接受世界的挑战。因此,积极探索适合我国金融机构的风险管理方法和体系成为当前重要而紧迫的任务。VaR(Value-at-Risk)风险度量方法自1993年提出以来,己成为金融机构和金融管理机构衡量市场风险的标准方法。VaR方法近年来非常流行,但研究结果和实践经验都表明,过于单纯的VaR风险度量方法存在严重缺陷。CVaR的提出,又弥补了VaR的缺陷。论文第二部分同时考虑这两种风险度量方法,以更全面地刻画尾部风险。VaR和CVaR提出以后,有不少的计算方法出现,但它们都有各自的缺陷。因为几乎所有的传统方法采用的观测值都集中在分布中部,实际上,分布尾部才是VaR和CVaR计算所最关心的。分布在尾部的点都是一些极少发生又具有显著影响的观测值,称为极值,而极值理论正是对这些极值提供统计分析的模型。论文第三部分介绍了极值方法的理论基础,极值方法根据极值的选取方式分为峰值法和阈值法,本文先从理论上推导了由阈值法和峰值法计算VaR和CVaR的公式,再对上证综指采用阈值法进行实证分析,并与正态分布下的结果进行了比较,而对S&P500指数用峰值法进行实证分析,结果表明,峰值法能很好地刻画金融回报的分布尾部,得到较精确的VaR和CVaR估计值。最后由返回检验的结果来确定最佳的分组方式及其VaR和CVaR值。论文最后一部分运用GARCH模型分析收益率序列的波动聚集现象,对其中的随机项分别采用正态分布、t分布和广义Pareto分布进行拟合,通过对证券指数的VaR实证计算发现GARCH-GPD模型在VaR计算研究中能够得到更好的结果。关键词:风险价值,条件风险价值,极值理论,广义极值分布,广义帕累托分布,GARCH模型山东科技大学硕士论文abstract2AbstractAbstractAbstractAbstractSinceChinahasjoinedWTO,itsfinancialindustrywillhavetobefacedwiththechallengesfromtheworld.Soitisanimportantandurgenttasktosearchforfitfulmethodsandsystemstomanagerisk.Value-at-Risk(VaR)approachhasbecomethestandardmethodforriskmanagementorganizationVaRisverypopularrecently.Butithasvarioustheoreticaldeficienciesasameasureofmarketrisk.ConditionalVaR(CVaR)isanalternativeriskmeasuretothequantilewhichovercomesthetheoreticaldeficienciesofVaR.Inparticular,thisriskmeasuregivessomeinformationaboutthesizeofthepotentiallossesgiventhatalossbiggerthanVaRhasoccurred.Thispaperestimatesandassessestail-relatedriskusingVaRandCVaRtogether.ThoughVaRandCVaRhavemanycomputingmethods,theyhavelimitations.Becausealmostallofthetraditionalmethodsestimatingtail-relatedriskVaRandCVaRfocusonthecentralobservationsor,inotherwords,onreturnsundernormalmarketconditions.However,VaRandCVaRareriskmeasuresthatrelatessolelytothetailsofthedistribution.Theextremevalueswhichliesinthetailaresomerarelyhappenedeventsthathavesignificantinfluence.ExtremeValueTheory(EVT)isthestatisticalmodeltostudythebehaviorofextremevalues.ThispaperintroducesthebasicknowledgeofEVTandestimatesVaRandCVaRusingEVT.TheapplicationmethodsofEVThaveblockmaximummethodandpeakoverthresholdmethodaccordingtorelatedwaysofidentifyingextremesinrealdata.ThispapergainstheformulascomputingVaRandCVaRusingpeakoverthresholdmethodandblockmaximummethod.Thenmakesempiricalana