Modeling Financial Volatilities ARCH, GARCH, CARR and Other Models.ppt_第1页
Modeling Financial Volatilities ARCH, GARCH, CARR and Other Models.ppt_第2页
Modeling Financial Volatilities ARCH, GARCH, CARR and Other Models.ppt_第3页
Modeling Financial Volatilities ARCH, GARCH, CARR and Other Models.ppt_第4页
Modeling Financial Volatilities ARCH, GARCH, CARR and Other Models.ppt_第5页
已阅读5页,还剩54页未读 继续免费阅读

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

modellingfinancialvolatilities arch garch carrandothermodels rayy chou周雨田academiasinica nationalchiao tunguniversitypresentedat南開大學經濟學院4 11 12 2007 2 planofthetalks 1 overviewofarchmodeling2 frontiersofarchmodeling3 carrandacarrmodels4 carrdccandcovarianceforecasting5 carrindynamichedgeratio 3 2003nobelprizewinnerrobertengle 4 2003nobellaureate engle snobelcitationwasexplicitly formethodsofanalyzingeconomictimeserieswithtime varyingvolatility arch 5 25yearsofarchmodeling archsurveypapers engle 1982 econometricabollerslev chouandkroner 1992 journalofeconometrics bollerslev engleandnelson 1994 handbooksofeconometricsengle 2002 journalofappliedeconometricsengle 2004 nobellecture americaneconomicreview 6 thefirstarchmodel rollingvolatilityor historical volatilityestimatorweightsareequalforjnwhatisn 7 1982archpaper weightscanbeestimatedarch p 8 whatisarch autoregressiveconditionalheteroskedasticitypredictive conditional uncertainty heteroskedasticity thatfluctuatesovertime autoregressive 9 thesimplestproblem whatisvolatilitynow oneansweristhestandarddeviationoverthelast5yearsbutthiswillincludelotsofoldinformationthatmaynotberelevantforshorttermforecastinganotheransweristhestandarddeviationoverthelast5daysbutthiswillbehighlyvariablebecausethereissolittleinformation 10 thearchanswer useaweightedaverageofthevolatilityoveralongperiodwithhigherweightsontherecentpastandsmallbutnon zeroweightsonthedistantpast choosetheseweightsbylookingatthepastdata whatforecastingmodelwouldhavebeenbesthistorically thisisastatisticalestimationproblem 11 rollingwindowvolatilitiesnumberofdays 5 260 1300 12 arch garchvolatilities 13 confidenceintervals 14 valueatrisk futurelossesareuncertain findalossthatyouare99 sureisworsethanwhateverwilloccur thisisthevalueatrisk onedayinadvancemanydaysinadvancethissinglenumber aquantile isusedtorepresentafulldistribution itcanbemisleading 15 calculatingvar forecasttheonedaystandarddeviation garchstylemodelsarewidelyused then assumingnormality multiplyby2 33withoutassumingnormality multiplybythequantileofthestandardizedresiduals fortheexample multiplier 2 65 16 multi dayhorizons ifvolatilitywereconstant thenthemulti dayvolatilitywouldsimplyrequiremultiplyingbythesquarerootofthedays becausevolatilityisdynamicandasymmetric thelowertailismoreextremeandthevarshouldbegreater 17 twoperiodreturns twoperiodreturnisthesumoftwooneperiodcontinuouslycompoundedreturnslookatbinomialtreeversionasymmetrygivesnegativeskewness highvariance lowvariance 18 multiplierfor10days fora10day99 valueatrisk conventionalpracticemultipliesthedailystandarddeviationby7 36forthesamemultiplierwithasymmetricgarchitissimulatedfromtheexampletobe7 88bootstrappingfromtheresidualsthemultiplierbecomes8 52 19 options tradedoptionsalwayshavemultipledaystoexpiration hencethedistributionoffuturepricelevelsisnegativelyskewed thustheblackscholesimpliedvolatilityshoulddependonstrikeifoptionsarepricedbygarch askewinimpliedvolatilitywillresultfromasymmetricgarch atleastforshortmaturities 20 impliedvolatilityskewfor10dayoption fromsimulated riskneutral finalvalues findaverageputoptionpayoffforeachstrike calculateblackscholesimpliedvolatilitiesandplotagainststrike noticethecleardownwardslope thiswouldbezeroforconstantvolatility 21 whataboutheteroskedasticity 22 exponentialsmoother anothersimplemodelweightsaredecliningnofinitecutoffwhatislambda riskmetrics 06 23 thegarchmodel thevarianceofrtisaweightedaverageofthreecomponentsaconstantorunconditionalvarianceyesterday sforecastyesterday snews 24 25 forecastingwithgarch garch 1 1 canbewrittenasarma 1 1 theautoregressivecoefficientisthemovingaveragecoefficientis 26 garch 1 1 forecasts 27 forecastingaveragevolatility annualizedvol squarerootof252timestheaveragedailystandarddeviationassumethatreturnsareuncorrelated 28 variancetargeting rewritingthegarchmodelwhereiseasilyseentobetheunconditionalorlongrunvariancethisparametercanbeconstrainedtobeequaltosomenumbersuchasthesamplevariance mleonlyestimatesthedynamics 29 thecomponentmodel engleandlee 1999 qislongruncomponentand h q istransitoryvolatilitymeanrevertstoaslowlymovinglongruncomponent 30 taylor schwert standarddeviationmodel 31 asymmetricmodels theleverageeffect engleandng 1993 followingnelson 1989 newsimpactcurverelatestoday sreturnstotomorrowsvolatilitydefinedasadummyvariablewhichis1fordowndays 32 newsimpactcurve 33 otherasymmetricmodels 34 newarchmodels gjr garchtarchstarchaarchnarchmarchswarchsnparchaparchtaylor schwert figarchfiegarchcomponentasymmetriccomponentsqgarchcesgarchstudenttgedsparch 35 financialeconometrics thismayalsobethebirthoffinancialeconometricsstatisticalmodelsdevelopedspecificallyforfinancialapplicationstodaythisisaverypopularandactiveresearchareawithmanyapplications 36 exogenousvariablesinagarchmodel includepredeterminedvariablesintothevarianceequationeasytoestimateandforecastonestepmulti stepforecastingisdifficult 37 examples non lineareffectsdeterministiceffectsnewsfromothermarketsheatwavesvs meteorshowersotherassetsimpliedvolatilitiesindexvolatilitymacrovariablesorevents 38 stochasticvolatilitymodels easytosimulatemodelseasytocalculaterealizedvolatilitydifficulttosummarizepastinformationsethowtodefineinnovation 39 svmodels taylor 1982 40 41 whathavebeenaccomplishedin25years stochasticpropertiesofarchmodelsthealphabetsoupofarchtypemodelsculminatinginfiglewski syaarchcomparisonwithlatentorstochasticvolatilitymacroapplicationsinflation policysimpleoptionstradingstrategiesefficiencyofoptionsmarketsmodelingtheriskreturntrade offassetpricing capmmeasuringrisk var 42 newfrontiers highfrequencyvolatilityclocktimeticktimeusehighfrequencydatatoimprovedailyvolatilityestimates 43 anotherfrontier multivariatearchhowgeneralshouldamultivariategarchmodelbe thedynamicconditionalcorrelationmodelengle 2002 engleandsheppard 2002 highlyrestrictedparameterizationseparatesthevolatilityandcorrelationparametershowtomeasurecorrelationswithhighfrequencydataepps 1972 zebedee 2001 44 themultivariateproblem assetallocationandriskmanagementproblemsrequirelargecovariancematricescreditrisknowalsorequiresbigcorrelationmatricestoaccuratelymodellossordefaultcorrelationsmultivariategarchhasneverbeenwidelyused itistoodifficulttospecifyandestimate 45 dynamicconditionalcorrelation dccisanewtypeofmultivariategarchmodelthatisparticularlyconvenientforbigsystems seeengle 2002 orengle 2004 46 dcc estimatevolatilitiesforeachassetandcomputethestandardizedresidualsorvolatilityadjustedreturns estimatethetimevaryingcovariancesbetweentheseusingamaximumlikelihoodcriterionandoneofseveralmodelsforthecorrelations formthecorrelationmatrixandcovariancematrix theyareguaranteedtobepositivedefinite 47 howitworks whentwoassetsmoveinthesamedirection thecorrelationisincreasedslightly whentheymoveintheoppositedirectionitisdecreased thiseffectmaybestrongerindownmarkets thecorrelationsoftenareassumedtoonlytemporarilydeviatefromalongrunmean 48 stillmorefrontiers optionspricingandhedgingsimulationmethods englemustafa 1992 treemethods trevorandrichken 1999 riskneutralizationbylocalquadraticapproximation duan 1995 empiricalpricingkernel rosenbergandengle 2001 pathdependentoptions 49 twomorefrontiers modelingnon negativeprocessesusingarch garchmodelsforawiderrangeoftimeseriesproblemssimulationmethodsformodelanalysis seeengle spaperfordetails 50 modelingnon negativeprocesses supposeisanon negativeprocesswhichhasnon zeroprobabilityofbeingzeroatanytimeamodelsuchasmightcommonlybeemployedwheretheconditionalmeanandvariancedependonpredeterminedandweaklyexogenousvariables howeverdcannothaveaconstantrangedisunlikelytohaveconstantvarianceleastsquaresisconsistentbutinefficient 51 multiplicativeerrormodel instead considerthemodelnowtheerrordistributioncanbei i d withoutviolatingtheassumptionsofthemodel ofcourseitstillmightnotbe forecastsofxdonotdependupontheerrordensity 52 estimationofmem specifythemean forexampleassumeittobelinearinlaggeddependentvariables laggedmeansandotherpredeterminedandweaklyexogenousvariableszspecifytheerrordensityandanyheteroskedasticityitmayhave forexample assumeitisunitexponential 53 loglikelihood inthiscase nowifthisissimplygarch p q withgaussianlikelihoodfunctionandexogenousvariablesz 54 estimationwithgarch thislikelihoodcanbemaximizedwithagaussiangarchprogramsuchaseviews simplyconsidersquarerootofxasthedependen

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论