




版權說明:本文檔由用戶提供并上傳,收益歸屬內容提供方,若內容存在侵權,請進行舉報或認領
文檔簡介
FinanceandEconomicsDiscussionSeries
FederalReserveBoard,Washington,D.C.
ISSN1936-2854(Print)
ISSN2767-3898(Online)
ThePricingKernelinOptions
StevenHeston,KrisJacobs,HyungJooKim
2023-053
Pleasecitethispaperas:
Heston,Steven,KrisJacobs,andHyungJooKim(2023).“ThePricingKernelinOptions,”FinanceandEconomicsDiscussionSeries2023-053.Washington:BoardofGovernorsoftheFederalReserveSystem,
/10.17016/FEDS.2023.053
.
NOTE:StafworkingpapersintheFinanceandEconomicsDiscussionSeries(FEDS)arepreliminarymaterialscirculatedtostimulatediscussionandcriticalcomment.TheanalysisandconclusionssetfortharethoseoftheauthorsanddonotindicateconcurrencebyothermembersoftheresearchstafortheBoardofGovernors.ReferencesinpublicationstotheFinanceandEconomicsDiscussionSeries(otherthanacknowledgement)shouldbeclearedwiththeauthor(s)toprotectthetentativecharacterofthesepapers.
ThePricingKernelinOptions*
StevenHeston
UniversityofMaryland
KrisJacobs
UniversityofHouston
HyungJooKim
FederalReserveBoard
April7,2023
Abstract
Theempiricaloptionvaluationliteraturespecifiesthepricingkernelthroughthepriceofrisk,ordefinesitimplicitlyastheratioofrisk-neutralandphysicalprobabilities.Instead,weextendtheeconomicallyappealingRubinstein-Brennankernelstoadynamicframeworkthatallowspath-andvolatility-dependence.Becauseoflowstatisticalpower,kernelswithdifferenteconomicpropertiescanproducesimilaroveralloptionfit,evenwhentheyimplycross-sectionalpricinganomaliesandimplausibleriskpremiums.Imposingparsimoniouseconomicrestrictionssuchasmonotonicityandpath-independence(recoverytheory)achievesgoodoptionfitandreasonableestimatesofequityandvarianceriskpremiums,whileresolvingpricingkernelanomalies.
*Heston:sheston@;Jacobs:kjacobs@;Kim:hyungjoo.kim@.WewouldliketothankCaioAlmeida,DavidBates,HiteshDoshi,Bj?rnEraker,XiaohuiGao,StefanoGiglio,MassimoGuidolin,AlexKostakis,PaolaPederzoli,Jean-PaulRenne,seminarparticipantsatthe2023AFAConference,the2022FinanceDownUnderConference,the2022SoFiEConference,the2022FMAConferenceonDerivativesandVolatility,K.U.Leuven,SyracuseUniversity,theUniversitiesofHoustonandLiverpool,andespeciallyourdiscussantsGurdipBakshi,MikhailChernov,andJeroenDalderopforhelpfulcomments.TheanalysisandconclusionssetfortharethoseoftheauthorsanddonotindicateconcurrencebytheFederalReserveBoardorothermembersofitsstaff.Co-authorHyungJooKimworkedonthisprojectpriortoemploymentattheFederalReserveBoard,whileaPh.D.candidateattheUniversityofHouston.DatasourceswereobtainedunderpurviewofUniversityofHoustonlicenses.
1
1Introduction
Thepricingkernelisacriticalconceptinassetpricing,becauseitdeterminesriskpremiaonallsecurities.Oneapproachtostudythepropertiesofthepricingkernelspecifiesitsrelationtoaggregateconsumptionandestimatestheresultingmodelusingconsumptiondataandreturnsonvariousassets.1Alternatively,buildingontheinsightsofBreedenandLitzenberger(1978),anextensiveliteratureestimatesthepricingkernelusingindexreturnsandindexoptionprices.2Indexoptionsareinterestingfromanempiricalperspectivebecausetheyidentifythepricingkernelundertheassumptionthattheequityindexlevelisequaltoaggregatewealth.However,thisliteraturehasgivenrisetopuzzlingnon-monotonic(U-shaped)estimatesofthepricingkernel.3
Thispapercomplementsthesetwoapproaches.Wespecifyeconomicallyintuitivepricingkernelsinparametricoptionpricingmodelsandestimatethemusingindexreturnsandindexoptionprices.Ourapproachdiffersfromtheexistingparametricliterature,whichtypicallyspecifiespricesofrisk,andtherebydefinesthepricingkernelimplicitlyastheratioofrisk-neutralandphysicalprobabilities.4Instead,weproposeaclassofeconomicallymotivatedpricingkernelsthatextendthepowerkernelsinRubinstein(1976)andBrennan(1979)tobefunctionsofthepathsoflatentvariancev(t)andtheindexlevelS(t).Thesekernelsalsonestpath-independentkernels(Ross,2015)asspecialcases,andareconsistentwiththeconventionalassumptionofaffinedynamicsunderthephysicalandrisk-neutralmeasureinthesquarerootstochasticvolatilitymodel(Heston,1993).5Byincludingseparatecomponentsforvolatilityandstockrisk,thesekernelsallowusto
1See,amongothers,HansenandSingleton(1982),MehraandPrescott(1985),HansenandJagannathan(1991),CampbellandCochrane(1999),BansalandYaron(2004),Gabaix(2012)andWachter(2013)forimportantcontri-butionstothisliterature.
2SeeforinstanceA¨?t-SahaliaandLo(1998),A¨?t-SahaliaandLo(2000),JackwerthandRubinstein(1996),Jack-werth(2000)andRosenbergandEngle(2002).
3ForevidenceonU-shapedpricingkernels,seeforinstanceJackwerth(2000),A¨?t-SahaliaandLo(2000),RosenbergandEngle(2002),Bakshi,Madan,andPanayotov(2010),Chabi-Yo(2012),Christoffersen,Heston,andJacobs(2013),SongandXiu(2016),andCuesdeanuandJackwerth(2018).Linn,Shive,andShumway(2018)andBarone-Adesi,Fusari,Mira,andSala(2020)ontheotherhandarguethatthepricingkerneliswell-behaved.
4Forexamplesofthisapproach,seetheseminalpapersinthisliteraturebyChernovandGhysels(2000),Pan(2002),andEraker(2004).
5Forsimplicity,weusethesimplestpossibleoptionpricingmodelwithastochasticvolatilityfactor,butourapproachcanbeeasilygeneralizedtomorecomplexmodels.
2
examinedistinctoriginsoftheequityandvarianceriskpremiums.
Westartouranalysiswiththeoften-usedspecificationofanequity(market)riskpremiumμv(t)andavarianceriskpremiumλv(t).Werefertothisspecificationas“completelyaffine”,adoptingtheterminologyinSingleton(2006,p.392)andthetermstructureliterature.Wecharacterizeapricingkernelthatisconsistentwiththeseriskpremia,andderivetheparameterrestrictionsconsistentwiththemartingaleconditionsandabsenceofarbitrage.Wethenexploreaspecificationwithanequityriskpremiumμ0+μ1v(t)andavarianceriskpremiumλ0+λ1v(t).Werefertothisspecificationas“affine”.Singleton(2006)pointsoutthatthisspecificationisproblematicbecauseitmayviolateno-arbitrageconditions,andnotesthatitwouldbeinterestingtocharacterizetheparameterrestrictionsthatpreventarbitrageforthisspecification.Becauseourkernelisformulatedasafunctionofthestatevariables,itisstraightforwardtospecifysuchrestrictions.
Ourempiricalanalysisusesajointlikelihoodbasedonindexreturnsandarichoptiondataset,usingdatafortheJanuary1996toJune2019period.6Weestimatethisjointlikelihoodforthepricingkernelscorrespondingtothecompletelyaffineandaffinepricesofrisk,andwealsoestimateitsubjecttovariousrestrictionsonparametervaluesandriskpremia.Becausethekernelsareformulatedasafunctionofthestatevariables,itisrelativelystraightforwardtoderivetheimplicationsofeachkernelforthe“marginal”kernelwhichspecifiesstatepriceasafunctionofS(t).Themarginalkernelplaysacentralroleinempiricalapplicationssuchasthepricingkernelpuzzle.
Afirstempiricalresultaddressesthefitandempiricalcontentofthekernelsthatsupportcompletelyaffineandaffinepricesofrisk.Unsurprisingly,wefindthattheaffinepriceofriskspec-ificationprovidessignificantlybetterfitthanthenestedcompletelyaffinespecification.However,theimprovedfitduetotheinterceptsintheaffinespecificationcomesatthecostofimplausibleSharperatiosand/orsignsoftheriskpremiums.Moreover,whilethemarginalpricingkernelforthecompletelyaffinespecificationiswellbehavedandeconomicallyplausible,thekernelinthe
6Muchofthemodernoptionpricingliteraturejointlyconsidersthetime-seriesofobservablereturnsandoptionprices.See,forinstance,Pan(2002),Eraker(2004),Bates(2006),A¨?t-SahaliaandKimmel(2007),Hurn,Lindsay,andMcClelland(2015),andAndersen,Fusari,andTodorov(2017).
3
affinecaseimpliesstatepricesthatareS-shapedasafunctionofwealth.Whenweimposeaddi-tionalparameterrestrictionsthatprecludearbitrage,themarginalpricingkerneliswell-behaved,butempiricalfitworsens.Weconcludethattheaffinespecificationcorrespondstoimplausibleeconomicassumptions,andthatsmallandseeminglyinnocuousmodificationstothepriceofriskspecificationsusedintheliteraturecorrespondtodifferentpricingkernelswithradicallydifferenteconomicimplications.Consequently,weadvocatetheuseofthecompletelyaffinepriceofriskspecification.
Asecondsetofresultscomparestheunrestrictedkernelthatsupportsthecompletelyaffinepriceofriskwithrestrictedversions.Werejectrestrictionsthattheequityorvarianceriskpremiumsarezero,butwecannotrejecttheindependenceofthepricingkernelfromeithervarianceormarketreturnshocks.7Weareunabletostatisticallypindowntheoriginsoftheseriskpremiumsbecauseinnovationstomarketreturnsandmarketvariancearehighly(negatively)correlated.Theequityandvarianceriskpremiumeachhavetwocomponents,oneduetovarianceaversionandanotherduetoindexlevelriskaversion.Ifwerestrictoneoftheriskaversionparameterstobezero,theotherparameterabsorbsmostofthateffect.Inotherwords,theseriskslargelyspaneachother.However,restrictingvarianceorindexlevelriskaversiontozeroimpliesradicallydifferentestimatesoftheequityandvarianceriskpremia,aswellaslargedifferencesinthestatepricesembodiedinthemarginalpricingkernelasafunctionofwealth.Therealizedtimeseriespathofthekernelwithoutvolatilityriskisalsosubstantiallylessvariablecomparedtotheunrestrictedkernel,especiallyincrisisperiods.Afinalobservationisthatwhilewecannotstatisticallyrejectthepath-independencerestrictionusedinRoss(2015),itimpliesanimplausibleestimateofthevarianceaversionparameter.
Ourthirdfindingshedslightonthepricingkernelpuzzle–thefindingthatthemarginalpricingkernelisU-shaped–intheexistingliterature.WeshowthatU-shapedmarginalkernelscanresultfromanunderlyingpricingkernelthatisamonotonicfunctionofvolatilityandthestockprice.Therefore,U-shapedpricingkernelsarenotanomalousnordotheyconstituteanassetpricing
7Notethatthehypothesisthatthevariance-aversionparameterequalszeroamountstotheabsenceofaninde-pendentvarianceriskpremium,whichamountstologarithmicutilityintheMerton(1973)ICAPM.
4
puzzle.
Ourfourthfindingaddressesestimationofriskpremiums.FollowingBreedenandLitzenberger’s(1978)insightthattherisk-neutraldensitycanbeinferredfromoptionprices,financialeconomistshaveemphasizedfittingoptionsandreturnsjointlytoidentifyriskpremia.Wefindthatthesedatahavelowpowertodistinguishdifferentpricingkernels,becauseidentifyingpricingkernelsisequivalenttotheestimationofconditionalriskpremia,anditisdifficulttoestimateaveragereturnsovershortperiods.8DifferentparameterrestrictionsleadtowidelydifferentSharperatiosandequityandvarianceriskpremia,butdonottranslateintolargedecreasesinthelikelihood.Merton(1980)convincinglyarguesthatverylongtimeseriesofreturnsarerequiredtoobtainreliableestimatesoftheequitypremium.OurfindingsextendMerton’sobservationtojointestimationofequityandvarianceriskpremia.WealsoreinforceMerton’sconclusionthateconomicrestrictionsincreasepowertoidentifymarketriskpremia.WhileMerton(1980)advocatesimposingapositivityrestrictiononthepathoftheconditionalequityriskpremium,wefindthatimposinganegativityrestrictiononthemarketvarianceriskpremiumleadstomoreplausibleandreliableestimates.OurfindingsalsoconfirmtheresultsinBakshi,Crosby,andGao(2022)thatsomeoptionmodelparametersarehardtoidentifybecauseof(darkmatter)unspannedrisksthataffectriskpremiums. Ourpaperisrelatedtoseveralotherstrandsofliteraturebesidesthoseontheestimationofparametricoptionpricingmodelsandthepricingkernelpuzzle.Severalstudiesuseconsumption-basedmodelstoanalyzehowpreferencesandpricingkernelsimpactindexoptionprices.9SomeofthesestudiesusetherecursivepreferencesofKrepsandPorteus(1978),EpsteinandZin(1989)andDuffieandEpstein(1992),whichresultinstochasticvolatilityofindexreturns.OurproposedpricingkernelsareextensionsofthepowerutilityofRubinstein(1976).Whileconsumptionisnotastatevariableinoursetup,ourapproachprovidesadirectrelationwithexistingempiri-
8Thisstatementisspecifictoplainvanillaoptionprices,whicharesensitivetotheprobabilitiesatexpirationbutnotveryinformativeaboutthepath-dependentpropertiesofthepricingkernel.PricingkernelswithwidelydifferenteconomicimplicationscanthereforeproducesimilarvaluesforEuropeanoptions.
9See,forinstance,Garcia,Luger,andRenault(2003),ErakerandShaliastovich(2008),Drechsler(2013),Shalias-tovich(2015),ErakerandYang(2019),andSeoandWachter(2019).Liu,Pan,andWang(2005)andErakerandWu(2017)userelatedmodelswiththedividendpayoutrateandcashflowrespectivelyasthestatevariable.
5
calimplementationsof(reduced-form)parametricdynamicoptionpricingmodels.Itisthereforestraightforwardtoimplementusingoptiondata,whichallowsustoexploretheimpactofstockindexvolatilityonthepricingkernel.
Fromanempiricalperspective,arelatedpaperisChernov(2003),whoreverseengineersthepricingkernelbasedonoptionsonvarioussecurities.Chernov(2003)alsostudiesthetimepathoftherealizedpricingkerneltolearnaboutstatevariablesandtherelationbetweenthepricingkernelandeconomicconditions.Ghosh,Julliard,andTaylor(2017)alsoexploretherelationbetweenthepricingkernelandbusinesscyclefluctuations,butdonotuseoptionstoestimatethekernel.Brennan,Liu,andXia(2006)specifyandestimatepricingkernelswithmultiplestatevariables.BeasonandSchreindorfer(2022)analyzetheimplicationsofoptiondataformacro-financemodels.Dew-BeckerandGiglio(2022)studytheimplicationsofsyntheticputsforthepropertiesofthemarginalpricingkernel.
Thepaperproceedsasfollows.Section2discussesthedata.Section3reviewstheHeston(1993)stochasticvolatilitymodelanddiscussesourestimationapproachbasedonreturnsandoptionsdata.Section4specifiestheclassofpricingkernelsthatconnecttherisk-neutralandphysicaldynamics.Section5presentstheestimationresultsandSection6discussestheireconomicimplications.Section7concludes.
2Data
Ourempiricalanalysisusesout-of-the-money(OTM)S&P500callandputoptionswithmaturitiesbetween14and365daysfortheJanuary1996toJune2019period.WeobtaintheoptiondatafromOptionMetrics.Weapplythefollowingfilters:
1.Discardoptionswithimpliedvolatilitysmallerthan5%orgreaterthan150%.
2.Discardoptionswithvolumeoropeninterestlessthantencontracts.
3.Discardoptionswithmidpricelessthan$0.50orbidpricelessthan$0.375toavoidlow-valuedoptions.
6
4.Discardoptionswithdataerrors–wherebidpriceexceedsofferprice,oranegativepriceisimpliedthroughput-callparity.
5.Discardoptionswithmoneyness<0.75or>1.25.
Thenwekeepthesixmostactivelytradedstrikepricesforeachavailablematurity.Itisimportanttouseaslongatimeperiodaspossibletoidentifykeyaspectsofthemodel,includingvolatilitypersistence.10Ontheotherhand,estimationusinglargeoptionpanelsandlongtimeseriesisverytime-intensive.Ratherthanusingashorttimeseriesofdailyoptiondata,weuseanextendedtimeperiod,butweselectoptioncontractsforonedayperweekonly.Followingseveralexistingstudies(see,e.g.,HestonandNandi,2000;Christoffersen,Heston,andJacobs,2013),weuseWednesdaydatabecauseitisthedayoftheweekleastlikelytobeaholiday.Itisalsolesslikelythanotherdaystobeaffectedbyday-of-the-weekeffects.Thesestepsresultinadatasetwith62,483optioncontracts.Table1presentsdescriptivestatistics.
WeobtainS&P500indexreturnsfromCRSP.WeusedatafortheJanuary1990toJune2019period.Thissamplestartsbeforetheoptionsampletohelpwiththeidentificationofthereturnparametersunderthephysicalmeasure,asinChristoffersen,Heston,andJacobs(2013).WealsousedataontheVIXfromJanuary1990toJune2019,whichweobtainfromtheFederalReserveBankofSt.LouisEconomicDatabase.Thetimeseriesfortherisk-freerateisproxiedbytheone-monthTreasuryBillrateobtainedfromCRSP.Followingexistingwork,optionsarevaluedusingamaturity-specificrisk-freerate.WeapplyacubicsplineinterpolationtothedataobtainedfromOptionMetrics.
3Return-BasedandOption-BasedParameterEstimates
WeestimatethestylizedaffineHeston(1993)stochasticvolatilitymodel.Weobtainparameterestimatesforthismodelunderthephysicalmeasure,exclusivelybasedonreturns,andundertherisk-neutralmeasure,exclusivelybasedonoptions.Thenwecomparetheresultingestimates.
10See,forinstance,Broadie,Chernov,andJohannes(2007)foradiscussion.
7
3.1TheModel
Wefocusonthesimplestpossiblestochasticvolatilitymodelwithasinglediffusivevolatilityfactor.Werecognizethattheexistingliteraturehasclearlyestablishedthatadditionalvolatilityfactors,jumpsinreturnsandvarianceand/ortailfactorsarerequiredtoimproveoptionfitandpricingperformance.However,wedeliberatelyfocusonthesimplestpossiblemodelbecauseitsufficestoillustrateourmainargumentandwewanttoavoidcomparisonsbetweenmodelsandfactors.Ouranalysiscanberepeatedusingmoregeneralmodels,butatthecostofmuchgreatercomplexity.Webelievethatmostoftheissueswehighlighthereusingasimplemodelareevenmorerelevantinmorecomplexmodels,butweleavethisanalysisforfuturework.
WeemploytheHeston(1993)continuous-timestochasticsquarerootvolatilitymodeltospecifystockpricedynamicsaswellasoptionprices.Foroptionvaluation,therisk-neutralstockpricedynamicissufficient.Thesquarerootstochasticvolatilitymodelspecifiestherisk-neutraldynamicsofthespotindexS(t)anditsstochasticvariancev(t)asfollows:
(1)
dS(t)/S(t)=rdt+^v(t)dz(t),
dv(t)=κ?(θ??v(t))dt+σ^v(t)dz(t),
wheredzanddzareWienerprocesseswithcorrelationcoefficientρ.Therisk-freeratercanbeeitherconstantortime-varying;thishasnegligibleimplicationsforourresults.Itisalsostraight-forwardtospecifyastochasticmodelfortherisk-freerate,butitiswell-knownfromtheexistingliteraturethatthisdoesnothaveamajorimpactonoptionvaluation(Bakshi,Cao,andChen,1997).Wethereforedeliberatelyfocusonthesimplestpossiblemodel.Consistentwithmostoftheexistingliterature,wefocusonaphysicaldynamicthathasthesamefunctionalformastherisk-neutraldynamic:
(2)
dS(t)/S(t)=[r+μ(v(t))]dt+^v(t)dz1(t),
dv(t)=κ(θ?v(t))dt+σ^v(t)dz2(t),
8
whereμ(v(t))denotestheequitypremiumasafunctionofv(t),anddz1anddz2areWienerprocessesunderthephysicalmeasure.Notethatσ,thevarianceofvarianceparameter,andρ,thecorrelationbetweenz1andz2,areassumedtobeidenticaltothecorrespondingparametersintherisk-neutraldynamics.However,thelong-runphysicalvarianceθandmeanreversionκdifferfromthelong-runrisk-neutralvarianceθ?andmeanreversionκ?.Thisspecificationisconsistentwiththeexistingliterature.Itrepresentsthemostgeneralcombinationofphysicalandrisk-neutraldynamicsthatareconsistentwiththeaffinespecificationandGirsanov’stheorem.Weanalyzethismappinginmoredetailbelowinourdiscussionof(the)pricingkernel(s).
3.2TheInstantaneousStochasticVarianceandtheVIX
IntheHeston(1993)model,aswellasinitsmanygeneralizationsstudiedintheliterature,thestochasticvarianceisunknown.Thislatencyistypicallyaddressedinestimationbyusingfiltering-orsimulation-basedtechniques(see,e.g.,Eraker,Johannes,andPolson,2003;Eraker,2004;Bates,2006;Christoffersen,Jacobs,andMimouni,2010).Itiswell-knownthattheimplementationofsuchtechniquesiscomputationallyverydemanding,especiallywhenusinglongtimeseriesand
largecross-sectionsofoptionpricesinestimation.
Toalleviatethiscomputationalburden,wefollowadifferentapproach.11Weusethefactthatthestochasticvariancev(t)canberepresentedasalinearfunctionofVIX2(t).Thisdirectlyfollowsfromthemodelspecification:Whenv(t)followsaCIRprocess,VIX2(t)isalinearfunctionofv(t).Specifically,themodel-impliedVIX2(t)isgivenby:
(3)
VIX2(t)=E[\tt+?1mv(u)du]
=θ?+e11(v(t)?θ?),
11SeeBates(2000)andAndersen,Fusari,andTodorov(2015)foralternativeapproaches.
9
w
where?1m≈30/365.Rearrangingequation(3)yields
VIX2(t)?θ?(1?w)
,
v(t)=(4)
wherew=(1?exp(?κ??1m))/(κ??1m).Inimplementation,wecanaddameasurementerror
becauseequations(3)and(4)usethemodel-impliedVIX2(t).Equation(3)inconjunctionwiththemeasurementerroryieldsameasurementequationwhichcanbeusedtofilterthelatentstatevariable.Jones(2003),Cheung(2008),andChernov,Graveline,andZviadadze(2018)usethismeasurementequationandaBayesianframeworkwithMarkovchainMonteCarlomethodstoestimateoptionpricingmodels.Wefurthersimplifythesetup:Wedonotusethemeasurementequation,butrelaxtherestrictionsonthecoefficientsinequation(4)andomitthemeasurement
error.Specifically,weassume:
v(t)=η0+η1VIX2(t).(5)
Wethenuseequation(5)inthevaluationformulaforalloptionsinthesample.Asaresult,optionsareafunctionnotonlyofthestochasticv(t),butalsooftheobservableVIX.ThisimplementationfollowsA¨?t-SahaliaandKimmel(2007),whouseitinasamplewhichcontainsasingleshort-maturityat-the-moneyoptionateachtimet.
Wenextdiscussthedetailsofthisestimationapproachwhenusingreturnsandwhenusingoptions.OuruseoftheVIXasaproxyforthestochasticvariancehasimplicationsforbothestimationexercises.
3.3Return-BasedEstimation
ThemainpurposeoftheassumptionthatthestochasticvarianceisanaffinefunctionofVIXistoalleviatethecomputationalburdenwhenestimatingthemodelusingoptiondata.However,thisassumptionalsohasimplicationsforthereturn-basedestimation.SinceweobservethetotalreturnofthestockindexandVIXateachtimet,wecanformulatethejointlikelihoodfunctionofthereturnandVIX2toestimatethephysicalparameters.Inmostexistingestimations,thevarianceis
10
insteadfilteredfromtheunderlyingreturns,andtheVIXisnotusedinestimation.
Tocharacterizethelikelihoodfunction,wefirstapplyIto’slemmaandtheEulerdiscretizationtoequation(2),whichresultsin:
(6)
logR(t+?)=[r+μ(v(t))?v(t)]?+?R(t+?),
v(t+?)?v(t)=κ(θ?v(t))?+?v(t+?),
whereR(t+?)=S(t+?)/S(t)representsthegrossreturnand?=1/252.12Theerrors?(t+?)=(?R(t+?),?v(t+?))\followajointnormaldistribution,andtheirmeanandvariance-covariance
matrixaregivenby
0=Σ(t)=?.
Thejointlog-likelihoodfunctionisgivenby:
logLR
=
=
=
T?1
之logf(logR(t+?),VIX2(t+?)|VIX2(t))
t=1
之logf(logR(t+?),v(t+?)|v(t))×J(t+?)
T?1
t=1
之?log(2π)?log|Σ(t)|??(t+?)Σ\?1(t)?(t+?)+logη1,
T?1
t=1
wheref(logR(t+?),v(t+?)|v(t))istheconditionaldensityofthediscretizedlogR(t+?)andv(t+?),J(t+?)istheJacobianbetweenVIX2(t+?)andv(t+?),whichisgivenbyη1fromequation(5),andtrepresentstimemeasuredindays.LetΘ={μ,κ,θ,σ,ρ,η0,η1}bethesetofphysicalparameters.ToestimateΘ,wesolvethefollowingoptimizationproblem:
maxlogLR.
Θ
(7)
12NotethatlogR(t+?)isthedailylogreturnbetweentandt+?whilev(t)istheannualizedvarianceattimet.
11
3.4Option-BasedEstimation
Therisk-neutralparametersforthedynamicinequation(1)canbeestimatedinvariousways,buteachimplementationrequiresanoptionvaluationtechnique.WefollowthefastFourierimplemen-tationofCarrandMadan(1999).ThepriceofacalloptionwithitsstrikepriceKandmaturityτisexpressedbyaquasiclosedformuptoanumericalintegration,anditisgivenby
C(S(t),v(t),t)=\0∞Re[e?iukψ(u)]du,(8)
wherekisthenaturallogofK.Thefunctionψ(u)istheFouriertransformofamodifiedcallprice,whichisthecallpricemultipliedbyeαkforα>0.Wefoundthatα=4workswell.Thefunctionψ(u)iscalculatedasfollows:
(α+iu)(α+1+iu),
ψ(u)=e?rτf(u?i(α+1)|S(t),v(t))
whereiistheimaginaryunit,andf(?|S(t),v(t))=E[ei?logS(t+τ)]istherisk-neutralcon-ditionalcharacteristicfunctionoflogS(t+τ).Theclosed-formexpressionoff(?|S(t),v(t))followsHeston(1993).13Thepriceofaputoptionwiththesamestrikepriceandmaturitycanbeobtainedthroughput-callparity.Notethattheoptionpricingformulainequation(8)doesnotaccountfordividends.Wefollowtheexistingliteratureanduseafuture-dividend-adjustedindexprice.Specifically,weuseS(t)e?qτ,whereqisthedividendyieldattimet.
13WhenlogS(t)andv(t)arecharacterizedby
dlogS(t)=[r+uv(t)]dt+^v(t)dz1(t),
dv(t)=(a?bv(t))dt+σ^v(t)dz2(t),
thecharacteristicfunctionsolutionisgivenby
f(?|S(t),v(t))=eC+Dv(t)+i?logS(t),
where
C=r?iτ+{(b?ρσ?i+d)τ?2log[]},D=[],
g=,andd=^(ρσ?i?b)2?σ2(2u?i??2).
(9)
(10)
12
Weusevega-weightedoptionpricingerrors.LetOandOd
溫馨提示
- 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
- 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯系上傳者。文件的所有權益歸上傳用戶所有。
- 3. 本站RAR壓縮包中若帶圖紙,網頁內容里面會有圖紙預覽,若沒有圖紙預覽就沒有圖紙。
- 4. 未經權益所有人同意不得將文件中的內容挪作商業或盈利用途。
- 5. 人人文庫網僅提供信息存儲空間,僅對用戶上傳內容的表現方式做保護處理,對用戶上傳分享的文檔內容本身不做任何修改或編輯,并不能對任何下載內容負責。
- 6. 下載文件中如有侵權或不適當內容,請與我們聯系,我們立即糾正。
- 7. 本站不保證下載資源的準確性、安全性和完整性, 同時也不承擔用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。
最新文檔
- 家具廠質量管理制度
- 應急處置室管理制度
- 強電室安全管理制度
- 律師兩結合管理制度
- 微生物培訓管理制度
- 心電圖質量管理制度
- 急診科被褥管理制度
- 總承包投資管理制度
- 患標本安全管理制度
- 成品倉收貨管理制度
- 基本藥物政策培訓
- 安徽省宣城市皖東南初中六校2024-2025學年上學期七年級期中考試數學試卷
- 古鎮旅游活動策劃方案
- 清遠市突發事件總體應急預案
- 通風與防排煙系統的施工方案
- 滬教版英語小學四年級上學期試卷與參考答案(2024-2025學年)
- 人工智能訓練師理論知識考核要素細目表二級
- 2024年人教版一年級數學(下冊)期末試卷及答案(各版本)
- 《卒中患者吞咽障礙護理規范》
- DL∕T 698.45-2017 電能信息采集與管理系統 第4-5部分:通信協議-面向對象的數據交換協議
- 2024版房屋租賃合同范本房屋租賃合同
評論
0/150
提交評論