JP摩根全球量化策略大数据与AI策略机器学习在股票衍生品中的应用201962439页

(1-212)[e-mail],PhDAC(1-212)[e-mail],PhD(1-212)[e-mail]ge36foranalystcertificationandimportantdisclosures.J.P.Morgandoesandseekstodobusinesswithcompaniescoveredinitsresearchreports.Asaresult,investorsshouldbeawarethatthefirmmayhaveaconflictofinterestthatcouldaffecttheobjectivityofthisreport.Investorsshouldconsiderthisreportasonlyasinglefactorinmakingtheirinvestmentdecision.Inthisreport,weillustratetheapplicationsofmachinelearningtechniquestoequityderivativestradingstrategies.Specifically,wefocusonthetopicsbelow:HowtoSystematicallyDesignOptimalHedgingPortfolios:Thequestionofhowtooptimallyincorporateoptionsintoaninvestmentportfoliodoesnotfiteasilyintotheexistingportfoliooptimizationframeworkduetothenon-normalityofoptionreturns.WeapplyandextendthemethodologyproposedbyFaiasandSanta-Clara(2017),anddemonstratethatwithproperinvestabilityconstraints,themethodologyisshowntoproduceintuitiveandeffectiveoptionhedgingportfolios.DynamicVIXTermStructureAllocationStrategy:Wedevelopanactiveportfoliostrategywhichusesavariationofmean-varianceoptimizationtodynamicallyallocatetotheVIXtermstructure.Wefindourstrategyoutperformsthemean-varianceoptimizationportfolioprocessandtheshortconstant1MVIXFuturesbenchmarkonanaftertransactioncostbasis.ANonlinearFactorModelforDispersionTrading:UsingS&P500dispersion(vs.top50)asthesubjectofourstudy,weapplyoursupportvectormachine(SVM)approachtoforecastdispersionP&L.AbespokeS&P500dispersionbasketisconstructedusingtheproposedmachinelearningframework.Pleaserefertoourpreviousreportsforadditionalresearchonsimilartopics:ApplicationsofMachineLearninginEquityDerivatives–May2018ApplicationsofMachineLearninginEquityDerivatives–Nov20182GlobalQuantitative&DerivativesStrategy24June2019PengCheng,CFA(1-212)[e-mail]ystematicallyDesignOptimalHedgingPortfolios.3DynamicVIXTermStructureAllocationStrategy................13ANonlinearFactorModelforDispersionTrading...............233GlobalQuantitative&DerivativesStrategy24June2019PengCheng,CFA(1-212)[e-mail]signOptimalHedgingPortfoliosThequestionofhowtooptimallyincorporateoptionsintoaninvestmentportfoliodoesnotfiteasilyintotheexistingportfoliooptimizationframeworkduetothenon-normalityofoptionreturns.Instead,weapplyandextendthemethodologyproposedbyFaiasandSanta-Clara(2017),onthreecasestudies:HedgingAT1portfolioswithSX7EoptionsHedgingLeveragedLoanportfolioswithHYGoptionsOptimalselectionofS&P500optionhedgingstrategiesWithproperinvestabilityconstraints,themethodologyisshowntoproduceintuitiveandeffectiveoptionhedgingportfolios.Figure1:BacktestperformancewithandwithoutoptimalSPXoptionoverlayIndexedperformance100200300200520102015LongSPXwithOptionoverlayLongSPXSource:J.P.Morgan4GlobalQuantitative&DerivativesStrategy24June2019PengCheng,CFA(1-212)[e-mail]signoptimalhedgingportfoliosAlthoughoptionpricingtheoryandmodernportfoliotheoryarebothhighlydeveloped,thequestionofhowtooptimallyallocatetooptionsinaninvestor’sportfolioisfarlessunderstood.Thereasonisperhapsthatportfoliotheoryfocusesonthemeanandvarianceoftheassetreturns,whichcannoteasilyaccommodatetheasymmetricnatureofoptionreturns.Inthisreport,weexploresystematicwaysofcreatingoptimaloptionhedgingstrategiesonlongonlyportfolios.Todoso,weemployandextendthemethodologyproposedbyFaiasandSanta-Clara(2017)1whichoptimizesamyopicutilityfunctiononaportfolioofoptions,heldtomaturity.Thereareseveraladvantagesoftheapproach:firstly,itaccountsforhighermomentsoftheportfolioreturnsintheoptimizationprocess;secondly,itisindependentofoptionpricingmodel;moreover,fromapracticalpointofview,itrequiresrelativelylittlehistoricaldataforestimation;andlastbutnotleast,themethodologydoesnotrequiretheoptionstohavethesameunderlyingassetastheportfolio,andthereforecanbeappliedtoproxyhedgeportfoliosonwhichthereisnolistedoption.Themaincaveatofthemethodologyisthatthereturnsareevaluatedonlyattheendoftheinvestmentperiod,andthereforetheoptionsarerequiredtobeheldtomaturity.Inthenextsection,wefirstdescribethemethodology,thengoontoshowthreedifferentcasestudies,eachwithprogressivelymorecomplexoptionstructures.Lastly,werunahistoricalbacktestofasamplestrategytodemonstratetheeffectiveness,andsuggestpotentialfurtherextensions.MethodologyAselaboratedbyFaiasandSanta-Clara(2017),theoptimaloptionportfoliostrategy(OOPS)methodologyismadeupofthefollowingsteps:1)Givenaninvestmenthorizon(e.g.3months),computethehistoricalrollingreturnsoftheunderlyingasset.Scalethetimeperiodreturnsbytheratioofacurrentvolatilitymeasureoverthepoint-in-timevolatilitymeasure(e.g.current3Mimpliedvolatility/historical3Mimpliedvolatility).Doingsono