资源描述:
《arma and arch models with infinite variance innovations》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、c15-ARMA-ARCHInfPage495Thursday,October26,20062:12PMCHAPTER15ARMAandARCHModelswithInfinite-VarianceInnovationsnChapter6wedescribedautoregressivemovingaverage(ARMA)Iprocessesandtheirpropertieswithregardstostationarityandestima-tion.Inthepreviouschapterweintroducedastablenon-Gaussiandis-tribu
2、tedrandomvariable(exhibitingheavytails)alongwithmajorpropertiesofstabilityandpowerlawdecayoftailswhichimplythatitssecondmomentisinfinite.1ARMAmodelscanthenbeextendedbycon-sideringerrortermsthatfollowastablenon-Gaussiandistribution,giv-ingriseto“infinitevarianceautoregressivemovingaveragemode
3、ls.”Suchheavy-tailedprocessesareencounteredineconomicsandfinanceanditisofpracticalinteresttoanalyzetheirproperties.However,thestatisticaltheoryoftheinfinitevariancemodelsisfundamentallydiffer-entfromthatofmodelswithfinitevariances.InthischapterwedescribeARMAmodelsandautoregressiveconditionalh
4、eteroskedastic(ARCH)modelswithinfinitevarianceinnovationsandoutlinerelevantproper-tiesalongwithestimationapproaches.INFINITEVARIANCEAUTOREGRESSIVEPROCESSESRecallfromequation(6.8)inChapter6thatastationaryautoregressive(AR(p))timeseries{yt}isrepresentedbydifferenceequationyt=a0+a1yt–1+a2yt–2+
5、…+apyt–p+εt(15.1)1Thereareversionsofthestablerandomvariablescalled“modifiedtemperedsta-ble”withfinitesecondmoment.JanRosinski,“TemperingStableProcesses,”inO.E.Barndorff-Nielsen(ed.),SecondMaPhyStoConferenceonLévyProcesses:TheoryandApplications,2002,pp.215–220.495c15-ARMA-ARCHInfPage496Thur
6、sday,October26,20062:12PM496FINANCIALECONOMETRICSwhere{εt}isasequenceofindependentandidenticallydistributed(IID)errorsandφ=(a0,a1,…,ap)′isanunknownparametervectorwithtrue2valueφ0.Whenthesecondmomentoftheerrorterm,E()εt,isfinite,wehaveshownthatvariousestimators(e.g.,least-squaresestimatorsan
7、dmaximumlikelihoodestimators)ofφ0areasymptoticallynormal2andseveralmethodsareavailableforstatisticalinference.WhenE()εtisinfinite,wecallthemodelgivenbyequation(15.1)theinfinitevari-anceautoregressivemodel(IVARmodel).Suchheavy-tailedmodelshavebeenencounteredineco