DB QIS Research：Long-Only Factor Investing

DeutscheBankResearch DBQISResearch Long-OnlyFactorInvesting #PositiveImpact Distributedon:12/03/202417:09:06GMT March2024 ClaytonGillespieGianpaoloTomasi ResearchTeam VivekAnandClaytonGillespieCaioNatividadeGianpaoloTomasi IMPORTANTRESEARCHDISCLOSURESANDANALYSTCERTIFICATIONSLOCATEDINAPPENDIX1.NotetoU.S.investors:US regulatorshavenotapprovedmostforeignlistedstockindexfuturesandoptionsforUSinvestors.Eligibleinvestorsmaybeabletogetexposurethroughover-the-counterproducts.DeutscheBankdoesandseekstodobusinesswithcompaniescoveredinitsresearchreports.Thus,investorsshouldbeawarethatthefirmmayhaveaconflictofinterestthatcouldaffecttheobjectivityofthisreport.Investorsshouldconsiderthisreportasonlyasinglefactorinmakingtheirinvestmen7tdTec2issioen.3MrC0IO(P)t0641k/1w0/2o02P3.a Long-OnlyMultifactorInvesting AimsofaLong-onlyEquityInvestor oMaximizetheoutperformanceoverthebenchmark Weseektoeffectivelycapturefactorpremiasystematically oRealizesimilar(orless)volatilitythanthebenchmark WeuseanMVOapproachpenalizingspecificriskonly oGeneratedrawdownslowerthanthebenchmark Weaddaconstraintontheportfolio’shistoricdrawdowns Constraint Value PortfolioSize Greaterthanorequalto200stocks TrackingError Lessthan4% IndividualStockWeights Between0.2%and2% RegionandSectorconstraints [Region,Sector]weight±5%vsbenchmark FundingConstraints Fullyinvested,prohibitingcash&leverage TakenfromAghassietal(2023) Long-onlydemandsadifferentapproach •WeincludeValue,QualityandMomentuminlinewithacademicresearch. •WefindourPCA-basedReversionconstructionmakesitprofitableinnetspaceanddiversifyingtootherfactors1. •WeexcludeLowBeta,becauseinanunleveredframeworktheportfolio willunderperformduetoabetabelow1 QualityandValue:OwnerSeries&Financials •OurValueandQualityscoresutilizeour“OwnerSeries”framework1,whichmakesaccountingadjustmentstofamiliarratiosinordertobetterrewardcashgenerationandprofitablegrowth. OwnerQuality OwnerEarningstoOperatingAssetsOwnerEarningstoSales OwnerBooktoOperatingAssets DividendtoOperatingAssets OwnerValueOwnerEarningstoPriceOwnerBooktoPriceDividendYield •WeaugmentthesescoreswithourresearchintofactorsinFinancials/RealEstate2,whichproxiescashflowusingdividendsandbuybacksandrewardslowcreditriskinsteadofprofitability. DeutscheBank Research 1SeeLeng2019andLeng2020Source:DeutscheBank 2SeeGillespie2023 DBQISResearch Clayton.gillespie@db.com4 AlphaAggregationRiskparity Σ𝐹𝑎𝑐𝑡𝑜𝑟�=𝐹′.Σ𝑆𝑡𝑜𝑐𝑘𝑠.� 𝐶𝑜𝑚𝑝𝑜𝑠𝑖𝑡�𝐴𝑙𝑝ℎ�𝑆𝑐𝑜𝑟𝑒𝑗,�=෍𝑤𝑖,𝑡𝐹𝑎𝑐𝑡𝑜𝑟𝑗,� � 1 2 •Weprefertousean‘Integrated’methodologyatthescorelevelandtocombinescoresusingarisk-parityapproach1. •Thisgivesmoreweighttolessvolatilefactors,butalsoconsiderscorrelationssuchthateachfactor’scontributiontototalriskisequal. DeutscheBank 1SeeRoncalli(2012)Source:DeutscheBank DBQISResearch MinimizingSpecificRisk 3 Systematic Risk SpecificRisk TotalRisk Σ𝑠𝑝𝑒𝑐𝑖𝑓𝑖�=Σ𝑇𝑜𝑡𝑎�−Σ𝑆𝑦𝑠𝑡𝑒𝑚𝑎𝑡𝑖� 𝑤′.�−�x𝑤′.Σ𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐.� 𝑎𝑟𝑔𝑚𝑎𝑥� Increasingfunctionoffactorexposure 4 Independent offactorexposure •InourMVOapproachwepenalizeonlythestock’sspecificrisk,ratherthantotalrisk. •Thispreventstheoptimizerfrompenalizingfactorrisk(allowinghigherfactorexposure). DeutscheBank Source:DeutscheBank,Factset DBQISResearch •Togetherthesechangesleadtoback-testedCAGRimprovementsof3.6%p.a.vsMSCIworld,netofcosts1,withthesamevolatility. •Thesamestrategyonlyachievesa1%p.a.CAGRimprovementin atotalriskframework. DeutscheBank 1LiquidityFilter:>$10milADV,TransactionCosts:2bps; Source:DeutscheBank,Factset DBQISResearch 𝑎𝑟𝑔𝑚𝑎𝑥�𝑤′.�−�x𝑤′.Σ𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐.� •Thefiguresshowtheimpactoftheriskaversionparameter,λ. •Fortotalrisk,allvaluesofλresultinastrategyvolatilityfarbelowthebenchmark.However,minimizingspecificriskallowstheinvestortoachievethesamelevelofvolatility. •Fromhereonwesetλ=1toequatethebenchmarkvolatility. DeutscheBank Source:DeutscheBank,Factset DBQISResearch •UsingAxioma’sriskmodel,weshow(left)thatthespecificriskapproachresultsinhigherValue,Size,MomentumandMarketBetaexposures. •Furthermore,intheworstcasescenario,thetotalriskapproachgeneratesveryconcentratedportfolioswhereasspecificriskresultsinamuchsmootherprofile. DeutscheBank Source:DeutscheBank,Factset,Axioma DBQISResearch •Weshowtherobustnessofourresultsabovebyrepeatingtheexercisewithanindependentsetofmultifactorscores. •Hereweusethescoresgeneratedbyalong-onlyconstruction ofourmachinelearningfactorrotationmodel(NLASR)1. •Weagainseespecificriskresultinginthebestperformance,generatingsimilar(ifslightlybetter)resultstoourmultifactor. DeutscheBank 1SeeTomasi(2020)Source:DeutscheBank,Factset DBQISResearch •Drawdownsaretypicallynotincludedintheoptimizationphase. •Duetodiversification,thedrawdown(DDp)ofalong-onlyportfolioislessthanorequaltotheweightedaverageoftheDDoftheunderlyingstocks. •DDpcanbemodelledasasetofconstraintsandthenaddedeitherinthecostfunctionorasconstraints(weusethesecondapproach). •Using5yofhistoricmonthlyreturns,wedefinethemaximumfraction(φ)ofthe benchmarkDDthattheportfoliocanexhibit. Av