EssaysinHousingMarketsandtheRealEconomyCopyright2020byChristopherMLako1AbstractEssaysinHousingMarketsandtheRealEconomybyChristopherMLakoDoctorofPhilosophyinBusinessAdministrationUniversityofCalifornia ID: 874578
Download Pdf The PPT/PDF document "EssaysinHousingMarketsandtheRealEconomyb..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1 EssaysinHousingMarketsandtheRealEconomyb
EssaysinHousingMarketsandtheRealEconomybyChristopherMLakoAdissertationsubmittedinpartialsatisfactionoftherequirementsforthedegreeofDoctorofPhilosophyinBusinessAdministrationintheGraduateDivisionoftheUniversityofCalifornia,BerkeleyCommitteeincharge:ProfessorNancyWallace,ChairProfessorMartinLettauProfessorEmiNakamuraProfessorDavidSraerSpring2020 EssaysinHousingMarketsandtheRealEconomyCopyright2020byChristopherMLako 1AbstractEssaysinHousingMarketsandtheRealEconomybyChristopherMLakoDoctorofPhilosophyinBusinessAdministrationUniversityofCalifornia,BerkeleyProfessorNancyWallace,ChairThisdissertationconsistsofthreechaptersontheeectsthathousingmarketshaveontherealeconomy.IntheUnitedStates,personalrealestatehadanaggregatemarketvalueof$30trillionin2019.1Additionally,dataonrealestatetransactionsandtheunderlyingloansandpropertiesareextremelyhighquality,therebymakingrealestateanidealempiricallaboratory.Therstchaptershowsthatmortgagecreditaccessisvitalforsmallbusinessnancingandalleviatingcreditconstraints.Todothis,Irelyonmicro-datafromamergeofthepersonalhomeequityextractionactivityofbusinessownerstothecondentialIRStaxrecordsoftheirbusinesses.Withdirectmeasurement,Indthatoneoutoffoursmallbusinessescreatedduringthemid-2000swerefundedbypersonalhomeequity,doubletheratepreviouslythoughtbasedonevidencefromsurveydata.Entrepreneursusetheirpersonalhomeequitytoalleviatecreditconstraints,whichthischapterndshasalong-runeectonboththesurvivalandemploymentlevelsofsmallbusinesses.Notonlyarenewbusinessescreditconstrained,butexistingsmallbusinessesalsofacecreditconstraintsthathaveapersistenteect.FollowingtheGreatRecession,restrictionstomortgagecreditaccesshavecausedone-thirdofthedeclineinrmentryratesinthepost-crisisperiod.Inthesecondchapter,Ishowhowvariationsinpersonalhousingwealthfeedsintoprofes-sionalbehaviorthroughstudyingmutualfundmanagers.Theliteratureonovercondenceinnancialmarketsh
2 asprimarilyfocusedonretailinvestors.Usin
asprimarilyfocusedonretailinvestors.Usingnoveldatathatiden-tiesthepersonalrealestateholdingsoffundmanagers,thischapterstudiesthedegreetowhichovercondenceaectsthereturnsandinvestmentbehaviorofinstitutionalinvestors.Positiveshockstothepersonalrealestateofmutualfundmanagersshouldnotaecttheirprofessionalbehavior.However,thischapterndsthataonestandarddeviationpositivehomepriceshockleadstoadeclinein4-factoralphaof37bpsperyear.Thisisduetofundmanagersbecomingovercondentintheirunderperformingtradingpositions,makingworsesellingchoices,andtradingmorefrequently.Fundmanagerswhoaremorelikelytobeaectedbyovercondence,suchaslesseducatedandlessexperiencedfundmanagers, 1Source:FederalReserveFlowofFunds 2showamuchstrongerresponse.Thischapterprovidesevidencethatovercondenceistimevaryingandshowshowinstitutionalinvestorsrespondtobehavioralshocksthatshouldbeorthogonaltotheirprofessionalduties.Inthethirdchapter,I,alongwithmyco-authors,AyaBellicha,RichardStanton,andNancyWallace,studyhowthevastoutstandingstockofmortgagedebtaectsU.S.Treasurybondyields.WeproposeanempiricaldurationmeasureforthestockofU.S.AgencyMBSthatappearstobelesspronetomodelriskthanmeasuressuchastheBarclaysEectiveDurationmeasure.Wendthatthismeasuredoesnotappeartohaveastrongeectonthe12-monthexcessreturnsoften-yearTreasuriesaswouldbeexpectedifshockstoMBSdurationleadtocommensurateshockstothequantityofinterestrateriskbornebyprofessionalbondinvestors(Hanson,2014;Malkhozov,Mueller,Vedolin,andVenter,2016).Giventhisnegativereducedformresult,wethenexplorethemortgageandtreasuryhedgingactivitiesoftheprimaryMBSinvestorssuchascommercialbanks,insurancecompanies,theagencies,theFederalReserveBank,mutualfunds,andforeigninvestors.WendthattheonlyinvestorsthatmayfollowthemodelsofHanson(2014)andMalkhozovetal.(2016)arelifeinsurancerms.Wealsondarelationwithbankshoweverwecannotruleoutthatthisismerelycorrelation.Li
3 feinsurancermmarketsharehasdeclined
feinsurancermmarketsharehasdeclinedovertheperiod,droppingbelow10%since1996andreaching4%in2016.Oftheinvestorswearenotabletostudy,hedgefundsandpensions/retirementfundsarethetwoinvestorgroupsthatmaytradealongtheHanson(2014)andMalkhozovetal.(2016)models.However,althoughthesetwoinvestorgroupsheldalmost25%oftheAgencyMBSmarket(includinghouseholdsandnonprotorganizations)inthelate1990s,postcrisistheirsharehasfallenbelow10%. iTomyfather iiContentsContentsiiListofFiguresivListofTablesvi1TheLong-RunEectsofMortgageCreditAccessonEntrepreneurship11.1Introduction....................................11.2DataandtheFundingofSmallBusinessesByHousingEquity........41.3IntensiveMarginImpactofCreditConstraints.................71.4DeclineinRenancingActivityandBusinessFormation............211.5Conclusion.....................................261.6FiguresandTables................................282PersonalWealthShocksandInvestmentManagerOvercondence542.1Introduction....................................542.2HomePriceVariation...............................562.3Data........................................582.4EmpiricalResultsonPerformance........................602.5TradingBehavior.................................642.6AttentiontoHomePriceGrowth........................682.7Conclusion.....................................692.8FiguresandTables................................713DurationMeasurementandHedgingChannelsforGSEInsuredMort-gageBackedSecurities893.1Introduction....................................893.2Duration......................................913.3EvidenceforMBSInvestorHedging.......................973.4Conclusions....................................1033.5FiguresandTables................................104Bibliography124 iiiAAppendixtoChapter1132A.1Merges.......................................132A.2HomeEquityExtractionDatafromATTOM..................135A.3AppendixFiguresforChapter1.........................137BAp
4 pendixtoChapter2139B.1AdditionalRobustne
pendixtoChapter2139B.1AdditionalRobustnessTests...........................139B.2AppendixTablesforChapter2.........................141 ivListofFigures1.1HomeEquityExtractionandFirmEntryRates..................281.2ShareofSmallBusinessesFoundedwithHomeEquityFundingByRegion...291.3IdenticationSetup..................................301.4IdenticationIllustration...............................311.5DierenceinDistancefromHometoFirmBetweenTreatedandControl....321.6DierenceinPropensitytoRenanceWithaTraditionalBankBetweenTreatedandControl......................................331.7HomePriceGrowthExample............................341.8LoanTerminationRates...............................352.1ComparisonofAverageSalaryandHomeValuesforInvestmentManagers...712.2BostonMetroAreaZipCodeLevelHomePriceGrowth..............722.3Comparisonof3-yearHomePriceGrowthforNeighboringZipCodes......722.4NumberandShareofFundManagersMerged...................732.5HeterogeneousEectofHomePriceGrowthonFundAlpha:1..........742.6HeterogeneousEectofHomePriceGrowthonFundAlpha:2..........752.7HeterogeneousEectofHomePriceGrowthonFundAlpha:3..........762.8Distributionof1-yearChangesin3-yearHomePriceGrowth...........772.9ShareofFundManagersExtractingHomeEquityandAmountofHomeEquityExtracted.......................................783.1ComparisonofBarclaysEectiveDurationandEmpiricalDuration.......1043.2ComparisonofDurationMeasures(EmpiricalandBarclayseective)AgainstTerminationMeasure.................................1053.3DistributionofAgencyandGSEBackedSecuritiesHoldingsbyInvestorGroup1063.4DistributionofUSTreasuryHoldingsbyInvestorGroup.............1073.5FannieMaeandFreddieMacDerivativesPortfolios................1083.6ComparisonofGSERetainedPortfolioReportingtoCongressvsFederalReserveFlowofFunds.....................................1093.7BankMBSPass-ThroughHoldings.........................1093.8BankRatioofTreasuri
5 estoAssets.........................1103.
estoAssets.........................1103.9ShareofUSTreasuryDebtHeldbyChinaandJapan...............111 vA.1ShareofSmallBusinessesFoundedwithHomeEquityFundingByFirmSize..137A.2ShareofEntrantSmallBusinessWithFewerThan10Employees........138A.3ShareofMortgageOriginationbyIndependentMortgageBanks.........138 viListofTables1.1SummaryStatisticsforLBD.............................361.2SummaryStatisticsofEntrantBusinessFinancingSources............371.3SummaryStatisticsBetweenTreatedandControlMembersforEntrantBusinessMatching.......................................381.4SummaryStatisticsofEntrantBusinessMatchedSample.............391.5EectofCreditAccessonEntrantBusinessSurvival...............401.6EectofCreditAccessonEntrantBusinessEmployment.............411.7EectofCreditAccessonEntrantBusinessEmploymentGrowthRates.....421.8EectofCreditAccessonEntrantBusinessPayroll................431.9SummaryStatisticsComparingEntrantBusinessesbyInitialFundingSource..441.10ComparingDierencesinBusinessesandEntrepreneursByInitialFundingSource451.11ComparingDierencesinPersonalForeclosureOutcomeByInitialFundingSource461.12SummaryStatisticsofContinuingBusinessMatchedSample...........471.13SummaryStatisticsBetweenTreatedandControlMembersforContinuingBusi-nessMatching.....................................481.14EectofCreditAccessonContinuingBusinessSurvival..............491.15EectofCreditAccessonContinuingBusinessEmployment...........501.16EectofCreditAccessonContinuingBusinessPayroll..............511.17ExtensiveMarginEectsofRenancingActivityonBusinessFormation....521.18RobustnessTestsforExtensiveMarginInstrument................532.1SummaryStatisticsComparingtheMergedandNotMergedPopulations....792.2EectofHomePriceGrowthonFundAlpha....................802.3EectofHomePriceGrowthonFundFlow....................812.4EectofHomePriceGrowthonFundAlphaforIndexandPassiveFunds...822.5Eectof
6 HomePriceGrowthonFundAlphabySecondHomeOw
HomePriceGrowthonFundAlphabySecondHomeOwnershipStatus832.6EectofHomePriceGrowthonReturnsforBuyandSellTrades:1.......842.7EectofHomePriceGrowthonReturnsforBuyandSellTrades:2.......852.8EectofChangesinHomePriceGrowthonActiveTradingPerformance....862.9EectofHomePriceGrowthonRiskTaking....................872.10EectofHomeEquityExtractiononFundPerformance.............883.1EectofMBSDurationonTreasuryExcessBondReturns............112 vii3.2EectofMortgageTerminationsonTreasuryExcessBondReturns.......1133.3EectofGSEHedgingPortfoliosonTreasuryExcessBondReturns.......1143.4EectofChangeinGSEHedgingPortfoliosonChangeinTreasuryExcessBondReturns........................................1153.5EectofMBSDurationonBankTreasuryBondHoldingsforBanksWithMBSHoldings........................................1163.6EectofMBSDurationonBankTreasuryBondHoldingsforBanksWithoutMBSHoldings.....................................1173.7EectofChangeinBarclaysEectiveMBSDurationonChangeinForeignTrea-suryBondHoldings..................................1183.8EectofChangeinEmpiricalMBSDurationonChangeinForeignTreasuryBondHoldings....................................1193.9EectofMBSDurationonMutualFundTreasuryBondHoldings........1203.10EectofChangeinMBSDurationonChangeinMutualFundTreasuryBondHoldings........................................1213.11EectofBarclaysEectiveMBSDurationonLifeInsuranceCompanyTreasuryBondPurchases....................................1223.12EectofEmpiricalMBSDurationonLifeInsuranceCompanyTreasuryBondPurchases.......................................123B.1EectWithinCommutingZonebyTimeofHomePriceGrowthonFundAlpha142B.2EectofHomePriceGrowthonVariousFundReturnMeasures.........143B.3EectofHomePriceGrowthonFundAlpha,UsingEx-AnteInformation....144B.4Eectof2and4-YearHomePriceGrowthonFundAlpha............145B.5EectofHomePriceGrowthonFu
7 ndAlphaforFundManagersWhoSurvivedAtLeast
ndAlphaforFundManagersWhoSurvivedAtLeast12(36)Months...............................146 viiiAcknowledgmentsIamdeeplygratefultothemanypeoplewhohaveassistedandguidedmeinmydoctoralstudies.First,IwouldliketothankNancyWallace,mydissertationchair,forherguidance,support,andimmensewealthofknowledge.Icannotimagineamoresupportive,helpful,andencouragingadvisor.Iamalsogratefultomyothercommitteemembers,MartinLettau,EmiNakamura,andDavidSraer,whoseinputandguidancehelpedshapedmyresearch.IalsooweaspecialthankyoutoAmirKermaniandChristopherPalmerwhoprovidedearlymentoringandtrainingwhileIwastheirResearchAssistant.Icouldnothavecompletedmythirdchapterwithoutmyco-authors,AyaBellicha,RichardStanton,andNancyWallace,whospentlonghoursdiscussingourresearch,whichhelpedtrainmeasaneconomist.Inadditiontothefacultymentioned,IamgratefulforallofthefacultyinboththeHaasSchoolofBusinessandtheEconomicsdepartmentwhotaughtme,attendedmyseminars,andwerealwayswillingtomeetoneononetodiscussmyresearch.Ithankmyparentswhoraisedmeandalwayspushedmetofocusonmyeducation.IamespeciallygratefultomylatefatherwhoencouragedandsupportedmydesiretopursueaPhD.Ialsothankmyhusband,whomovedtoCaliforniaafterhispost-doc,forhelpingprovideasourceofsupportthroughmyPhD.Iapologizeforthelonghoursandforoftenbeingdistracted.MyfriendsoutsidethePhDprogramwereimmenselyhelpfulinprovidinganoutletoutsideacademiaanddistractionfromresearch.Priortomydoctoralstudies,IworkedinindustryatFreddieMac.Thisperiodhelpedmegainthetechnicalexpertiseandinstitutionalknowledgethatassistedmeinmydoctoralwork.IwouldliketothanktheeconomiststhatIworkedwith,ZhengLiu,JanLuytjes,FrankNothaft,andHanqingZhou,aswellasmydirectorCharlesMcKinney,whoencouragedmetopursuemyPhD.Iamalsoverythankfultomyfellowclassmates.IamluckytohavehadmycohortasastudygroupduringthecourseworkphaseofthePhDandforthefriendshipsthatweforged.Fromfeedbackinpre-seminar,totalkingcasuallyintheoceorovercoee,todinnerconversations,andeverythinge
8 lseinbetween,myfellowclassmateshavebeeni
lseinbetween,myfellowclassmateshavebeeninvaluable.Lastly,IamgratefulforhavingaccesstoboththePhDoceatHaasandtheFisherCenterforRealEstateandUrbanEconomics.LindaAlgazalli,ThomasChappelear,MelissaHacker,CharlesMontague,andLisaVillalbaprovidedexceptionalsupportnavigatingtheadministrativeissuesduringmyPhD.Additionally,ThomasChappelearprovidedexcep-tionaleditingservicesformythesis,aswellasbeingagreatresourceoverall.PauloIsslerhasbuiltanamazingresearchlabanddatainfrastructure,withoutwhichIwouldnothavebeenabletocompletemyresearch,andwasalwayswillingtodiscussmyresearch.Lastly,IwouldliketothankAngelaAndruswhoprovidedguidancewiththeapprovalprocessforCensusdataandassistanceworkingwiththedata. 1Chapter1TheLong-RunEectsofMortgageCreditAccessonEntrepreneurship1.1IntroductionFormosthouseholds,personalhousingequityisthelargestsourceofsavings(Campbell,2006).Incorporatenance,pledgingdurableassetssuchasrealestateisacommonwaytoalleviatecreditconstraints(KiyotakiandMoore,1997),makingthelinkbetweenmortgagecreditaccessandsmallbusinessnancinganaturalone.Thispapershowsthatthepersonalhomeequityofentrepreneurswasusedtofundoneoutoffoursmallbusinessescreatedduringthemid-2000s,highlightingthevitalroleofmortgagecreditasafundingsourceforsmallbusinesses.Thisrolewaspreviouslygreatlyunderestimatedintheliterature.SincetheGreatRecession,theshareofnew(entrant)smallbusinessesfundedwithpersonalhomeequityhasfallentooneoutoftwenty.Variationinthesupplyofmortgagecreditaectstheabilityofsmallbusinessestofundprojects.Ontheintensivemargin,thispaperndsthatbothentrantsmallbusinesses(thoseintheirrstyearofoperation)andcontinuingsmallbusinesses(thosethathavesurvivedatleastthreeorfouryears)faceseverecreditconstraintsthatleadtoapermanentdisadvantage.1Furthermore,tightermortgagecreditstandardsexplainone-thirdofthedeclineinrmentryratessince2006.Explicitlyshowingthelinkbetweenmortgagecreditandentrepreneurshipisdicultduetoal
9 ackofindividualleveldatathatshowsbothent
ackofindividualleveldatathatshowsbothentrepreneursextractingpersonalhousingequityandalongitudinalpanelofbusinessoutcomes.Thishasledpastresearchtoproxyforentrepreneursextractinghousingequitywithchangesinhousingwealth,andtolargelyfocusontheextensivemarginofentryintoentrepreneurship(CorradinandPopov,2015;Kerr, 1Theintensivemargineectforentrantrmsshowshowinitialsize,long-runsize,andsurvivalareaectedbyexogenousrelaxationsincreditconstraintsintherstyearofoperation(conditionalonentry).Asopposedtotheextensivemargineect,whichwouldshowhowthechoicetostartabusinessisaectedbyexogenousrelaxationsincreditconstraints.Theintensivemargineectforcontinuingrmsshowshowsize,long-runsize,andsurvivalareaectedbyexogenousrelaxationsincreditconstraintsinyearthreeorfour(forrmsthathavesurvivedforatleastthreeoffouryears,respectively). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP2Kerr,andNanda,2015;Adelino,Schoar,andSeverino,2015).Consequently,thendingsofthesepapershavebeenmixedandasaresult,thedegreetowhichsmallbusinessesarecreditconstrainedremainsanopenquestionintheacademicliterature(HurstandLusardi,2004).Thispapercontributestothisliteraturebyconstructinganovelhousehold-rmlinkedpaneldatasetfrommultipleadministrativedatasets,includingtherestricteduseLongitudinalBusinessDatabase(LBD).Withthesedata,thispapermorepreciselyestimatestheresponseofsmallbusinessestorelaxationsintheircreditconstraints,estimateslong-runeects,andseparatelyshowstheeectforbothentrant(conditionalonentry)andcontinuingsmallbusinesses.Thispapershowsthatasmallexogenousincreaseinavailablecreditatentrycausessmallbusinessestostartlarger,growfaster,permanentlyremainlarger,andhavehigherhighersurvivalrates(measuredouttoveyears).Initialfundinghasanirreversibleeectonsmallbusinesses.Forevery$88,000increaseinasmallbusiness'screditaccessatentry,oneadditionaljobiscreated.Similarly,conti
10 nuingsmallbusinessesstronglyandimmediate
nuingsmallbusinessesstronglyandimmediatelyrespondtocreditshocks.Whensmallcontinuingbusinessesreceiveapositivecreditshock,theyimmediatelyexpandandpermanentlyremainlarger.Forcontinuingrms,a$153,000creditshockcreatesoneadditionaljob(implyingthattheyarelessconstrainedthanentrantrms).Schmalz,Sraer,andThesmar(2017)establishthatthereisapersistentimpactofcreditconstraintsatentryforbusinessesregisteredin1998inFrance.Thispapercomple-mentstheirsbyshowingthatthepersistenteectatentryholdsforthemorerecentperiodintheUnitedStates,whichhasaverydierentbankingmarket,andistrueforcontinuingrmsaswell.2Identicationisachallengeforstudyingtheseverityofcreditconstraintsgiventhehigh-dimensionalheterogeneityacrossboththeassociatedhouseholdsandtheirbusinesses.Toestablishacausalimpactofcreditshocksonthelong-runoutcomesofbusinesses,thispaperutilizesvariationintheavailabilityofhomeequityviaamatchedpairframework.Atahighlevel,twosimilarbusinessesthatarelocatedinthesamezipcodeandthatareownedbysimilarentrepreneurswholiveindierentzipcodesfromthebusinessesandfromeachotherarecompared.Thisisolatesexogenousvariationintheavailabilityofcreditstemmingfromdierentialzipcodelevelhomepricegrowthwithinanarrowgeographicalarea.AsimilarapproachhasbeenusedbyBernstein,McQuade,andTownsend(2018)tostudyinnovationandStoman,Pool,Yonker,andZhang(2018)tostudyassetpricing.Asathirdcontribution,thispaperndscausalevidencethatthetighteningofmort-gagecreditstandardspost-crisishasbeenanimportantchannelforthelackofrecoveryinbusinessformationrates.Smallbusinessesrelyonhousingwealthtoalleviatetheircreditconstraints.Consequently,whenmortgagesbecomehardertoobtain,businessentryisnega-tivelyimpacted.Despitehomepricesrecoveringbetween2009and2016,cash-outrenancingvolumeremainsdepressedduetoatighteningofmortgagecreditstandards.Fromitspeakin 2InFrance,homeownerswithamortgagecannotextracthousingequity,whileintheUnitedStatestheycan.In
11 addition,mortgagesinFrancegenerallycanno
addition,mortgagesinFrancegenerallycannotbeprepaidandhousingwealthisnotviewedassomethingthatwouldbeborrowedagainst. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP32005,theamountofhomeequityextractedbyhouseholdshasdeclinedmorethan60%,withvolumecurrentlyat2001levels(Figure1.1).Toidentifyacausallink,thispaperconstructsaBartikshiftshareinstrumentthatexploitsaninstitutionaldetailaboutindependentmort-gagebanks(IMBs).IMBs,suchasQuickenLoans(oneofthelargestmortgagelenders),aggressivelytargettheirexistingcustomerstopersuadethemtorenance.Thisprovidesasourceofexogenousvariationinlocalrenancingactivity.RobustnesstestsruleoutthatthepredictedIMBsharedirectlycorrelateswithmeasuresthatdrivebusinessformation,otherthanrenancing.Tofurthershowthattheeectisdrivenbyalackofmortgagecreditaccessandnotchangesinlocaldemand,theeectisshowntoholdforindustriesthatarenotreliantonlocaldemand.Pastresearchlargelyfocusedontheimpactsofhousingwealthonbusinessentryduringthehousingbubbleofthemid-2000s.However,theimpactofalackofaccesstohousingwealthonbusinessentryinthedecadesincethe2008nancialcrisishasnotbeenstudied.Ifmortgagedebtwasasubstituteforotherformsofcreditthenreductionsinaccesstoliquidhousingwealthshouldnothaveanoticeableimpactonbusinessentry.RecentworkbyDavisandHaltiwanger(2019)looksattheimpactthathomepriceshavehadontheshareofrmsthatarelessthanveyearsoldbetween1999and2014.Thispaperdiersinthatitshowstheimpactsthatatighteninginmortgagecreditsupplyhashadspecicallyonrmentryintheperiodsince2009|aperiodduringwhichhomepricesrosedramaticallyandmortgagerenancingvolumestagnated.Thebenetsofhomeequityextractionfoundinthispaperareinincontrasttotheresultsfortheaveragehouseholdthatextractedhomeequityduringthemid-2000s.Gener-ally,householdsthatextractedhomeequityhadlowercreditscoresandusedtheextractedhomeequitytoincreasetheirconsumption(BhuttaandKeys,2016).However,homeeq
12 uityextractionwasalsousedbyalargesubseto
uityextractionwasalsousedbyalargesubsetofthepopulationwithprimecreditscorestostartandgrowbusinesses,whichledtojobandGDPgrowth.Thiswouldbeirresponsibleifentrepreneurswhousehomeequitystartweakerbusinesses,butthisisnotthecase.Thispaperndsthathomeequityfundedbusinessesaresimilartoandfacesimilarsurvivalratesastheirnon-homeequityfundedcounterparts.Alleviatingcreditconstraintsviahousingcollateralallowsentrepreneurstostartproductivebusinessesatamoreoptimalsize,whichinrecentyearshasnotbeenpossible.RecentworkbyBahaj,Foulis,andPinter(2017)ndsthatincreasesinpersonalhomevaluesofrmdirectorsintheUnitedKingdomleadstoacontemporaneousincreaseincorporateinvestment(viaapersonalguaranteeforbusinessdebt).Incontrast,thispaperestimatesthelong-runresponsetocreditsupplyshocksbyageforsmallrmsintheUnitedStates(forsurvival,employment,andpayroll).Additionally,thispapershowsmorebroadlytheimpactsthatrestrictionsonmortgagecredithavehadonrmentryanddirectlyshowstheactualratesofhomeequityusebysmallbusinesses.Thispaperbuildsontheimpactsofcreditallocationinthecorporatenanceliterature.Earlyresearchshowedthatcreditmayberationedduetoinformationalasymmetries(StiglitzandWeiss,1981).Onewaytoovercomeasymmetricinformationisthroughlendingonsoftinformation,whichisacquiredviarelationshipsbetweenbusinessownersandbankers(Pe- CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP4tersenandRajan,1994;BergerandUdell,1995).However,acquiringabankingrelationshiptakestimeandasaconsequenceoftenisnotanoptionforentrantsmallbusinesses.Toovercomethis,entrepreneursgenerallyrelyonpersonalassetstofundtheirentrantbusi-nesses(RobbandRobinson,2014).Earlyresearchoncreditconstraintsstudiedwealthandfoundthatentryintoentrepreneurshipincreaseswithwealth(EvansandJovanovic,1989).Theliteratureprogressedtostudyingtransitionsintoentrepreneurshipafterreceivinganinheritance(Blanch owerandOswald,1998;HurstandLusardi,2004).Lately,thefocusofthislite
13 raturehasshiftedtostudyinghousingwealths
raturehasshiftedtostudyinghousingwealthshocks.Morebroadly,inrecentyearsthecorporatenanceandhouseholdnanceliteratureshavefocusedontheimpactsthatcreditsupplyshockshaveonbusinessesandhouseholds.3Thispapertsintotheselitera-turesbyshowingthemacroimplicationsthatdisruptionstomortgagecreditaccesshaveonentrepreneurshipandthelong-runintensivemargineectsofcreditconstraintsbyrmage.Lastly,thispaperbuildsontheliteratureofbusinessdynamismbyidentifyingoneofthechannelsbehindthedeclineinrmentryrates.4Thepaperproceedsasfollows.Section2describesthedataandprovidesnewstylizedfactsontheuseofhomeequitybysmallbusinessowners.Section3providestheempiricalmethodologyandresultsfortheimpactofcreditconstraintsontheintensivemargin.Section4showscausalevidencethatthelackofrecoveryinhomecash-outrenancingactivitysincetheGreatRecessionhasnegativelyimpactedbusinessformationrates.Lastly,Section5concludes.1.2DataandtheFundingofSmallBusinessesByHousingEquity1.2.1Micro-dataOverviewEmpiricalresearchonsmallbusinesseshasbeenhinderedbyalackoftime-seriesdataatthebusinesslevellinkingbusinessoutcomesandthehouseholdbalancesheetoftheentrepreneur.Thisistrueevenforadministrativedatasets.Inthestudyofpersonalhomeequityasafundingsourceforsmallbusinesses,andmorebroadly,theeectofcreditconstraintsonsmallbusinesses,therehavebeenawiderangeofndingsduetothislackofdata.Thispaperreliesonnovelmergestocreatea\big"datasetofsmallbusinessadministrativedataintheUnitedStateslinkedtothemortgageandrenancingactivityofthebusinessowner. 3SeeChodorow-Reich(2013),Greenstone,Mas,andNguyen(2014),Krishnan,Nandy,andPuri(2014),LauferandPaciorek(2018),BenmelechandRamcharan(2017),DeFusco,Johnson,andMondragon(2017),Goodman(2017),Bord,Ivashina,andTaliaferro(2018),GeteandReher(2018),Mondragon(2018),Nguyen(2019).4Pastworkhaslargelyidentiedtheissueofthelong-termdeclineinrmentryratesandtherelatedeconomicconsequences.SeeDecker,Haltiwanger,Ja
14 rmin,andMiranda(2014),HathawayandLitan(2
rmin,andMiranda(2014),HathawayandLitan(2014),Gourio,Messer,andSiemer(2016),Siemer(2016).Recentworkhasidentiedimportcompetitionandchangingdemographicsastwochannelscausingthedeclineinrmentryrates(PugsleyandS,ahin,2018;Karahan,Pugsley,andSahin,2019). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP5ThedatasetconstructedinthispaperisauniquemergebetweenadministrativedataonbusinessoutcomesfromtherestricteduseLongitudinalBusinessDatabase(LBD)andadministrativedataforrealestatetransactionsfromATTOM.TheLBDisacondentialdatasethousedwithinFederalStatisticalResearchDataCentersoftheUSCensusBureau.TheLBDprovidesannualestablishmentleveldataonsurvival,employment,andpayroll.5ThedataareavailableforallbusinesseswithatleastonepaidemployeeintheUnitedStatesandareconstructedfromIRStaxreturndata.AkeyfeatureoftheLBDistheLBDNUMvariable,whichallowsanestablishmenttobelongitudinallytrackedovertime.Inthispaper,smallbusinessesaredenedasbusinesseswithinitialemploymentoftenorfeweremployeesandthatstartedassingle-unitrms.DespitetherichnessoftheLBDdata,itdoesnotcontaininformationoncreditaccess.Forthis,datafromATTOMareutilized.ATTOMisacomprehensivepropertyandtrans-actionleveldatasetonresidential(andcommercial)realestatepurchasesandrenances,thesuccessortoDataquick.IntheUnitedStates,countyrecorderocestrackeveryrealestatetransaction,includingresidentialrenances.ATTOMconsolidatedtheserecordsandcreatedadatasetofthisinformation,fromwhichtheamountofhomeequityextractedfromapropertycanbeconstructed(describedintheOnlineAppendix).Atahighlevel,thispa-perlongitudinallylinksrealestatetransactionsthroughtimeatthepropertybyhomeownerlevel.Usingthepurchasetransactionandpastrenancingrecords(ifany),theoutstand-ingmortgagebalanceisestimatedatthetimeofrenanceandtheamountofhomeequityextractedisconstructedfromthisbasedontherenanceloanamount.TheLBDdoesnotcontainpersonallyidentiableinforma
15 tionanddoesnotidentifythebusinessowner.C
tionanddoesnotidentifythebusinessowner.Consequently,ATTOMisrstmergedtoNETS.TheNETSdataarebasedonDunandBradstreet(D&B)data,whichcontaintheuniverseofallbusinesses(includingnon-employerbusinesses)intheUnitedStates.TheadvantageofNETSisthatitliststhebusinessname,alongitudinalpanelofaddresses,andthenameofthebusinessowner(for60%ofbusinesses).6ATTOMislinkedtoNETSbymergingthenameofthehomeownerinATTOMtothenameofthebusinessownerinNETS(describedintheOnlineAppendix).WiththelinkofATTOMtoNETS,anentrantbusinessisclassiedasbeingfundedwithhomeequityifthebusinessownerextractspersonalhousingequityintheyearthatthebusinessiscreatedortheprioryear.Lastly,NETSismergedtotheLBD(describedintheOnlineAppendix)forsmallbusi-nessesfoundedbetween2001and2011.Thisprovidesadirectlinkoflongitudinalhomeequityextractionactivityofbusinessownerstolongitudinaloutcomesoftheirbusinesses.Fromwhichthispapershowsthepersistentimpactthatcreditconstraintshaveonbothentrantandcontinuingbusinesses.Fornotation,entrantrmsarermsintheirrstyearof 5JarminandMiranda(2002)provideadetailedoverviewoftheLBD(andthecompaniondataset,theSSEL).Revenuemeasuresarenotcurrentlyavailable,butshouldbeavailableoncetherevenueenhancedLBDiscompleted(Haltiwanger,Jarmin,Kulick,andMiranda,2016).6Thedatasetalsocontainsannualemploymentinformation.However,analysisonthereliabilityofthedatahasledtoaviewthatitshouldnotbereliedonforhighfrequencytimeseriesinformationonbusinesses(Neumark,Zhang,andWall,2007;Barnatchez,Crane,andDecker,2017). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP6operation.Whereascontinuingrmsarematurermsthathavesurvivedforatleastthreeorfouryears.Table1.1comparestheoverallLBDsmallbusinesspopulationtothenalpopulationutilizedinthispaper.461,000smallentrantrmsaresuccessfullymergedtoalldatasets.Thesmallbusinessesinthemergedsamplearebiasedtowardslargerbusinesseswithhighersurvivalrates.Assuch,thisshouldbiastheresultsfo
16 rtheeectofcreditconstraintstowardsz
rtheeectofcreditconstraintstowardszero(sincestrongerrmsshouldbelessimpactedbycreditconstraints).Unfortunately,themicro-dataendwithbusinessesformedin2011.Tostudytheeectofacontractioninmortgagecreditsupplyonbusinessentryratessincethe2008nancialcrisis,county-leveldatabetween2009and2016areused.ThisisdescribedintheDeclineinRenancingActivityandBusinessFormationsection.1.2.2TheRoleofHousingEquityinSmallBusinessFinancingPastresearchthatlookedatthefundingstructureofentrantsmallbusinessesreliedonsurveydata.RobbandRobinson(2014)providedtherstinsightintothefundingofentrantsmallbusinessesbyusingcondentialsurveydatafromtheKaumanFamilySurvey(KFS).7Theyndthat16%ofentrantbusinessesin2004werefoundedwithhousingequity.Similarly,Kerretal.(2015)usedthepublicuse2007SurveyofBusinessOwners(SBO)micro-data,whichisbasedonsurveyresultsfrombusinessesaliveasof2007.Theyndthat12to14%ofsmallbusinesseswerefundedwithhomeequityatentrybetween2003and2007.Surveydatahasissueswithaccuracy(Hurst,Li,andPugsley,2014)andnon-response(Campbell,2006).Asasolutiontotheseissues,thispaperdirectlyobservesifanentrepreneurextractshousingequityinanarrowperiodaroundwhentheirbusinessisformed.8Table1.2comparesthedegreetowhichsmallbusinessownersfundtheirbusinesswithhousingequitytotheratesfoundinpriorresearch.Theactualuseofhousingequityisalmostdoublewhathasbeenreportedonsurveys.Despitehousingwealtheectsnotbeingparticularlylargeduringthemid-2000s(Guren,McKay,Nakamura,andSteinsson,2018),thispapershowsthathousingequitywasusedbyroughlyoneoutofeveryfoursmallbusinessstartupsduringthisperiod.Figure1.2showstheuseofhomeequityacrossregionsandtimeforentrantsmallbusinessbytheyeartheyarefounded.9Nationally,oneoutoffoursmallentrantrmswerefundedwiththepersonalhomeequityoftherm'sownerinthemid-2000s,butsince2008this 7Ballou,Barton,DesRoches,Potter,Zhao,Santos,andSebastian(2007)provideadetailedoverviewoftheKFS.8Whileitisnot
17 observediftheextractedhomeequityisinvest
observediftheextractedhomeequityisinvestedinthebusiness,itisunlikelythatanoticeableamountofthehomeequityextractedwasusedforpurposesotherthanfundingbusinesses.Inlatersections,strongresultsfromtheextractedhousingequityonthebusinessarefound.9Theunderlyingdata(ATTOMmergedtoNETS)isrestrictedtobusinessownerswhoalsoownahome,whichwouldbiasupwardsthestatisticsonthepercentofbusinessownerswhousepersonalhomeequity.Tocorrectforthis,thestatisticsareadjustedbythepercentageofthetotalpopulationofbusinessownerswhoownahomewithinthesamecohortthatthestatisticiscalculatedfor.HomeownershipratesarecalculatedfromtheAmericanCommunitySurvey(ACS)micro-data. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP7hasfallentooneoutoftwenty.10Between2001and2003theuseofhomeequitybysmallbusinessownersrosesharplyinallregions.After2003,therewerestrongregionaldierencesinthepatternsofhomeequityuse.OntheWestCoast,homeequityusewasfairlyuniformbetween2003to2006,whileintheMidwesttheuseofhomeequitypeaksin2003.ThepatternfortheMidwestissimilartotheoverallnationalpatternofhomeequityextractions,whichalsopeakedin2003.BhuttaandKeys(2016)attributethedeclineinhomeequityextractionratesafter2003totheriseininterestrates,highlightingtherolethatfundingcostshaveinthedecisiontoextracthomeequity.Followingthe2008crisis,homeequitysharplydeclinedasasourceofcreditforentrantsmallbusinesses.Whilethedatainthispaperendin2011,itisunlikelythattheuseofhousingequityhasrecoveredgiventheaggregatelackofrecoveryincash-outrenancingvolume(Figure1.1).Additionally,comparingwithinsurveydatafromtheCensus,theuseofhomeequityhasfallenbyalmosttwo-thirdsto5%in2016from15%in2005(Table1.2).Thisnewndinghighlightstheimportanceofunderstandinghowmortgagecreditaccesspropagatesthroughtheeconomy.1.3IntensiveMarginImpactofCreditConstraintsTounderstandwhysmallbusinessownersrelysoheavilyontheirpersonalhousingcollateral,itisnecessarytounderstandthedegreetowhichtheyarecreditconstr
18 ained.Thissectionestimatestheintensivema
ained.Thissectionestimatestheintensivemarginlong-runimpactofcreditconstraintsonsmallbusinesses.Section3.1explainsthegeneralempiricalmethodologytoisolateexogenousvariationincreditaccess.Sections3.2and3.4explainthespecicapproachtostudyingentrantandcontinuingbusinesses,respectively.Sections3.3and3.5presenttheresults.1.3.1GeneralEmpiricalMethodologyAkeychallengeinstudyinghowsmallbusinessesrespondtocreditsshockswithmicro-dataisthatbothbusinessownersandtheirbusinessesarehighlyidiosyncratic.Toisolatethelongrunresponseofabusinesstoexogenouscreditshocks,variationsinboththebusinessownerandthebusinesshavetobecontrolledfor.Simplecontrolsandxedeectsareunlikelytobesucientincontrollingforthehigh-dimensionalidiosyncraticdierences.Thispaperusesvariationintheamountofhomeequityextractedfromidiosyncratichomepricegrowthasacreditshock.Toisolateidiosyncratichomepricegrowth,businessownerswholiveinsimilarzipcodesarestudied.Simplycomparingtheamountofhomeequityextractedwouldleadtoestimatingaregressionacrossheterogeneouspopulations,astheamountofhomeequityextractedbyahomeownerisstronglycorrelatedtotheirhomevalue.Thisinturnisassociatedwiththeirwealth,amongothercovariates.Ahomeownerwhoextracts$250,000 10TheNETSdataincludednon-employerrms.Toshowthatthisndingholdsforemployerrmsaswell,thetimeseriesareshownforsmallrmsofvariousinitialsizes.Inallcases,theratiosofrmsfundedwithhomeequityatentryarethesameacrosstime(Figure1intheOnlineAppendix). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP8ofhomeequitywillvaryonobservablequalitiesbetweenahomeownerwhoextracts$100,000ofhomeequity.Forexample,linearornon-parametriccontrolsforhomepriceswouldnotcontrolfordierencesinbusinessownerswiththesamehomevalueacrossdierentcommutingzones.Abusinessownerwhoownsa$200,000homeinSanFranciscoisdistinctfromabusinessownerwitha$200,000homeinCleveland.EvenwithinSanFrancisco,abusinessownerwitha$200
19 ,000homein2001wouldbeverydierentfro
,000homein2001wouldbeverydierentfromabusinessownerwitha$200,000homein2007.Unlesshomevaluesareinteractedwithcontrolsforlocationandtime,theestimatedresponsetocreditshockswouldcapturedierencesinex-antewealthacrossbusinessowners.Additionally,businesscharacteristicsexhibitstrongheterogeneityaswell.Abusinessownerwhostartsarestaurantin2002inthe02120zipcodeofBostonisverydierentfromabusinessownerwhostartsarestaurantinthatsamezipcodein2006.Withoutcontrollingfortheinteractionofzipcode,yearofcreation,andindustry,theestimatedeectofcreditshockswouldpickupdierencesinendogenouswealth,riskaversion,skill,etc.Basedonthesedimensionsalone,controlsforhomevaluesatthesamepointintimewithinthesameregionamongbusinessownerswhostartabusinessinthesameindustryinthesameyearandwithinthesamezipcodewouldneedtobeincludedintheregressionmodel.Thisstrongheterogeneityacrossmanydimensionslendsitselfnaturallytoacoarsenedexactmatchingapproach.Matchingnon-parametricallyisolatesthecausaleectofcreditconstraintsfromidiosyn-craticvariationintheamountofhomeequityextractedamongaheterogeneouspopulationinobservationaldata.AssumebusinessYihadcreditaccessof$Xattimet.Ideally,businessoutcomesforYicouldalsobeobservedifYiinsteadhadcreditaccessof$X+attimet.However,itisnotpossibletoobserveYiinbothcases.Thecausaleectofcreditconstraintscanbedirectlyestimatediftheamountofhousingequityextractedwasrandomlyassigned.Inreality,betweentwobusinessowners,theonewhoextractedgreaterhomeequitywillbecorrelatedwithvariouscovariates,suchasrmindustry,rmlocation,yearofrmen-try,homevalue,housingleverage(combinedloantovalueratio),etc.Dierencesinthesecovariateswilldirectlyaectbusinessoutcomes.Controllingforthesedierenceswithmatchingwillallowforacausalinterpretationtobeuncovered(CardandSullivan,1988;Angrist,1998).11However,exactmatchingwillnotbefeasiblesincemanyofthecovariatesarecontinuousandthecovariatevectorishigh-dimensi
20 onal.Instead,acoarsened-exactmatchingapp
onal.Instead,acoarsened-exactmatchingapproachisutilized(seerecentworkbySarsons,2017;Iacus,King,andPorro,2019).Atahighlevel,thematchingprocedureutilizedinthispaperselectspairsofbusinesseswherebothbusinessownersandtheircorrespondingbusinessesaresimilaracrossasetofcharacteristics.Thematchedpairsarerestrictedtobusinessesthatarelocatedinthesamezipcode(z1).However,thebusinessownersliveindierentzipcodesfromeachotherandfromtheirrms(z2andz3),wherez16=z26=z3andthethreezipcodesarealllocatedwithinthesamecommutingzone.Thisallowsforthetwo 11Forearlyworkonmatchingsee:Rubin(1973),Rubin(1974),Rubin(1977),andRosenbaumandRubin(1983). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP9businessownersinthepairtoexperiencethesamelocalshockstotheirbusinessesinzipcodez1.Figure1.3visualizesthisapproach.Thepairmemberthatexperiencesgreater(less)lagged3-yearzipcodelevelhomepricegrowthislabeledasthetreated(control)member.12Fornotation,thetreatedbusinessowner(A)istheonewholivesinz2andthecontrolbusinessowner(B)istheonewholivesinz3.BusinessownersAandBaresimilarandliveinsimilarzipcodes,butbusinessownerAreceivesadditionalhomepricegrowthgrowthbecauseofthezipcodeinwhichtheylive.Zipcodelevelhomepricegrowthexhibitsstrongcross-sectionalandtime-seriesvariationwithinacommutingzone.Acrosssimilarzipcodes,variationinhomepricegrowthisnotlarge,buteven5%dierential3-yearhomepricegrowthwouldtranslateintoa$15,000creditshockforahomeworth$300,000.Itistestableifzipcodelevelhomepricegrowthdoesnotexhibitstrongvariation,asthiswillleadtoaweakrststage.Itmaybepossibletopredictthatfuturehomepricegrowthwillbehigherforsomezipcodes,forexampleagentrifyingzipcodethatattractsyoungerpeople.However,afterconstructingasampleofsimilartreatedandcontrolbusinessowners,thisisunlikelytobethecase.Asaresult,dierentialhomepricegrowthbetweenthetreatedandcontrolmembersofthepaircanbeusedasanexogenoussourceofcreditthatisorthogonaltobusinessoutcomes.A
21 similaridenticationapproachhasbeenu
similaridenticationapproachhasbeenusedbyBernsteinetal.(2018)andStomanetal.(2018).Thispaperstudieshowbothentrantandcontinuingsmallbusinessesrespondtocreditshocks.Forbothanalyses,thepairsareexactmatchedon:yearofbusinesscreation,industry(SICdivision),zipcodeofthebusiness,andthezipcodesofthehomesbeingdierentfromoneanotherandthebusinesses(thoughthehomesarelocatedwithinthesamecommutingzone).13Giventhestrongandpersistenteectsthatwillbeshownfortheinitialconditionsofentrantrms,continuingrmsarealsoexactmatchedoninitialemployment.Theexactmatchingcriteriacreatesasampleofsimilarrms,whilethecoarselymatchedvariableswillcreateasamplewherethebusinessownersaresimilaraswell.Thecoarsenedsectionofthematchingprocedureisadaptedfortheentrantandcontinuinganalysesandisexplainedbelow.1.3.2EmpiricalMethodologyforEntrantBusinessesThissectionstudiesexogenousvariationintheamountofhomeequityextractedforentrantsmallrms.Thesampleisrestrictedtobusinessesfundedwithhomeequityatentrytocontrolforthefactthatarmnotfundedwithhomeequitylikelyhasdierentialaccesstootherformsoffunding,suchasfamilyorpersonalsavings.Figure1.4agraphicallyillustratestheidenticationset-up.Asanexample,AandBaretwosimilarbusinessownerswhobothstartrestaurantsinzipcode94610inyeartandusedpersonalhomeequitytofundthe 12ZipcodelevelhomepricegrowthdataisprovidedbyZillow.13SICdivisionsarebroadindustrycategoriesof:agriculture/forestry/shing,mining,construction,man-ufacturing,transportation/communications/electric/gas/sanitaryservices,wholesaletrade,retailtrade,-nance/insurance/realestate,andservices.2-digitSICindustryxedeectsareincludedintheregressionstocontrolfordierencesacrosstheindustrieswithineachSICdivision. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP10business.Threeyearsearlier,AandBbothownedhomesofsimilarvalueinneighboringzipcodes94611and94612.Overthesubsequentthreeyears,Areceiv
22 ed10%additionalhomepricegrowth,whichallo
ed10%additionalhomepricegrowth,whichallowedAtoextract$30,000ofadditionalhomeequity.Ifcreditconstraintshavealong-termeectonentrantrms,Ashouldstarttheirbusinessatalargersize(threeinsteadoftwoemployees)andsubsequentlygrowatafasterratecomparedtoB.Forthisanalysis,anentrantbusinessisdenedasbeingfundedwithhomeequityiftheentrepreneurextracts$10,000(or5%ofhomevalueforlessvaluablehomes)intheyearthatthebusinessiscreatedortheprioryear.Thisrestrictionremovesbusinessownerswhoarenotextractingasizableamountofhomeequityandassucharebothunlikelytorespondtosmallvariationsinhomepricegrowthandaremorelikelytohaveotherlargersourcesoffundingsecured.Whiletheuseoftheextractedhomeequityisnotobserved,itcanbeassumedthatthemajorityofthefundsareputtowardsthebusinessgiventhestrengthoftheresultsandthesizablecostofstartingabusiness.Thisintroducesmeasurementerrorintotheresult,whichwilldownwardsbiasthecoecientstowardszeroinOLS.Businessesfundedwithhomeequitylikelyuseothersourcesoffundsaswell(i.e.,savings,creditcards,businessloans,etc.),whichwillintroduceasecondsourceofmeasurementerrorthatwillalsodownwardsbiastheresultsinOLS.Toformmatchedpairs,businessownersandtheirbusinessesareexactmatchedonthecriterialistedintheGeneralEmpiricalMethodologysection.Inaddition,thepairsarecoarsenedmatchedtocaseswherethehomevalues(measuredthreeyearspriortobusinesscreationyear)arewithin20%or$100k(forlessvaluablehomes)ofeachotherandthecombinedloantovalueratios(CLTV)atpurchasearewithin20bpsofeachother.14Thecoarsenedmatchcriteriarestrictstosimilarrmowners,afterhavingalreadybeenrestrictedtosimilarrmswiththeexactmatchcriteria.Thetreatedrmowneristheownerwithineachpairwhoexperiencedgreaterhomepricegrowthinthe3-yearperiodpriortothebusinessbeingformed.Tomakesurethattheentrepreneurspersonallyexperiencethehomepricegrowth,theentrepreneursmusthaveboughttheirhomesatleastthreeyearspriortostartingthebusinesses.Lastly,forcaseswheremore
23 thanonecontrolrmismatchedtoatreated
thanonecontrolrmismatchedtoatreatedrm,thecontrolrmthathasthemostsimilarhomevalueasthetreatedrmisselected.Abusinesscanbeacontrolrmmultipletimes(andcanbebothacontrolandtreatedrmindierentpairs)butcanonlybeatreatedrmonce(themajorityofrmsareonlyinonepair).Thematchingalgorithmformsasampleof5,100rms(basedon4,600distinctrms).MatchedrmstendtobeslightlylargerandtheirownershavehigherCLTVratiosandlowerhomevaluescomparedtotheoverallpopulationofhomeequityfundedrms(Tables1.4andTables1.9).Thestartingsampleis124,000homeequityfundedrms,approximately60%ofwhichareownedbyentrepreneurswholiveindierentzipcodesfromtheirrms,and45%ofwhichareownedbyentrepreneurswhoboughttheirhomeatleastthreeyearspriorto 14CLTVatpurchaseistheratioofthesumofallliensattimeofhomepurchaserelativetopurchaseprice.LTVonlyaccountsforthe1stlien,howeverduringthehousingbubblemanyhouseholdsincludedsecondaryliens(socalled"piggyback"secondliens,LeeandTracy,2012). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP11startingtheirbusinesses.15Alargenumberofobservationsarenotincludedintheanalysisinordertocreateasamplewhereboththetreatedandcontrolrmownersaresimilaronobservables,exceptforthedierentialhomepriceshock.Forhomepricegrowthtobeexogenouswithinapair,thetreatedandcontrolownersmustonaveragebesimilaronobservablecharacteristicsandliveinsimilarzipcodesaseachother.Thisisthegoalofmatching.Table1.3providessummarystatisticsforthedierenceinvaluesbetweenthetreatedandcontrolowners.Onaverage,dierencesacrossthevariablesareclosetozero.Covariatesforhomezipcodesarenotmatched,butarealsosimilarforboththetreatedandcontrolrmowners.Forrobustness,amatchedpairsampleisconstructedfromtheATTOMtoNETSmergedsamplewiththesamecriteriathatisusedtoformtheentrantmatchedpairswiththeLBDdata.Withthisdataitisshownthatthetreatedandcontrolrmownersbothlivethesamedistancefromtheir
24 12;rmsonaverage(Figure1.5).Thisisnotshow
12;rmsonaverage(Figure1.5).ThisisnotshownusingthesampleformedfromtheLBDdataduetodisclosurerestrictions.Oneconcernisthatifthehomezipcodesofeitherthetreatedorcontrolownersareconsistentlymorecorrelatedtoshockstothermzipcode,thenbusinessoutcomescouldbecorrelatedtotheidiosyncratichomepricegrowth.Sincethehomezipcodesaresimilarandthermownerslivesimilardistancesfromtheirrms,thisisunlikelytobethecase.Despitethetreatedandcontrolownersbeingfromsimilarzipcodes,thetreatedownersreceive,onaverage,6.5%greaterhomepricegrowth.Theaveragehomevalueforthepopulationis$206,000,whichleadstoanaverageincreaseof$13,000inavailablecreditforthetreatedownersrelativetothecontrolowners.Withthismatchedpairsample,theeectofcreditshocksonrmoutcomesisestimatedusingtwo-stageleastsquares:Yi;j;t+k=+\ln($AmountExtractedi;t)+ Xj+!Controlsi+i;j;t+k(1.1)rmsaredenotedbyi,rmcreationyearbyt,andthepairthatrmsbelongstobyj.Xjarexedeectsforeachpair.Yi;j;t+karetheoutcomevariables:survivalinyears1,3,and5(estimatedasalinearprobabilitymodel),employmentandemploymentgrowthinyears1through5,andpayrollinyears1through5.Employmentgrowthiscalculatedusingthestandardapproachintheentrepreneurshipliterature16:Ei;tEi;t1 :5(Eit+Ei;t1)(1.2)\ln($AmountExtractedi;t)istheinstrumentedamountofhomeequitythatanentrepreneurextractsintheyearthatthebusinessiscreatedortheprioryear.Evenwithpairxedef-fects,theamountofhomeequityextractedisendogenous.Toisolateexogenousvariationintheamountofhomeequityextracted,thefollowingrststageisestimated: 15ThesestatisticsarecompiledfromtheATTOMtoNETSmergedsampleduetodisclosurerestrictions.16SeeTornqvist,Vartia,andVartia(1985),Davis,Haltiwanger,andSchuh(1996),andHaltiwanger,Jarmin,andMiranda(2013). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP12ln($AmountExtractedi;t)=c+3ln(ZipCodeHomePricei;t)+ Xj+!Controlsi+i;t(1.3)Therststag
25 einstrumentstheamountofpersonalhomeequit
einstrumentstheamountofpersonalhomeequityextractedwiththelogofzipcodelevelhomepricegrowthforthe3-yearperiodpriortobusinessentryyeart.Xjpartialsoutthecommontrendinhomepricegrowthwithineachpair,leavingidiosyncratichomepricegrowthbetweentwosimilarandgeographicallyclosezipcodes.Ifgreaterlaggedhomepricegrowthleadstomorehomeequitybeingextracted,then0.StandarderrorsareclusteredattheSICdivisionbyrmzipcodelevel,whichistheprimaryexactmatchcriteriaofthepairs.Theclusteringleveldoesnotincludermcreationyeartoallowforarbitrarycorrelationacrossrmcreationyearsamongrmslocatedwithinthesamezipcodeandbroadindustrycategory.Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.17Thecontrolvariablesattempttocorrectforanyheterogeneitywithinthepairthatisnotaccountedforbythecoarsenedexactmatchingalgorithm.Thersttwocontrols,homevalueandCLTV,arecoarselymatched,howevertheyareincludedascontrolsincaseanyheterogeneityremains.Likewise,2-digitSICindustryxedeectsareincludedincasehet-erogeneitywithinSICdivisionremains.Thenumberofmonthsbetweenhomepurchaseandrmcreationyeartisincludedtocontrolfordierencesbetweenhowlongabusinessownerwaitstostartabusinessafterbuyingahome.LFOxedeectscontrolfordierencesamongincorporationtypesofbusinesses.Lastly,thehomezipcodecharacteristiccontrolsremovevariationfromthechoiceofhomezipcodewithinacommutingzone,ifanyremainaftermatching.1.3.3EntrantBusinessEmpiricalResultsIfanewbusinessiscreditconstrainedandreceivesanexogenousrelaxationintheircreditlimit,howarethebusiness'ssurvival,size,andgrowthaected?Ifthebusinessisex-antecreditco
26 nstrainedandtheadditionalcreditisspentpr
nstrainedandtheadditionalcreditisspentproductively,thenlooseningthecreditconstraintsshouldresultinthebusinessstartinglarger,growingfaster,andsurvivinglonger.Second,itisimportanttounderstandifbusinessesthathavelessaccesstocreditatentryareabletocatchupfromtheirinitialdisadvantage.Theeectonbusinesssurvivalisshownrst.Table1.5column1showstherawOLSeectonthe5-yearsurvivalratewhenonlythepairxedeectsareincluded.Havingadditional 17Zipcodelevelcontrolvariablesarefromthe2000DecennialCensus. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP13accesstocreditleadstoanincreasedsurvivalrate.Theresultsurvivesandremainsstablewhenthevectorofadditionalcontrolsareincluded,implyingthatthesecontrolsarenotasourceofheterogeneitybetweenthetreatedandcontrolpairmembers.Column3teststherststageandshowsthatwithinthepair,thebusinessthatreceivesadditionalpersonalhomepricegrowthinthe3-yearperiodbeforethebusinessisfoundedextractsadditionalhousingequity.TheinstrumentisstronglysignicantandtheF-statisticisabove10,indicatingthattheinstrumentisnotweak.Thecoecientof0.88onlaggedhomepricegrowthmeansthatthetreatedmemberextracts88%oftheadditionalhomeequitythattheyhaveaccesstofromvariationinhomepricegrowth.Themeandierenceinhomepricegrowthbetweenthetreatedandcontrolrmsis6.5%,implyingthattheaveragetreatedrmhas5.7%ofadditionalrealizedfundingatcreationrelativetothecontrolrm.Sincethetreatedrmownerextractsmostoftheadditionalhomeequitythattheyhaveaccessto,thisispreliminaryevidencethatrmsarecreditconstrained.Otherwise,therewouldnotbeasstrongofaresponsetoexogenousvariationinhomepricegrowth.Incolumn4,theregressionmodelisestimatedwiththeamountofhomeequityextractedinstrumentedwith3-yearlaggedhomepricegrowth.Thecoecientshowsastrongpositivecausaleect.18A10%increaseincreditincreasesarm's5-yearsurvivalrateby5.1%.Forthesampleusedintheseregressions,theaverageamountofhomeeq
27 uityextractedis$100,000,whichmeansa10%cr
uityextractedis$100,000,whichmeansa10%creditshocktranslatesinto$10,000ofadditionalcredit.The5.1%increaseinthe5-yearsurvivalratefromthissmallcreditshockwouldleadtothesurvivalrateincreasingfrom67%to71%,onaverage.Thisresultshowsthatthereisalongtermeectonsurvivalfromapositiveexogenouscreditshock.Therefore,theinitialcreditaccessofabusinessisstronglyimportantandsmallbusinessesarecreditconstrainedatentry.Schmalzetal.(2017)alsofoundapersistentimpactofinitialhousingwealthonsmallbusinessesforsmallbusinessesfoundedin1998inFrance.However,thisisthersttimeithasbeenconclusivelyshownforsmallentrantbusinessesintheUnitedStates.Duetostrongdierencesinthebankingandloanmarketsandentrepreneurialdemand/preferencesbetweenthetwocountries,itisnotobviousthattheresultsshouldapplytotheUnitedStates.Anaturalquestioniswhethertheeectonsurvivalisapparentimmediatelyorifitslowlyaccumulates.Columns7and8showtheeectinyears1and3,respectively.Theeectispresentevenintherm'sinitialyear,withthesurvivalrateincreasingfromhavingaccesstoadditionalcredit.Theeectinyear3islargerthaninyear5,implyingaconcaverelationship.Afteryear3,thecontrolrmstartstocatchupbutremainsatadisadvantageinyear5.Unfortunately,survivalbeyondthe5thyearcannotbetestedduetorestrictionsonCensusRDCdisclosure.However,a5-yearsurvivalrateeectimpliesthattheeectispermanent.Eectsonotherbusinessoutcomesareestimatedtounderstandwhybusinesssurvivalincreaseswithadditionalcredit,andtoshowrobustnessthattheeectisnotconnedto 18Asarobustnesscheck,ifthestandarderrorsareclusteredattheSICdivisionbyrmzipcodebycreationyearlevel,theresultremains(column6). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP14survival.InTable1.6,theregressionisestimatedfortheeectofcreditconstraintsonem-ployment.Columns1-5showthatemploymentincreasesinyears1through5fromhavingaccesstoadditionalcreditatentry.Similartotheeectonsurvival,th
28 eeectforemploy-mentispersistentandr
eeectforemploy-mentispersistentandremainsthroughyear5.Thesamplesizedeclinesfortheresultsonemploymentafteryear1,duetobusinessesbeingomittedfordisclosurereasonsiftheyhavemissingdataaftertheirinitialyear.19A10%increaseincreditleadstoa4.9%increaseinemploymentatcreation.IntheLBD,5.6millionsmallentrantrmswerefoundedbetween2001and2011andcreated13.5millionjobsatentry.Approximately22%ofthesermswerefundedwithhomeequityatentry,leadingto2.9millionjobscreatedatentrybyhomeequityfundedrms.20Forevery10%increaseinhousingcollateral,abackoftheenvelopecalculationshowsthat140,000additionaljobswouldbecreated.Additionally,the0.49coecientfortheeectoninitialemploymenttranslatestooneadditionaljobbeingcreatedfromevery$88,000positivecreditshockasmallentrantbusinessreceives.Inadditiontostartinglargerinyear1,thetreatedbusinessesgrowfasterbetweenyears1and2(Table1.7).A10%increaseininitialcreditincreasesinitialrmgrowthratesby18%betweenyears1and2.Afteryear2,growthratesbetweenthetreatedandcontrolbusinessesstabilizeandthisallowsthetreatedbusinessestoremainlargerthroughyear5.Theeectsonemploymentshowthatsmallincreasesininitialcredithaveanimportanteectonarm'sabilitytostartatamoreoptimalsize.Inaddition,greaterinitialcreditaccessallowsbusinessestooperatemoreproductively,whichleadstolargerinitialgrowthrates.Businesseswithlessaccesstocreditatentryareneverabletocatchupfromtheirinitialdisadvantage.Asalasttest,theeectonpayrollisestimated(Table1.8).Initialpayrollisnotsta-tisticallysignicantlylargerinyear1fromacreditshockatentry.Inlateryears,payrollincreasesfromthecreditshockatentry.Apossiblereasonforthisisthathavingadditionalinitialcreditallowsbusinessestoattractmorecustomersthroughincreasedadvertisingoramoreattractiveinterior.Thisinturnincreasesrevenueanddemand,whichnecessitatesaneedtohiremoreworkers.However,businessesmayhavetodecreasethewagesofworkersinordertohiremoreemployees.Studyinghowwagesa
29 reaectedbyarm'saccesstocrediti
reaectedbyarm'saccesstocreditisleftforfutureresearch.Aconcernwiththeempiricalapproachisthatdierentialhomepricegrowthmightaectthedecisiontostartabusiness.Ifthisisthecase,dierentialhomepricegrowthbetweenthetwopairmembersmightdirectlyaectbusinessoutcomes.Thendingsarebasedonsmalldierencesinhomepricegrowthbetweenthetreatedandcontrolentrepreneurs(6.5%onaverage),whichmakesthislessofaconcern.However,ifthetreatedentrepreneursrequired 19Duetorestrictionsondisclosure,afteryear1pairsaredroppedifoneofthebusinessesinapaireverhasmissingemployment/payrolldatainyears2through5.Inaddition,ifarmisinmultiplepairsandoneofthepairsisexcludedthenalloftheirotherpairsareexcludedaswell.20ThisstatisticisbasedonthemergedsampleofATTOMtoLBD.Itisadjustedbythepercentofsmallbusinessownerswhoarehomeownerstoaccountforthefactthatthemergedsampleonlyincludeshomeowners.ThepercentofsmallbusinessownerswhoarehomeownersiscalculatedfromACSmicro-data. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP15additionalcreditinordertobeconvincedtostartabusinesstheywouldlikelybeofalowerskilltypecomparedtothecontrolentrepreneurs.Thiswouldmakethetreatedentrepreneurslesssuccessful,whichwoulddownwardsbiastheeect.Second,ifthecontrolentrepreneurswereseverelyconstrained,becausetheyhadlesshomepricegrowth,andstilldecidedtostartabusinessitcouldrevealthattheyareofahigherskilltype.Thiswouldagaindownwardsbiastheeect.Inbothcasesthebiasfromthispotentialconcernwoulddecreasetheeectthatisfound.Thepopulationofsmallbusinessesisrestrictedtobusinessesthatstartedasasingle-establishmentandwithtenorfeweremployees.Thethresholdoftenwaschosenbecauselargerbusinessesrelylessonhomeequity,duetothegreatercostsinvolvedinstartinglargerbusinesses.Adelinoetal.(2015)ndthatrmswithmorethantenemployeesdonotrespondtohomepricegrowthshocks.Inaddition,Patnaik(2017)showsthatrmswithtenorfeweremployeesrelyonhousingwealt
30 h,whilermswithmorethantenemployeesr
h,whilermswithmorethantenemployeesrelyonbankloans.Additionally,fewerthan10%ofentrantbusinessesbetween2001and2011startedwithmorethantenemployees(Figure2intheOnlineAppendix).Includingthissmallpopulationoflargerrmswouldaddnoisetotheresultssincetheyarelikelynotrelyingonhousingwealthandareinherentlydierentfromsmallerrms.Anadditionalconcernisiftreatedrmownersareconsistentlymoreorlesslikelytooriginatetheircash-outrenancemortgageswithtraditionalbanks,suchasBankofAmer-ica.Traditionalbankscross-sellcustomers,asaresultbankersmightencouragermownersobtainingbusinessloanstoalsoobtaincash-outrenancemortgagesinordertohaveaddi-tionalcredit.Non-traditionalbanks,suchasQuickenLoans,donotoriginatebusinessloans.Iftreatedrmownersarebiasedtowardsorawayfromtraditionalbanks,therecouldbeabiasintheirtotalamountoffundingrelativetocontrolrmowners.UsingthematchedpairsamplethatisconstructedfromtheATTOMtoNETSmergedsample,itisshownthatthetreatedandcontrolrmownershavethesamelikelihoodofhavingoriginatedtheircash-outrenancemortgagesfromtraditionalbanksonaverage(Figure1.6).ThisisnotshownusingthesampleformedfromtheLBDdataduetodisclosurerestrictions.1.3.4HomeEquityFundedBusinessesAkeyassumptiontogeneralizingthendingstoallsmallbusinessesisthathomeequityfundedbusinessesaresimilartonon-homeequityfundedbusinesses.Table1.9comparesthepopulationsofhomeequitytonon-homeequityfundedbusinesses.Overall,homeequityfundedbusinesseshavesimilarsurvivalratesandtendtostartslightlylargercomparedtonon-homeequityfundedbusinesses.Theentrepreneursofhomeequityfundedbusinesseshavegreaterhousingleverage(attimeofhomepurchase),largerhomevalues3-yearspriortowhenthebusinessisformed,andboughttheirhomesmorerecently.Theaveragebusinessownerwhouseshomeequityasafundingsourceextracts$100,000ofhomeequity.Aregressionmodelisutilizedtorigorouslytestthedierencesbetweenthesetwopopulations:Yi=+I(HomeEquityFundedi)
31 +!Controlsi+i(1.4) CHAPTER1.THELONG
+!Controlsi+i(1.4) CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP16whereiindexesabusiness.Asarsttest,thispaperchecksifhousingcharacteristicsvaryacrossthetwofundingtypes,byincludingtheinteractionofhomezipcodebyhomepurchaseyearxedeectsintheregressionmodel.Additionalcontrolsincludelegalformoforganization(LFO)and2-digitSICindustryxedeects.Standarderrorsareclusteredatthehomezipcodelevel.Table1.10columns1and2rigorouslyshowthathomeequityfundedrmstendtobestartedbyentrepreneurswhoownslightlymoreexpensivehomes(5%greaterhomevalues)andwhohaveslightlygreaterhousingleverage(4pointhigherCLTVratios)withinazipcodeatagivenpointintime.Next,thispapertestsifhomeequityfundedbusinessesdieroninitialsizeandsurvival,byincludingthetripleinteractionof2-digitSICindustrybycreationyearbyrmzipcodexedeectsintheregressionmodel.Additionalcontrolsincludelogpurchasevalue,CLTV,andxedeectsforLFO.Standarderrorsareclusteredatthe2-digitSICindustrybyrmzipcodelevel.Table1.10columns3and4showthathomeequityfundedrmsstartslightlylargerintermsofemployment,butnotpayroll.Interestingly,thereisnodierentialoutcomeforsurvival.Togeneralizetheresultsfortheexistenceofcreditconstraints,survivalisthemostimportantvariabletotest,sinceaninsignicantdierenceinsurvivalbetweenhomeequityandnon-homeequityfundedrmsimpliesthatrmsfundedwithhomeequityatentryarenotweakerrms.Adelinoetal.(2015)presentssuggestiveevidencethathomeequityfundedrmswerenotmorelikelytoexitduringthe2008nancialcrisiscomparedtormsnotfundedwithhomeequity.Theresultsofthissectionshowfurtherproofthatthisisindeedthecase.Inresultsavailableuponrequest,itisfoundthatduringthe2007-2009period,smallbusinessesfundedwithhomeequityatentrywerelesslikelytoclosecomparedtosmallbusinessesnotfundedwithhomeequityatentry.21Lastly,themortgagedefaultratesofsmallbusinessownerswhoextractpersonalhome
32 equitytofundtheirbusinessesareshown.Over
equitytofundtheirbusinessesareshown.Overall,itiswellestablishedthathomeownerswhoextractedhomeequityduringthe2000srealizedlargermortgagedelinquentratescomparedtohomeownerswhodidnot.BhuttaandKeys(2016)ndthathomeownerswithacash-outrenancebetween2001and2003experienced20%greater4-yeardefaultrates.By2006,the4-yeardefaultrateofhomeownerswhoextractedhomeequitywasdoublethedefaultrateofhomeownerswhodidnotextracthomeequity.Toshowmortgagedefaultratesforbusinessowners,Equation4isestimatedasalinearprobabilitymodelwithadependentvariableequaltooneifthebusinessownerexperiencesaforeclosureontheirpersonalhomewithinfouryearsofstartingthebusiness.22Controlsincluderiskcharacteristicsforcombinedloantovalue(CLTV)atpurchase,FICOatpur-chase,andinitialmortgagerate.23Theinteractionofhomepurchaseyear,rmcreation 21Duetodisclosurereasons,thisresultisshownusingsurvivalinformationfromNETSandnottherestricted-useLBD.22ThisanalysisusesthedatafromthemergedATTOM-NETSsamplethatisfurthermergedtoMcDash.ForeclosureisestimatedfromATTOM.FICOandinitialmortgageratearefromMcDash.23Homepurchasevalueandinitialrmemploymentareincludedasadditionalcontrols. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP17year,homezipcode,rmzipcode,and2-digitSICindustryisincludedinordertoestimatethemortgagedefaultprobabilitydierentialamongsimilarbusinessowners.Itisimportanttocompareamongborrowerswhoboughttheirhomeinthesameyearasmortgagedefaultratesvarysubstantiallyacrosshomepurchaseyears(Palmer,2015).Overall,thereisnodierentialprobabilityofforeclosureforbusinessownerswhofundtheirbusinesseswithhomeequity(column1,Table1.11).Eventhoughintheoverallpop-ulation,homeownerswhoextracthomeequitydefaultontheirmortgagesatgreaterrates,businessownerswhorelyonhomeequitydonotdefaultathigherratescomparedtosimilarbusinessownerswhodonotextracthomeequitytofundtheirbusiness.Forbusinessownerswhostartedabusinessbetween2005and2007,thereisanelevate
33 dprobabilityofforeclosureifhomeequityise
dprobabilityofforeclosureifhomeequityisextractedtofundthebusiness(column3,Table1.11).Thedefaultraterisesby25.6%,fromthepopulationmeandefaultrateof24.8%to31.2%.Whilethede-faultrateiselevatedduringthisperiod,theincreaseindefaultismuchsmallercomparedtotheincreaseindefaultratesfortheoverallpopulationofhomeownerswhoextractedhomeequityduringthisperiod.BhuttaandKeys(2016)notethatdefaultratesfortheoverallpopulationofhomeownerswhoextractedhomeequityduringthisperiodroseby80-100%ofthepopulationmean.Lastly,businessownerswhofundedtheirbusinesseswithhomeequitysince2008havehadlowermortgagedefaultratescomparedtoallbusinessownerswhostartedabusinessduringthisperiod(column4,Table1.11).Thisisconsistentwiththenotionoftightermortgagecreditstandardsdeterringrmentry,whichisexploredinSection4.Ingeneral,mortgagedefaultratesareelevatedforbusinessowners,withmeandefaultratesonmortgagedebtof19%duringthe2001to2011period(column1,Table1.11).Thisislikelyaresultoftheriskinessofsmallbusinessownership.Ifthebusinessfails,thebusinessownerisatincreasedriskoflosingtheirhome.However,theuseofhomeequitytofundbusinessesdoesnotnoticeablyaectthealreadyhighmortgagedefaultrateofbusinessowners.Thisisadditionalevidence,beyondlowsurvivalrates,totheriskinessofsmallbusinessownership.1.3.5EmpiricalMethodologyforContinuingBusinessesContinuingbusinessesareoftenignoredintheliteratureoncreditconstraints.Entrantbusinesses,intheory,haveamorediculttimeobtainingcreditduetoalackofbothsoftandhardinformationandbusinesscollateral.However,continuingbusinesseshavehardinformation(nancialstatements),theabilitytobuildarelationshipwithabankerforsoftinformation,andpotentiallyhavebusinesscollateral.Additionally,dataonthelong-runresponseofcontinuingbusinessesisdiculttoobtain.Thispaperhastheidealdatasettotestifcontinuingbusinessesarealsocreditconstrained.Toobtainameasureofcreditconstraintsforcontinuingbusinesses,exogenousvariationintheamountofpersonalhomeequit
34 yextracted(ifany)inyears3or4ofarm's
yextracted(ifany)inyears3or4ofarm'slifeisstudied.Figure1.4bgraphicallyillustratestheidenticationset-upandshowstheoutcomeifcontinuingbusinessesarecreditconstrained.AandBaretwosimilarbusinessownerswho CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP18bothstartrestaurantswiththesamenumberofemployeesinzipcode94610inyeart.Inyeart,AandBalsoownedhomesinneighboringzipcodes94611and94612,respectively,withthesamehomevalue.Overthenexttwoyears,theirrmsgrewatasimilarrateintermsofemployment.Inyeart+3,Ahasaccumulated10%additionalhomepricegrowthandbecauseofthisextracts$100,000ofhomeequity,whileBdoesnot.(ItcouldalsobethatBextractshomeequityaswell,butextractslessduetoexperiencinglesshomepricegrowth.)Withthisexogenousvariationincreditaccessinyeart+3,Aexpandstheirrmandpermanentlyremainslarger.Theexampleshowsthematchingexerciseforextractioninyear3(referredtoastheeventyear)|however,matchingisalsoperformedforrmsinyear4.Forarmtobematched,itmusthavesurvivedthroughtheeventyear.Firmscanbematchedinboththeagethreeandfoureventyearcohortsandtheresponsetocreditshocksisjointlyestimatedwithdatafrombotheventyears,whichassumesthatbusinessesdonotdeferentiallyrespondtocreditshocksatagethreeversusagefour.Itismorediculttoisolateexogenousvariationforcontinuingbusinessescomparedtoentrantbusinessesduetotheadditionaldimensionsofthebusiness'spastrecordthatalsomustbematchedon.Inthesectionforentrantbusinesses,itisshownthatinitialconditionsforabusinessplayapersistentroleinthebusiness'ssuccessandsize.Therefore,abusiness'sinitialconditionsmustbecontrolledforinthecoarsened-exactmatchalgorithmforcontinuingbusinessesbyalsoexactmatchingoninitialemployment.Continuingrmsarecoarselymatchedontwocovariates.First,homevalues(asmeasuredthreeyearspriortotheeventyear)mustbewithin20%or$100k(forlessvaluablehomes)ofeachother.Second,thenumberofemployeesoneyearpriortotheeventyearmustbewithinthreeemployeesofeachother.Thi
35 ssecondconstraintrestrictstormsthat
ssecondconstraintrestrictstormsthatareonsimilargrowthpaths(afterstartingatthesamesize).Thetreatedrmsaretheonesthatexperiencegreaterhomepricegrowthwithinthe3-yearperiodpriortotheeventyear.Toconrmthattheentrepreneurspersonallyexperiencethehomepricegrowth,theymusthaveboughttheirhomesatleastthreeyearspriortotheeventyear.Lastly,incaseswheremorethanonecontrolrmismatchedtoatreatedrmforagiveneventyear,thecontrolrmthathasthemostsimilarhomevalueasthetreatedrmisselected.Abusinesscanbeacontrolrmmultipletimes,butcanonlybeatreatedrmoncewithineachofthetwoeventyears.Roughlyhalfofthermsinthesampleareonlyinonematchedpair.Fromthematchingalgorithm,17,500rmsarematched(11,500uniquerms).Thestartingpopulationforthissamplearermsthathavesurvivedthroughthematchedeventyear,whichrestrictsthesampletoatmost345,000rms.Ofthese,approximately60%ofthermownersliveindierentzipcodesfromtheirrmand45%boughttheirhomeatleastthreeyearsbeforetheeventyear.24Alargenumberofobservationsareremovedinordertocreateasampleinwhichboththetreatedandcontrolrmownersaresimilaron 24Duetodisclosurerestrictions,exactnumbersarenotprovidedandthesestatisticsarecompiledfromtheATTOMtoNETSmergedsample. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP19observables,exceptforthedierentialhomepriceshock.Onaverage,$7,909ofhomeequityisextractedintheeventyear(Table1.12).Theaverageincludeszerosforthebusinessesthatdidnotextracthomeequityintheeventyear.Toensurethatthematchingdoesnotbiasthetreatedorcontrolmemberstowardscertaincharacteristics,averagedierencesbetweenkeyvariablesarecalculated.Therearenonoticeabledierencesbetweenthetreatedandcontrolbusinesses/ownersbasedonasetofobservablecharacteristics(Table1.13).Thetreatedownersreceiveanaverage10%ofadditionalhomepricegrowth,sotheaveragehomevalueof$228,000translatesintoacreditshockofroughly$23,000.Toestimatetheeec
36 tofcreditconstraintsoncontinuingbusiness
tofcreditconstraintsoncontinuingbusinesses,aregressionmodelsimilartoequation1.1isutilized:Yi;j;t+p+k=+\ln($AmountExtractedi;t+p)+ Xj+!Controlsi+i;j;t+p+k(1.5)wherepisthenumberofyearssincefoundingtthathomeequityispotentiallyextractedforexpansion(twoorthree,whichcorrespondstormsofeitheragethreeorfour).Thedierencebetweenequation1.5andequation1.1isthevectorofcontrolvariables,sincecontinuingbusinesseshaveadditionaldimensionsthatneedtobecontrolled.Forcontinuingbusinesses,Controlsialsoincludesaxedeectforifthebusinesswasinitiallyfundedwithhomeequity,anddependingontheregressionalsoincludesnon-parametricxedeectsforinitialpayrollandemploymentandpayrollgrowthbetweenyears1and2.Theselasttwocontrolsattempttocontrolforvariationinex-antebusinesssuccess.Theinitialpayrollcontrolisinadditiontomatchingoninitialemployment.Toisolateexogenousvariationintheamountofhomeequityextracted,thefollowingrststageisestimated:ln($AmountExtractedi;t+p)=c+3ln(ZipCodeHomePricei;t+p)+ Xj+!Controlsi+i;t+p(1.6)Therststageinstrumentstheamountofpersonalhomeequityextractedwiththelogofzipcodelevelhomepricegrowthforthe3-yearperiodpriortotheeventyeart+p.Xjpartialsoutthecommontrendinhomepricegrowthwithineachpair,leavingidiosyncratichomepricegrowthbetweentwosimilarandgeographicallyclosezipcodes.Ifgreaterlaggedhomepricegrowthleadstomorehomeequitybeingextractedthen0.measuresthecausalimpactofhowcontinuingbusinessesrespondtoexogenousavail-abilityofhomeequityatagethreeorfour(t+p).Theresponseisthenstudiedforoneyearbeforetothreeyearsafteryeart+p.Ifthetreatedandcontrolrmsaresimilarbeforeyeart+p,thereshouldbenodierenceintheiroutcomesinyeart+p1.Ifhomeeq-uityextractedinyeart+p,duetoexogenousvariationinpasthomepricegrowth,relaxesthecreditconstraintsofcontinuingrmsandtheyproductivelyusethecredit,apositiveshouldbefoundinyeart+ponward. CHAPTER1.THELONG-RUNEFFECTSO
37 FMORTGAGECREDITACCESSONENTREPRENEURSHIP2
FMORTGAGECREDITACCESSONENTREPRENEURSHIP201.3.6ContinuingBusinessEmpiricalResultsThissectionstartswithtestingifexogenouscreditshockstocontinuingbusinessesaectrmsurvival.DuetorestrictionsonRDCdisclosure,onlyonesurvivalperiodisstudied,whichisthreeyearsaftertheeventyear(year6or7ofthebusiness'slifedependingonwhichagetheeventyearcorrespondsto).Firmshadtosurvivetotheeventyear(agethreeorfour)tobeincluded,whichrestrictsthepopulationtoasetofstrongerbusinessesgivenlowinitialsurvivalrates.Table1.14columns1and2estimatetheregressionmodelusingOLS.NoeectisfoundwhenusingOLSandthecoecientisveryclosetozero.Thisimpliesthatthereisnoraweect.TherststageshowsaverystrongpositiveeectfrompasthomepricegrowthonhomeequityextractionintheeventyearandtheF-statisticisabove10(column3),indicatingthattheinstrumentisnotweak.Inthesecondstage,nosignicanteectisfoundforsurvivalandthecoecientisstillroughlyzero(column5).Totestforrobustnessaroundtheclusteringlevel,columns6and7clusteratrmzipcodebySICdivisionbyrmcreationyearandrmcommutingzonebySICdivisionlevels,respectively.Thecoecientremainsinsignicant.Relaxingcreditconstraintsoncontinuingbusinessesdoesnotaecttheircontinuedsurvivallikelihood.Thisislikelytheresultofsurvivalratesbeingconvex,inthatmanyrmsclosewithintheirrstfewyears.Followingwhichtherateofsurvivallevelso.Giventhatthesermshavealreadysurvivedthroughtheirmostvolatileyears,havingadditionalcreditwillnotaecttheirfuturesurvival.Thisdoesnotimplythatcontinuingrmsarenotcreditconstrained.Totestifrmsizeisaected,theregressionisestimatedwithemploymentasthedependentvariable.First,column1ofTable1.15showsthatinyearpriortotheeventyear,yeart+p1,thereisnosignicantdierenceinemploymentbetweenthetreatedandcontrolrms.Intheeventyear,employmentfortreatedrmssignicantlyexceedtheemploymentofcontrolrms(column2).Thisimpl
38 iesthatbetweenyearst+p1andt+p,treate
iesthatbetweenyearst+p1andt+p,treatedrmsexpandedmorerapidlythancontrolrms.Column6conrmsthisbyshowingthattheemploymentgrowthrateislargerfortreatedrms.Inaddition,treatedrmsremainlargerforthefollowingthreeyears,whichiswhenthesampleends(columns3through5).Similarly,Table1.16showsthatpayrollbetweenthetreatedandcontrolrmsissimilarbeforetheeventyearandimmediatelyincreasesforthetreatedrmsintheeventyear.Overall,whencontinuingbusinessesreceiveacreditshock,theyimmediatelyexpandandpermanentlyremainlargerasaresult.ThendingsinTable1.15showthatwhencontinuingrmsextractadditionalhomeequitybecauseofexogenousavailabilityofhomeequity,theyimmediatelyexpandandsub-sequentlypermanentlyremainlarger.Theseresultsprovideevidencethatcontinuingbusi-nessesarecreditconstrained.Havingaccessto10%ofadditionalcreditleadstoanincreaseinemploymentof1.1%,withtheeectincreasingto1.6%afterthreeyears.Thecoecientof0.114translatesintoacreditshockof$153,000immediatelycreatingoneadditionaljobatasmallcontinuingrm.Sincecontinuingrmsrequirealargerpositivecreditshocktohireoneadditionalworker, CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP21continuingrmsarelesscreditconstrainedthanentrantrms.Toseewhythisimpliesthatcontinuingrmsarelesscreditconstrained,itishelpfultothinkintermsofa\marginalpropensitytohire(MPH)",similartotheconceptofmarginalpropensitytoconsumeorborrow.Oneworkerhiredfroman$88,000creditshockimpliesthat0.011jobsarecreatedforevery$1,000ofcreditthatisextended,thisistheMPHforentrantrms.WhilecontinuingrmshaveanMPHof0.0065forevery$1,000ofcreditthatisextended.SincecontinuingrmshavealowerMPH,thisimpliesthattheyhavelessofaneedtohirefromacreditshock.Thesearetherstresultstotestifcontinuingbusinessesarecreditconstrained.Contin-uingbusinesseshavesurvivedtheirearlyyears,whentheyaremostpronetofailure.Thissegmentsthepopulationtoastrongergroup
39 ofbusinesses.Assuchtheyareoftenignoredin
ofbusinesses.Assuchtheyareoftenignoredinthestudyofcreditconstraintsonsmallbusinessessinceitisassumedtheycanmoreeasilyobtainexternalnancing.However,theresultsinthissectionshowthatbusinessesremaincreditconstrainedastheyage.Whencontinuingbusinessesreceiveanexogenousincreaseincredit,theyimmediatelyexpandandpermanentlyremainatalargersize.1.4DeclineinRenancingActivityandBusinessFormationAsshownearlier,entrantsmallbusinessesreliedheavilyonhomeequityfornancingduringthemid-2000s.However,startingin2008,thiscreditchannelwasvirtuallyeliminated(Fig-ure1.2andTable1.2).Mortgagelendinghasbecomeverytightandasaresult,borrowersarehavingadiculttimeobtainingmortgagecredit.Goodman(2017)ndsthatamongallhomebuyers,creditstandardshavebecometwiceasrestrictiveastheywerein2001,whichwaspriortothelooserlendingstandardsofthesubprimebubbleofthemid-2000s.Recentresearchhasshownthatthistighteningofmortgagecreditsupplyhascausedfewermortgageoriginations|particularlyamongyounger,middleincome,andblackbor-rowers|andhigherrentalprices(LauferandPaciorek,2018;GeteandReher,2018).Giventherelianceofsmallbusinessownersonhomeequityandthestrongcreditconstraintsthattheyface,anotherpossibleeectofatighteningofcreditstandardsisareductioninbusinessformationrates.DuringtheGreatRecession,householdrenancingactivityandbusinessformationratessimultaneouslyfell,andbothhaveremainedpermanentlylowersincetheGreatRecession(Figure1.1).Thisisdespitestronghomepricegrowthoverthepost-crisisperiod|between2009and2018,homepriceshavegenerallyreachedtheirpre-crisislevelsandinmanycasesexceededthem(Figure1.7).Asaresult,householdsarelikelytohavesignicanthomeequity.Correlationsbetweenhomeequityextractionactivityandrmen-tryratesimplythatalackofrenancingactivityofsmallbusinessownersisachannelforthesteepdeclineinrmentryratessincetheGreatRecession.Thissectionwillshowthiscausally.IntheUnitedStates,increasedhousingwealthisrealizedbyentrepre
40 neursthroughex- CHAPTER1.THELONG-RUNEFFE
neursthroughex- CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP22tractinghomeequityandthenusingthecashtofundtheirbusinesses.Normally,housingwealthvariationsandhomeequityextractionactivityarehighlycorrelated.Thishasal-lowedpastresearchtoproxyhomeequityextractionactivitywitheitherhomepricegrowthoravailablehousingwealthinstudyingtheeectsonbusinessformation.However,thepe-riodsince2009hasseenarecoveryinhomepricegrowthwithoutaconcurrentrecoveryincash-outrenancingrates,therebybreakingthislink.Instead,thissectionteststheeectofhomeequityextractionactivityonbusinessentry.Themicro-datathatisutilizedearlierinthispaperdoesnothavedataonrmsthatarecreatedafter2011.Therefore,thissec-tionutilizescounty-levelmeasuresofcash-outrenancingactivitytostudytheimpactsonbusinessformationduringthe2009to2016period.Unlikedataforhomepricegrowth,localmeasuresofcash-outrenancingactivityarenotreadilyavailable.Usingmortgagetrans-actionleveldatafromATTOM,thispaperconstructs1-yeargrowthratesinthenumberofhouseholdsextractinghomeequityatthecounty-levelbetween2009and2016.Totestthehypothesisthatincreasedlocalrenancingactivityleadstoanincreaseinbusinessformation,1-yeargrowthratesincounty-levelhomeequityrenancingratesareregressedonto1-yeargrowthratesincounty-levelbusinessformationrates.However,thiswillsuerfromomittedvariablebiasduetoconfoundingdemandandbankingshocksthataectbothrenancingactivityandbusinessformationactivity.Inaddition,theexplanatoryvariableofcounty-levelgrowthinrenancingactivitysuersfrommeasurementerror.Themeasureofrenancingactivityisforallhouseholdsinthecounty,notonlyentrepreneurs.Sincegrowthinrenancingactivityandgrowthinbusinessformationratesareassumedtobepositivelyrelated,thecoecientwillbebiaseddownwardsinOLS.Toisolateacausalrelationshipthatisdrivenbythecollateralchannel,twostepsareun-dertaken.First,aninstrumentthatisolatesexogenousvari
41 ationinrenancingactivitythatdoesnot
ationinrenancingactivitythatdoesnotdirectlyaectbusinessformationisutilized.Theinstrumentwillalsocorrectformeasurementerror.Second,tofurtherisolatethecollateralchanneleectfromthedemandchanneleect,industriesthatarereliantonlocaldemandareremoved.Forrobustness,itisshownthattheresultgeneralizestoallindustries.Toconstructaninstrument,thispaperexploitstheinstitutionaldetailthatindependentmortgagebanks(IMBs,whichincludesmortgageoriginatorssuchasQuickenLoans)arepronetochurningtheiroriginations.ThismeansthatIMBsaggressivelytargettheirpriorcustomerstoencouragethemtorenancetheirexistingmortgage.Consequently,greaterIMBmarketsharewillleadtogreaterrenancingactivity.25Figure1.8showsthiseect 25InaSEClingfrom2015,EllingtonFinancialnotedtheaggressivetargetingofcustomersbyIMBstorenanceasarisktoMBSholdings:\Asimprovedtechnologyspreadsthroughoutthelendingindustry,webelievethatthelendingindus-trywillchangeinanumberofimportantways...borrowersshouldstarttoprepaymoreeciently,asdemonstratedbythehigherprepayspeedsofmortgageloansservicedbyQuicken,whichreportedlyusesproprietaryalgorithmstotargetborrowerswhoaremorelikelytorenance,andisparticularlyquickatcontactingborrowersaboutrenancingincentives...Atechnology-driven,broad-basedin-creaseinprepaymenteciencymayputpressureonMBSpricesand/orreducetheexcessspreadenjoyedbyMBSinvestors." CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP23graphically.PurchasemortgagesthatwereoriginatedbyQuickenLoansaremuchmorelikelytoterminatewithinoneyearcomparedtopurchasemortgagesoriginatedbythetopthreebanklenders.Thisrelationshiphasbeentruesincethestartofthedatain2001andisnotanewphenomenon.Buchak,Matvos,Piskorski,andSeru(2018)alsonotedtheincreasedprepaymentsofloansoriginatedbytheseIMBs.Basedonthisinstitutionaldetail,thispaperconstructsashift-shareinstrumentinthespiritof(Bartik,1991;Blanchard,Katz,Hall,andEichengreen,1992).Foreachc
42 ountyc,thepredictedannual(t)IMBmarketsha
ountyc,thepredictedannual(t)IMBmarketshareforpurchasemortgagesiscalculatedusingthepre-existingIMBmarketshareforthecountyinteractedwithnationalgrowthratesinIMBshare(excludingcountyc):Avg%IMBc;t1%IMBc;t(1.7)Thepre-existingshares(Avg%IMBc;t1)exhibitstrongcross-sectionalvariation.26Thesesharesareinteractedwith1-yeargrowthratesinnationalIMBshare(excludingcountyc)toconstructaBartikstyleinstrument,whichisbasedonthepremisethatnationalshockstoIMBmarketsharewillaectagivencountymorewhentheirpre-existingIMBmarketshareislarger.IfagreaterpredictedshareofrecentpurchasemortgagesinacountyareoriginatedbyIMBs,therewillbegreaterrenancingactivitywithinthecountyduetothosehouseholdsreceivingaggressivemarketingtorenance.Therefore,thiswillprovideexogenousvariationinrenancingactivity.Forrobustness,asecondversionofthisinstrumentbasedonmanyexogenousshocksfromeachlargeIMBisalsoimplemented.Additionalrobustnesstestsareshownattheendofthissectiontoruleoutthatwithin-countychangesintheinstrumentstemfromreasonsrelatedtolocalbusinessactivity.Asasecondconcerntoestimatingthecausalimpactofrenancingactivityonbusinessformationrates,increasesinrenancingactivitymayleadtoanincreaseinbusinessformationbecauseofalocaldemandchannel.Householdsinaggregateextracthomeequity,notonlyentrepreneurs,whichmayleadtoaconfoundinglocalconsumptionshock.Anincreaseinlocalconsumptionmightinturnleadtoincreasedbusinessformationinindustriesreliantonlocaldemand.Tofurtherisolateacausallinkdrivenbythecollateralchannel,industriesthatrelyonlocaldemandareremovedfromthecalculationofbusinessformationgrowthrates.First,industriesthatarenon-tradeablebasedonMianandSu(2014)areremoved|retailtradeandaccommodation/foodservices(NAICS44,45,and72).Second,rmsintheconstruction,nance,insurance,andrealestateindustries(NAICS52and53)arealsoremoved(Adelinoetal.,2015).PubliclyavailabledatafromtheCensusStatisticsofUSBusinesses(SUSB)areusedto
43 calculatemeasuresoflocalestablishmentcre
calculatemeasuresoflocalestablishmentcreation.TheSUSBisanannualdatasetthatmeasuresthenumberofnewestablishmentscreatedatthecountyby2-digitNAICSindustry 26Figure3intheOnlineAppendixshowsAvg%IMBc;tforLosAngeles,Broward,andSuolkcounties.EachcountyexperiencesdieringgrowthratesinIMBshareovertime. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP24level.Toestimatethecausaleect,weighted2SLSareestimatedasfollows:ln(#NewEstablishmentsc;t+1)=+\ln(#Extractedc;t+1)+!c+t+1+Zc;t+c;t+1(1.8)where!candt+1arecountyandannualtimexedeects,respectively,whichareincludedtoisolatetheeectwithin-countywhilecontrollingforcommonannualshocks.Zc;tisavectoroftime-varyingcounty-levelnon-parametricgrowthratecontrols(betweent1andt),including:homeprices,numberofhomepurchases,unemploymentrate,smallbusinessloanvolume,andthetotalnumberofestablishments.27Thesamplecovers1,493counties(c)forthe8-yearperiodbetween2009and2016.tstartsin2009inordertofocusonthepost-crisisperiodduringwhichrenancingandbusinessformationrateshavestagnated.Standarderrorsareclusteredatthecounty-levelandtheregressionsareweightedbythecounty'spopulationfromthe2010Census.Resultsincolumns1through7ofTable1.17arebasedongrowthratesofbusinessformationthatexcludeindustriesreliantonlocaldemand.Columns1and2showthatgreaterrenancingactivityisassociatedwithgreaterbusinessentry(column2addsthevectoroftimevaryingcounty-levelcontrolsZc;t)usingWLS.Totestthestrengthoftheinstrument,column3estimatestherststage:ln(#Extractedc;t+1)=+Avg%IMBc;t1%IMBc;t+!c+t+1+Zc;t+c;t+1(1.9)Therststageisolatesvariationwithincountyforcash-outrenancingactivitybasedonpredictedIMBmarketshare,whilecontrollingfornationalshocks.IfIMBmarketsharepositivelyaectscash-outrenancingactivitywithincounty,willbepositiveandsigni-cant.Column3ndsthata1pointinc
44 reaseinIMBshareincreasesrenancingac
reaseinIMBshareincreasesrenancingactivityby0.25points.Theinstrumentissignicantatthe1%levelandproducesanF-statisticover10,indicatingthatitisnotaweakinstrument.Column4estimatesthesecondstage.Thecoecientfromweighted2SLSislargerthanthecoecientestimatedwithWLS,likelyaresultofthemeasurementerrordiscussedearlier.A10%increaseinrenancingactivitycausesanincreaseinbusinessentrygrowthratesof2.5%.Forrobustness,column5showsthatthecoecientissimilarwhentheexplanatoryvariableofgrowthinthenumberofhouseholdsextractinghomeequityisreplacedwithgrowthinthedollaramountofhomeequityextracted.28Ifthedependentvariableisreplacedwithgrowthinestablishmententryrates(theratioofthenumberofenteringestablishmentstothepopulationofestablishments)theresultisunaected(column6).Toshowthatthe 27County-levelhomepricedataarefromFHFA.VolumeofhomepurchasesdataarecalculatedfromHMDA.UnemploymentratedataarefromBLSLAU.SmallbusinessloanvolumedataarefromtheCRAandrestrictedtoloanamounts$100,000.DataonthetotalnumberofestablishmentsisfromtheCensusSUSB.28Growthinthedollaramountofhomeequityextractedmixeseectsfromchangesinthenumberofhouseholdsextractinghomeequityandtheamounthouseholdsextract. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP25resultgeneralizes,thecoecientisshowntobelargelyunaectedwhenonlybusinessesinthenon-tradeablesectorsareexcluded(column8)andwhenbusinessfromeveryindustryareincluded(column9).Sincethecoecientisstablewhentheregressionmodelisestimatedforallindustriesandalsoforindustriesthatarenotreliantonlocaldemand,theeectisfromthecollateralchannel(asopposedtothedemandchannelfromanincreaseindemandforbusinesses).Thecoecient'smagnitudeislarge.Aonestandarddeviationshocktocash-outre-nancingactivity(0.53)wouldcauseanincreaseinbusinessentrygrowthratesof11%.This53%shockwouldincreaseannualcash-outrenancingactivityby$65billion,whichwouldleadtoarecoveryincash-outre
45 ;nancingvolumetolevelsseenin2002,aperiod
;nancingvolumetolevelsseenin2002,aperiodofnormallendingstandardspriortothehousingbubble.Althoughtherecoveryinrenancingactivityismodest,the11%increaseinbusinessentrygrowthratesfromthisshockwouldrecoverone-thirdofthedeclineinbusinessentryratesexperiencedsince2006.Apossibleconcernwiththeinstrumentisthatashifttowardsnon-banksformortgagelendingmightincreasesmallbusinesslendingbybanks.Ifbanksholdtheiroriginatedmort-gagesonbalancesheetanddonotsecuritizethemortgagesthenbankmortgageoriginationscanpotentiallyreducenon-mortgagelending.AsIMBsincreasetheirmarketshareofmort-gages,banksmayasaresulthaveincreasedlendingcapacityfornon-mortgageloansandsubsequentlyoriginatemorebusinessloans.Thisinturnmaystimulatebusinessformationrates.Giventheextenttowhichbankssecuritizetheirmortgageoriginations,thisislikelynotaconcern.TotestthatsmallbusinesslendingisnotaectedbyIMBshare,theeectoftheinstrumenton1-yeargrowthratesinsmallbusinesslendingisestimatedusingWLS(countyandtimexedeectsandZc;t[excludinglaggedsmallbusinesslendingvolume]areincluded).Theregressionshowsthattheinstrumentisnotcorrelatedwithchangesinsmallbusinesslendingvolume(column1,Table1.18).Anotherconcernisifwithin-countychangesinIMBmarketsharearecorrelatedwithchangesintheunemploymentrate,shareofresidentswhoarewhite,orshareofresidentswithlessthanahighschooleducation.Buchaketal.(2018)ndthatIMBshareishigherinthecross-sectionforcountieswithlargervaluesforthesevariables.Theirndingiscross-sectional,althoughitispossiblethatIMBsharewithinacountyevolveswithchangesinthesevariables,whichwouldsubsequentlyaectbusinessentrygrowthrates.Toshowthatthisisnotaconcern,aWLSregressionisestimatedtotestif1-yearcounty-levelgrowthratesinthesevariablesforecastpredictedIMBshare.Columns2to4,Table1.18showthatthereisnorelationshipbetweenpredictedIMBshareandchangesinthesevariables,therebyalleviatingthisconcern.Therobustnesstestsattemptedtoruleoutpotentialissueswiththein
46 strument.Asanadditionaltest,theinstrumen
strument.Asanadditionaltest,theinstrumentistransformedtoexploitexogenousvariationfromtheshocksincaseendogeneityremainsaconcern.TheinstrumentusesnationalshockstooverallIMBmarketshare,excludingcountyc,sooneexogenousshockisutilizedforeachcountyc.Withmanyexogenousshocks,theestimatedeectforisconsistentevenifthesharesareendogenous(Borusyak,Hull,andJaravel,2018;Goldsmith-Pinkham,Sorkin,andSwift,2018).Totransformtheinstrumenttomanyexogenousshocks,foreachcountyc,national CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP26growthratesinmarketshareforeachIMB(excludingcountyc)arecalculated.OnlylargeIMBs,thosethatoperateinatleast100counties,areincludedtominimizethechancethatgrowthratesinIMBmarketsharearecorrelatedwithlocaleconomicactivity.29EachIMBmarketsharegrowthrateisweightedbytheIMB'sex-anteshareoftotalpurchasemortgageoriginationvolumeincountycinyeart1:XjfIMBc;t1g!c;t1%jc;t,where!c;t1=$PurchaseOriginationjc;t1 $PurchaseOriginationc;t1(1.10)wherethesumisindexedoverthesetofalllargeIMBsthatoperateincountycinyeart1and%jc;tisthenationalgrowthrateofIMBj'smarketshareexcludingcountycbetweenyearst1andt.Eveninthepresenceofendogenouslaggedshares,themanyexogenousshockswillallowforaconsistentestimationof.Withthisinstrument,theestimatedcoecientdoesnotnoticeablychange(column7,Table1.17).Inresultsavailableuponrequest,thestandarderrorapproachfromAdao,Kolesar,andMorales(2019),whichaccountsforcorrelationacrosscountieswithsimilarIMBshares,isshowntonotreducethesignicanceoftheresult.1.5ConclusionUsingnoveldata,thispaperprovidesevidencethatbothentrantandcontinuingsmallbusi-nessesarenegativelyimpactedbycreditconstraints.Atentry,positivecreditshockscausesmallbusinessestostartlarger,growfaster,permanentlyremainlarger,andhaveagreaterchanceofsurvival.Similarly,continuingbusinessesimmediatelyrespondwhentheircreditconstraintsareloosenedandperman
47 entlyremainlarger.Creditconstraintshavea
entlyremainlarger.Creditconstraintshaveaper-manenteectonsmallbusinesses,highlightingtheeconomicbenetsofalleviatingcreditconstraintsearlierinarm'slife.Intermsofdirectandimmediateemploymenteects,oneadditionaljobiscreatedfromapositivecreditshockof$88,000toentrantrmsand$153,000tocontinuingrms.Toalleviatethesecreditconstraints,businessesownershavehistoricallyheavilyreliedontheirpersonalhousingequity.Inrecentyears,thischannelofcredithasalmostentirelydisappeared,withonlythemostcreditworthyentrepreneursabletotapintotheirpersonalhomeequity.Atighteningofcreditstandardssince2008hasbeenanimportantreasonforthelackofrecoveryinbusinessformationratessincethe2008nancialcrisis.Whilehomepriceshaverecovered,personalhomeequityextractionhasnotrecoveredduetomorestringentlendingstandards,whichinturnhasledtofewersmallbusinessesbeingstarted.Theresultsinthispapershedlightonanotheravenuethroughwhichdisruptionsinlendingsincethe2008nancialcrisishaveaectedtheeconomy.Whilerestrictionsonlending 29Inthesimplestofcases,ifanIMBoperatesinonecountythengrowthratesinthatIMB'ssharewillberelatedtolocaleconomicconditions.AsthenumberofcountiesthattheIMBoperatesinincreases,thesmallertherelationthatgrowthrateswillhavetolocaleconomicconditions. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP27tobusinessesandhouseholdshavebeenstudiedseparately,theyhavenotbeenstudiedjointly.Thestrongandpreviouslyunderappreciatedrolethatmortgagecredithadonsmallbusinessformationpriortothe2008crisishighlightstheneedtounderstandtheimplicationsthatmortgagecreditaccesshasonbusinessformation.Whiletherehasbeenadeclineinrmentryratesfordecades,thedeclinesince2006wasperhapsthesharpestinrecenthistory.Thispapershowsthatatighteningofmortgagecreditavailabilitytopotentialentrepreneursisanimportantreasonforthisdecline.Arecoveryofcash-outrenancingactivitytoitslevelin2002wouldrecoverone-thirdofthedeclineinrmentryr
48 atessince2006.Whiletightermortgagestanda
atessince2006.WhiletightermortgagestandardsintheaftermathoftheGreatRecessionareoftenviewedasnetpositive,thenegativeexternalitythatthiscanhaveonthecreation,size,andstrengthofsmallbusinessesisoverlooked. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP281.6FiguresandTables Figure1.1:HomeEquityExtractionandFirmEntryRatesThesolidlineshowsvolumeofhomeequityextractedbyyearbetween2001and2016,constructedfromATTOM(inbillionsofdollarsadjustedto2016prices).ThedashedlinereportsrmentryratesfromtheCensusBusinessDynamicStatisticsdata.Firmentryrateiscalculatedastheratioofthenumberofprivatesectorrmscreatedinagivenyeartothenumberofallactiveprivatesectorrmsintherespectiveyear(PugsleyandS,ahin,2018). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP29 Figure1.2:ShareofSmallBusinessesFoundedwithHomeEquityFundingByRegionTheshareofentrantsmallbusinessfundedbypersonalhomeequitybyyearofformation.ThedataareconstructedfromamergeofATTOMtoNETS.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Abusinessisclassiedasbeingfundedbyhomeequityiftheownerextractsover$5,000ofhomeequityintheyearthatthebusinessiscreatedortheprioryear.RegionsfollowtheCensusregionclassication.Therawunderlyingdataonlyincludesbusinessownerswhoownahome.Tocorrectforthis,thetimeseriesareadjustedbythehomeownershiprateofbusinessownersbasedonthepopulationofbusinessownersbyregionandyearfromtheAmericanCommunitySurveymicro-data. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP30 Figure1.3:IdenticationSetupIllustrationofidenticationsetupbasedonlocationmatching. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP31 (a) (b)Figure1.4:IdenticationIllustrationFiguresAandBshowtheidealset-upformeasuringexogenouscreditshockstoentrantandcontinuingbusinesses,respectively. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP32
49 Figure1.5:DierenceinDistancefromHo
Figure1.5:DierenceinDistancefromHometoFirmBetweenTreatedandControlDensityplotofthelndierencebetweenentranttreatedandcontrolrmsforthedistancebetweenarmowner'shomeandrm.TheentrantmatchedpairsforthisplotareconstructedfromtheATTOMtoNETSmergedsamplewiththesamecriteriatoformmatchedpairsfromthesamplemergedtoLBDforentrantrms.Themeandierenceis-0.0338,withastandarddeviationof1.191. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP33 Figure1.6:DierenceinPropensitytoRenanceWithaTraditionalBankBetweenTreatedandControlDensityplotofthedierencebetweentheentranttreatedandcontrolrmowners'propensitytooriginatetheircash-outrenancemortgagethroughatraditionalbank(=1iftraditionalbank).TheentrantmatchedpairsforthisplotareconstructedfromtheATTOMtoNETSmergedsamplewiththesamecriteriatoformmatchedpairsfromthesamplemergedtoLBDforentrantrms.Themeandierenceis0.007,withastandarddeviationof0.671. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP34 Figure1.7:HomePriceGrowthExampleQuarterlyalltransaction(purchaseandrenance)homepricegrowthbetweenJanuary2000andDe-cember2018forCalifornia,Florida,Texas,Colorado,andnationally.DataarefromFRED.Valuesarenormalizedto100inJanuary2009. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP35 Figure1.8:LoanTerminationRatesProbabilityofloanterminationwithinoneyearoforigination.DataarefromamergeofloanlevelpublicuseFreddieMacandFannieMaedatatoATTOM.Thedataarerestrictedtopurchasemortgageswithanoriginationbalancebetween$100kand$400k. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP36Table1.1:SummaryStatisticsforLBD SmallBusinessesinLBDAllMergedtoAllDatasetsDidNotMergeNMeanSDNMeanSDNMeanSD InitialEmployment5,656,0002.372.247461,0002.5732.2285,195,0002.3522.248InitialPayroll($000s)5,656,00057.7587.07461,00078.51102.45,195,00055.9185.34SurvivedtoYear55,656,0000.4888-461,0
50 000.6386-5,195,0000.4755- Summarystatist
000.6386-5,195,0000.4755- SummarystatisticsforsmallbusinessesintherestricteduseLongitudinalBusinessDatabase(LBD)thatwerefoundedbetween2001and2011.Note,thisonlyincludesemployerbusinesses.Duetorestrictionsondisclosure,onlyroundedsamplecounts,mean,andstandarddeviationareshownforselectvariables.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Abusinessiscountedasmergedtoalldatasetsifitwassuccessfullymergedto:theSSEL,NETS,ATTOM,andzipcodeleveldatafromZillowandthe2000decennialCensus. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP37 Table1.2:SummaryStatisticsofEntrantBusinessFinancingSources Summarystatisticsonsourcesofstart-upnancingfrom:2007SurveyofBusinessOwners(SBO),the2013-2014,2014-2015,and2015-2016wavesoftheAnnualSurveyofEntrepreneurs(ASE),the2004KaumanFamilySurveybasedonresultsinRobbandRobinson(2014),andtheNETS-ATTOMmergeddatasetutilizedinthispaper.The2007SBOincludesbusinessesthatsurvivedthrough2007.TheASEisbasedonsurveysin2014,2015,and2016andprovidesstatisticsforbusinessesfoundedwithinthetwoyearperiodbeforethesurvey.TheSBOdataarerestrictedtormswithbetween1and25employees,theamountofstartupcapitalreported,andwithownerswhofoundedthebusiness.Tabulationweightsareused.FortheASE,theserestrictionsarenotpossibleasamicro-datasetisnotavailable.TheKaumanFirmSurveyisbasedonrmsofallsizes,butisprimarilycomprisedofrmswithtenorfeweremployees.RobbandRobinson(2014)report16%utilizepersonalbankloans(whichincludeshomeequity)asinitialfundingcapital.Thisstatisticrestrictstormsthatdonothavemissingsurveyresultsoverthe2004to2007period(orwereknowntohavegoneoutofbusiness).TheNETS-ATTOMmergeddatasetutilizedinthispaperisrestrictedtobusinesseswithtenorfeweremployeesinitiallyandtobusinessownerswhoownahome.Tocorrectforthislastrestriction,theratiosfromtheNETS-ATTOMpopulationareadjustedbythehomeownershiprateforthepopulationofbusinessownerseachyear
51 between2003and2007usingmicro-datafromthe
between2003and2007usingmicro-datafromtheAmericanCommunitySurvey(ACS). CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP38Table1.3:SummaryStatisticsBetweenTreatedandControlMembersforEntrantBusinessMatching (1)(2)(3)NMeanSD LNDierenceinHomeValueatFirstYear-32,500-0.015650.261LNDierenceinCLTVatPurchase2,5000.0019520.05182LNDierenceinNumberofMonthsfromPurchaseUntilFirstYearofFirm2,5000.0038150.5278LNDierenceinMedianHomeZipCodeIncome2,500-0.070320.3436DierenceinHomeZipCodePercentofResidentsWhoAreWhite2,500-0.020460.205DierenceinHomeZipCodePercentofHouseholdsBelowPovertyLine2,5000.011420.0714DierenceinHomeZipCodePercentofHouseholdsWhoRent2,5000.027110.2038 RatioofLagged3-YearHomeZipCodeHPIGrowth2,5000.065250.06252 Summarystatisticsatthepairlevelforentrantsmallbusinesspairscreatedbasedontheentrantsmallbusinessmatchingalgorithm.Pairsareformedfromexactmatchesonthezipcodeofthebusiness,thezipcodesofthehomesbeingdierentfromeachother(andthebusinesses),SICdivision,andyearofbusinesscreation.Thepairsarefurtherrestrictedtocaseswherethehomevalues(asmeasuredthreeyearspriortobusinesscreationyear)arewithin20%or$100k(forlessvaluablehomes)ofeachotherandtheCLTVratiosatpurchasearewithin20bpsofeachother.Thetreatedrmistheonethatexperiencedgreaterhomepricegrowth.Foreachtreatedrm,acontrolrmthathasthemostsimilarhomevalueasthetreatedrmisselectedafterthematchingexercise.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Firmsinthesamplewerefoundedbetween2001and2011.Roundedsamplesize,mean,andstandarddeviationarereportedforthedierencesinvariablesbetweenthetreatedandcontrolmemberofeachpair. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP39Table1.4:SummaryStatisticsofEntrantBusinessMatchedSample (1)(2)(3)NMeanSD InitialEmployment4,6002.7782.32InitialPayroll($000s)4,600$94.72$125SurvivedtoYear14,6000.9494-
52 SurvivedtoYear34,6000.7899-SurvivedtoYea
SurvivedtoYear34,6000.7899-SurvivedtoYear54,6000.6732-AmountofHomeEquityExtracted4,600$100,800$102,200HomeZipCode2000MedianFamilyIncome4,600$64,33021330HomeZipCodePercentWhite4,6000.74740.1833HomeZipCodePercentBelowPovertyLine4,6000.087520.06056HomeZipCodePercentRenter4,6000.30930.1564CLTVatPurchase4,6000.87160.1648HomeValueatPurchase($000s)4,600$206.8122.13-yearHPIGrowthBeforeFirmCreation4,6001.5070.2728#MonthsBetweenHomePurchaseandFirmCreation4,60076.8734.18 Summarystatisticsatthebusinesslevelforentrantsmallbusinessthatwerematchedtoanotherbusiness.Thematchedentrantrmsampleisconstructedassmallbusinessesthatwerefundedwithhomeequityatentry,wheretheentrepreneurlivesinadierentzipcodefromthebusiness,andwherethehomewaspurchasedatleastthreeyearspriortothermentryyear.Pairsareformedfromexactmatchesonthezipcodeofthebusiness,thezipcodesofthehomesbeingdierentfromeachother(andthebusinesses),SICdivision,andyearofbusinesscreation.Thepairsarefurtherrestrictedtocaseswherethehomevalues(asmeasuredthreeyearspriortobusinesscreationyear)arewithin20%or$100k(forlessvaluablehomes)ofeachotherandtheCLTVratiosatpurchasearewithin20bpsofeachother.Thetreatedrmistheonethatexperiencedgreaterhomepricegrowth.Foreachtreatedrm,acontrolrmthathasthemostsimilarhomevalueasthetreatedrmisselectedafterthematchingexercise.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Firmsinthesamplewerefoundedbetween2001and2011andeachrmisonlyincludedonceevenifmatchedmultipletimes.Roundedsamplesize,mean,andstandarddeviationarereported. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP40 Table1.5:EectofCreditAccessonEntrantBusinessSurvival (1)(2)(3)(4)(5)(6)(7)(8)Survival5-year5-year5-year5-year5-year5-year1-year3-yearOLS,OnlyPairFEOLS1stStage2SLS2SLS2SLS2SLS2SLS ln($AmountExtractedi;t)0.05510.05320.5130.5130.5130.20
53 20.712(3.02)(2.89)(2.19)
20.712(3.02)(2.89)(2.19)(1.88)(2.24)(1.73)(2.69)3ln(ZipCodeHomePricei;t)0.882(3.21) #Obs51005100510051005100510051005100R-squared0.5250.5530.6550.30.30.30.332-0.187F-statistic--10.32----- Clusterrmzip*SICdivision?YYYYNNYYClustercommutingzone*SICdivision?NNNNYNNNClusterrmzip*SICdivision*rmentryyear?NNNNNYNN tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifentrantsmallbusinesseshavedierentialsurvivaloutcomesbasedoninitialcreditaccess:I(Survivedi;j;t+k)=+\ln($AmountExtractedi;t)+ Xj+!Controlsi+i;j;t+kFirmsaredenotedbyi,rmcreationyearbyt,andthepairthatthermbelongstobyj.Xjarexedeectsforeachpair.TheregressionsarealinearprobabilitymodelwithI(Survivedi;j;t+k)equaltooneifthebusinesssurvivedatleastkyears.ln($AmountExtractedi;t)istheamountofhomeequitythattheentrepreneurextractedintheyearthatthebusinessiscreatedortheprioryear.Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsareestimatedontheentrantmatchedpairsampleandincludessmallbusinessesthatwereinitiallyfundedwithhomeequityandwerefoundedbetween2001and2011.Columns1and2estimatetheeecton5-yearsurvivalrateusingOLS(onlycontrollingforpairxedeectsincolumn1).Column3estimatestherststagewiththeinstrumentbeinglaggedlog3-yearhomezipcodelevelhomepricegrowthfromZillow.Columns4through6estimatethecausalimpacton5-yearsurvival.Forrobustness,column5clustersatthecommutingzonebySICdivisionlevelandcolumn6clustersatthermzipcodebySICdivisionb
54 yrmcreationyearlevel.Columns7and8es
yrmcreationyearlevel.Columns7and8estimatetheeecton1-yearand3-yearsurvival,respectively. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP41 Table1.6:EectofCreditAccessonEntrantBusinessEmployment (1)(2)(3)(4)(5)(6)ln(Employment)Year1Year2Year3Year4Year5Year52SLS2SLS2SLS2SLS2SLSOLS ln($AmountExtractedi;t)0.4881.3801.4581.4211.2410.111(1.84)(2.39)(2.24)(2.21)(2.00)(2.75) #Obs510043004300430043004300R-squared0.4-0.141-0.0790.05610.2230.57 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifentrantsmallbusinesseshavedierentialemploymentbasedoninitialcreditaccess:ln(Employmenti;j;t+k)=+\ln($AmountExtractedi;t)+ Xj+!Controlsi+i;j;t+kFirmsaredenotedbyi,rmcreationyearbyt,andthepairthatthermbelongstobyj.Xjarexedeectsforeachpair.Theoutcomevariablesarethelogofemploymentinyears1through5.ln($AmountExtractedi;t)istheamountofhomeequitythattheentrepreneurextractedintheyearthatthebusinessiscreatedortheprioryear.Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsareestimatedontheentrantmatchedpairsampleandincludessmallbusinessesthatwereinitiallyfundedwithhomeequityandwerefoundedbetween2001and2011.Foremploymentinyears2through5(columns2through6)thesampleisrestrictedtopairswherebothmembersneverhadmissingemployment/payrolldataintheLBDwithintheirrstveyears(ifarmisinmultiplepairsandoneofthepairsisexcludedthenalloftheirotherpairsareexcludedaswell).Columns1through5est
55 imatetheeectonemploymentinyears1thr
imatetheeectonemploymentinyears1through5,respectively,using2SLSwiththesecondstagereported.Column6estimatestheeectonyear5employmentusingOLS. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP42 Table1.7:EectofCreditAccessonEntrantBusinessEmploymentGrowthRates (1)(2)(3)(4)EmploymentGrowthYears1to2Years2to3Years3to4Years4to5 ln($AmountExtractedi;t)1.8090.9700.2070.231(2.18)(1.45)(0.50)(0.52) #Obs4300430043004300R-squared-0.2990.3980.7020.750 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifentrantsmallbusinesseshavedierentialemploymentgrowthratesbasedoninitialcreditaccess:Employmenti;j;t+k+1Employmenti;j;t+k :5(Employmenti;j;t+k+1+Employmenti;j;t+k)=+\ln($AmountExtractedi;t)+ Xj+!Controlsi+i;j;t+kFirmsaredenotedbyi,rmcreationyearbyt,andthepairthatthermbelongstobyj.Xjarexedeectsforeachpair.Theoutcomevariablesare1-yearemploymentgrowthratesforthebusiness'srstfouryears.ln($AmountExtractedi;t)istheamountofhomeequitythattheentrepreneurextractedintheyearthatthebusinessiscreatedortheprioryear.Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsareestimatedontheentrantmatchedpairsampleandincludessmallbusinessesthatwereinitiallyfundedwithhomeequityandwerefoundedbetween2001and2011.Thesampleisrestrictedtopairswherebothmembersneverhadmissingemployment/payrolldataintheLBDwithintheirrstveyears(ifarmisinmultiplepairsandoneofthepairsisexcludedthenalloftheirotherpairsareexcludedaswell).C
56 olumns1through4estimatetheeecton1-y
olumns1through4estimatetheeecton1-yearemploymentgrowthinforthebusinessesrstfouryears,respectively,using2SLSwiththesecondstagereported. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP43 Table1.8:EectofCreditAccessonEntrantBusinessPayroll (1)(2)(3)(4)(5)(6)ln(Payroll)Year1Year2Year3Year4Year5Year52SLS2SLS2SLS2SLS2SLSOLS ln($AmountExtractedi;t)0.5572.6853.6643.3062.9720.296(0.92)(2.13)(2.25)(2.13)(1.96)(2.98) #Obs510043004300430043004300R-squared0.5660.128-0.07520.1500.2780.586 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifentrantsmallbusinesseshavedierentialpayrollbasedoninitialcreditaccess:ln(Payrolli;j;t+k)=+\ln($AmountExtractedi;t)+ Xj+!Controlsi+i;j;t+kFirmsaredenotedbyi,rmcreationyearbyt,andthepairthatthermbelongstobyj.Xjarexedeectsforeachpair.Theoutcomevariablesarethelogofpayrollinyears1through5.ln($AmountExtractedi;t)istheamountofhomeequitythattheentrepreneurextractedintheyearthatthebusinessiscreatedortheprioryear.Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsareestimatedontheentrantmatchedpairsampleandincludessmallbusinessesthatwereinitiallyfundedwithhomeequityandwerefoundedbetween2001and2011.Forpayrollinyears2through5(columns2through6)thesampleisrestrictedtopairswherebothmembersneverhadmissingemployment/payrolldataintheLBDwithintheirrstveyears(ifarmisinmultiplepairsandoneofthepairsisexcludedthenalloftheirotherpairsareexcluded
57 aswell).Columns1through5estimatethee
aswell).Columns1through5estimatetheeectonpayrollinyears1through5,respectively,using2SLSwiththesecondstagereported.Column6estimatestheeectonyear5payrollusingOLS. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP44Table1.9:SummaryStatisticsComparingEntrantBusinessesbyInitialFundingSource (1)(2)(3)(4)(5)(6)HomeEquityFundedNotHomeEquityFundedNMeanSDNMeanSD InitialEmployment124,0002.6612.264337,0002.5412.213InitialPayroll($000s)124,000$81.04$102.9337,000$77.59$102.2SurvivedtoYear1124,0000.9453-337,0000.9378-SurvivedtoYear3124,0000.7703-337,0000.764-SurvivedtoYear5124,0000.6382-337,0000.6387-AmountofHomeEquityExtracted124,000$100,600$123,70033,7000--forInitialCapitalHomeZipCode2000MedianFamilyIncome124,000$65,470$21,660337,000$63,480$21,900HomeZipCodePercentWhite124,0000.79450.1761337,0000.78930.1871HomeZipCodePercentBelowPovertyLine124,0000.081090.05934337,0000.086640.06415HomeZipCodePercentRenter124,0000.28680.1561337,0000.29560.1621CLTVatPurchase124,0000.77680.3322337,0000.70060.3852HomeValueatPurchase($000s)124,000$277.5$223.7337,000$264.9$230.73-yearHPIGrowthBeforeFirmCreation124,0001.3620.296337,0001.1720.3339#MonthsBetweenHomePurchase124,00052.2141.5337,00059.5847.39andFirmCreation Summarystatisticscomparinghomeequityfundedbusinessestonon-homeequityfundedbusinesseswithinthesampleofsmallbusinessesfoundedbetween2001and2011intheLBDthatweremergedtotheotherdatasets.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Abusinessislabeledashomeequityfundediftheentrepreneurextractedatleast$5,000ofpersonalhomeequityintheyearthatthebusinessiscreatedortheprioryear.Duetorestrictionsondisclosure,onlyroundedsamplecounts,mean,andstandarddeviationareshownforselectvariables. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP45 Table1.10:ComparingDierencesinBusinessesandEntrepreneursByInitialFundingSource (1)(2)(3)(4)(5)(6)(7)ln(PurchaseValu
58 e)CLTVln(Employmentt)ln(Payrollt)1-year3
e)CLTVln(Employmentt)ln(Payrollt)1-year3-year5-yearSurvivalSurvivalSurvival I(HomeEquityFundedi)0.05350.03970.01610.009540.00220.00114-0.000473(22.23)(28.1)(2.1)(0.48)(0.64)(0.19)(-0.07)ln(PurchaseValuei)0.02950.1580.004670.009070.0120(5.69)(11.66)(2.01)(2.3)(2.73)CLTVi-0.0141-0.0582-0.00559-0.0171-0.0184(-1.37)(-2.17)(-1.2)(-2.16)(-2.06) #Obs461000461000461000461000461000461000461000R-squared0.670.450.7890.7840.7530.770.776 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifhomeequityfundedbusinessesdierfromnon-homeequityfundedbusinesses:Yi=+I(HomeEquityFundedi)+!Controlsi+iwhereiindexesabusiness.I(HomeEquityFundedi)isanindicatorvariable=1ifabusinessisinitiallyfundedwithhomeequity(entrepreneurextracted]TJ/;ø 9;.962; Tf; 18.;ʖ ; Td; [00;$5,000intheyearthatthebusinessiscreatedortheprioryear).ThepopulationisthemergedLBDtoATTOMdatasetandincludessmallbusinesses(startedwithtenorfeweremployeesandinitiallysingle-unitrms)foundedbetween2001and2011.OutcomevariablesarefromATTOMandtheLBD.Columns1and2usehousinginformationfromATTOMasoutcomevariablesandControlsiincludesxedeectsfortheinteractionofhomezipcodebyhomepurchaseyear,LFO,and2-digitSICindustry.Standarderrorsareclusteredatthehomezipcodelevel.Columns3through7usedataonthebusinessesfromLBDasoutcomevariablesandControlsiincludeslogpurchasevalue,CLTV,andxedeectsforthetripleinteractionof2-digitSICbycreationyearbyrmzipcodeandLFO.Standarderrorsareclusteredatthe2-digitSICbyrmzipcodelevel. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP46 Table1.11:ComparingDierencesinPersonalForeclosureOutcomeByInitialFundingSource (1)(2)(3)(4)2001-20112001-20042005-20072008-2011 I(HomeEquityFundedi)0.0156-0.01010.0636-0.0239
59 ;(1.59)(-0.22)(4.14)(-1.82) #Obs55188757
;(1.59)(-0.22)(4.14)(-1.82) #Obs55188757501158062334300R-squared0.8800.9240.8650.881OverallDefaultRate0.1930.1430.2480.177 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifhomeequityfundedbusinessesdierfromnon-homeequityfundedbusinessesinforeclosureprobability:I(Foreclosurei)=+I(HomeEquityFundedi)+!Controlsi+iwhereiindexesabusiness.I(HomeEquityFundedi)isanindicatorvariable=1ifabusinessisinitiallyfundedwithhomeequity(entrepreneurextracted]TJ/;ø 9;.962; Tf; 18.;ʖ ; Td; [00;$5,000intheyearthatthebusinessiscreatedortheprioryear).ThepopulationisthemergedNETStoATTOMdatasetthatisfurthermergedtoMcDashandincludessmallbusinesses(startedwithtenorfeweremployeesandinitiallysingle-unitrms)foundedbetween2001and2011.ForeclosureisestimatedfromATTOMandthedependentvariableequals1ifabusinessownerhasaforeclosureontheirhomewithinfouryearsofstartingthebusiness.Controlsiisvectorofcontrolsthatincludes:homepurchaseyear*rmcreationyear*homezipcode*rmzipcode*2-digitSICindustryFE,purchasevalue,combinedloantovalue(CLTV)atpurchase,initialrmemploymentFE,originalmortgageinterestrate,andFICOattimeofhomepurchase.Standarderrorsareclusteredatthehomezipcodelevel.Column1isbasedonthefull2001-2011period,whilecolumns2to4segmentintothepre-crisis(2001-2004rmentryyears),crisis(2005-2007rmentryyears),andpost-crisis(2008-2011rmentryyears)periods,respectively. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP47Table1.12:SummaryStatisticsofContinuingBusinessMatchedSample (1)(2)(3)NMeanSD InitialEmployment11,5001.7111.096InitialPayroll($000s)11,500$62.46$70SurvivedtoEventYear+311,5000.7581-AmountofHomeEquityExtractedinEventYear11,500$7,909$36,350HomeZipCode2000MedianFamilyIncome11,500$62,800$20,710HomeZipCodePercentWhite11,5000.76640.1921HomeZipCodePercentBelowPovertyLine11,5000.088690.06358HomeZipCodePercentRenter1
60 1,5000.30570.168CLTVatPurchase11,5000.73
1,5000.30570.168CLTVatPurchase11,5000.73340.3258HomeValueatPurchase($000s)11,500$227.8$152.93-yearHPIGrowthBeforeEventYear11,5001.0880.3848#MonthsBetweenHomePurchaseandFirmCreation11,50059.7744 Summarystatisticsof(unique)continuingsmallbusinessesinthematchedcontinuingrmsample.Thematchedcontinuingrmsampleisconstructedassmallbusinessesthathavesurvivedthrougheitheryear3or4(theeventyear),wheretheentrepreneurlivesinadierentzipcodefromthebusiness,andwherethehomewaspurchasedatleastthreeyearspriortotheeventyear.Pairsareformedfromexactmatchesonthezipcodeofthebusiness,thezipcodesofthehomesbeingdierentfromeachother(andthebusinesses),SICdivision,yearofbusinesscreation,andsameinitialemployment.Thepairsarerestrictedtocaseswherethehomevalues(asmeasuredthreeyearspriortotheeventyear)arewithin20%or$100k(forlessvaluablehomes)ofeachotherandemploymentoneyearpriortotheeventyeariswithinthreeemployeesofeachother.Withinthepair,thetreatedrmistheonethatexperiencedgreaterhomepricegrowthwithinthe3-yearperiodpriortotheeventyear.Thecontrolrmisselectedastheonethathasthemostsimilarhomevalueasthetreatedrm.Abusinesscanbeacontrolrmmultipletimes(andcanbebothacontrolandtreatedrmindierentpairs)butcanonlybeatreatedrmonceforeachofthetwoeventyears.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Duetorestrictionsondisclosure,onlyroundedsamplecounts,mean,andstandarddeviationareshownforselectvariables.Theamountofhomeequityextractedintheeventyearissetto0forthebusinessesthatdonotextracthomeequityintheeventyear. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP48Table1.13:SummaryStatisticsBetweenTreatedandControlMembersforContinuingBusi-nessMatching (1)(2)(3)NMeanSD LNDierenceinHomeValue8,900-0.0048870.2766LNDierenceinCLTVatPurchase8,900-0.0046170.2975LNDierenceinNumberofMonthsfromPurchaseUntilFirstYearofFirm8,900-0.0054211.082L
61 NDierenceinMedianHomeZipCodeIncome8
NDierenceinMedianHomeZipCodeIncome8,9000.01290.3745DierenceinHomeZipCodePercentofResidentsWhoAreWhite8,9000.00420.2238DierenceinHomeZipCodePercentofHouseholdsBelowPovertyLine8,9000.0018660.07808DierenceinHomeZipCodePercentofHouseholdsWhoRent8,9000.014980.2202DierenceinNumberofEmployeesinYearPriortoEventYear8,9000.0011291.413LNDierenceinPayrollinYear18,9000.015651.125 RatioofLagged3-YearHomeZipCodeHPIGrowth8,9000.10160.115 Summarystatisticscomparingdierencesinvariablesbetweenthetreatedandcontrolmembersofthematchedcontinuingrmsample.Thematchedcontinuingrmsampleisconstructedassmallbusinessesthathavesurvivedthrougheitheryear3or4(theeventyear),wheretheentrepreneurlivesinadierentzipcodefromthebusiness,andwherethehomewaspurchasedatleastthreeyearspriortotheeventyear.Pairsareformedfromexactmatchesonthezipcodeofthebusiness,thezipcodesofthehomesbeingdierentfromeachother(andthebusinesses),SICdivision,yearofbusinesscreation,andsameinitialemployment.Thepairsarerestrictedtocaseswherethehomevalues(asmeasuredthreeyearspriortotheeventyear)arewithin20%or$100k(forlessvaluablehomes)ofeachotherandemploymentoneyearpriortotheeventyeariswithinthreeemployeesofeachother.Withinthepair,thetreatedrmistheonethatexperiencedgreaterhomepricegrowthwithinthe3-yearperiodpriortotheeventyear.Thecontrolrmisselectedastheonethathasthemostsimilarhomevalueasthetreatedrm.Abusinesscanbeacontrolrmmultipletimes(andcanbebothacontrolandtreatedrmindierentpairs)butcanonlybeatreatedrmonceforeachofthetwoeventyears.Smallbusinessesaredenedashavingtenorfeweremployeesandbeingsingle-unitrmsatentry.Duetorestrictionsondisclosure,onlyroundedsamplecounts,mean,andstandarddeviationareshownforselectvariables. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP49 Table1.14:EectofCreditAccessonContinuingBusinessSurvival (1)(2)(3)(4)(5)(6)(7)Survivaltot+p+3O
62 LSOLS1stStage2SLS2SLS2SLS2SLS ln($Amount
LSOLS1stStage2SLS2SLS2SLS2SLS ln($AmountExtractedi;t+p)-0.000121-0.000168-0.005330.006690.006690.00669(-0.08)(-0.11)(-0.10)(0.13)(0.13)(0.12)3ln(ZipCodeHomePricei;t+p)1.162(3.76) #Obs17500175001750017500175001750017500R-squared0.5240.5320.5430.5230.5310.5310.531F-Statistic--14.11---- Includecontrolsforinitialpayroll,payroll/employmentgrowth?NYYNYYYClusteronFirmZip*SICDivision?YYYYYNNClusteronFirmZip*SICDivision*CreationYearNNNNNYNClusteronCommutingZone*SICDivisionNNNNNNY tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifcontinuingsmallbusinesseshavedierentialsurvivaloutcomesbasedontheamountofhomeequityextracted:I(Survivedi;j;t+p+3)=+\ln($AmountExtractedi;t+p)+ Xj+!Controlsi+i;j;t+p+3Firmsaredenotedbyi,rmcreationyearbyt,thepairthatthermbelongstobyj,pisthenumberofyearssincefoundingtthathomeequityisextractedforexpansion(eithertwoorthree,whichcorrespondstothermsofagethreeorfour).Xjarexedeectsforeachpair.TheregressionsarealinearprobabilitymodelwithI(Survivedi;j;t+p+3)equaltooneifthebusinesssurvivedthroughyeart+p+3.ln($AmountExtractedi;t+p)istheamountofhomeequitythattheentrepreneurextractedinyeart+p(setto0ifhomeequityisnotextracted).Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart+p3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,xedeectforifthebusinesswasinitiallyfundedwithhomeequity,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.Dependingontheregression,additionalcontrolsincludenon-parametricxedeectsestimatedby2-digitSICforinitialpayrollandemploymentandpayrollgrowthbetweenyears1and2.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsarees
63 timatedonthecontinuingrmmatchedpair
timatedonthecontinuingrmmatchedpairsampleandincludessmallbusinessesthatsurvivedthroughtheeventyeart+p.Columns1and2estimatetheeectusingOLS,column1doesnotincludethenon-parametriccontrolsforinitialpayrollandemployment/payrollgrowthbetweenyears1and2.Column3estimatestherststagewiththeinstrumentbeinglaggedlog3-yearhomezipcodelevelhomepricegrowthfromZillow.Columns4and5estimatethecausalimpactonsurvival,column4doesnotincludethenon-parametriccontrolsforinitialpayrollandemployment/payrollgrowthbetweenyears1and2.Forrobustness,columns6and7replicatecolumn5withclusteringatthermzipcodebySICdivisionbyrmyearlevelandcommutingzonebySICdivisionlevel,respectively. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP50 Table1.15:EectofCreditAccessonContinuingBusinessEmployment (1)(2)(3)(4)(5)(6)(7)ln(Employment)EmploymentGrowthRatest+p1t+pt+p+1t+p+2t+p+3t+p1tot+p+1t+p+1tot+p+3 ln($AmountExtractedi;t+p)-0.03830.1140.1730.1490.1590.2080.0576(-0.99)(2.10)(2.20)(1.70)(1.69)(1.73)(0.48) #Obs17500175001750017500175001750017500R-squared0.7270.5380.3420.4350.4120.3150.53 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifcontinuingsmallbusinesseshavedierentialemploymentoutcomesbasedontheamountofhomeequityextracted:ln(Employmenti;j;t+p+k)=+\ln($AmountExtractedi;t+p)+ Xj+!Controlsi+i;j;t+p+kFirmsaredenotedbyi,rmcreationyearbyt,thepairthatthermbelongstobyj,pisthenumberofyearssincefoundingtthathomeequityisextractedforexpansion(eithertwoorthree,whichcorrespondstothermsofagethreeorfour).Xjarexedeectsforeachpair.Theoutcomevariablesarelogemploymentkyearsaroundtheeventyeart+p,wherekrangesfrom-1to3.ln($AmountExtractedi;t+p)istheamountofhomeequitythattheentrepreneurextractedinyeart+p(setto0ifhomeequityisnotextracted).Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:l
64 ogofCLTVatpurchase,logofhomevalueatyeart
ogofCLTVatpurchase,logofhomevalueatyeart+p3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,xedeectforifthebusinesswasinitiallyfundedwithhomeequity,non-parametricxedeectsestimatedby2-digitSICforinitialpayroll,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.Exceptforcolumns1and6,whoseoutcomevariablesarebasedondatafromyeart+p1,controlsalsoincludenon-parametricxedeectsestimatedby2-digitSICforemploymentandpayrollgrowthbetweenyears1and2.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsareestimatedonthecontinuingrmmatchedpairsampleandincludessmallbusinessesthatsurvivedthroughtheeventyeart+p.Columns1through5estimatethecausalimpactoftheamountofhomeequityextractedonlogemploymentfromyearst+p1throught+p+3,respectively.Columns6and7estimatetheimpactonemploymentgrowthfromyearst+p1tot+p+1andt+p+1tot+p+3,respectively. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP51 Table1.16:EectofCreditAccessonContinuingBusinessPayroll (1)(2)(3)(4)(5)ln(Payroll)t+p1t+pt+p+1t+p+2t+p+3 ln($AmountExtractedi;t+p)-0.1110.2070.3000.2180.202(-1.42)(2.02)(1.63)(1.01)(0.81) #Obs1750017500175001750017500R-squared0.7870.6850.5220.5710.559 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingifcontinuingsmallbusinesseshavedierentialpayrolloutcomesbasedontheamountofhomeequityextracted:ln(Payrolli;j;t+p+k)=+\ln($AmountExtractedi;t+p)+ Xj+!Controlsi+i;j;t+p+kFirmsaredenotedbyi,rmcreationyearbyt,thepairthatthermbelongstobyj,pisthenumberofyearssincefoundingtthathomeequityisextractedforexpansion(eithertwoorthree,whichcorrespondstothermsofagethreeorfour).Xjarexedeectsforeachpair.Theoutcomevariablesar
65 elogpayrollkyearsaroundtheeventyeart+p,w
elogpayrollkyearsaroundtheeventyeart+p,wherekrangesfrom-1to3.ln($AmountExtractedi;t+p)istheamountofhomeequitythattheentrepreneurextractedinyeart+p(setto0ifhomeequityisnotextracted).Controlsiisavectorofzipcodelevelandrmlevelcontrolsthatincludes:logofCLTVatpurchase,logofhomevalueatyeart+p3,2-digitSICindustryxedeects,thelogofthenumberofmonthsbetweenhomepurchaseandyeart,legalformoforganization(LFO)xedeects,xedeectforifthebusinesswasinitiallyfundedwithhomeequity,non-parametricxedeectsestimatedby2-digitSICforinitialpayroll,andnon-parametriccontrolsforhomezipcodecharacteristicsofpercentwhite,percentrenter,percentbelowpovertyline,andmedianincome.Exceptforcolumn1,whoseoutcomevariableisfromyeart+p1,controlsalsoincludenon-parametricxedeectsestimatedby2-digitSICforemploymentandpayrollgrowthbetweenyears1and2.StandarderrorsareclusteredatthermzipcodebySICdivisionlevel.Theregressionsareestimatedonthecontinuingrmmatchedpairsampleandincludessmallbusinessesthatsurvivedthroughtheeventyeart+p.Columns1through5estimatethecausalimpactoftheamountofhomeequityextractedonlogpayrollfromyearst+p1throught+p+3,respectively. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP52 Table1.17:ExtensiveMarginEectsofRenancingActivityonBusinessFormation (1)(2)(3)(4)(5)(6)(7)(8)(9)ExcludingNon-tradeable,Construction, Excluding AllIndustriesFinance,Insurance,RealEstate Non-tradeable Y=ln(EntryManyShocks WLSRatec;t+1)Instrument WLSWLS1stStage2SLS2SLS2SLS2SLS 2SLS 2SLS ln(#Extractedc;t+1)0.003830.01820.2470.2240.2360.2220.200(0.68)(3.26)(2.26)(2.23)(1.74)(2.30)(2.25)Avg%IMBc;t10.301%IMBc;t(3.18)ln($Amt0.197(Extractedc;t+1)(2.00) #Obs971992659173917391739173916891739174R-squared0.2690.3810.4370.2000.005580.2270.2160.2960.368F-stat--10.09------Zc;tFE?NYYYYYYYY
66 tstatisticsinparenthesesp0:10,&
tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingtheeectofcash-outrenancingactivityonbusinessentryinaggregate:ln(#NewEstablishmentsc;t+1)=+\ln(#Extractedc;t+1)+!c+t+1+Zc;t+c;t+1Countiesareindexedbyc(1,493counties)andyearsbyt(2009to2015).#NewEstablishmentsc;t+1arefromtheCensusSUSBdataset,whichprovidesannualestablishmententrycountsbycountyandindustry.#Extractedc;t+1iscalculatedfromATTOMandisacountylevelcountofthenumberofhouseholdswhoextractedpersonalhomeequityinyeart+1incountyc.Zc;tisavectorofnon-parametriccontrolsfortime-varyingzipcodelevellaggedgrowthrates(t1!t)of:homeprices,numberofhomepurchases,unemploymentrate,smallbusinessloanvolume,andthetotalnumberofestablishments.Regressionsareweightedbythecounty's2010Censuspopulation.Columns1-2useWLS,column3showstherststage,andcolumns4-9useweighted2SLS.Columns1-7estimatetheeectexcludingestablishmentsinthenon-tradeable,construction,nance,insurance,andrealestateindustries.Column4showsthebaselineeectforindustriesnotreliantonlocaldemand,column5estimatestheeectforgrowthinthedollaramountextracted(insteadof#ofhouseholdsextractinghomeequity),column6replacesthedependentvariablewithlngrowthinestablishmententryrates,andcolumn7usestheinstrumentwithmanyexogenousshocks.Column8estimatestheeectexcludingestablishmentsinnon-tradeableindustries.Column9includesallindustries. CHAPTER1.THELONG-RUNEFFECTSOFMORTGAGECREDITACCESSONENTREPRENEURSHIP53 Table1.18:RobustnessTestsforExtensiveMarginInstrument (1)(2)(3)(4)BusinessLoanVolumeUnemploymentRateShareofResidentsShareofResidentsWhoareWhiteWithoutaHighSchoolEducation Avg%IMBc;t1%IMBc;t0.04440.0364-0.002990.0111(1.45)(1.31)(-0.64)(0.34) #Obs9266926644804474R-squared0.5490.9240.5650.0739 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionestimatestestingiftheinstrumentiscorrelatedwithcoun
67 tylevelvariablesacrosstime:Yc;t=
tylevelvariablesacrosstime:Yc;t=+Avg%IMBc;t1%IMBc;t+!c+t+Zc;t+c;tCountiesareindexedbyc(1,493counties)andyearsbyt(2009to2015).Zc;tisavectorofnon-parametriccontrolsfortime-varyingzipcodelevellaggedgrowthrates(t1!t)of:homeprices,numberofhomepurchases,unemploymentrate(notincludedincolumn2),smallbusinessloanvolume(notincludedincolumn1),andthetotalnumberofestablishments.Regressionsareweightedbythecounty's2010Censuspopulation.Incolumn1,thedependentvariableisgrowthinsmallbusinessloanvolume(businessloansforlessthan$100k)betweentandt+1.Thisdependentvariableisbetweentandt+1sincethepotentialconcernisimpactsonbusinessformationbetweentandt+1duetobusinessloanfunding(whichiscontemporaneous).LoanvolumedataarefromtheCommunityReinvestmentAct(CRA)dataset.Incolumn2,thedependentvariableisgrowthintheunemploymentratebetweent1andtfromBLSLAU.Incolumns3and4,thedependentvariablesaregrowthintheshareofresidentswhoarewhite(andnon-Hispanic)andwhohavelessthanhighschooleducation,respectively,fromCensusACSbetweent1andt. 54Chapter2PersonalWealthShocksandInvestmentManagerOvercondence2.1IntroductionOneofthepredominatequestionsinnanceisifprofessionalinvestmentmanagersaddvalue.Giventhatactiveequitymutualfundsinvestinover$5trillionofassets,thisissuehaslargeeectsonbothreturnstoclientsandtheoveralleciencyofthemarket.Mostofthepriorliteraturehasfocusedonex-antestaticmeasuresthatareusedtodeterminewhichfundmanagershaveskill,andhaslargelyassumedrationality.Studiesofdeviationsfromrationalbehavior,suchasovercondence,havelargelybeenrestrictedtoretailinvestors(BarberandOdean,2001;Kumar,2009;GrinblattandKeloharju,2009).Thispaperinvestigatespositiveshockstothepersonalhousingwealthoffundmanagersasachanneloftime-varyingovercondenceformutualfundmanagers.Homepricesvarydramaticallywithinnarrowgeographicalareas,whichprovideslargeid-iosyncraticvariationinreturns.Iffundmanagersinferth
68 eirprofessionalskillfromreturnstotheirpe
eirprofessionalskillfromreturnstotheirpersonalhousingwealth,positiveshockstohousingwealthmayleadtoovercondencebeliefs.Localhomepricegrowthshouldhavenoeectonfundperformancebasedonratio-nalmodels.Surprisingly,astrongandrobustrelationshipisfoundbetweenpositivelaggedchangesinzipcodelevelhousingpricesandforecastsofriskadjustedperformance.Thisalonedoesnotsignalovercondence.Todisentanglethechanneldrivingtheanomaly,thetradingbehavioroffundmanagersisanalyzed.Astunningpatternemergesfromstudyinghowfundmanagertradingchangesashomepriceshocksevolve.Followingpositivehomepriceshocks,fundmanagersbecomemorelikelytoleveruponexistingstockholdingsthathaveperformedpoorly,shifttheirbuyingtowardsleveringuponcurrentlyheldstocksandawayfromselectingnewstocks,makeworsechoicesinchoosingexistingpositionstoliquidate,andbecomelesslikelytofullyliquidateexistingpositions.Interestingly,fundmanagersdonotmakeworsedecisionsinselectingstockstobuythatarenotcurrentlyheldintheirportfolioafterapositivehomepriceshock.Instead,theybecome CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE55biasedtowardstheirpastchoicesandbecomemorelikelytobelievetheirpoorpaststockpickswillbecomewinners.Akepanidtaworn,Mascio,Imas,andSchmidt(2018)ndthatfundmanagersunderperformbecauseofpoorsellingchoices.Theovercondencechannelshowninthispapermakesthesellingchoicesoffundmanagersevenworse.Inadditiontohavingashort-runimpactontrading,overcondencealsohasalong-runeect.Toshowthis,thefund'sownex-anteportfolioholdingsareusedasabenchmark,similartotheapproachinBarberandOdean(2001)forretailinvestors.Ashomepricegrowthpositivelyaccelerates,thefund'sactivereturndecreasesandbecomesworsethantheirpassiveportfolio.Additionally,tradingcostsincrease,whichispartiallytoblameforthedeclineinperformance.Totestthisfurther,thefund'sannualturnoverisanalyzedandfoundtoincreaseashomepricegrowthincreases|acommonsignofovercondence.Tofurthershowthattheresul
69 tisdrivenbyovercondence,thispaperbo
tisdrivenbyovercondence,thispaperborrowsfromtheretailinvestorandcorporatenanceliteratures.Theeectisstrongestamonggroupsoffundmanagerswhoaremorelikelytobesusceptibletoovercondencebeliefs.Thisincludesfundmanagerswhoaremale(BarberandOdean,2001),lessexperienced(GreenwoodandNagel,2009;Chernenko,Hanson,andSunderam,2016),havegreaterhousingleverage(Cronqvist,Makhija,andYonker,2012;LiuandYermack,2012),whoboughttheirhomemorerecently,andhavelesseducation.Pastresearchhasshownthatfund owsaectreturns(BerkandGreen,2004;Song,2019)andthatinvestorsinvestlocally(CovalandMoskowitz,1999;IvkovicandWeisbenner,2005;Pool,Stoman,andYonker,2012).Toruleoutthatthedeclineinperformanceisdrivenbylocalincreasesinfund owscausedbylocalhousingwealth,threetestsareperformed.First,itisshownthatfund owsarenotaectedbyhomepriceshocks.Second, owsareincludedasanadditionalcontroltotheperformanceregression,whichisfoundtonotaecttheresults.Lastly,pastperformanceisaddedasanadditionalcontroltotheperformanceregressionandisalsofoundtonotaecttherelationshipbetweenpositivehomepriceshocksandfuturefundperformance.Thistestistoruleoutthatfuturereturnsaredecreasingduetostrongpriorreturns.Iflaggedperformancewasstrong,fund owsmightincrease,whichsubsequentlywoulddecreasereturns.Toruleoutaspuriousrelationship,threeplacebotestsareperformed.First,itisshownthatindexfundsdonotrespondtothesehomepriceshocks(Pastor,Stambaugh,andTaylor,2017).Inadditiontoindexfunds,fundswithalowtrackingerrororActiveShareshouldalsonotrespondtohomepriceshocks.Thelogicbeingthatifafundmanagerisaclosetindexer,theywillmaintainthesamepassivestrategybeforeandaftertheshock.Itisfoundthatbothdenitionsofclosetindexersdonotrespondviathischannel,andtheeectisstrongerfortrulyactivefundmanagers.Thesetestsruleoutthepossibilitythattheresultisdrivenbyanunobservedlocalfactorthatin uencesfundreturns.Oneconcernwiththeresultsinthispaperisthatfundmanagersmi
70 ghtnotbepayingattentiontochangesintheirh
ghtnotbepayingattentiontochangesintheirhomevalue.Toprovideevidenceagainstthisconcern,fundmanagerswhoextractpersonalhomeequityviacashoutrenancingareshowntohaveadeclineinperformancerelativetotheirperformancepriortoextractinghomeequity.Cash-outrenancinghasalargeeectonthehouseholdbalancesheet(GreenspanandKennedy, CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE562008;BhuttaandKeys,2016;DeFusco,2018)andtheseresultsshowthatthereisalinktoprofessionalperformanceaswell.Withthisresult,homepricegrowthisusedasaninstrumentalvariableforthedecisiontoextracthomeequitytoshowthatfundmanagerswhoextracthomeequitydecidetorenancewhenhomepricesarehigher.Theuseoftheinstrumentalvariableistoprovideevidencethatfundmanagerspayattentiontotheirpersonalhomevalueandisnotintendedforuseasaninstrumentinthetraditionalsense.Thispaperprovidesresultsshowinghowprofessionalbehaviorcanbeaectedbyhomepriceshocks,focusingontheeectfrompositivehousingwealthshocks.ThisisrelatedtorecentworkbyStomanetal.(2018)andBernsteinetal.(2018),whostudytheimpactofnegativehousingwealthshocksformutualfundmanagersandinnovativeworkers,re-spectively.Stomanetal.(2018)ndthatinresponsetonegativehousingwealthshocks,fundmanagersdecreasetheirportfolioriskduetocareerconcerns.Theyndnoeectonrisk-adjustedreturnsandamarginallysignicanteectonrawreturns.Thispaperndsasimilareectonreturnsandrisktakingduringperiodsofnegativehomepriceshocks.Whenextendedtoabroadertime-series,theeectforchangesinriskduringperiodsofpositivehomepricegrowthsurvives,whiletheresultduringperiodsofnegativehomepricegrowthdoesnot.Thispaperbuildsontheliteratureforsortinginstitutionalinvestmentmanagersbyex-antemeasuresofskill.Mostofthepriorworkinthisbodyofresearchhasfocusedonstaticmeasures(Khorana,Servaes,andWedge,2007;Chen,Goldstein,andJiang,2008;Evans,2008;CremersandPetajisto,2009;Cremers,Driessen,Maenhout,andWeinbaum,2009;Gre
71 enwoodandNagel,2009;HongandKostovetsky,2
enwoodandNagel,2009;HongandKostovetsky,2012;Pool,Stoman,andYonker,2015;Chernenkoetal.,2016;CremersandPareek,2016).Recentwork,suchasGuptaandSachdeva(2017)andPastoretal.(2017),hasfounddynamicmeasuresthatforecastperformance.Thispaperfocusesonanoveldynamicmeasurethathasnotbeenexploredpreviously.Tostudyhousingwealthshocks,dataonthepersonalrealestateholdingsandtransactionsofmutualfundmanagersaremergedtodataontheirprofessionalportfolioholdingsandreturnsoftheirfundsforasampleofover1,200activelymanageddomesticequitymutualfundswithdatabetween2001and2018.Theremainderofthispaperisorganizedasfollows.Section2providesanoverviewonhomepricevariationandtheapproachusedinthispaper.Section3describesthedatasetsusedinthispaper,aswellashowtheyaremerged.Section4describestheempiricalmethodologyandresultsfortheeectonperformanceandrelatedtests.Section5studieshowtradingbehaviorisaectedbyhousingwealthshocks.Section6presentsevidencethatfundmanagerspayattentiontothevaluesoftheirhomes.Lastly,Section7concludes.2.2HomePriceVariationForthevastmajorityofhouseholds,thehomeisthesinglelargestinvestment(Campbell,2006).Fundmanagersareawealthiersubsetofthepopulation,buttheyownlargerhomesthantheaveragehouseholdandtheirhomevaluesdwarftheirannualsalary(Figure2.1). CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE57Theaveragehomevalueoffundmanagersisgenerallyover$800k,whiletheiraverageannualsalaryisbetween$200-300k.A20%idiosyncratichomepriceshockcanequalmorethanoneyearofsalary.Ifhomepricegrowthisabnormallylargethenfundmanagersmightusethisasfeedbackreinforcingabeliefintheirsuperiorabilitytopickassets.Atthesametime,thisrepresentsalargewealthshock.Whilethewealtheectcannotberuledout,testsshowingwhichtypesoffundmanagersrespondtotheseshocksandthechangeintheirtradingbehaviorfromtheseshockslendcredencetotheovercondencechannel.Totestthehypothesisthathomepriceshocksaectfundmanagerperformance,lagged3-yearzipcodelevelh
72 omepricechangesoftheprimaryhomeoffundman
omepricechangesoftheprimaryhomeoffundmanagersareregressedonforecastsofperformanceandinvestmentselections.1Inordertoobservetherealestatedataforfundmanagers,amergeisutilizedbetweenfundmanagersinMorningstarandrealestatetransactionandassessordatafromATTOM(discussedintheDatasection).Fundmanagersareincludedinthesamplewhenatleastthreeyearshaspassedsincethepurchaseoftheirhome.Thisrestrictionensuresthatfundmanagersrealizedthehomepricegains.2Theprimaryregressionmodelinthispaperutilizesfundxedeects(foreachmergednon-disjointperiodoftime)andmonthlytimexedeects.ItispossiblethatthemergebetweenMorningstarandATTOMproducesdisjointsamplesacrosstimeforafund.Forthesefunds,eachgroupofmergedoverlappingfundmanagersthatproducesanon-disjointsampleacrosstimeisseparatelystudied.Ifmorethanonefundmanagerismergedforagivenmonth,thehomepricegrowthofthefundmanagersisaveraged.Thesexedeectsremovetimeinvariantpreferencesoffundmanagers,includinghousinglocationchoice(therebycontrollingforsortingintohomezipcodeswithinageographicalarea).Withthesepreferencesremovedandaggregateeconomicconditionsabsorbedbythetimexedeects,theremainingvariationinhomepricegrowthisunlikelytobeendogenous.3Homepricegrowthismeasuredatthezipcodelevel.Thismicro-levelmeasurementofhomepricegrowthprovidesdramaticvariationwithinsmallgeographicalareas.Figure2.2showshomepricegrowthatthezipcodelevelfortheBostonarea,apopularregionformutualfundmanagers.Thevolatilityinhomepricegrowthwithinsmallgeographicalareasmakesitunlikelythatmicrolevelhomepricegrowthiscorrelatedwithinvestmentchoicesduetoanomittedvariableoflocaleconomicactivity.Tofurthershowthemicrolevelvariationinhomepricegrowth,Figure2.3showschangesinhomepricegrowthfortwoneighboringzipcodesinthePhiladelphiasuburbs.Thesetwozipcodesarepopularforfundmanagers. 1Forrobustness,2-yearand4-yearhomepricechangesand3-yearhomepricegrowthontheirsecondaryhomesarealsotested.2Fundmanagerswhoboughtfewerthanthr
73 eeyearsagoareincludedinthesampleiftheyow
eeyearsagoareincludedinthesampleiftheyownedanotherpropertythatwasmergedandspent,inaggregate,atleastthreeyearsinthetwoproperties.Inthiscase,thehomepricegrowthofthetwopropertiesisaveragedbasedontheamountoftimeeachpropertywaslivedin.3Intheappendix,asecondmodelutilizingcommutingzonebytimexedeectsandabatteryofcontrolsisestimated,andasimilareectisfound.Theadditionalcontrolsattempttocorrectforfundmanagerpreferences,includinglocationpreferencewithinacommutingzone.Thisleavesidiosyncratichomepricegrowthwithinacommutingzone.AsimilarmethodisutilizedbyStomanetal.(2018)andBernsteinetal.(2018). CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE58Therstobservationisthatidiosyncratichomepricevariationisnotpersistent.Initiallyzipcode19087outperforms19405,butafteracoupleyearstherelationship ipsandzipcode19405outperforms19087.Overthe2001to2018period,thewinnerbetweenthesetwozipcodesswitchesmultipletimes.Itisnotthatonezipcodealwaysoutperforms.Thedynamicnatureofwhichzipcodesoutperformmakesitlesslikelythatfundmanagersboughtinazipcodeinanticipationthatitwilloutperformaneighboringzipcode.Second,variationbetweenneighboringzipcodesissubstantial.In2003,zipcode19405hadlagged3-yearhomepricegrowthof20%,whilezipcode19087hadgrowthof40%.Inotherwords,ahomeashortdistanceawayexperienceddoubletheamountofhomepricegrowth.Giventhatmanyfundmanagershavehomevaluesinexcessof$1million,thiscreatesverylargeidiosyncraticshocks.2.3DataTostudytheeectofpersonalwealthshocksonprofessionalbehavior,dataonboththehouseholdbalancesheetandchangesinperformanceonthejobareneeded.Forconden-tialityreasons,thiscombinationofdataischallengingtoobtain.Mutualfundmanagersandtheirpersonalrealestatearetheperfectset-uptostudythis.Securitieslawsmaketheprofessionalbehaviorofmutualfundmanagerspublicinformation.Afund'sperformance,managementteam,andholdingsarepublicknowledge.Thiscombinationofdatapresentsextremelyrichdataonprofessionalbehavi
74 or.IntheUnitedStates,realestatetransacti
or.IntheUnitedStates,realestatetransactionsandtheassociatedbuyersarealsopublicrecordandarecollectedbycountyrecorderof-ces.Whenmergedtogetherthiscombinationofdatapresentsauniqueopportunitytodynamicallystudychangesinprofessionalbehaviorfrompersonalwealthshocks.FundcharacteristicsandperformancedataarefromCRSPandinformationonwhoman-agesamutualfundisobtainedfromMorningstar.UsingamethodsimilartoLoutskinaandStrahan(2015),CRSPandMorningstararemergedonCUSIPandticker.Roughly70%ofthefundsinCRSParemergedtoMorningstar.QuarterlyholdingsdataarefromThomsonReuters,whichismergedtoCRSPonWFICNusingthem inkles.4Thismergeisonlyusedforanalysisonmutualfundholdings;forthemajorityoftheresults,thedataarenotrestrictedtofundsmergedtoThomsonReuters.5Fundsareonlyincludedinthesampleif 4TheholdingsdataaresupplementedwithCRSPholdingsdatawhenthedataaremissinginThomsonReuters.5AdditionaldataonstockcharacteristicsareobtainedfromCRSP.FactordataisobtainedfromFama-French-CarhartinWRDSandbenchmarkreturndataisobtainedfromFRED. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE59theyareactivelymanageddomesticequityfundsthatpredominantlyinvestinUSequities.6Indexfundsareidentiedbytheindexfund aginCRSPbeingpopulatedorhavingtheword\index"inthenameofthefundandareremovedfromtheactivelymanagedpopulation.Thepopulationisfurtherrestrictedtofundmonthswithatleast$15millionintotalnetassets(TNA,measuredinin ationadjusted2000dollars)andwith10orfewermanagers.RealestatetransactiondataisobtainedfromATTOM,whichisanadministrativedatasetofrealestatetransactionsintheUnitedStates.ATTOMcollectsanddigitizesthedatafromcountyrecorderoces.Thedatasetincludesthelocationofthehome,loanamounts,andthevalueofthehomeattimeofpurchase.7Toobtainchangesinhomevalueovertime,monthlyzipcodelevelhomepricegrowthdatafromZillowareused.Thedatastartin1997,withdataforsomezipcodesnotavailableuntillater.Asaresult,fundperformancedatafrom2001toJune2018areutilized.Therelati
75 velyunexploreddataformutualfundmanagerpe
velyunexploreddataformutualfundmanagerpersonalrealestateisobtainedbyamergebetweenATTOMandMorningstar.ATTOMincludesthe(non-standardized)namesofhomebuyers,whicharemergedtofundmanagernamesinMorningstar.Namesarestandardizedtotrytoaccountforspellingdierences(i.e.,ChristopherismappedtoChris).Twopassesareattemptedtomergethesetwodatasets.Therstpassmergesonthefullname(includingmiddleinitial,ifavailable).While,thesecondpassmergesontheinitialsoftherstandmiddlenamesandthefulllastname.Otherattemptstomergeonpartialnamesornotusingmiddleinitialyieldedlimitedadditionalsuccessfulmerges.Amergeisonlykeptifthefundmanagernamemergedtoauniquehomebuyernamewithinthecommutingzoneofthefundlocation.FundlocationisobtainedfromthezipcodesofthefundoceslistedineitherCRSPorMorningstar.ThealgorithmissimilartotheoneusedinBernsteinetal. 6ThesetypesoffundsareidentiedbyrestrictingtheLipperProspectusobjectivecode,StrategicInsightobjectivecode,andtheWeisenbergerobjectivecodetothevalueslistedinCremersandPareek(2016).TherestrictionontheCDA/SpectrumcodeisnotusedasthisvalueisinThomsonReuters,whichwouldrequireanadditionalmergethatwouldlimitthepopulationsize.Additionalrestrictionsareincludedtobesurethatonlyactivelymanageddomesticequityfundsareincluded.TheCRSPobjectivecodeisrestrictedtodomesticequityfundsofcap-basedorstylefunds(therstthreecharacterssettoEDCorEDY).Thisrestrictstodomesticequitylarge/mid/small/microcap,growth,income,hedged,short,andincomefunds.Additionally,theMorningstarcategoryhastobeinthe3-by-3size/valuegrid,asinStomanetal.(2018),orhaveacategorytypeof85%+equityallocation.TheMorningstarcategoryisalsousedtoassignfundsabenchmark.Thebenchmarksareassignedasfollows:USFundLargeGrowth(Value)isRussell1000Growth(Value)TotalReturn,USFundLargeBlendisRussell1000TotalReturn,USFundMid-CapGrowth(Value)isRussellMid-capGrowth(Value)TotalReturn,USFundMid-CapBlendisRussellMadcapTotalReturn,USFundSmallGrowth(Value)isRussell2000
76 Growth(Value)TotalReturn,USFundSmallBlen
Growth(Value)TotalReturn,USFundSmallBlendisRussell2000TotalReturn,andUSFundAllocation{85%+EquityisRussell3000TotalReturn.7AnadditionalfeatureofATTOMisinformationonthetaxaddressofproperties(basedonannuallesfromcountyassessoroces).Withthisinformation,secondaryhomesarelinkedtotheprimaryhomesofthefundmanagers.Roughly16%offundmanagersinthesampleownmorethanonehome,andatmostfourhomesareownedbyasinglefundmanageratanygivenpointintime.Thetop25zipcodesforthesesecondarypropertiesareoverwhelminglyinvacationdestinationssuchasNaplesFlorida,LakeTahoeCalifornia,skitownsinColoradoandtheBerkshires,andvariouslakeandbeachtowns(suchastheHamptonsandLakeWinnipesaukee).Thelocationsofthesehomesprovidesvalidationthatthesearevacationpropertiesownedbythefundmanager. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE60(2018).Mergesaredroppediftheyresultinanon-disjointperiodoflessthan24monthsoffunddataforwhichatleastonefundmanagerismergedtoATTOM.Thisisinordertoaccuratelyestimaterisk-adjustedreturns(robustnessisprovidedaroundthisrestriction).40%ofthemanagersinMorningstararesuccessfullymergedtoATTOM,whichislowerthanthe52%ratereportedinBernsteinetal.(2018)forinnovativeworkerswhoproducepatents.Theirpatentdataincludestheperson'shomezipcode,whileMorningstaronlyincludesthezipcodeofthefundanddoesnotincludeinformationonthelocationofthefundmanager'shome.Itisassumedthatafundmanagerliveswithinthesamecommutingzoneastheirfund.Managerswhowerenotmergedeitherrent,haveacommonname,liveoutsidethecommutingzoneoftheiroce,usedatrusttopurchasetheirhome(whichusuallypartiallyorfullyshieldstheiridentity),orhadamisspellingineitherdataset.Thenalpopulationcomprises1,241fundsbasedoninformationfor1,368fundmanagers,foratotalpopulationof87,719monthlyfundpairs.Figure2.4showsthepercentageandnumberofmanagersmergedforeachfundinthenalpopulation.Manyfundsaremanagedbyateamoffundmanagers,butforthemajorityoffundsonlyonemanagerismerged.Generally
77 ,atleast50%ofthemanagersaremergedforeach
,atleast50%ofthemanagersaremergedforeachfund.Incaseswhenmorethanonemanagerismerged,thefundmanagercharacteristicsandhousinggrowthinformationisaveraged.Nothavingallofthemanagersmergedleadstonoiseontherighthandside,whichshouldbiastheresultstowardszero|robustnessisprovidedaroundthemergerate.Table2.1providessummarystatisticscomparingthepopulationofactivedomesticeq-uitymutualfundsmergedtoATTOMagainstthepopulationnotmerged.Onaverage,bothpopulationshaveanegativealphaafterfeesofabout10bpspermonth.Themergedpopu-lationisbiasedtowardslargerfunds(bothintermsofTNAandnumberofmanagers)withlowerturnover.Onthefundmanagercharacteristicsqualities,almostallmanagersaremale(92%),mosthaveanMBAorCFA(57and56%,respectively),andmosthadsubstantialcareerexperienceatthetimetheybecamefundmanager(18yearsonaverage).Experiencedataareonlyavailableforasubsetofmanagersandissupplementedwithyearofbirth+22ortheyearofundergraduategraduationforthecareerstartdate(whenthesevariablesareavailable).2.4EmpiricalResultsonPerformance2.4.1MainSpecicationTheprimaryhypothesisofthispaperisthatexogenouspositivereturnstothepersonalrealestateoffundmanagersaectsfundmanagersviaovercondence.Totestthis,aregressionmodeloflaggedzipcodelevelhomepricegrowthonforecastsofperformanceisestimated.Thebaselinemodelanalyzes1-monthaheadforecastsofriskadjustedreturnsonlagged3-yearhomepricechangesforthevaluesoftheprimaryhomesoffundmanagers.Additionaltestsareperformedtolinkthiseecttoovercondence. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE61TheregressionframeworkissimilartotheoneutilizedbyGuptaandSachdeva(2017).ThefourfactorFama-French-Carhartmodelisestimatedatthefundlevel:RitRft=MKT;iMKTt+SMB;iSMBt+HML;iHMLt+UMD;iUMDt+it(2.1)Returnsarenetoffees(robustnessisprovidedaroundthis),whichrepresentsthereturnsexperiencedbyinvestors.Thefactorloadingsareusedtoobtainmonthlyalphas:bFFCit=RitRftbiFF
78 CtwhereFFCt=(MKTtSMBtHMLtUMDt)(2.2)
CtwhereFFCt=(MKTtSMBtHMLtUMDt)(2.2)ThebFFCitareusedinpanelregressionstoestimatethepredictiveabilityoflaggedhomepricegrowthtoforecast1-monthaheadriskadjustedperformance:bFFCi;t;a=c+ln(3yrHPIi;t1)+i+t+a+ita(2.3)whereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects,andaarenon-parametricxedeectsforTNA.Standarderrorsareclusteredattheleveloffundbyeachmergednon-disjointperiodoftime(robustnessisprovidedaroundthis),whichallowsforarbitrarycorrelationoftheerrortermswithinfundovertime.Unboundedvariablesarewinsorizedatthe1%and99%level.Totesttheeectoverperiodsofpositiveandnegativehomepricegrowth,twoapproachesareused.Therstsimplydividesthesampleintotwoperiods,onewhere3yrHPIi;t10andtheotherwhere3yrHPIi;t10.Thesecondapproachusestheentirepopulationandestimatesthecoecientsforpositiveandnegativehomepricegrowthseparatelyinasingleregression.ln(3yrHPIi;t1)isinteractedwithindicatorvariablesforpositiveandnegativehomepricegrowth,denotedasln(3yrHPI+i;t1)andln(3yrHPIi;t1),respectively.Forthisregression,thetimexedeectsarealsointeractedwithindicatorvariablesforpositiveandnegativehomepricegrowthtocontrolforvariationinreturnsamongfundsmanagedbymanagersexperiencingpositiveandnegativehomepricegrowthinagivenmonth,denotedby+tandt,respectively.TheresultsforthemainregressionmodelarepresentedinTable2.2.Overall,homepricegrowthsignicantlynegativelyimpactsfutureperformancewhenusingdatafromtheentire2001to2018period(column1).Incolumns2and3,thesampleisdividedintoperiodsofpositiveandnegativehomepricegrowth,respectively.Fromthissegmentation,itisshownthattheeectisisolatedtoperiodsofpositivehomepricegrowth,withthemagnitudeincreasingwhenperiodsofnegativehomepricegrowthareremoved.Interestingly,duringperiodsofnegativehomepricegrowththereisnoe
79 ect.Thisisinlinewiththendingso
ect.ThisisinlinewiththendingsofStomanetal.(2018),whoalsondnoimpactonriskadjustedreturnsfromhomepriceshocksduringtheGreatRecession.Thisalsoshowshowthetwoshocksoperateviadierentchannels,overcondenceduringperiodsofpositivehomepricegrowthand,asshownbyStomanetal.(2018),careerconcernsduringperiodsofnegativehomepricegrowth.When CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE62theregressionisestimatedusingtheentiresampleandthehomepricegrowthvariableisbifurcatedintoperiodsofpositiveandnegativehomepricegrowth,similarresultsarefound(column4).Theaveragestandarddeviationin3-yearhomepricegrowthwithinfundis12%,whichyieldsadeclineinalphaof37bpsperyearforaonestandarddeviationpositivehomepriceshock.Giventhedramaticriseinhomepricegrowthoverthemid-2000sandlate2010s,thereisthepotentialforadramaticdecreaseinperformancefromexogenouspositiveshockstohomepricegrowth.Theresultsaresignicantatthe1%levelandthet-statisticsaregenerallyabove3topassthethresholdofHarvey,Liu,andZhu(2016).Incolumn5,1-yearlaggedalpha(calculatedfromaregressionofFama-French-Carhartonthepreceding12months)and1-monthforwardfund owsareaddedascontrols.Withthesevariablesincluded,thesamplesizeisreducedsincetherehavetobe12-monthsoflaggeddataavailable.Theresultisstrengthenedwiththeinclusionofthesecontrols.Tofurthershowthatfund owsarenotaectedbyhomepriceshocks,theregressionisestimatedwiththelefthandsidereplacedwith1-monthand1-yearforecastsoffund ows,calculatedusinggrossreturns(beforefees).Table2.3showsthatthehomepriceshocksdonothaveanimpactonfund ows.Combinedwiththeearlierresultshowingthattherelationshipbetweenhomepriceshocksandreturnsisnotaectedbycontrollingforfund ows,fund owsarenotaconcernforthendingsinthispaper.AdditionalrobustnesstestsareprovidedintheAppendix.Acommonplacebotestistore-estimatetheeectonperformanceforasampleofindexfunds.Table2.4showsthatwhenthesampleofactivefundsisreplacedwithasamp
80 leofindexfunds,theresultbecomesinsigni
leofindexfunds,theresultbecomesinsignicant.Inadditiontousingindexfundsasaplacebo,closetindexfundsrepresentanotherpotentialplacebotest.Forrobustness,twoapproachesareusedtodeneafundmanagerasbeingaclosetindexer.Therstapproachlabelsafundmanagerasaclosetindexeriftheirfund'strackingerrorisinthelowestquantile.Column4showsthatforthispopulationofclosetindexfundsnoeectisfound.Thisapproachcanalsobeusedtorestrictthesampletothemostactivefunds.Whenrestrictedtoveryactivefunds,alargerresponseisfound(column5).Forthesecondapproach,afundisdenedasaclosetindexfundifthefund'sActiveShareislessthan60%,followingChernenkoetal.(2016).ActiveShareisprovidedbyCremersandPetajisto(2009).Similarly,whenusingtheActiveSharedenitionofaclosetindexernoeectisfound(column6).UsingActiveSharetorestricttothemostactivefunds(thosewithanActiveShare60%),astrongereectisfoundonceagain.2.4.2Whichfundmanagers/fundsrespondtohomepricegrowth?Thissectionprovidessubsampleanalysistoseewhichtypesoffundmanagersandfundsrespondtohomepriceshocks.Ifovercondenceisdrivingtheresultthenfundmanagerswithlesseducation,lessexperience,whoboughttheirhomesmorerecently,withgreater CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE63housingleverage,andwhoaremaleshouldrespondmoretohomepriceshocks.Totestthis,theregressionmodelisestimatedwiththehomepricegrowthvariableinteractedwithindicatorvariablesfordierentmeasuresofcharacteristicsoffunds/fundmanagersduringperiodsofpositivehomepricegrowth.Theomittedcategoryisthegroupoffunds/fundmanagerswhoareleastlikelytobeaectedbyhomepriceshocks(i.e.thosewithanMBAwhensplittingthesamplebyeducation).Figures2.5,2.6,and2.7presenttheseresultsgraphically.Figure2.5showsthatnothavinganMBA,havinglessexperience,andbuyingahomemorerecentlyallproduceastrongereect.Theseareexactlythetypesoffundmanagersonewouldexpecttobuyintoovercondencefromhomepriceshocks.Whilemoreexperiencedfundma
81 nagersgenerallydonotrespondtohomepricesh
nagersgenerallydonotrespondtohomepriceshocks,experiencedfundmanagerswhoboughttheirhomemorerecentlyrespondstronglytohomepriceshocks.Buyingmorerecentlyshouldinduceastrongerresponseashomeownerswouldbemorelikelytopayattentiontotheirhomevalueshortlyaftertheybuytheirhome.Thisisduetotworeasons.First,theirhomevalueismorelikelytobeontheirmind.Second,short-runmovementswillhavealargereectontheoverallreturnontheirpurchase.Buyinginthedistantpastmakesshort-runmovementsinhomepriceslessimportantduetoalreadyhavingaccumulatedhomepricegains.Figure2.6alsoconrmsthatfundmanagerswithoutaCFAdesignation(notsigni-cantly),withmoreexpensivehomes,andwithgreaterhousingleverage(asmeasuredbyloantovalue[LTV]atpurchase)alsohaveastrongerresponsetohomepriceshocks.ThisisrelatedtotheCEOliterature,namelyCronqvistetal.(2012)andLiuandYermack(2012),whoshowthatCEOswithlargerhomesandhigherLTVratiosengageinriskierbehavior.ConsistentwiththeovercondenceresultsofBarberandOdean(2001)andLuandTeo(2018),fundswithonlymalemanagers(amongthemergedmanagers)haveamuchstrongerresponse.Inadditiontosegmentingthepopulationonfundmanagercharacteristics,thepopulationissegmentedonfundcharacteristicstoidentifythetypesoffundsthataremorelikelytohavemanagersaectedbyhomepriceshocks(Figure2.7).Thesizeofthefund(asmeasuredbyTNA)doesnotproduceadierentialresponse,indicatingthattheresultisnotdrivenbysmallfunds.Fundtypeandcapitalizationfocusdo,however,providestrongsegmentation.Valueandsmallcapfundmanagersarelessaectedbyhomepriceshocks,whilegrowth/blendandlargeandmid-capfundsarestronglyaected.Lastly,fundswithlowexpenseratioshaveastrongerimpact.Thiscouldbeexplainedbythefactthatfundsthatareabletocommandahighexpenseratioarelikelymanagedbyseniormanagerswithsignicantexperience,andassucharelesslikelytobeswayedbyhomepriceshocks.Asalasttest,thesampleisdividedintofundsmanagedbyfundmanagerswhohaveneverownedasecondhomeandfundmanagerswhohave.InTable2.5,ho
82 mepricegrowthforsecondaryhomesisincluded
mepricegrowthforsecondaryhomesisincludedasanadditionalregressor.Ifthepopulationisrestrictedtofundmanagerswhoeverownedasecondhome(columns1and3)theresultbecomesinsignicant,whilefocusingonfundmanagerswhoneverownedasecondhomepreservestheresults(columns2and4).Thisisadditionalevidenceofovercondence,asfundmanagerswhoownasecondhomearelikelytohavegreaterexperience.Whileotherchannelscannot CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE64beruledout,suchasincreasedrisktakingfromawealtheect,theresultsareconsistentwithovercondencebeingthechannel.Inthenextsection,tradingbehaviorisexploredandtheresultsprovideadditionalevidenceoftheovercondencechannel.2.5TradingBehaviorInordertounderstandwhyperformancedeclinesfollowingpositivehomepriceshocks,thetradingbehavioroffundmanagersisanalyzed.First,short-runchangesintradingbehavioroffundmanagersisanalyzed.Second,longruntradingbehaviorisstudiedusingthefund'sownbenchmarkreturnsasapassiveportfolio.Lastly,portfoliorisktakingisanalyzed.2.5.1TradingReturnAfterestablishingthatperformancedeclinesandtradingincreasesfrompositivehomepriceshocks,itiscrucialtounderstandthemechanismsthataredrivingthis.Inthissection,thetradesperformedeachquarterarestudiedusingquarterlyequityholdingsdata.8Withthisdata,stocksareidentiedaspurchased,heldconstant,orsoldbetweenquarters.Atradeislabeledabuytradeifthefund'spositioninthestockincreasesbyatleast5%.Istartbyestimatingtheeectofhomepriceshocksonthe1-quarteraheadreturntothesetrades.Returnsareweightedbythevalueofthechangeinthenumberofsharesusingtheaverageofthestock'spriceatthebeginningandendofthequarterinwhichthestockistraded.Toestimatetheeectthefollowingregressionisestimated:Yi;t+1;a=c+ln(3yrHPIi;t)+i+t+a+i;t;a(2.4)whereiarefundxedeects(foreachmergednon-disjointperiodoftime),tarequarterlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepr
83 icegrowth),andaarenon-parametric
icegrowth),andaarenon-parametricxedeectsforTNA.Theforecastedreturntothenettradingdecisions(thereturntothebuytradeslessthereturntotheselltrades)offundmanagersdeclinesashomepricegrowthincreases(Table2.6,column1).Thisimpliesthatfundmanagersaremakingworsetradingchoicesfrompositivepersonalwealthshocks.A12%positiveshocktohomepricegrowthdecreasesthereturndierencebyroughly7bps,whichisalmost100%ofitsmean.Tounderstandwhythereturndeclines,thereturnisdecomposedintobuyandsellchoices.Broadlyspeaking,afundmanagerhasvedierenttradingchoices:buyanewstocknotcurrentlyheld,increasethepositionofastockcurrentlyheld,nottradeastockcurrentlyheld,fullyliquidateaposition,orpartiallyliquidateaposition.Usingtheseveactions,the1-quarteraheadreturnslessthebenchmarkreturnarecalculatedforeach. 8ThedataarefromThomsonReuterss12,supplementedwithdatafromCRSP.Thedataarereportedasoftheendofeachquarter.Itispossiblethatstocksareboughtandsoldwithineachquarter,inwhichcasetheywouldnotbecaptured. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE65Table2.6column2showsthatfundmanagersarenotmakingdierentialchoiceswhenselectingnewstockstopurchase.Instead,fundmanagersmakeworsechoicesinpickingstockstoleverupon(column3).A12%positiveshockdecreasesthereturntoleveringupby35%ofitsmean.Themeanreturntothebuytradesofbuyingnewstocksandleveringuponexistingstocksarebothapproximately16bps,indicatingthatonaveragefundmanagersdonotmakebadchoicesinselectingstockstobuy.Foranalyzingthereturntosales,homepricegrowthisnotdividedintoperiodsofpos-itiveandnegativehomepricegrowthduetothestrongoveralleect.Theliteraturehasfoundthatakeyreasonforfundunderperformancestemsfrommakingpoorsellingchoices(Akepanidtawornetal.,2018).Theyattributethistoinattentioninthestocksellingdecisionandfollowingaheuristicofsellingstockswithrecentextremepricemovements.Fundman-agersmakeworsechoicesinfullyliquidatingexistingpositionsashomepriceshocksincrea
84 se(column6|thepositivecoecientimpli
se(column6|thepositivecoecientimpliesthatthestocksoldsubsequentlyperformsbetter).Thisresultprovidesanotherchannelthatdrivesthedeteriorationinthechoiceofstockstosell.Lastly,column7showsthat owsarenotaected,indicatingthattheresultsarenotdrivenbyfundmanagerstradingduetovariationsin owashomepriceshocksevolve.Thedeclineinreturnstoincreasingpositionsinexistingstocksisexploredfurtherbyaugmentingtheaboveregressionstotlinearprobabilitymodelsofthelikelihoodofastockbeinglevereduponhavingex-anteperformedworseinthe2-quarterperiodduringandpriortothetrade.Table2.7showsthatthestocksleveredupbecomemorelikelytohaveex-anteperformedworseashomepriceshocksincrease.Theresultisstrongestincolumn2wheretheprobabilityofbuyingmoreofstocksthathaveperformedatleast20%worsethanthebenchmarkintheprior6-monthperiodincreases.Asfundmanagersexperiencepositivehomepricegrowth,theybecomemorelikelytoincreasetheirpositionsinstocksthathaveperformedpoorly,relativetoboththebenchmarkandthestockstheyholdbutchoosenottoincreasetheirstakesin.Column4showsthatthisisnottruefornewstockspurchasedthatarenotcurrentlyheld.Fundmanagersdonotbecomeovercondentinpoorlyperformingstocksthattheydidnotpreviouslybuy.However,conditionalonhavingalreadyownedthestock,fundmanagersbecomemorelikelytodoubledownontheirpositionifthestockhadaperiodofpoorperformance.Columns5and6estimatetheeectontheshareofbuy(sell)tradesthatareincreasesinexistingpositions(fullliquidations).Ashomepricegrowthincreases,fundmanagersshifttheirbuytradestowardsleveringupandshifttheirselltradestowardspartiallyliquidating(insteadoffullyselling).Notonlydofundmanagersleveruponlosersbuttheydevotelessattentiontopickingnewstocks.Inaddition,theybecomemorelikelytopartiallyholdontotheirpositions.Inthenextsection,along-runimpactontheactivetradingreturnsoffundmanagersfromtheirpersonalhomepriceshocksisestablished.2.5.2ComparisontoaPassiveStrategyInthissection,evidenceispresentedthatpositivehomepricesho
85 cksalsohavealong-runimpactonperformance.
cksalsohavealong-runimpactonperformance.Toshowthis,actualreturnsarecomparedagainstown-benchmark CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE66("passive")returnsinasimilarspiritofBarberandOdean(2001),whostudyretailinvestors.EachJune(foreveryyeart)thechangeinln(3yrHPIi;t)overtheprior1-yearperiod(therateofchange)iscalculated.Figure2.8showsthedistributionofthesevalues.Overaoneyearperiodtherecanbeadramaticchangeintherateofhomepricegrowth.Thisrateofchangeisinteractedwithindicatorvariablesforln(3yrHPIi;t)beingpositiveornegative.Therateofchangeinhomepricegrowthisused,aschangesintradingbehaviorduetochangesinhomepricegrowthisstudied.Toconstructtheperformanceofa"passive"portfolio,the2-year(24monthsofdata)alphaisconstructedfromaFama-French-Carhartregressionofthereturnstothestocksheldinquarter3ofyeart1betweenquarter3ofyeart1andquarter2ofyeart+1.Fortheactiveperformance,analphaoverthesametimeperiodiscalculatedusingreturnstotheactualquarterlyholdings,updatedeachquarter.Thesampleisrestrictedtofundswhosemanagerssurvivedfortherespective24monthperiod,haveatleast80%oftheirassetsrepresentedbytheseholdings,andhaveacorrelationbetweenactualreturnsandtheconstructedactivereturnsofatleast99%(80%oftherecords).Theselasttworestrictionsaretoensurethatthereturnsfromtherawholdingsdatarepresenttheactualreturnstothefunds.Withoutthisrestriction,itmightbethatthefundmanagertradesinotherassetclasses,forwhichholdingsdataarenotavailable(i.e.cash,derivatives,etc.),andthiswouldaecttheirpositionsinequitiesandtheiroverallreturns.Toestimatetradingcosts,theapproachofKiyotakiandMoore(1997)isused.9Sincethedataareannual,commutingzonebytimexedeectsareusedinlieuoffundxedeects.Includedarethefollowingadditionalcontrolvariables:Morningstarfundcategoryxedeects,numberofmanagerxedeects,anindicatorforifthemanagerownsasecondhome,anindicatorforifthemanagerhasanadvanceddegree(aboveundergradua
86 te),andnon-parametricxedeectsf
te),andnon-parametricxedeectsforTNA,homevalue,combinedLTV(CLTV),averagezipcodelevelincome(fromthe2010IRSSOI),andthepercentofnon-whitehouseholdsatthezipcodelevel(fromthe2000Census).Thesevariablescontrolforfundandhomelocationtimeinvariantqualities.Thefollowingregressionmodelisestimated:Activei;t12mo!t+12mo;czPassivei;t12mo!t+12mo;cz=c+(ln(3-yrHPIi;t)ln(3-yrHPIi;t12mo))+cz;t+Controlsit+i;t;cz(2.5)wherecz;tareannual(tisannual)timebycommutingzonexedeects(interactedwithindicatorvariablesforpositiveandnegativehomepricegrowth).capturestheeectoftherateofchangeinhomepricegrowthontheperformanceofthefundrelativetotheperformanceifthefundmanagerdidnottrade.Apositivecoecient 9KiyotakiandMoore(1997)estimateinstitutionalstockleveltradingcostsasafunctionof:tradesize,marketcapitalization,ifthefundisatechnicalfund,ifthestockisNasdaqlisted,andifthefundisanindexfund.Thislastindicatorvariableissettozeroforthepopulationofactivemanagersinthispaper.FollowingBusse,Chordia,Jiang,andTang(2016),thetechnicalfundindicatorissetto0.45forbuysand0.6forsales.Forobviousreasonstradingcostsarenotsubtractedfromthereturnstothepassiveportfolio. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE67wouldimplythatincreasesinhomepricegrowthleadtothefundmanager'stradingaddingvaluerelativetodoingnothing.Standarderrorsareclusteredbyfund(foreachmergednon-disjointperiodoftime).Table2.8column1estimatestheeectonactivereturnslessestimatedtradingcostsandtheirown-benchmarkpassivereturn.Theresultisnegativeandsignicantatthe5%levelforperiodswherehomepricegrowthispositive.Aonestandarddeviationshocktotherateofchangeinhomepricegrowthis0.1,whichequatestoadeclineinactiverisk-adjustedperformanceof45bpsperyearrelativetotheirpassivebenchmark.Thereisnoeectwhenhomepricegrowthisnegative.Column2replacesthelefthandsidewithanindicatorvariableforiftheactivealpha,ne
87 toftradingcosts,beatsthepassivealpha.Aon
toftradingcosts,beatsthepassivealpha.Aonestandarddeviationshocktotherateofchangeinhomepricesdecreasestheprobabilityofbeatingthepassivebenchmarkby8%.Thisimpliesthatnotonlydoestheactiveperformancedeclinebutitalsomakesthemanagerlesslikelytoaddvalueovertheirown-benchmarkpassiveholdings.Onaverage,only35%offundmanagersbeattheirpassivebenchmarkwhenaccountingfortradingcosts.Again,thereisnoeectfromnegativehomepriceshocks.Whentradingcostsarenotincluded,theresultforperiodsofpositivehomepricegrowthlosesmagnitudeandsignicancebutremainssignicantatthe10%level.Partofthereasonforthedeclineinperformanceisduetoanincreaseintradingcosts.Column4showsthattradingcostsincreaseashomepricegrowthincreases.Anincreaseintradingisacommonsignofovercondenceintheliterature.Toconrmthattradingisincreasing,theeectonannualturnoverdatafromCRSPisestimated.Forthisregression,thefullsampleofmergedfundsisused(notonlytheonesthatfurthermergedtoThomsonReuters).Sincethedataareannual,theregressionusesthespecicationabovewithcommutingzonebytimexedeectsandadditionalcontrolvariables.Laggedlogturnoverisincludedasacontrolvariableaswell.Column5showsthatturnoverdoesindeedincreaseashomepricegrowthincreases.Forretailinvestors,itiswellestablishedthatexcessivetradingleadstoworsereturns(Odean,1999;OdeanandBarber,1999;BarberandOdean,2001),whileforinstitutionalinvestorstheresultsaremixed.Overall,Pastoretal.(2017)ndthathighertradingleadstohigherreturns,withtheeectbeingstrongerforhighfeefunds.Conversely,CremersandPareek(2016)ndthatpatientinvestmentstrategiesofinstitutionalinvestorsleadstohigherreturns.Thispaperarguesthatifturnoverisdrivenbyovercondenceduetohomepriceshocks,performancewillsuer.Thisisduetotheincreaseintradingbeingdrivenbyirrationalreasons.2.5.3RiskTakingInthissection,thelinkbetweenhomepriceshocksandrisktakingisinvestigated.ThissectionrelatestoStomanetal.(2018)andextendstheirndi
88 ngstoabroadertimeseries.Intheirwork,thep
ngstoabroadertimeseries.Intheirwork,theprimaryfocusisontherisktakingmeasureofthestandarddeviationofreturnsduringtheGreatRecession.Usingalongerperiod(2001to2018),thestandarddeviationofreturnsiscalculatedusingmonthlyreturnsforthesubsequent12-monthperiodfollowingeachmonthlyobservationoflagged3-yearhomepricegrowth.Additionally,results CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE68fortrackingerrorandCAPMbeta(marketrisk)areshown.Theregressionusesthebaselinespecicationinthispaperandincludesfund(foreachmergednon-disjointperiodoftime)xedeects,timexedeects,andnon-parametricTNAxedeects.Table2.9columns1and2showthatpositivehomepricegrowthhastheoppositeeectonthestandarddeviationofreturnscomparedtonegativehomepricegrowth.Forperiodsofpositivehomepricegrowth,theeectisnegative.Whileforperiodsofnegativehomepricegrowth,theeectispositive.BoththispaperandStomanetal.(2018)ndapositiveandsignicanteectonrisktakingduringperiodsofnegativehomepricegrowth.Whenrunningahorseraceonthebroader2001to2018period,theeectfrompositivehomepricegrowthsurviveswhiletheeectfromnegativehomepricegrowthdoesnot.InStomanetal.(2018)apositiveeectisfoundfortrackingerrorandnoeectisfoundonmarketrisk.Similarly,thispaperndsapositiveeectfortrackingerror(column5)andnoeectformarketrisk(column8)duringperiodsofnegativehomepricegrowth.Similartotheeectforthestandarddeviationofreturns,inahorseracetheeectonmarketriskonlyremainsforperiodsofpositivehomepricegrowth.Forperiodsofpositivehomepricegrowth,trackingerrorandmarketriskdecline(columns4and7).Overall,positivehomepriceshocksleadtoadeclineinrisktaking.Giventheearlierresults,thisismostlikelyreconciledwithadeclineintakingpositionsonnewstocksandrelyinginsteadonexistingpositions.2.6AttentiontoHomePriceGrowthAkeyassumptionoftheresultsinthispaperisthatfundmanagerspayattentiontohomepricegrowth.T
89 otestthis,personalhomeequityextractionso
otestthis,personalhomeequityextractionsoffundmanagersarestudied.Cashoutrenancingwascommonduringthehousingboomofthemid-2000s,withmanypaperswrittenaboutthisinthehouseholdnanceliterature.10ForadescriptionofhowtheamountofhomeequityextractedisconstructedusingATTOMdatapleaseseetheOnlineAppendixtoChapter1.Thispaperuseshomeequityextractionstostudyiffundmanagersarepayingattentiontohomeprices,amongtheselectedgroupofhomeownerswhoextracthomeequity.Iffundmanagersaremorelikelytoextractequitywhenhomepricesaregrowingatafasterrate,thisisevidencethatmanagersareawareofthechangesintheirhomevalue.Fundmanagersareawealthiersegmentofthepopulationbuttheyalsoheavilyengageincashoutrenancing.Itisleftasanopenquestionastowhyfundmanagersextracthomeequity(i.e.,isittosupplementincome,forconsumption,tostartasmallbusiness,toinvest,etc.).Around5-15%offundmanagersextracthomeequityinanygivenyear,withtheproportionoffundmanagersextractinghomeequitydecreasingovertime(Figure2.9a).Amongfundmanagerswhoextracthomeequity,themedianfundmanagerextractsroughly$36,000and10%offundmanagersextractover$340,000(Figure2.9b).Totesttheimpactofhomeequityextractiononfundperformance,thefollowingregressionmodelisestimated: 10SeeGreenspanandKennedy(2008),BhuttaandKeys(2016),andDeFusco(2018),amongothers. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE69 i;t+1;a i;t1;a=c+\I(ExtractHomeEquityi;t)+i+t+a+ i;t1;a+i;t;a(2.6)where i;t;aistheaveragemonthlyalphainquartertandI(ExtractHomeEquityi;t)isanindicatorequalto1ifafundmanagerextractedhomeequityinquartert.iarefundxedeects(foreachmergednon-disjointperiodoftime),tarequarterlytimexedeects,andaarenon-parametricxedeectsforTNA.tismeasuredasquarterlyobservationsandstandarderrorsareclusteredatthefundlevel(foreachmergednon-disjointperiodoftime).Ifisnegative,homeequityextractioninquartertforecastsadec
90 lineinrisk-adjustedperformanceinquartert
lineinrisk-adjustedperformanceinquartert+1relativetoperformanceinquartert1.Theseresultsareforecastsandprovideex-anteinformationthatcanbeusedtoscreenfunds.Thechangeinlagged3-yearzipcodelevelhomepricesisusedasaninstrument.Thepurposeofthisinstrumentisnottocorrectforendogeneity,buttoshowthatvariationinhomepricegrowthaectstheprobabilityofhomeequityextraction.Table2.10,column1showsoverallthereisnosignicanteectforhomeequityextractionsonfutureperformance.Whenthechoiceofhomeequityextractionisdrivenbyvariationinhomepricegrowth(columns3and5)thereisalargenegativeeectonforecastedfundperformance.Additionally,laggedhomepricegrowthpredictsahigherpropensitytoextracthomeequityintherststage(columns2and4).Thisisevidencethatfundmanagersareawareofhomepricegrowthandthatextractionsofhomeequitysignalasubsequentdeclineinfundperformance.2.7ConclusionNumerouspapershavefocusedonovercondencebehaviorofretailinvestors,butthemajor-ityofthesestudieshaveonlylookedatstaticmeasures,andtheliteratureonovercondencebehaviorofinstitutionalinvestorsisscant.Thispaperprovidesin-depthandhighlyro-bustevidenceofatimevaryingmeasureofovercondenceanddetailedndingsonhowthistranslatesintolowerperformanceamonginstitutionalinvestors.Usingmutualfundhold-ingsdata,thetradingbehaviorforfundmanagersisbackedout,whichprovidesdetailedevidenceexplaininghowfundmanagersreacttopositivehomepriceshocks.Fundmanagersinterpretchangesinthevalueoftheirpersonalhomesasfeedbackontheirabilitytopickinvestments.Thispaperndsthatpositiveshockstothevalueofthepersonalrealestateoffundmanagersforecastsadeclineinperformance.Fundmanagerswhoaremoresusceptibletoovercondencehaveamuchstrongerresponse,particularlyinexperienced,lesseducated,andmalefundmanagers.Thedeclineinperformancestemsfromanincreaseintrading(andtheassociatedtradingcosts)andmakingworsechoicesinpickingstockstoleveruponandtosell.Interestingly,followingpositivehomepriceshocks,f
91 undmanagersdonotmakeworsechoiceswhenpick
undmanagersdonotmakeworsechoiceswhenpickingnewstocksbutinsteadbecomemorelikelytobuymoreoftheirexistingpositionsthathaveunderperformed.Fundmanagersdonotengageinthistypeoftradingbehaviorduringperi-odsofnegativehomepriceshocks.Overall,housingshocksaectbehaviorbutviadierent CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE70channelsdependingonthesignoftheshock.Thechannelofovercondenceduringperiodsofpositiveshocksleadstoastrongeectonrisk-adjustedreturns.Homeequityextractionssignalasignicantdeclineinfutureperformanceaswell,highlightingtheimportanceofscreeningfundsonthepersonalqualitiesoftheunderlyingfundmanagers.Whilethispaperfocusespurelyonpersonalrealestate,otherpersonalnancemeasuressuchasFICOandoverallhouseholddebtareotherpotentiallyrelevantmetrics.Althoughthesemeasurescouldpotentiallyraiseprivacyissuesastheyarenotpublicinformation,unlikerealestate.Overall,thispaperdocumentsastrongfeedbackloopoffundmanagersbecomingovercondentinprofessionalbehaviorfromtheinvestmentselectionofpersonalrealestate.Thiseectverylikelygeneralizestotheassetsboughtbythefundmanagerfortheirfund(s)andotherassetclasses(suchasart),therebypotentiallyindicatingfeedbackloopsthatcouldleadtoaspiralofovercondence. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE712.8FiguresandTables Figure2.1:ComparisonofAverageSalaryandHomeValuesforInvestmentManagersAcomparisonofaveragesalaryandaveragehomevalueforinvestmentmanagersbetween2001and2016.DataonaverageannualsalaryisfromtheQCEWandaveragehomevalueisfromATTOMforthesamplemergedtoMorningstar. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE72 Figure2.2:BostonMetroAreaZipCodeLevelHomePriceGrowthMedianhomepricegrowthatthezipcodelevelforzipcodesintheBostonmetroareabetweenJanuary1997andDecember2017.DataarefromZillowandusesingle-familyresidentialandcondo/co-op.ValuesarenormalizedtothevalueinJanuary1997. Figure2.3:Comparisonof3
92 -yearHomePriceGrowthforNeighboringZipCod
-yearHomePriceGrowthforNeighboringZipCodesComparisonof3-yearlaggedzipcodelevelhomepricegrowthforneighboringzipcodes19405and19087.BothzipcodesareinthesuburbsofPhiladelphia. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE73 (a) (b)Figure2.4:NumberandShareofFundManagersMergedFigureA(B)showsthenumber(percentage)offundmanagersmergedforeachfundinthenalpopu-lation. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE74 Figure2.5:HeterogeneousEectofHomePriceGrowthonFundAlpha:1Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+Xiiln(3yrHPIi;t1)I(ExperienceMeasurei)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.ln(3yrHPIi;t1)isinteractedwithindicatorvariablesformeasuresofeducation,experience,orhomepurchasetiming(measuredbetweenpurchaseandwhenthefundmanagerstartedatthefund).Theomittedcategoryistheoneforthegroupofmanagerswhoareleastlikelytobeaectedbyhomepricegrowth(i.e.thosewithanMBA).Percentageslistednexttovaluesonthex-axisrepresentthepercentofthepopulationthatfallsintothecategorylisted.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fun
93 dmonthsaredroppediftherearegreaterthan10
dmonthsaredroppediftherearegreaterthan10managers,ifTNA,measuredin2000dollars,fallsbelow$15million,orifhomepricegrowthisnegative(onlymonthswithpositivehomepricegrowtharekept).OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE75 Figure2.6:HeterogeneousEectofHomePriceGrowthonFundAlpha:2Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+Xiiln(3yrHPIi;t1)I(ExperienceMeasurei)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.ln(3yrHPIi;t1)isinteractedwithindicatorvariablesformeasuresofcertication(CFAstatus),ifallofthemergedfundmanagersaremale,homevalueattimeofpurchase,orLTV(includingsecondliens)attimeorpurchase.Theomittedcategoryistheoneforthegroupofmanagerswhoareleastlikelytobeaectedbyhomepricegrowth(i.e.thosewithaCFA).Percentageslistednexttovaluesonthex-axisrepresentthepercentofthepopulationthatfallsintothecategorylisted.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managers,ifTNA,measuredin2000dollars,fallsbelow$15million,orifhomepricegrowthisnegativ
94 e(onlymonthswithpositivehomepricegrowtha
e(onlymonthswithpositivehomepricegrowtharekept).OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE76 Figure2.7:HeterogeneousEectofHomePriceGrowthonFundAlpha:3Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+Xiiln(3yrHPIi;t1)I(FundMeasurei)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.ln(3yrHPIi;t1)isinteractedwithindicatorvariablesformeasuresoffundstyle,fundsizefocus,expenseratio,orTNA.Theomittedcategoryistheoneforthegroupofmanagerswhoareleastlikelytobeaectedbyhomepricegrowth(i.e.lowexpenseratio),exceptforfundsizewheretheomittedcategoryisrandomlypicked.Percentageslistednexttovaluesonthex-axisrepresentthepercentofthepopulationthatfallsintothecategorylisted.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managers,ifTNA,measuredin2000dollars,fallsbelow$15million,orifhomepricegrowthisnegative(onlymonthswithpositivehomepricegrowtharekept).OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE
95 77 Figure2.8:Distributionof1-yearChanges
77 Figure2.8:Distributionof1-yearChangesin3-yearHomePriceGrowthDistributionofthe1-yearchangeinln(3yrHPIi;t),measuredasofJuneeachyear. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE78 (a) (b)Figure2.9:ShareofFundManagersExtractingHomeEquityandAmountofHomeEquityExtractedFigureAshowstheproportionoffundmanagersextractinghomeequityinagivenyear.FigureBisahistogramoftheamountofmoneyextractedamongfundmanagerswhoextracthomeequityviaacashoutrenance.Valuesarewinsorizedatthe10%level,forreadability. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE79 Table2.1:SummaryStatisticsComparingtheMergedandNotMergedPopulationsNMeanP1P25P50P75P99StdDev MergedPopulation ManagerStatistics YearsExperienceWhenStarted77619185131722397PercentFemale889980.08000010.17PercentWithCFA889980.5600.330.57110.36PercentWithMBA862070.5700.330.50110.36 FundStatistics TNA($millions,2000dollars)8899816111778271978320555515AnnualExpenseRatio862011.12%0.20%0.91%1.11%1.33%2.15%0.37%Turnover106230.630.020.260.470.812.920.61Monthly88962-0.09-3.88-0.79-0.080.623.501.41NumberofManagers889983.261234102.05PercentofManagersMerged889980.560.130.330.500.831.000.29 NotMergedPopulation ManagerStatistics YearsExperienceWhenStarted191584184131722397PercentFemale2360050.10000010.23PercentWithCFA2360050.59000.67110.41PercentWithMBA2226260.5800.140.63110.40 FundStatistics TNA($millions,2000dollars)23600510401774226781143703230AnnualExpenseRatio2284361.18%0.29%0.95%1.15%1.39%2.25%0.38%Turnover276240.750.030.300.550.973.480.82Monthly231528-0.10-4.16-0.84-0.080.663.831.46NumberofManagers2360052.36112381.59Statisticscomparingthemergedpopulationtotheun-mergedpopulationforactivedomesticequitymutualfunds.Dataareaggregatedfromthefundbymonthlevel. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE80 Table2.2:EectofHomePriceGrowthonFundAlpha (1)(2)(3)(4)(5)AllPeriodsHPI0HPI0AllPeriodsAllPeriods ln(3yrHPIi
96 ;t1)-0.171-0.236
;t1)-0.171-0.236-0.0286(-3.03)(-2.73)(-0.21)ln(3yrHPI+i;t1)-0.258-0.301(-3.46)(-3.45)ln(3yrHPIi;t1)-0.148-0.129(-1.35)(-1.01) #Obs8771959799279208771971581R-squared0.08770.09150.1140.09010.0903 ControlsforFlowtandt13!t1?NNNNY tstatisticsinparenthesesp0:10,p0:05,p0:01Baselinemodelformonthlyforecastsofalpha,withfundandtimexedeects.Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE81 Table2.3:EectofHomePriceGrowthonFundFlow (1)(2)(3)(4)(5)1-monthFlowForecast1-yearFlowForecastAllPeriodsHPI0HPI0AllPeriodsAllPeriods ln(3yrHPIi;t1)0.004990.004650.00887(0.98)(0.72)(1.06)ln(3yrHPI+i;t1)0.005210.0435(0.82)(0.43)ln(3yrHPIi;t1)0.009170.0163(1.
97 03)(0.11) #Obs8383157202266298383171833R
03)(0.11) #Obs8383157202266298383171833R-squared0.1820.2290.2340.1840.424 tstatisticsinparenthesesp0:10,p0:05,p0:01Theeectofhomepriceshocksonfund ow.Estimationoftheregressionmodel:Flowi;t;a=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.Flowiscalculatedusinggrossreturnsandiscalculatedboth1-monthand1-yearahead.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE82 Table2.4:EectofHomePriceGrowthonFundAlphaforIndexandPassiveFunds (1)(2)(3)(4)(5)(6)(7)IndexFundsLowestQuantileHighestQuantileActiveShareActiveShareTrackingErrorTrackingError60%60%HPI0HPI0AllPeriodsAllPeriodsAllPeriodsAllPeriodsAllPeriods ln(3yrHPIi;t1)0.0746-0.0259(0.42)(-0.07)ln(3yrHPI+i;t1)-0.0206-0.217-0.585-0.127-0.365(-0.13)(-1.28)(-2.56)(-0.79)(-3.18)ln(3yrHPIi;t1)-0.000272-0.1130.0799-0.284-0.0198(-0.00)(-0.49)(0.19)(-1.39)(-0.13) #Obs63263362968817616174671080647913R-squared0.1770.1500.1470.1160.1520.1160.119 tstatisticsinparenthesesp0:10,p0:05,p0:01Placebotestforthebaselinemodelformonthlyforecastsofalphausingindexfundsandsplit
98 sbytrackingerrorandActiveShare.Estimatio
sbytrackingerrorandActiveShare.Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE83 Table2.5:EectofHomePriceGrowthonFundAlphabySecondHomeOwnershipStatus (1)(2)(3)(4)HPI0HPI0HPI0HPI0HasSecondDoesNotHaveHasSecondDoesNotHaveHomeSecondHomeHomeSecondHome ln(3yrHPI,PrimaryHomei;t1)-0.313-0.213-0.3340.0796(-1.47)(-2.18)(-0.97)(0.54)ln(3yrHPI,SecondHome(s)i;t1)0.2410.159(1.47)(0.87) #Obs893150868475123169R-squared0.1240.09230.1310.116 tstatisticsinparenthesesp0:10,p0:05,p0:01Robustnessofthebaselinemodelformonthlyforecastsofalpha,withfundandtimexedeects,usingdataforallresidentialrealestateholdingsoffundmanagers.Estimationoftheregressionmodel:bFFCi;t;a=c+1ln(3yrHPIPrimaryHomei;t1)+2ln(3yrHPISecondaryHome(s)i;t
99 0;1)+i+t+a+itawhere&
0;1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.Homepricegrowthiscalculatedseparatelyforprimaryandsecondaryhome(s).bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE84 Table2.6:EectofHomePriceGrowthonReturnsforBuyandSellTrades:1 (1)(2)(3)(4)(5)(6)(7)RBuyi;t+1RSelli;t+1RNewPurchasesi;t+1RBuysofExistingi;t+1RNotTradedi;t+1RPartiallySelli;t+1RFullySelli;t+1FlowRBenchi;t+1RBenchi;t+1RBenchi;t+1RBenchi;t+1RBenchi;t+1 ln(3yrHPI+i;t)-0.5780.0132-0.502-0.0767.0393(-2.43)(0.05)(-1.79)(-0.32)(1.35)ln(3yrHPIi;t)-0.349-0.1510.185-0.242-.014(-0.85)(-0.32)(0.49)(-0.61)(-0.27)ln(3yrHPIi;t)-0.2260.613(-1.13)(2.78) #Obs27460233842281323519229812331128383R-squared0.07160.1830.2040.1860.1870.1650.279MeanofDepVar0.07480.1520.1680.2010.1250.128-0.0057SDofDepVar2.5252.7622.5782.5242.5373.0200.181 tstatisticsinparenthesesp0:10,p0:05,
100 p0:01Modelofeectofhomepricegrowthon
p0:01Modelofeectofhomepricegrowthontradingbehaviorandperformance.Estimationoftheregressionmodel:Yi;t+1;a=c+ln(3yrHPIi;t)+i+t+a+i;t;awhereiarefundxedeects(foreachmergednon-disjointperiodoftime),tarequarterlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowth),andaarenon-parametricxedeectsforTNA.Yi;t+1;aare:thedierenceinreturnbetweenbuyandselltradeslessthebenchmarkreturn,thereturnonpurchasesofstocksintthatwerenotheldinquartert1lessthebenchmarkreturn,thereturnonpurchasesofstocksintthatwereheldinquartert1lessthebenchmarkreturn(atleast5%increaseinnumberofshares),thereturnonstockscurrentlyheldbutnottradedinquartertlessthebenchmarkreturn,thereturnonstockscompletelysoldinquartertint+1lessthebenchmarkreturn,thereturnonstockspartiallysoldinquartertint+1lessthebenchmarkreturn,and owbetweenquarterst1andt.Dataarequarterlyduetothetheregressionsusingquarterlyholdingsdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundquartersaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE85 Table2.7:EectofHomePriceGrowthonReturnsforBuyandSellTrades:2 (1)(2)(3)(4)(5)(6)I(RBuysofExistingi;t1!tI(RBuysofExistingi;t1!tI(RBuysofExistingi;t1!tI(RBuysofNewi;t1!tExistingBuys AllBuysFullLiquidations AllSalesRBenchi;t1!t)RBenchi;t1!t:80)RNotTradedi;t1!t)RBenchi;t1!t) ln(3yrHPI+i;t)0.09470.1140.08670.03060.0744-0.0711
101 (1.99)(2.36)(1.76)(0.71)(2.03)(-2.30
(1.99)(2.36)(1.76)(0.71)(2.03)(-2.30)ln(3yrHPIi;t)-0.0887-0.0762-0.0709-0.00181-0.0241-0.00613(-1.18)(-1.01)(-1.02)(-0.03)(-0.45)(-0.13) #Obs297782977829778297782977829778R-squared0.2800.2870.1290.2510.2430.219MeanofDepVar0.3520.3140.5350.2790.4870.344SDofDepVar0.4780.4640.4990.4480.3340.313 tstatisticsinparenthesesp0:10,p0:05,p0:01Modelofeectofhomepricegrowthontradingbehaviorandperformance.Estimationoftheregressionmodel:Yi;t+1;a=c+ln(3yrHPIi;t)+i+t+a+i;t;awhereiarefundxedeects(foreachmergednon-disjointperiodoftime),tarequarterlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowth),andaarenon-parametricxedeectsforTNA.Yi;t+j;aare:indicatorvariablesforifthereturnonexistingstockspurchasedinquartertbetweenquarterst1andtislessthanthebenchmarkreturnoverthesameperiod,80%ofthebenchmarkreturn,orthereturntostocksheldbutnottradedinquartert,indicatorvariableforifthereturnonnewstockspurchasedinquartertbetweenquarterst1andtisworsethanthebenchmarkreturnoverthesameperiod,thepercentofbuytradesthatwereincreasesinexistingpositions,andthepercentofselltradesthatwerefullliquidations.Dataarequarterlyduetothetheregressionsusingquarterlyholdingsdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundquartersaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE86 Table2.8:EectofChangesinHomePriceGrowthonActiveTradingPerformance (1)(2)(3)(4)(5),NetofT
102 radingCostsI(ActivePas
radingCostsI(ActivePassive)GrosslnTradingCostsi;t1!t+1 TNAitln(Turnoveri;t!t+1) ln(3-yrHPI+it)ln(3-yrHPIi;t12mo)-0.374-0.763-0.2750.991(-2.15)(-1.92)(-1.67)(2.09)ln(3-yrHPIit)ln(3-yrHPIi;t12mo)0.06800.2080.0314-0.0285(0.39)(0.63)(0.19)(-0.08)ln(3yrHPI+i;t)0.254(1.96)ln(3yrHPIi;t)-0.0966(-0.66) #Obs19641964196419646160R-squared0.3800.3610.3770.9260.793MeanofDepVar-0.07780.351-0.00182-13.80-0.734SDofDepVar0.2350.4770.2261.7680.908 tstatisticsinparenthesesp0:10,p0:05,p0:01Modelofeectofchangesinhomepricegrowthonthelong-runactivetradingperformancerelativetothefund'sownbenchmark.Estimationoftheregressionmodel:Activei;t12mo!t+12mo;czPassivei;t12mo!t+12mo;cz=c+(ln(3-yrHPIi;t)ln(3-yrHPIi;t12mo))+cz;t+Controlsit+i;t;czwherecz;tareannual(tisannual)timebycommutingzonexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowth)andControlsitiscomprisedof:Morningstarfundcategoryxedeects,numberofmanagerxedeects,anindicatorforifthemanagerownsasecondhome,anindicatorforifthemanagerhasanadvanceddegree(aboveundergraduate),andnon-parametricxedeectsforTNA,homevalue,combinedLTV(CLTV),averagezipcodelevelincome(fromthe2010IRSSOI),andthepercentofnon-whitehouseholdsatthezipcodelevel(fromthe2000Census).Activei;t12mo!t+12mo;czisthemonthlyalphaobtainedfroma2-yearregressionoftheactivereturnsmeasuredbetweenquarter3ofyeart1andquarter2ofyeart+1.Passivei;t12mo!t+12mo;czisthemonthlyalphaobtainedfroma2-yearregressionofthereturnstoholdingxedtheportfolioofstocksheldinquarter3ofyeart1betweenquarter3ofyeart1andquarter2ofyeart+1.Thelefthandsideisreplacedwithanindicatorvariableforiftheactivealphanetoftradingcostsbeatsthepassivealpha(column3)andthelogratiooftradingcostsoverthe2-year
103 periodrelativetoTNAinJuneofyeart(column4
periodrelativetoTNAinJuneofyeart(column4).Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsanddidnotsurvivetheperiodbetweenquarter3ofyeart1andquarter2ofyeart+1aredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundswithmorethan10managersorifTNA,measuredin2000dollars,fallsbelow$15millionarealsodropped.OnlyactivelymanagedUSdomesticequityfundsareincluded.Column5usesthesamplefromTable2andestimatestheeectofhomepricegrowthon1-yearaheadturnover,includeslnlaggedturnoverasacontrol,andusesannualdata. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE87 Table2.9:EectofHomePriceGrowthonRiskTaking (1)(2)(3)(4)(5)(6)(7)(8)(9)(Returns)TrackingErrorCAPMBeta(MarketRisk)HPI0HPI0AllPeriodsHPI0HPI0AllPeriodsHPI0HPI0AllPeriods ln(3yrHPIi;t1)-0.1100.0670-0.1250.0769-0.09320.0561(-2.92)(1.92)(-2.75)(1.94)(-2.47)(1.65)ln(3yrHPI+i;t1)-0.0769-0.0905-0.0804(-2.27)(-2.16)(-2.48)ln(3yrHPIi;t1)0.006280.01780.00586(0.17)(0.40)(0.17) #Obs486102475773367486232478173404486232478173404R-squared0.8790.9410.9000.8200.9220.8660.5760.7050.568MeanofDepVar1.2381.5441.3410.9161.2921.0430.01430.03550.0214SDofDepVar0.4080.3880.4270.3890.3800.4250.2280.1610.208 tstatisticsinparenthesesp0:10,p0:05,p0:01Regressionsshowingtheeectofhomepriceshocksonrisktaking.Estimationoftheregressionmodel:ln(RiskTakingi;t!t+12;a)=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswhereh
104 omepricegrowthissegmentedonbeingpositive
omepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.Risktakingvariablesare:standarddeviationofreturns,trackingerror,andCAPMbeta.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. CHAPTER2.PERSONALWEALTHSHOCKSANDINVESTMENTMANAGEROVERCONFIDENCE88 Table2.10:EectofHomeEquityExtractiononFundPerformance (1)(2)(3)(4)(5)AllManagers,OLS Extractors,1stStageExtractors,2ndStage AllManagers,1stStageAllManagers,2ndStage I(extractedit)-0.0354 -2.526 -3.179(-1.64) (-2.03) (-1.73)ln(3yrHPIi;t) 0.0787 0.0473 (2.69) (2.14) #Obs26647 1608816083 2664726638R-squared0.558 0.1170.144 0.1370.137 tstatisticsinparenthesesp0:10,p0:05,p0:01Modelofeectofhomeequityextractiononchangesinperformanceandtestsforiffundmanagersarepayingattentiontohomepricegrowth.Estimationoftheregressionmodel: i;t+1;a i;t1;a=c+\I(extractedit)+i+t+a+ i;t1;a+i;t;awhere i;t;aistheaveragemonthlyalphainquartert,I(extractedit)isanindicatorvariableequalto1ifafundmanagerextractedpersonalhomeequityinquartert.iarefundxedeects,tarequarterlytimexedeects,andaarenon-parametricxedeectsforTNA.tismeasuredasquarterlyobservationsandstandarderrorsareclusteredatthefundforeachmergednon-disjointperiodoftimelevel. i;t;aaretheaverageofthemonthlyalphasinquartertwherethemonthlyalphasareobtainedfrombackingoutthemonthlypricingerrorsfromafu
105 ndlevel(foreachmergednon-disjointperiodo
ndlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.2SLSregressionisestimatedincolumns3and5,wherethe1ststageinstrumentstheindicatorforextractinghomeequityinquartertwith3-yearhomepricegrowth(therststageresultsarepresentedincolumns2and4).Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. 89Chapter3DurationMeasurementandHedgingChannelsforGSEInsuredMortgageBackedSecurities13.1IntroductionInmostacademicterm-structuremodels,theinterestrateriskofanyparticularbondcanbeexactlyreplicatedwithaportfolioofbondsofothermaturities,soshockstosupplyordemandforanybondmaturitymustaecttheentireyieldcurve.However,underthe\preferredhabitat"viewofinterestrates,rstdescribedbyCulbertson(1957)andmodeledtheoreticallybyVayanosandVila(2009),investorclienteleshavepreferencesforparticularmaturities,soshockslocaltoaparticularmaturitymayaectthatinterestratewithoutaectinginterestratesofothermaturities.Thisviewisconsistentwithalargeliteratureovertheyearsshowingtheimportanceoflocalsupplyanddemandshocksforthelevelofinterestrateswithspecicmaturities(e.g.,GreenwoodandVayanos,2010,2014;ModiglianiandSutch,1966;Ross,1966;Wallace,1967;KrishnamurthyandVissing-Jrgensen,2011;Gagnon,Raskin,Remache,andSack,2011).Historically,thisliteraturehasfocusedonregulationchangesorothergovernmentactionsovertheyearsthathavesignicantlyaectedthesupplyordemandforbondsofaparticularmaturity.Duetothesheersizeoftheoutstandingstockofmortgage-backedsecuritiesintheU.S.(about$13t
106 rillionpre-crisis)andrecentcrisis-relate
rillionpre-crisis)andrecentcrisis-relatedstabilizationpolicyinitiativesonthepartoftheFederalReserveBoard,suchasQuantitativeEasingI{IIIandOperationTwist,thathavespecicallytargetedthepurchaseofmortgage-backedsecurities,itisim-portantbothforourgeneralunderstandingofthetermstructureandforevaluatingandmodifyingtheseinterventionpoliciestofullyunderstandtheexactchannelsthroughwhichshockstomortgagebackedsecuritiesaecttheTreasuryyieldcurve.Recently,someauthors(inparticular,Hanson,2014;Malkhozovetal.,2016)havear- 1Co-authoredwithAyaBellicha,RichardStanton,andNancyWallace. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES90guedthatthenegativeconvexityofmortgagesandmortgage-backedsecuritiesmeansthatthechangeinaggregateinterestrateriskcausedbydurationshiftsinthesesecuritiesisofcomparablemagnitudetothatcausedbygovernmentinterventions,andthatthiscanleadtoanamplicationofbond-marketshocks.Empirically,theseauthorslookforarelation-shipbetweenbondriskpremiaandmortgagedurationbyregressingexcessbondreturnsonmortgagedurationandothercontrols,suchastheCochraneandPiazzesi(2005)factor.Whenwestudytheseregressionsinmoredetail,wendthattheevidenceisratherweakerthanitrstseems,beingverydependentontheexactsampleperiodchosen,andalsodrivenatleastinpartbymodelingchangesbyBarclaysduringtheperiod.Additionally,Barclaysdurationislessanchoredtorealitybasedonhavingaweakrelationshiptounderlyingmort-gageprepayments.However,amorefundamentalproblemisthatthemechanismproposedtoexplaintheseresultsreliesonMBSholdersbuyinglongtermTreasurieswhendurationislow.AnalyzingthebehaviorofthemajorityofMBSinvestors,wendthatinvestorsgenerallydonotactinthisway.ToidentifytheimpactofdurationandMBSholdingsontheTreasurypositionsofbanksweusecallreportdata.WeareabletoestimatethechangeinTreasuryholdingsseparatelyforbanksthathedgeandbanksthatdonothedge,identiedbytheirholdingsofinterestratederivatives,inresponseto
107 duration.Forforeigninvestorsweusedatafro
duration.ForforeigninvestorsweusedatafromtheTreasuryInternationalCapital(TIC)datasetthatlistsmonthlyTreasuryholdingsbycountryformajorforeignholders.TheGSEsholdaverysmallamountofTreasurydebt,howevertheyareknowntobethemostaggressivehedgers.UsingquarterlydatafromFHFA/OFHEOregulatoryreportswestudytheimpactoftheGSEhedgingpositionsonexcessbondreturns.Unfortunately,forpensionsandretirementfundswearenotabletoobserveTreasuryorMBSholdings.Theannualcomprehensiveannualnancialreports(CAFRs)forpensionsandretirementfundsonlyreportthepercentageoftheirassetsinUSbonds,whichincludesbothTreasuriesandMBS(amongotherassets).WeestimatetheeectofdurationonTreasuryholdingsofmutualfundsusingdatafromCRSP.Lastly,usingNAICSdataweestimatetheprobabilityoflifeinsurancecompaniesbuyingTreasuriesinagivenmonthbasedonlaggedMBSduration.WendthattheonlyinvestorsthatmayfollowthemodelsofHanson(2014)andMalkho-zovetal.(2016)arelifeinsurancerms.Wealsondarelationwithbankshoweverwecannotruleoutthatthisismerelycorrelation.Lifeinsurancermmarketsharehasdeclinedovertheperiod,droppingbelow10%since1996andreaching4%in2016.Oftheinvestorswearenotabletostudy,hedgefundsandpensions/retirementfundsarethetwoinvestorgroupsthatmaytradealongtheHanson(2014)andMalkhozovetal.(2016)models.Thesetwoinvestorgroupsheldalmost25%oftheAgencyMBSmarket(includinghouseholdsandnonprotorganizations)inthelate1990s,howeverpostcrisistheirsharehasfallenbelow10%. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES913.2DurationDurationisanamegiventoanumberofmeasuresofinterest-ratesensitivity.Foraxed-couponbond,therstsuchmeasure,Macaulayduration(Macaulay,1938),isaweightedaverageofthetimetoeachpayment,theweightsproportionaltothePVofeachpayment(discountingusingthebond'syieldtomaturity,y).Forabondwithyieldtomaturityycompoundedntimesperyear,itsMacaulayduration,Dmac,isDmac=1 PnXi=1tiCi (1+y=n)nti;wheretiisthetime(inyears)u
108 ntiltheithpayment.Amorecommonlyusedmeasu
ntiltheithpayment.Amorecommonlyusedmeasureinpracticeismodiedduration,Dmod(Hicks,1939):2Dmod=Dmac (1+y=n):ItisasimplemattertoshowthatDmod=Dmac (1+y=n)=1 P@P @y;(3.1)soforsmallchangesiny,thebond'spricechangesbyapproximatelyP PDmac 1+y=ny=Dmody:Tohedgeaportfolio,weaddanothersecurityuntiltheportfolio'soveralldurationiszero.ThedenitionofbothMacaulayandmodieddurationrequiresthecash owsonthesecuritytobexed.Durationcan,however,beextendedtosecuritieswithinterest-rate-dependentcash ows(e.g.,securitieswithembeddedoptions)byusingEquation(3.1)asthedenitionof\eectiveduration,"De=1 P@P @y:3Allofthesemeasuresrelateabond'spricetochangesinitsownyield,whichcancauseproblemswhenaggregatingtotheportfoliolevel,sincetheyieldsondierentbondsdonotnecessarilymoveexactlytogether.Moreconsistentistomeasurethesensitivityofallbondstomovementsinthesameunderlyingstatevariable.FisherandWeil(1971)duration 2NotethatMacaulayandmodieddurationcoincidewithcontinuouscompounding,wheren!1.3Whenmodieddurationcanbecalculated,itisalwaysequaltoeectiveduration.However,eectivedurationcanbecalculatedforawiderrangeofsecurities. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES92issimilartoMacaulayduration,buteachcash owisdiscountedusingtheappropriate-maturityspotinterestrate.Itthusmeasuresabond'ssensitivitytoparallelshiftsintheentire(notnecessarily at)yieldcurve.Similarly,inmodelsbasedonthedynamicsoftheshort-termrisklessrate,r(e.g.,Vasicek,1977;Cox,Ingersoll,andRoss,1985),wecancalculatethedurationofasecurityrelativetor,1 P@P @r;andinmodelswithmorethanonefactor,wecancalculatedurationswithrespecttoeachoftheunderlyingstatevariables.4Givenaninterest-ratemodel,thederivativesabovecanbeevaluatedeitherinclosedformornumerically.However,anyerrorsinthemodelwillgiverisetoerrorsintheresultingdurationsandhedgeratios.Analternative,model-fr
109 ee,techniqueistocalculateasecurity's\emp
ee,techniqueistocalculateasecurity's\empiricalduration"(see,forexample,Hayre,2001,Chapter14),inwhichreturnsonthesecurityareregressedonchangesinoneormorestatevariablestoestimatedirectlytheaveragechangeinpriceforagivenchangeintheunderlyingvariable.5Anextensionofthisideaiskey-rateduration(Ho,1992),whereamultivariateregressionisrunofreturnsagainstchangesinseveraldierentinterestrates,thusestimatingthesensitivityofthesecuritytochangesineachofthese\keyrates"keepingtheotherratesconstant.3.2.1MeasuringthedurationofGSEMBSHanson(2014)andMalkhozovetal.(2016)focusontheroleofprepaymentrelatedshockstothedurationofoutstandingresidentialmortgagebackedsecuritiesthat,inturn,leadtolarge-scaleshockstothequantityofinterestrateriskbornebyprofessionalbondinvestors.BothpapersusedurationmeasuresobtainedfromDatastreamthataretheproductofpro-prietaryprepaymentmodelsdevelopedBarclaysCapital,formerlyLehmanBrothers.TheBarclaysU.S.MBSindexcoversmortgagebackedpass-throughsecuritiesguaranteedbyGovernmentNationalMortgageAssociation(GNMA),theFederalNationalMortgageAs-sociation(FannieMae),andtheFederalHomeLoanMortgageCorporation(FreddieMac),collectivelyknownasU.S.AgencyMBS.Theindexiscomposedofpass-throughsecuritiesbackedbyconventionalxed-ratemortgages.TheMBSindexdoesnotincludenon-agencyorprivate-labelMBS(e.g.,MBSbackedbyJumbo,Alt-A,orsubprimemortgages).Malkhozovetal.(2016)useaduration-to-worstmeasure(LHMNBCK(DU)inDatas-tream)whichisanMBSdurationcomputedusingthebond'snearestcalldateormaturity,whichevercomesrst.Theythenscaletheirduration-to-worstmeasurebytheaverageunitpriceofU.S.agencyMBSwhichproducesadollardurationperunitofMBSinthemarketnottheaggregateMBSdollarduration.Alimitationofthisdurationmeasureisthatit 4Hedgingaportfolioina(say)two-factorworldinvolvesaddingatleasttwoadditionalsecuritiesuntilbothdurationsequalzero.5Thisassumesthatthesensitivityremainsxedovertheperiodoftheregression(seeBoudoukh,Richard-son,Stanton,andWhite
110 law,1995,foradiscussionandextensions). C
law,1995,foradiscussionandextensions). CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES93ignoresfuturecash ow uctuationsduetoembeddedoptionality.Sincetheintentoftheirempiricalexerciseistomeasuretheimpactofprepaymentrelatedshocksonthequantityofinterestraterisk,aduration-to-worstmeasurewouldnotbeanaccuratecontrolforthechannelofinterest.Hanson(2014)usestwomeasuresofdurationalsoconstructedusingdatafromBarclaysCapitalmodelsandobtainablefromDatastream.Therstoftheseisaneectivedura-tion(correspondingto(LHMNBCK(DM)inDatastream)fortheBarclaysMBSIndexandmeasuresthepercentagechangeinU.S.agencyMBSmarketvaluefollowingashiftintheyieldcurve.HissecondpreferreddurationmeasureisthecontributionofMBSbondstotheBarclaysAggregateIndexduration.ThismeasureisconstructedbyweightingtheeectivedurationmeasurebytheratioofthemarketvalueofMBS,usingBarclaysU.S.MortgageBackedSecurities{MarketValue(MM),totheBarclaysmeasureoftheU.S.Aggregate{MarketValue(MM).ThisscaleddurationmeasureisthereforeEffectiveDurtMBSMVt AGGMVtwhereMBSMVisU.S.MortgageBackedSecurities{MarketValue(MM)andAGGMVtistheU.S.Aggregate{MarketValue(MM).Themeasure,capturesthefactthatshiftsinMBSdurationintheU.S.havehadagrowingimpactonaggregatebondmarketdurationduetothegrowthoftheMBSmarket.ThemeasureproxiesforthetransientcomponentofaggregatebondmarketdurationduetoMBSandconstituteshispreferredforecastingvariable.Weapplya\prepaymentmodelfree"empiricaldurationmeasureusingtheuniverseofoutstanding30-yearFannieMae(FNMA),FreddieMac(FHLMC),andGinnieMae(GNMA)MBS.OurempiricaldurationestimatesthesensitivityofdailyMBSpricechangestodailychangesin10-yearTreasuryyields.Weuse10-yearzerocouponTreasuryyieldsfromGurkay-nak,Sack,andWright(2007)andTBApricesattheagency,maturity,andcouponlevelfromEMBStoestimatethefollowingequation:TBAPricet;c;pTBAPricet1;c;p TBAPricet1;c;p=+yieldtyieldt1 100+t;c;p(3.2)where1ist
111 heempiricalduration,tistime(daily),cisco
heempiricalduration,tistime(daily),ciscoupon(in50bpsincrements),andpisprogram(i.e.FNMA30-yearorGNMA15-year).ThefollowinganalysisusesdataforFNMA,FHLMC,andGNMA30-yearMBSwithacouponbetween2.5and10%.Forourseconddurationmeasure,wescaleourempiricaldurationbythemarketvalueoftheoutstandingstockofU.S.MBSandthemarketvalueoftheBarclaysAggregatefollowingHanson(2014).6Oursecondmeasureisthus,EmpiricalDurtMBSMVt AGGMVt,whereMBSMVtistheEMBSmeasureofthemarketvalueoftheoutstandingstockofU.S.agencyMBSandAggMVtistheU.S.Aggregate{MarketValue(MM).OurnalpreferreddurationmeasureisourempiricaldurationmeasuretimesU.S.MortgageBackedSecurities{MarketValue(MM)toBarclaysaggregateeective,durationobtainedfromDataStream,timesU.S.Aggregate{MarketValue,EmpDurtMBSMVt AggDurtAggMVt.ThisistherelativecontributionofMBSdollardurationtotheaggregatedollarduration. 6OurEMBSmeasureofthemarketvalueoftheoutstandingstockofU.S.agencyMBSexactlymatchestheBarclaymeasureU.S.MortgageBackedSecurities{MarketValue(MM). CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES94Malkhozovetal.(2016)notesthatthedurationchannelisstrongerwhenGSEshareishigher.TheybasethisonthecorrelationbetweentherollingR-squaredfromtheregressionofexcessbondreturnondurationandtheshareofMBSheldbytheGSEs.ThiscontrastswiththeimplicationsofthemodelinHanson(2014).HansonnotesthattheMBSbuyersthatdeltahedgebearaconstantamountofinterestrateriskandarenotimportantforthechannel.AsshownintheGSEsection,theGSEsdonotin uenceexcessbondreturnsbytheirhedgingactivity.TheGSEsprimarilyuseswaps,swations,Treasuryfuturesoptions,Eurdollarfuturesoptions,andotherinterestratederviatiesaswellastheirissuanceofAgencydebttohedgetheirdurationexposure.TheydonotuseTreasurydebtforthispurpose.ConsequentlyanincreaseinGSEshareshouldreducetheimportanceofthedurationchannelonbondreturnsiftradinginthederivativesmarketsdoesnotin uenceexcessbondreturns.Onasimilarnote,themodelinMalkhozovetal.(2016)
112 reliesonthedierencebetweentheaverag
reliesonthedierencebetweentheaverageMBScouponandthe5-yearswaprate.NotethattheaverageMBScoupondoesnotaccountforthechangeindistribution.Italsodoesnotaccountforvariousmeasuresthatareparamountinprepaymentmodeling,suchas:SATO,thepercentunderwater,FICO,etc.3.2.2BarclaysModelingChangesStartinginNovember2008,Barclaysregularlyupdatedtheirprepaymentmodeltocapturechangesinthemarketandregulatoryenvironment(Risa,Ibanez-Meier,Fan,andMaoui(2008)andSrinivasanandVelayudham(2010)).Primarily,Barclaysisinterestedincapturingfrictionsassociatedwithmortgageterminations.Themodelchangesareineectstructuralbreaksinthedata.Thesechangescanhavedramatic(andpersistent)eectsonduration.TheBarclayseectivedurationmeasurechangedby1.64(from1.29to2.93versusachangefrom1.2to1.48intheempiricaldurationmeasure)betweenAugustandSeptember2010,primarilyduetothemodelchangeinSeptember2010.Figure3.1ashowsthedierencebetweenBarclayseectivedurationandempiricaldu-ration.Theseriesisroughlywhitenoisearoundzerountil2008.Startinginlate2008,themodelstartstodeviatefromthezerotrend,withthebiggestbreakinSeptember2010whenBarclaysintroducedchangestotheirprepaymentmodelthathadlargeeectsontheirdurationmeasures.Thisisalsoshowningure3.1b,whichcomparesthelevelsofthetwodurationseries.Asisclearfromthetwographs,Barclayseectivedurationmeasureisnearlyuniformlyhigher(implyingthebondsarelonger)thantheempiricaldurationmea-surebetweenSeptember2010andOctober2015.TheseresultssuggestthatrevisionstotheBarclaysprepaymentmodelledtopersistentunderpredictionsofprepaymentrelativetoactualGSEprepaymentspeedsformostofthelaterpartofthesample.3.2.3In uenceofExtremeObservationsGiventheissueswiththeBarclaysdurationmeasureandtherelativelystablenatureofdurationexceptforafewperiodsofextremeandpersistentchangesindurationitisnatural CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES95toseeiftheresultisdrivenbyafewextremeobservatio
113 nsforthesubsamplethatdoesnotincludemodel
nsforthesubsamplethatdoesnotincludemodelchanges.Werstconsiderourtwomeasuresofeectiveduration,Barclayseectivedurationandourempiricalduration.Column1ofTable3.1replicatesthendingofHanson(2014)butfortheperiodbetweenFebruary1996andOctober2015.Wendasimilarresult.Column2replicatesthisusingempiricaldurationandweagainndasimilarresult,albeitslightlysmaller.However,giventheissueswithmodelingchangesstartingintheendof2008wecannotbecertainthattheresultoverthisperiodisduetoportfoliorebalancingormodelingchanges.Toaccountforthis,welookattheperiodbetweenJanuary1989andAugust2008andincludeperiodxedeectsforthetworenanceperiods(August1998-January1999andAugust2002-June2003).Usingthisdaterangeandwiththesexedeects,asreportedincolumn3ofTable3.1,wendtheBarclayseectivedurationmeasureisnolongersignicant.Furthermore,onlytheearliestAugust1998toJanuary1999renanceepisodeissignicant.FortheempiricaldurationserieswecanincludethedatathroughOctober2015,howevertheseresultsmaybebiasedbytheactionsoftheFederalReserve.WewouldexpecttheFederalReserveinterventionstodirectlyin uenceexcess10yearTreasuryreturnsastheFederalReserveenteredthemarkettoin uencebondyields,especiallyduringOperationTwist.OperationTwiststartedinSeptember2011andlastedthrough2012,however,unliketheotherthreequantitativeeasingprogramsOperationTwist'ssoleintentionwastobuylongtermTreasurydebtandsellshorttermTreasurydebt.TheconsequencesoftheFed'stradingactivityshouldbeadecreaseintheexcessreturnmeasuresincethelongrateshoulddeclineaslongtermdebtpricesarebidupandtheshortrateshouldincreaseasshorttermdebtpricessello.Sinceitseemsunlikelythattheexcessbondreturndynamicsoverthisperiodareonlyrespondingtothe uctuationsinMBSduration,wealsoestimateaspecicationthatintroducesperiodcontrolsforQE1,QE2,QE3,OperationTwist,andthetaperingperiod,aswellasthexedeectsforthetwoearlierrenanceepisodes.FortheFederal
114 Reservecontrols,weincludevariablesequalt
Reservecontrols,weincludevariablesequaltothedollaramountoflong-termTreasuries(greaterthan5yearsmaturity)heldbytheFederalReserveduringtherespectiveperiod(and0otherwise).Asshownincolumn4ofTable3.1,wendthattheempiricaldurationvariableisnolongersignicant.Aswewilldiscuss,giventheissueswithmodelriskfortheBarclaysdurationmeasureoverthislaterperiodweputmoreemphasisontheempiricaldurationresultforthefullsample.Columns5-8ofTable3.1repeattheregressionsincolumns1-4butusethefractionofdollarMBSdurationtodollarmarketdurationtoaccountforgrowthinMBSinterestrateriskrelativetogrowthinoverallinterestraterisk.Overthelongperiodstudied,MBSandoveralldebtvolumesincreaseddramatically.Wendthattheempiricaldurationmeasureisnotsignicantwhenincludingthepreviouslymentionedperiodxedeects,howeverBarclaysdurationisalwayssignicant. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES963.2.4MortgageTerminationsAkeyreasonthatMBSduration uctuatesfromchangesininterestratesisthatunderlyingmortgageholdersprepaytheirmortgageasinterestratesfallorbecomelesslikelytoprepaytheirmortgageasinterestratesrise.Basedonthis,wecomputeameasureofthepercentofoutstandingmortgagesthatprepayeachmonth.ThismeasureshouldcloselyalignwithMBSduration,basedonthemechanismthroughwhichMBSduration uctuates.ToconstructameasureweusedatafromMcDash,whichprovidesuswithmonthlymort-gagepaymentinformationforeveryoutstandingmortgagefromtheeightlargestmortgageservicers.Werestrictto30yearxedraterstlienmortgagesheldinFreddieMac,FannieMae,orGinnieMaeMBSandremoveconstructionloans.ForeachmonthbetweenJanuary1992andApril2018,wecalculatethepercentofmortgagesthatvoluntarilyterminateasafractionofoutstandingmortgages,weightedbytheoriginalloanamount.Wedonotincludemortgagesthatterminateduetoforeclosure,servicingtransfers,orthatgomissinginthedata,astheseterminationsdonotcorrelatetochangesininterestrates.Inagureavailableuponre
115 quest,weshowthatthedynamicsofthismeasure
quest,weshowthatthedynamicsofthismeasurematchthedynamicsofasimilarmeasurecalculatedfromEMBSloanperformancedata.EMBSloanperformancedatain-cludesdataforeverymortgageinaGSEorGinnieMaepool(notjustfromtheeightlargestservicers).However,EMBSdatadonotprovideuswiththereasonforloantermination,whichincludesnon-voluntaryterminationsandservicingtransfers.Figure3.2comparesnormalizedversionsoftheempiricalandBarclaysdurationmeasuresagainstthevoluntaryterminationmeasure.Theterminationmeasuremovescloselywiththedurationmeasures,asexpected.Interestingly,empiricaldurationappearstomorecloselyalignwiththemeasureoftheunderlyingmortgageterminationscomparedtoBarclaysdura-tion.ThisisconrmedbyndinganR-squaredof0.35whenregressingempiricaldurationontotheterminationmeasure,comparedtoanR-squaredof0.19whenregressingBarclaysdurationontotheterminationmeasure.Thisshowsthateventhoughtheempiricaldurationmeasureis"modelfree",itmorecloselytracksrealitycomparedtothemodelbasedBarclaysdurationmeasure.Next,wereplicatetheearlierresultsusingtheterminationmeasure.Inaddition,McDashprovidesuswithmorerecentdata,whichallowsustoextendtheanalysisthroughApril2018.WendthatfortheJanuary1992throughApril2018period,bothBarclaysdurationandtheterminationmeasurearesignicantatthe5%level.However,whenweincludeperiodxedeectsforthelate1990sandearly2000srenancecyclesandthevariousquantitativeeasingmeasurestakenbytheFederalReserve,wendthatthemortgageterminationmeasurebecomesinsignicant.Barclaysdurationremainssignicantatthe10%level.However,thisresultispartlybasedondataduringtheperiodforwhichBarclaysactivelyadjustedtheirmodel.Lastly,whenwerestricttotheperiodafterthequantitativeeasingmeasuresbytheFederalReserveended(January2014throughApril2018),wendthatthemortgageterminationmeasuresbecomesinsignicant,whiletheBarclaysmeasureremainssignicant.Theseregressionsshowtheimportanceofthechoiceinameasuretore ectMBSinterestrater
116 isk.Withameasureofrealizedmortgagetermin
isk.Withameasureofrealizedmortgageterminations,wendaveryweakrelationship CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES97thatisbasedonafewuniqueepisodesandthathassincebecomeirrelevantinrecentyears.However,usingamodelderivedmeasureofMBSinterestraterisk,wendastrongereectthatiscurrentlyafactorin uencingTreasuryyields.BasedontheinherentweaknessesoftheBarclaysmodelderivedmeasureandtheconsistencyoftheresultsbasedontwoindependentmeasuresofMBSinterestrateriskthatarebasedonfundamentals,wendlimitedsupportoftheclaimthatMBSdurationhedgingaectsTreasuryyields.Tofurthershowthis,inSection3werigorouslyshowthatmostclassesofinvestorsdonotbuyTreasuriestohedgeMBSdurationrisk,whichisthechannelhighlightedbypreviousworktoexplainarelationshipbetweenMBSdurationandexcessTreasuryyields.3.3EvidenceforMBSInvestorHedgingUsingtheFederalReserveFlowofFunds,AgencyMBSinvestorsarepresentedinFigure3.3.TheinvestorswecanaccountforaretheFederalReserve,banks,foreigninvestors,theGSEs,mutualfunds,andlifeinsurancecompanies.Wedonotaccountforbrokers/dealers,fed-eral/state/localgovernment,property/casualtyinsurancecompanies,householdsandnon-protorganizations(includeshedgefunds),retirementandpensionfunds,issuersofABS,andREITs.Overall,weareabletoestimatethehedgingresponseofinvestorscomprisingatleast70%oftheMBSmarketsincetherstquarterof2001(barringafewmonthswheretheirsharedippedslightlybelow70%)andinsomemonthsmorethan80%ofthemarket.3.3.1TheFederalReserveSinceNovember2008theFederalReservehaspurchasedover$2.3trilliondollarsofmortgagebackedsecuritiesthroughitsquantitativeeasingprograms(FederalHousingFinanceAgencyOcerofInspectorGeneral,2014).Bytherstquarterof2014,theFederalReserveheld$1.5trillionFreddieMacPCs,FannieMaemortgagebackedsecurities,andGNMAmortgagebackedsecuritiesonitsbalancesheet(Patrabansh,Doerner,andAsin2014).TheFederalReservedoesnothedgenordotheycareaboutportfoliodurat
117 ionsothisimportantinvestorcannotimportan
ionsothisimportantinvestorcannotimportantforthemeasurementoftheoverallU.S.GSEMBSduration.Malkhozovetal.(2016)ndsthattheFederalReserve'smarketshareisnegativelycorrelatedwiththestrengthofthedurationchannel.ThisisattributedtotheFederalReserveabstainingfromhedging.However,theperiodwhenFederalReserveMBSholdingsarehigheristheperiodduringunconventionalmonetarypolicyandthusitisdiculttoattributethereductioninthestrengthofthedurationchanneloverthisperiodtoFederalReserveMBSholdingsorFederalReserveopenmarketoperationsmorebroadly(suchasOperationTwist).3.3.2GovernmentSponsoredEnterprisesMalkhozovetal.(2016)notesthatthedurationchannelisstrongerwhenGSEshareishigher.TheybasethisonthecorrelationbetweentherollingR-squaredfromtheregressionofexcess CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES98bondreturnondurationandtheshareofMBSheldbytheGSEs.ThiscontrastswiththeimplicationsofthemodelinHanson(2014).HansonnotesthattheMBSbuyersthatdeltahedgebearaconstantamountofinterestrateriskandarenotimportantforthechannel.Asshownbelow,theGSEsdonotin uenceexcessbondreturnsbytheirhedgingactivity.TheGSEsprimarilyuseswaps,swations,Treasuryfuturesoptions,Eurdollarfuturesoptions,andotherinterestratederivativesaswellastheirissuanceofAgencydebttohedgetheirdurationexposure.TheydonotuseTreasurydebtforthispurpose.ConsequentlyanincreaseinGSEshareshouldreducetheimportanceofthedurationchannelonbondreturnsiftradinginthederivativesmarketsdoesnotin uenceexcessbondreturns.Figure3.4showstheholdersofTreasurydebt.TheGSEshareisnotvisible,risingtoamaximumofonly1.8%overtheperiod.Figure3.5showsthedistributionoftheGSEsderivativesportfoliorevealingthatinterestrateswapsarethedominantinstrumentusedforhedging.Note,Figure3.5doesnotshowthenon-mortgageinvestmentsportfolio,whichcomprisesTreasurydebt.However,TreasurydebtisnotbrokenoutseparatelyandisrolledintotheOthercate-gory,showingthatitisaverysmallinvestmentcategoryfortheGSEs.Th
118 eGSEhedgingportfoliodataincludesholdings
eGSEhedgingportfoliodataincludesholdingsofinterestratederivativeproductsbytheGSEs(asshowninFigure3.5).UsingquarterlydatafromtheFHFA/OFHEOregulatoryreportswetesttheeectoftheGSEderivativesportfolioonlogexcessreturnfor10-yearzerocouponbonds.Thefollowingregressionisestimated:rx(10)t+12m=+1GSEHedgingPortfoliot+(10)t+12m(3.3)EquationsincludingBarclayseectiveandempiricaldurationarealsoestimated.ThedataarequarterlyandstandarderrorsareNeweyWestallowing18monthsofserialcorrelation.Regressionsofchangesindurationandinterestratederivativeholdingsonthechangeinexcessreturnarealsoestimated.IftheGSEsbuymoreinterestratederivativeswhendurationdeclinesandthebuyingofthesecontractsin uencesexcessbondreturnsthen1shouldbenegative.Tables3.3and3.4showthattheGSEhedgingportfoliodoesnothaveanimpactontheexcessreturnandthecoecientispositive.Includingdurationorestimatingtheregressionwithdierencesinsteadoflevelsdoesnotchangethending.Note,theFlowofFundsdata(L.211)measuresGSEholdingsofagencyandGSEbackedsecurities.However,theGSEsholdalargeamountofwholeloansandprivatelabelsecurities(PLS),whicharenotincludedintheFlowofFundsdata(Figure3.6).Furthercomplicatingmatters,inQ1of2010therewasanaccountingpolicychangethatdramaticallyreducedreportedGSEholdings(andincreasedtheirliabilities).TheretainedportfolioreportingdatadoesnotshowasimilardropinQ12010.UsingFHFA/OFHEOretainedportfolioreportingdatawendthatbefore2002theretainedportfoliodynamicsfollowedtheFlowofFundsdatabutdivergedafterwards.AgencyMBSholdings(FNMA,FHLMC,andGNMAMBS)intheFHFAdatacloselytracktheFlowofFundsdata.HowevertheFHFAdatadonotshowadeclinein2010fromtheaccountingchange. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES99Overall,theGSEsdonotdirectlyholdTreasuries,howevertheydoholdverylargeinterestratederivativeportfolios.Wetestwhetherthesein uenceexcessTreasuryyieldsandndthattheydonot.3.3.3BanksUsingca
119 llreportdatawetesttheimpactofdurationonT
llreportdatawetesttheimpactofdurationonTreasurybondholdings(andmorebroadlynon-mortgagedebtholdingsbymaturity)ofbanks.WedividethesampleintobankswithMBSpass-throughholdingsintheirnon-tradingaccountsthathaveandhaveneverheldinterestratecontractsintheirnon-tradingaccountsaswellasbanksthatneverheldMBSsecuritiesasacomparison.Figure3.7showstheholdingsofMBSpass-throughbyaccountforbanksthathaveheldinterestratecontractsandforbanksthathavenotheldinterestratecontracts.Mostbankshaveneverheldaninterestratecontract.Themajorityofthebankpass-throughholdingsareheldbythesmallersubsetofbanksthathaveheldinterestratecontractsandarekeptintheavailableforsaleaccount.However,thiswasnottruebeforetheearly2000s.Thetradingassetsaccountisomitted,howeveranegligibleamountofMBSisheldinthisaccount.Weusefourdierentdependentvariables.First,welookattheoverallTreasurytoassetratio(thesumofRCFD0211[heldtomaturityamortizedcost]andRCFD1287[availableforsalefairvalue]overRCFD2170[totalbalancesheetassets]).However,thisdoesnotallowustodecomposebymaturity.Forthatwelookatthenon-mortgagedebttoassetratio(thesumofRCFDA549/550/551/552/553/554[securitiesissuedbytheU.S.Treasury,U.S.Governmentagencies,andstatesandpoliticalsubdivisionsintheU.S.;othernon-mortgagedebtsecurities;andmortgagepass-throughsecuritiesotherthanthosebackedbyclosed-endrstlien1-4familyresidentialmortgages,byremainingmaturityornextrepricingdate]overRCFD2170).Forthisvariablewealsolookattheratiofordebtwiththreeorfeweryearsremainingmaturityandfordebtwithmorethanthreeyearsremainingmaturity.Figure3.8showstheunweightedaverageTreasuryandnon-mortgagedebt(asdenedabove)toassetratiosforallbanks(includingthosethatdonotholdMBS).Wendthatsince2002,bankshaveheldaverysmallportfolioofTreasuries,howevertheydohavesizableholdingsofothernon-mortgagedebt(suchasagencydebtandmunicipaldebt).Thisislogicalastheseformsofdebtarealsoverysafebutprovideahigheryield.Thisisalsoconrmedbythe owoffundsdata(Fi
120 gure3.4)whichshowedthatbankshaveheldaneg
gure3.4)whichshowedthatbankshaveheldanegligibleamountofTreasuriessince2002.Thefollowingequationisestimated:Debtt;i Assetst;i=+1Durationt12m+2MBSSecuritiest12m;i Assetst12m;i+3MBSSecuritiest12m;i Assetst12m;iDurationt12m+Xi;t12m+t;i(3.4)whereiisabank,andXiarebankxedeectsandtimevaryingcontrolsfortheratioofresidentialloanstototalloans,loantoassetratio,capitalandliquidityratios,deposit CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES100andfundingcosts,anddeposittoassetratio.ThemeasureofMBSsecuritiesisthesumofRCFDG300/304/308/303/307/311(heldtomaturityamortizedcostandavailableforsalefairvalueresidentialpass-throughsecuritiesissuedbyGNMA/FNMA/FHLMC/other).ThedataarequarterlybetweenJune1997andDecember2013andstandarderrorsareclusteredatthebanklevelallowingforcorrelationinthestandarderrorswithinbanks.Wereportresultsusingonlyempiricalduration,asitdoesnotsuerfrommodelrisk.IfbanksincreaseTreasuryholdingsfollowingadeclineindurationwewouldexpect1tobenegative.Wewouldalsoexpect3tobepositiveifbanksthatholdmoreMBSaremoresensitivetoduration(3isexpectedtobepositiveinthiscaseas2isnegative,aswillbeshownintheregressionsbelow).TheeectshouldbemuchstrongerforbanksthatdonothaveinterestratederivativeholdingsandforTreasurieswithlongerduration(assumingbanksholdinginterestratederivativeshedgedtheirMBSpositions).Table3.5showsresultsusingdataforbanksthathaveheldMBSintheirnon-tradingaccountsandthathaveneverheldinterestratederivativesintheirnon-tradingaccountsincolumns1-4andbanksthathaveheldinterestratederivativesintheirnon-tradingaccountsincolumns5-8.Columns1and5showthatTreasuryholdingsoverallincreaserelativetoassetswhendurationdeclines.Columns2and6showadierentpatternforallnon-mortgagedebtrelativetoassets.Banksthatdonotholdinterestratederivativesdecreasethisvariablewhendurationdeclines,howeverlon
121 gdurationdebtstillincreasesfollowingdecl
gdurationdebtstillincreasesfollowingdeclinesinduration(column3).Shortdurationbondsseemtodominateleadingtotheoverallpositiverelationshipfoundincolumn2.Banksthathaveheldinterestratederivativesincreasethisvariableinresponsetodeclinesinduration,withanegativecoecientforlongdurationnon-mortgagedebttoassets(column7).Theinteractiontermisnotsignicantincolumns1and6-8indicatingmixedresultsforthesensitivityofbanksrelativetothesizeoftheirMBSholdings.Overall,wendanegativerelationbetweenMBSholdingsandTreasuryholdingsindicatingapossiblecrowdingouteectorsubstitutioneect.Asaplacebotest,wereplicatetheseregressionsforthepopulationofbanksthatneverheldMBSsecuritiesintheirnon-tradingaccounts.MBSdurationshouldhavenoimpactondebtholdingsforthesebanks,howeverwendsimilarresults.Table3.6showsthatthesebanksdoinfactincreasetheirTreasuryandlongdurationnon-mortgagedebtwhendurationdeclines(columns1and3).Thisnding,alongwiththeinsignicantinteractiontermsintable3.5indicatesthatthereispotentiallyanotherchanneldrivingtheseresults.WhendurationdeclinesitistypicallywhentheFederalReserveislooseningpolicythusindicatingheightenedriskaversionandlowin ation.Thisisanenvironmentconducivetobuyingsafelongdurationassets.BasedontheseregressionswecannotconcludethatMBSdurationresultsarecausal.3.3.4MajorForeignHoldersSince2002,foreigninvestorshaveheldbetween10and22%ofoutstandingAgencyMBS(Figure3.3).Furthermore,foreigninvestorsaremajorholdersofUSTreasurydebt,holding CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES101morethan40%inrecentyears(Figure3.4).ThisisaninvestorgroupthatholdsasizableMBSportfolioandalsoholdsasignicantportionofTreasurydebt.Figure3.9showsthatChinaandJapaneachholdroughly10%oftotalUSTreasurydebt.UsingdatafromtheTreasuryInternationalCapital(TIC)datasetthefollowingequationisestimated:TSYHoldingst;t+12m TSYHoldingst=+1Durationt12m;t+2
122 ;FXRatet12m;t+t;t+12m(3.5)where
;FXRatet12m;t+t;t+12m(3.5)wheretheforeignexchangerateishowmanyunitsofthecountry'scurrencybuysoneUSDollar.ThedataaremonthlyandthestandarderrorsareNeweyWestallowingfor18monthlags.RegressionsareestimatedusingpercentagechangeinTreasuryholdingsinsteadofthedierenceinTreasuryholdingstocontrolforthemassiveincreaseinholdingsovertheperiod.IfthesecountriesbuyTreasurieswhendurationdecreasestoextendthedurationoftheirportfoliosweshouldnd1tobenegative.InvestorsinthesecountriesmayalsobuyTreasurieswhentheircurrencyappreciates,inwhichcasewewouldexpect2tobenegative.However,ifthecountry'scurrencyfreely oatsthenlowdurationmayalsocoincidewithamorevaluableforeigncurrencyforthecountryaslowdurationmaybefromalowFedFundsrate,whichwoulddepreciatetheUSdollarrelativetoforeigncurrencies.AmorevaluableforeigncurrencymayleadtoincreasedbuyingofUSTreasurydebt.Table3.7showsresultsusingBarclayseectiveduration.Thechangeindurationvariableisneversignicant.Usingempiricaldurationwendsimilarresults(Table3.8).Thus,thelargestforeignholdersofUSTreasuriesdonotbuyTreasurieswhendurationdeclines.3.3.5MutualFundsUsingdatafromCRSP,wetesttheimpactofdurationonTreasurybondholdingsofmutualfunds.ThedataareavailablequarterlybetweenQ12010(whenthegovernmentbondholdingpercentagesrststartedbeingreported)andQ22016.Thefollowingequationisestimated:PerGovtBondst;i=+1Durationt12m+Xi+t;i(3.6)whereiisamutualfund,durationiseitherempiricalorBarclayseectiveduration,andXiaremutualfundxedeects.Standarderrorsareclusteredatthemutualfundlevel.Regressionsarealsoestimatedonchangesinsteadoflevels.WeestimatetheequationforallmutualfundsinthesampleandformutualfundsthateverhadMBSholdings.DataonMBSholdingsbecameavailableinOctober2010.IfmutualfundsincreaseTreasuryholdingsfollowingadeclineindurationwewouldexpect1tobenegative.Table3.9reportsresultsfortheestimatesoftheequationonlevelsandshowsthat1ispos
123 itiveandsignicantacrossallspeci
itiveandsignicantacrossallspecications.Table3.10reportsresultsforchangesandnds1ispositiveandsignicantacrossallspecications.TheseresultsshowthatmutualfundsdonotincreasetheirTreasuryholdingswhendurationdeclines. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES1023.3.6LifeInsuranceCompaniesLifeinsurancecompanyliabilitiesaregenerallylongdurationandrequireaminimumlevelofreturnwithminimalrisk.Tosatisfytheirliabilitieslifeinsurancecompaniesprimarilyholdxedincomeassets.Berends,McMenamin,Plestis,andRosen(2013)ndthat75%oflifeinsurancegeneralaccountassetsareinbonds.Ofthebondholdings60%areincorporatebonds,18%areinMBS(bothprivatelabelandGSE),andonly7%areinTreasuries.Cor-roboratingthisnding,gure3.4showsthatoverthisperiodinsurancecompanies,includingpropertyandcasualtyinsurancecompanies,heldaverysmallshareofTreasurydebt.Figure3.3showsthatinsurancecompaniesheldroughly10%ofoutstandingAgencyMBSin1985howeverthissharehasdeclinedovertime(70{80%oftheseholdingsarefromlifeinsurancecompanies).UsingdatafromNAICSweareabletoidentifylifeinsurancermsthatholdMBS(eitherGSEorprivatelabel)andtheirmonthlytradingofTreasurysecurities.Weusedatafromthe300les,ScheduleDPart1,whichreportssecuritiesheldasofyearend.Thereisalsoa303/304le,ScheduleDPart3,whichreportsalltradesinayear.Howeverdataforthe303/304lesendsin2007,whiledataforthe300lesendsin2012.Thedownsidetousingthe300lesisthatthereportingwillnotcapturesecuritiesheldandthensoldafterashortperiodoftimeorsecuritiesthatmaturebeforetheendoftheyearthattheyarepurchased.However,sinceweareinterestedinTreasuryholdingsthatareboughtforthepurposeofextendingportfoliodurationduringperiodsofpersistentdeclinesinMBSdurationthisislikelytobelessofanissue.IntermsofabsolutenumbersofTreasurypurchased,the300lescontain97%ofthenumberofTreasurypurchasescomparedtothe303/304les,whilethe300lesonlycont
124 ain79%ofalltradescomparedtothe303/304
ain79%ofalltradescomparedtothe303/304les.ThusitseemsthatTreasuriesarelikelytobeheldforextendedperiodsoftime.However,comparingcountsbytuplesofrmandyearandmonthofthetradeonly82%ofTreasuriesinthe303/304lesarematchedtothe300les,thusthereisvariationinthehighfrequencydata.Forrobustness,wereportregressionsfortheoverlappingperiodof2001{2007forboththe300and303/304les.TotestiflifeinsurancecompaniesbuyTreasurieswhenMBSdurationislowweestimatealinearprobabilitymodeloftheprobabilityofbuyingTreasuriesinagivenmonthbasedonlaggeddurationwithrmxedeects.Theequationestimatedis:1t;i=+1Durationtj+Xi+t;i(3.7)where1t;iisequalto1ifrmiboughtTreasuriesinmontht,DurationtjiseitherBarclayseectiveMBSdurationorempiricalduration(wherejisthelagrelativetothemonthofbuyingTreasuriesandiseither1,6,or12months),andXiarermxedeects.Thestandarderrorsareclusteredatthermlevel.OuridenticationofTreasuriesincludesTIPSandexcludeswhenissuedandSTRIPS.TheregressionsareestimatedusingdataforrmsthatheldMBSsecurities(identicationofMBSsecuritiesincludesGSEdebt)atanypointinthedata. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES103IflifeinsurancecompaniesbuyTreasurieswhendurationislowwewouldexpect1tobenegative.Table3.11showsthat1isnegativeandsignicant,indicatingthatlifeinsurancermsbuyTreasurieswhenMBSdurationislow.However,thecoecientissmallindicatingasmallrelativeincreaseinTreasurypurchases.ManyrmsinfrequentlybuyTreasuries.13%neverboughtTreasuriesand65%boughtTreasuriesin12orfewermonthsoverthe12yearperiod.WeimputethematurityoftheTreasuriespurchasedbasedontheyearofthetradeandthestatedmaturityoftheTreasury.Roughly53%haveamaturityof5orfeweryearsand44%haveamaturityofmorethan5years(wearemissingthematurityfor3%ofthebonds).Furthermore,only33%haveamaturitygreaterthan8years,thustheTreasuriespurchasedte
125 ndtobeofshortermaturity.Column9showsresu
ndtobeofshortermaturity.Column9showsresultsforthelinearprobabilitymodelestimatedwiththelefthandsideequaltooneifaTreasurywith5yearsispurchasedinthegivenmonthand0otherwise.Conversely,column10isestimatedwiththelefthandsideequaltooneifaTreasurywith5yearsmaturityispurchasedinthegivenmonth.Bothcoecientsarenegativeandsignicant.Asarobustnesscheck,columns11{12showthatweobtainsimilarresultsifweuseeitherthe300or303/304lesoverthe2001{2007period.Table3.12showsresultsusingempiricaldurationandresultsaresimilar.3.4ConclusionsWeproposeanempiricaldurationmeasureforthestockofU.S.AgencyMBSthatappearstobelesspronetomodelriskthanmeasuressuchastheBarclaysEectiveDurationmeasure.Wendthatthismeasuredoesnotappeartohaveastrongeectonthe12-monthexcessreturnsoften-yearTreasuriesaswouldbeexpectedifshockstoMBSdurationleadtocommensurateshockstothequantityofinterestrateriskbornebyprofessionalbondinvestors(Hanson,2014;Malkhozovetal.,2016).Giventhisnegativereducedformresult,wethenexplorethemortgageandtreasuryhedgingactivitiesoftheprimaryMBSinvestorssuchascommercialbanks,insurancecompanies,theagencies,theFederalReserveBank,mutualfunds,andforeigninvestors.WendthattheonlyinvestorsthatmayfollowthemodelsofHanson(2014)andMalkhozovetal.(2016)arelifeinsurancermsandpossiblybanks.Lifeinsurancermmarketsharehasdeclinedovertheperiod,droppingbelow10%since1996andreaching4%in2016.Oftheinvestorswearenotabletostudy,hedgefundsandpensions/retirementfundsarethetwoinvestorgroupsthatmaytradealongtheHanson(2014)andMalkhozovetal.(2016)models.However,althoughthesetwoinvestorgroupsheldalmost25%oftheAgencyMBSmarket(includinghouseholdsandnonprotorganizations)inthelate1990s,postcrisistheirsharehasfallenbelow10%. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES1043.5FiguresandTables (a) (b)Figure3.1:ComparisonofBarclaysEectiveDurationandEmpiricalDurationFigureAshowsthedi
126 ;erencebetweenBarclayseectivedurati
;erencebetweenBarclayseectivedurationandempiricaldurationfromFebruary1996toOctober2015.TheverticaldashedgraylinerepresentsSeptember2010.FigureBcomparestheempiricaldurationandBarclaysdurationseriesinlevels. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES105 (a) (b)Figure3.2:ComparisonofDurationMeasures(EmpiricalandBarclayseective)AgainstTerminationMeasureFiguresA(empiricalduration)andB(Barclayseectiveduration)comparedurationmeasuresagainstthevoluntaryterminationmeasureconstructedfromMcDashdata.Thetimeseriesarenormalized(de-meanedanddividedbystandarddeviation)andthevoluntaryterminationmeasureismultipliedby-1tobecomparabletoduration. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES106 Figure3.3:DistributionofAgencyandGSEBackedSecuritiesHoldingsbyInvestorGroupThedataarefromtheFederalReserve owoffundsandarequarterlyfromQ11985toQ12016(leftaxis).Thesolidline(rightaxis)showstotaloutstandingAgencyandGSEbackedsecuritiesinUSDmillions. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES107 Figure3.4:DistributionofUSTreasuryHoldingsbyInvestorGroupThedataarefromtheFederalReserve owoffundsandarequarterlyfromQ11985toQ12016(leftaxis).Otherincludesnon-nancialcorporatebusiness,non-nancialnon-corporatebusiness,closedendfund,exchangetradedfund,issuerofABS,securitybrokeranddealer,andholdingcompanyholdingsofUSTreasuries.Thesolidline(rightaxis)showstotaloutstandingUSTreasurydebtinUSDmillions. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES108 (a) (b)Figure3.5:FannieMaeandFreddieMacDerivativesPortfoliosFiguresA(FannieMae)andB(FreddieMac)showthetotaldollaramount(notional,inmillions)andcompositionofthenancialderivativesportfoliosusingquarterlydatafromtheFHFA/OFHEOannualreportstoCongressbetweenQ12000andQ42015.ThecategoriesofsecuritiesarenotconsistentbetweenFannieMaeandFr
127 eddieMac.ThedataarenotavailableforFreddi
eddieMac.ThedataarenotavailableforFreddieMacbetweenQ12003andQ32003. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES109 Figure3.6:ComparisonofGSERetainedPortfolioReportingtoCongressvsFederalReserveFlowofFundsComparestheaggregatesizeoftheGSEretainedportfolios(forFannieMaeandFreddieMac)asreportedbytheFHFA/OFHEOannualreportstoCongress(dottedline)andtheAgencyandGSEbackedholdingsoftheGSEsasreportedbytheFederalReserve owoffundsdata(solidline).WithintheportfolioreportedbytheFHFA/OFHEOannualreportstoCongressthegureshowsthecompositionbysecuritytype(AgencyMBS,wholeloans,privatelabelMBS,andmortgagerevenuebonds).Thedataareannualbetween1998and2015. Figure3.7:BankMBSPass-ThroughHoldingsBankMBSpass-throughholdingsgroupedbyaccountandifthebankeverheldaninterestratecontract.Thedataarefromthecallreportsandarequarterlybetween1995and2013. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES110 Figure3.8:BankRatioofTreasuriestoAssetsUnweightedaverageTreasurytoassetratioandnon-mortgagedebttoassetratiofromthecallreports.ThedataarequarterlybetweenJune1975andDecember2013andincludesbanksthatdonotholdMBS. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES111 Figure3.9:ShareofUSTreasuryDebtHeldbyChinaandJapanThepercentageoftotaloutstandingUSTreasurydebt(fromtheFederalReserve owoffundsdata)heldbymainlandChinaandJapanasreportedintheTreasuryInternationalCapitalSystem(TIC)data.DataarequarterlybetweenQ22000andQ12016. CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES112 Table3.1:EectofMBSDurationonTreasuryExcessBondReturns (1)(2)(3)(4)(5)(6)(7)(8) BarclaysDurationt2.5772.589(3.01)(1.66)EmpiricalDurationt1.7231.186(2.43)(1.87)DatesinAug98-Jan99-13.61-16.44-11.54-16.80(-4.37)(-9.82)(-3.61)(-11.01)DatesinAug02-Jun032.755-2.0454.688-2.888(0
128 .64)(-1.14)(1.15)(-1.87)QE11.474
.64)(-1.14)(1.15)(-1.87)QE11.4741.242(2.44)(2.17)QE22.2032.233(10.43)(10.43)QE3-0.297-0.239(-2.11)(-1.43)Tapering0.1010.189(1.21)(2.04)OperationTwist-0.299-0.356(-1.52)(-1.89)BarDurtMBSMVt AggDurtAggMVt55.9156.42(3.66)(2.33)EmpDurtMBSMVt AggDurtAggMVt22.0812.42(1.96)(1.54)Constant-3.684-0.222-4.2191.576-8.1210.223-8.8402.502(-1.07)(-0.08)(-0.70)(0.61)(-2.02)(0.07)(-1.38)(1.09) #Obs237237236237237237236237DateRangeFeb96-Oct15Feb96-Oct15Jan89-Aug08Feb96-Oct15Feb96-Oct15Feb96-Oct15Jan89-Aug08Feb96-Oct15 tstatisticsinparenthesesp0:05,p0:01 RegressionsmeasuringtheeectofeectiveMBSdurationandMBSinterestrateriskrelativetoaggregateinterestraterisk(theratioofMBSdollardurationtoaggregatexedincomedollarduration)on10yearTreasury12monthexcessbondreturns.ForMBSdurationweuseeitherBarclayseectivedurationorempiricalduration.ForaggregatexedincomedurationweusetheaggregatexedincomedurationprovidedbyBarclays.MBSandtotalxedincomemarketvalue(AGG)areprovidedbyBarlcays.TheMBSmarketvalueonlyincludesGSEandGinnieMaeMBS.BarclayseectivedurationbeginsinJanuary1989,whiletheempiricaldurationseriesbeginsinFebruary1996.BothserieshavedatathroughOctober2015.ThemodelisttedonmonthlydataandstandarderrorsareNeweyWestwith18lagstoaccountfortheoverlappingstructureofthedata.Theregressionmodelis:rx(10)t+12m=+1Durationt+(10)t+12m CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES113 Table3.2:EectofMortgageTerminationsonTreasuryExcessBondReturns (1)(2)(3)(4)(5)(6) %VoluntaryPayot1003.6373.6094.204(2.19)(1.44)(0.68)BarclaysDurationt2.4522.5453.422(2.69)(1.92)(3.73)DatesinAug98-Jan99-16.25-13.23(-9.73)(-5.25)DatesinAug02-Jun031.2093.089(0.32)(0.88)QE14.1185.049(
129 1.61)(2.29)QE214.9711.92
1.61)(2.29)QE214.9711.92(9.01)(7.43)QE3-1.486-6.090(-0.50)(-2.86)Tapering3.335-0.717(2.55)(-0.30)OperationTwist-1.267-1.745(-0.41)(-0.77)Constant7.260-4.1766.936-4.5075.031-12.96(5.30)(-1.19)(4.75)(-0.91)(0.92)(-4.67) #Obs3163163163165252DateRangeJan92-Apr18Jan92-Apr18Jan92-Apr18Jan92-Apr18Jan14-Apr18Jan14-Apr18 tstatisticsinparenthesesp0:10,p0:05,p0:01RegressionsmeasuringtheeectofBarclayseectiveMBSdurationandunderlyingvoluntarymortgageterminationson10yearTreasury12monthexcessbondreturns.VoluntarymortgageterminationsarecalculatedfromMcDashandarebasedon30-yearxedraterstlienmortgages,excludingconstructionloans,thataresecuritizedintoGSEorGinnieMaeMBS.ThedataaremonthlybetweenJanuary1992andApril2018andstandarderrorsareNeweyWestwith18lagstoaccountfortheoverlappingstructureofthedata.Theregressionmodelis:rx(10)t+12m=+1Durationt+(10)t+12m CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES114Table3.3:EectofGSEHedgingPortfoliosonTreasuryExcessBondReturns (1)(2)(3)(4)(5) BarclaysDurt2.3842.880(2.95)(2.63)EmpiricalDurt1.7441.857(2.12)(2.39)HedgingPortfoliot1.2303.1751.815(0.87)(1.79)(1.33)Constant-2.5080.4363.499-9.525-2.946(-0.84)(0.14)(1.25)(-1.41)(-0.88) #Obs6363636363 tstatisticsinparenthesesp0:05,p0:01RegressionsmeasuringtheeectoftheGSEhedgingportfolioson10yearTreasury12monthexcessbondreturns.MBSdurationisalsoincludedincolumns1,2,4,and5.ForMBSdurationweuseeitherBarclayseectivedurationorempiricalduration.TheGSEhedgingportfoliodataarequarterlyandwereretrievedfromthehistoricalFHFA(OFHEOpriortothecreationofFHFA)reportstoCongress.GSEhedgingportfoliodataarethesumofnancialderivativesholdingsofFannieMaeandFreddieMac.ThedatastartinQ12000andendinQ32015.StandarderrorsareNeweyWestw
130 ith18lagstoaccountfortheoverlappingstruc
ith18lagstoaccountfortheoverlappingstructureofthedata.Theregressionmodelis:rx(10)t+12m=+1HedgingPortfoliot+2Durationt+(10)t+12m CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES115Table3.4:EectofChangeinGSEHedgingPortfoliosonChangeinTreasuryExcessBondReturns (1)(2)(3)(4)(5) BarclaysDurt;t12m-1.573-1.665(-1.70)(-1.85)EmpiricalDurt;t12m-1.446-1.452(-1.75)(-1.73)GSEHedgingPortt;t12m-1.144-1.836-1.237(-0.31)(-0.48)(-0.32)Constant-0.0806-0.266-0.177-0.00385-0.220(-0.04)(-0.14)(-0.10)(-0.00)(-0.11) #Obs5656565656 tstatisticsinparenthesesp0:05,p0:01Regressionsmeasuringtheeectofthe12monthchangeinGSEhedgingportfoliosonthe12monthchangein10yearTreasury12monthexcessbondreturns.The12monthchangeinMBSdurationisalsoincludedincolumns1,2,4,and5.ForMBSdurationweuseeitherBarclayseectivedurationorempiricalduration.TheGSEhedgingportfoliodataarequarterlyandwereretrievedfromthehistoricalFHFA(OFHEOpriortothecreationofFHFA)reportstoCongress.GSEhedgingportfoliodataarethesumofnancialderivativesholdingsofFannieMaeandFreddieMac.ThedatastartinQ12000andendinQ42014.StandarderrorsareNeweyWestwith18lagstoaccountfortheoverlappingstructureofthedata.Theregressionmodelis:rx(10)t+24m;t+12m=+1GSEHedgingPortfoliot;t12m+2Durationt;t12m+(10)t+24m;t+12m CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES116 Table3.5:EectofMBSDurationonBankTreasuryBondHoldingsforBanksWithMBSHoldings (1)(2)(3)(4)(5)(6)(7)(8) EmpiricalDurt12m-0.0008720.000715-0.004410.00513-0.00136-0.00155-0.003850.00230(-3.96)(3.31)(-21.57)(22.36)(-14.71)(-5.25)(-15.07)(9.56)MBSHoldingst12m Assetst12m-0.0783-0.480-0.188-0.292-0.
131 0564-0.412-0.180
0564-0.412-0.180-0.232(-12.27)(-29.02)(-13.31)(-24.64)(-8.04)(-16.89)(-9.19)(-13.44)MBSHoldingst12m Assetst12m-0.000141-0.0141-0.00979-0.004290.003140.002640.00275-0.000102EmpiricalDurt12m(-0.15)(-5.02)(-3.88)(-2.41)(3.43)(0.60)(0.79)(-0.04)Constant-0.02770.2600.1650.09450.001200.1730.1050.0687(-5.19)(17.53)(13.97)(8.89)(0.21)(8.99)(8.48)(4.35) #Obs327440327440327440327440115350115350115350115350R-squared0.5720.8160.7120.6720.4650.7590.6800.619MaturityofDepVariableTreasAll3yrs3yrsTreasAll3yrs3yrsBankHasHeldIRNoIRNoIRNoIRNoIRHasIRHasIRHasIRHasIR tstatisticsinparenthesesp0:05,p0:01Regressionsmeasuringtheeectof12monthlaggedMBSdurationontheratioofdebtholdingstototalassetsatthebanklevelforbanksthathaveneverheldinterestratederivativesintheirnon-tradingaccountsincolumns1-4andforbanksthathaveheldinterestratederivativesintheirnon-tradingaccountsincolumns5-8.AllbanksinthissamplehaveheldMBSintheirnon-tradingaccounts.ThebankleveldataareretrievedfromthebankcallreportsandarequarterlyfromJune1997toDecember2013.ForMBSdurationweuseempiricalduration.Bankxedeectsandbanktimevaryingcontrolsarealsoincluded.Standarderrorsareclusteredatthebanklevel(rssd9001).Theregressionmodelis:Debtt;i Assetst;i=+1Durationt12m+2MBSSecuritiest12m;i Assetst12m;i+3MBSSecuritiest12m;i Assetst12m;iDurationt12m+Xi;t12m+t;i CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES117Table3.6:EectofMBSDurationonBankTreasuryBondHoldingsforBanksWithoutMBSHoldings (1)(2)(3)(4) EmpiricalDurt12m-0.0006150.00122-0.005100.00632(-1.70)(2.55)(-11.69)(12.93)Constant0.02680.282
132 0.1300.152(1
0.1300.152(1.40)(9.83)(7.66)(6.77) #Obs47227472254722747225R-squared0.7240.8750.7430.759MaturityofDepVariableTreasAll3yrs3yrs tstatisticsinparenthesesp0:05,p0:01Regressionsmeasuringtheeectof12monthlaggedMBSdurationontheratioofdebtholdingstototalassetsatthebanklevelforbanksthathaveneverheldMBSintheirnon-tradingaccounts.ThebankleveldataareretrievedfromthebankcallreportsandarequarterlyfromJune1997toDecember2013.ForMBSdurationweuseempiricalduration.Bankxedeectsandbanktimevaryingcontrolsarealsoincluded.Standarderrorsareclusteredatthebanklevel(rssd9001).Theregressionmodelis:Debtt;i Assetst;i=+1Durationt12m+Xi;t12m+t;i CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES118Table3.7:EectofChangeinBarclaysEectiveMBSDurationonChangeinForeignTreasuryBondHoldings (1)(2)(3)(4) BarclaysDurationt12m;t-0.0142-0.00844-0.0198-0.0207(-0.66)(-0.40)(-0.75)(-0.74)FX=1USDt12m;t-0.00461-0.0977(-2.17)(-0.35)Constant0.1060.1050.2430.228(2.60)(2.73)(4.86)(3.58) #Obs175175175175CountryJapanJapanChinaChina tstatisticsinparenthesesp0:05,p0:01Regressionsmeasuringtheeectof12monthchangesinMBSBarclayseectivedurationonthe12monthpercentagechangeinUSTreasuryholdingsofmainlandChinaandJapan.ChangeinFXratesarealsoincludedasacontrolinsomespecications.CountrylevelUSTreasuryholdingdataarefromtheTreasuryInternationalCapital(TIC)dataset.ThedataaremonthlyfromJune2000toDecember2014.StandarderrorsareNeweyWestwith18lagstoaccountfortheoverlappingstructureofthedata.Theregressionmodelis:TSYHoldingst;t+12m TSYHoldingst=+1Durationt12m;t+2FXRatet12m;t+t;t+12m CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES119Table3.8:EectofChangeinEm
133 piricalMBSDurationonChangeinForeignTreas
piricalMBSDurationonChangeinForeignTreasuryBondHoldings (1)(2)(3)(4) EmpiricalDurationt12m;t-0.0108-0.004290.002380.00176(-1.08)(-0.42)(0.11)(0.08)FX=1USDt12m;t-0.00462-0.0881(-2.11)(-0.31)Constant0.1050.1040.2410.229(2.62)(2.76)(4.57)(3.47) #Obs175175175175CountryJapanJapanChinaChina tstatisticsinparenthesesp0:05,p0:01Regressionsmeasuringtheeectof12monthchangesinMBSempiricaldurationonthe12monthpercentagechangeinUSTreasuryholdingsofmainlandChinaandJapan.ChangeinFXratesarealsoincludedasacontrolinsomespecications.CountrylevelUSTreasuryholdingdataarefromtheTreasuryInternationalCapital(TIC)dataset.ThedataaremonthlyfromJune2000toDecember2014.StandarderrorsareNeweyWestwith18lagstoaccountfortheoverlappingstructureofthedata.Theregressionmodelis:TSYHoldingst;t+12m TSYHoldingst=+1Durationt12m;t+2FXRatet12m;t+t;t+12m CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES120 Table3.9:EectofMBSDurationonMutualFundTreasuryBondHoldings (1)(2)(3)(4) BarclaysDurt12m0.1430.313(12.23)(12.31)EmpiricalDurt12m0.05560.118(7.18)(7.16)Constant7.64115.568.05616.49(168.70)(157.23)(416.20)(399.54) #Obs724356250620724356250620R-squared0.9340.9100.9340.910OnlyFundswithMBSPositions?NYNYDurationMeasureBarclaysBarclaysEmpiricalEmpirical tstatisticsinparenthesesp0:05,p0:01RegressionsmeasuringtheeectofMBSdurationonthepercentageofassetsheldinTreasurybondsbymutualfunds.MutualfunddataarefromCRSPandarequarterlyfromQ12010toQ22016.MBSdurationiseitherBarclayseectivedurationorempiricalduration.Standarderrorsareclusteredatthemutualfundlevel.Theregressionmodelis:PerGovtBondst;i=+1Durationtj+Xi+t;i CHAPTER3.DURATIONMEAS
134 UREMENTANDHEDGINGCHANNELSFORGSEINSUREDMO
UREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES121 Table3.10:EectofChangeinMBSDurationonChangeinMutualFundTreasuryBondHoldings (1)(2)(3)(4) BarclaysDurt12;t0.06640.206(8.25)(11.59)EmpiricalDurt12;t0.03850.115(6.01)(8.08)Constant0.1490.2150.1650.264(67.13)(44.42)(400.30)(293.21) #Obs571324199230571324199230R-squared0.1400.1280.1400.127OnlyFundswithMBSPositions?NYNYDurationMeasureBarclaysBarclaysEmpiricalEmpirical tstatisticsinparenthesesp0:05,p0:01RegressionsmeasuringtheeectofthechangeinMBSdurationonthechangeinthepercentageofassetsheldinTreasurybondsbymutualfunds.MutualfunddataarefromCRSPandarequarterlyfromQ12010toQ22015.MBSdurationiseitherBar-clayseectivedurationorempiricalduration.Standarderrorsareclusteredatthemutualfundlevel.Theregressionmodelis:PerGovtBondst!t+j;i=+1Durationtj!t+Xi+t!t+j;i CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES122 Table3.11:EectofBarclaysEectiveMBSDurationonLifeInsuranceCompanyTreasuryBondPurchases (1)(2)(3)(4)(5)(6)(7)(8)(9) BarclaysDurationt1m-0.00510-0.00541-0.00837-0.00270-0.00579-0.00515-0.00562(-4.36)(-2.20)(-3.41)(-2.70)(-6.59)(-3.48)(-3.70)BarclaysDurationt6m-0.00520(-4.34)BarclaysDurationt12m-0.00912(-8.05)Constant0.1310.1320.1440.1300.2420.08140.07470.1310.133(37.27)(36.13)(40.98)(17.51)(32.50)(27.08)(28.23)(30.23)(29.84) #Obs11940211940211940227741499171194021194027698878867R-squared0.1540.1540.1550.01910.1050.1160.1250.1680.169YearsAllAllAllAllAllAllAll2001-20072001-2007NumMoWithTSYTradeAllAllAll1212A
135 llAllAllAllFile3003003003003003003003003
llAllAllAllFile300300300300300300300300303/304TSYMaturityAllAllAllAllAllSTNLTDLTDAll tstatisticsinparenthesesp0:05,p0:01 RegressionsestimatealinearprobabilitymodeloftheprobabilityoflifeinsurancermsbuyingTreasuriesduetolaggedMBSduration.LifeinsuranceholdingdataarefromNAICSandaremonthlyfrom2001to2012.MBSdurationisBarclayseectiveduration.Firmxedeectsarealsoincluded.Standarderrorsareclusteredatthermlevel.Modelswith1,6,and12monthlagsontheMBSdurationvariableareestimated.Theregressionmodelis:1t;i=+1Durationtj+Xi+t;i CHAPTER3.DURATIONMEASUREMENTANDHEDGINGCHANNELSFORGSEINSUREDMORTGAGEBACKEDSECURITIES123 Table3.12:EectofEmpiricalMBSDurationonLifeInsuranceCompanyTreasuryBondPurchases (1)(2)(3)(4)(5)(6)(7)(8)(9) EmpiricalDurationt1m-0.00574-0.00603-0.0116-0.00380-0.00562-0.00518-0.00534(-5.35)(-3.11)(-5.17)(-4.46)(-7.09)(-3.92)(-3.95)EmpiricalDurationt6m-0.00215(-2.09)EmpiricalDurationt12m-0.00735(-7.28)Constant0.1320.1220.1380.1310.2500.08430.07350.1320.133(42.82)(40.21)(44.87)(23.52)(38.80)(34.32)(32.14)(31.98)(31.40) #Obs11940211940211940227741499171194021194027698878867R-squared0.1540.1540.1550.01930.1060.1160.1250.1680.169YearsAllAllAllAllAllAllAll2001-20072001-2007NumMoWithTSYTradeAllAllAll1212AllAllAllAllFile300300300300300300300300303/304TSYMaturityAllAllAllAllAllSTNLTDLTDAll tstatisticsinparenthesesp0:05,p0:01 RegressionsestimatealinearprobabilitymodeloftheprobabilityoflifeinsurancermsbuyingTreasuriesduetolaggedMBSduration.LifeinsuranceholdingdataarefromNAICSandaremonthlyfrom2001to2012.MBSdurationisempiricalduration.Firmxedeectsarealsoincluded.Standarderrorsareclusteredatth
136 ermlevel.Modelswith1,6,and12monthla
ermlevel.Modelswith1,6,and12monthlagsontheMBSdurationvariableareestimated.Theregressionmodelis:1t;i=+1Durationtj+Xi+t;i 124BibliographyAdao,Rodrigo,MichalKolesar,andEduardoMorales,2019,Shift-sharedesigns:Theoryandinference,TheQuarterlyJournalofEconomics134,1949{2010.Adelino,M.,A.Schoar,andF.Severino,2015,Houseprices,collateral,andself-employment,JournalofFinancialEconomics117,288{306.Akepanidtaworn,Klakow,RickDiMascio,AlexImas,andLawrenceSchmidt,2018,Sell-ingfastandbuyingslow:Heuristicsandtradingperformanceofinstitutionalinvestors,WorkingPaper.Angrist,D,1998,Estimatingthelabormarketimpactofvoluntarymilitaryserviceusingsocialsecuritydataonmilitaryapplicants,Econometrica.Bahaj,S.A.,A.Foulis,andG.Pinter,2017,Homevaluesandrmbehaviour,WorkingPaper.Ballou,J.,T.Barton,D.DesRoches,F.Potter,Z.Zhao,B.Santos,andJ.Sebastian,2007,Kaumanrmsurvey(kfs)baselinemethodologyreport,WorkingPaper.Barber,BradM.,andTerranceOdean,2001,Boyswillbeboys:Gender,overcondence,andcommonstockinvestment.,Thequarterlyjournalofeconomics116,261{292.Barnatchez,K.,L.D.Crane,andR.Decker,2017,Anassessmentofthenationalestablish-menttimeseries(nets)databases,WorkingPaper.Bartik,T.J.,1991,Whobenetsfromstateandlocaleconomicdevelopmentpolicies?,WEUpjohnInstituteforEmploymentResearch.Benmelech,MeisenzahlR.R.,E.,andR.Ramcharan,2017,Therealeectsofliquidityduringthenancialcrisis:Evidencefromautomobiles,TheQuarterlyJournalofEconomics132,317{365.Berends,KyalR.,RobertMcMenamin,ThanasesPlestis,andRichardJ.Rosen,2013,Thesensitivityoflifeinsurancermstointerestratechanges,EconomicPerspectives37.Berger,A.N.,andG.F.Udell,1995,Relationshiplendingandlinesofcreditinsmallrmnance,JournalofBusiness351{381. BIBLIOGRAPHY125Berk,JonathanB.,andRichardC.Green,2004,Mutualfund owsandperformanceinrationalmarkets,JournalofPoliticalEconomy112,1269{1295.Bernstein,S.,T.McQuade,andR.R.Townsend,2018,Dohouse
137 holdwealthshocksaectproductivity?ev
holdwealthshocksaectproductivity?evidencefrominnovativeworkersduringthegreatrecession,NBERWorkingPaper(No.24011).Bhutta,N.,andB.J.Keys,2016,Interestratesandequityextractionduringthehousingboom,AmericanEconomicReview106,1742{74.Blanchard,O.J.,L.F.Katz,R.E.Hall,andB.Eichengreen,1992,Regionalevolutions,Brook-ingsPapersonEconomicActivity1992,1{75.Blanch ower,D.G.,andA.J.Oswald,1998,Whatmakesanentrepreneur?,JournalofLaborEconomics16,26{60.Bord,V.M.,V.Ivashina,andR.D.Taliaferro,2018,Largebanksandsmallrmlending,NationalBureauofEconomicResearch(No.w25184).Borusyak,Kirill,PeterHull,andXavierJaravel,2018,Quasi-experimentalshift-sharere-searchdesigns,NBERWorkingPaper(No.w24997).Boudoukh,Jacob,MatthewRichardson,RichardStanton,andRobertF.Whitelaw,1995,Anewstrategyfordynamicallyhedgingmortgage-backedsecurities,JournalofDerivatives2,60{77.Buchak,G.,G.Matvos,T.Piskorski,andA.Seru,2018,Fintech,regulatoryarbitrage,andtheriseofshadowbanks,JournalofFinancialEconomics130,453{483.Busse,JereyA.,TarunChordia,LeiJiang,andYuehuaTang,2016,Mutualfundtransac-tioncosts,WorkingPaper.Campbell,J.Y.,2006,Householdnance,TheJournalofFinance61,1553{1604.Card,D.,andD.G.Sullivan,1988,Measuringtheeectofsubsidizedtrainingprogramsonmovementsinandoutofemployment,Econometrica.Chen,Qi,ItayGoldstein,andWeiJiang,2008,Directors'ownershipintheusmutualfundindustry,TheJournalofFinance63,2629{2677.Chernenko,Sergey,SamuelG.Hanson,andAdiSunderam,2016,Whoneglectsrisk?investorexperienceandthecreditboom,JournalofFinancialEconomics122,248{269.Chodorow-Reich,G.,2013,Theemploymenteectsofcreditmarketdisruptions:Firm-levelevidencefromthe2008{9nancialcrisis,TheQuarterlyJournalofEconomics129,1{59. BIBLIOGRAPHY126Cochrane,JohnH.,andMonikaPiazzesi,2005,Bondriskpremia,AmericanEconomicReview95,138{160.Corradin,S.,andA.Popov,2015,Houseprices,homeequityborrowing,andentrepreneur-ship,TheReviewofFinancialStudies28,2399{2428.Coval,JoshuaD.,andTob
138 iasJ.Moskowitz,1999,Homebiasathome:Local
iasJ.Moskowitz,1999,Homebiasathome:Localequityprefer-enceindomesticportfolios,TheJournalofFinance54,2045{2073.Cox,JohnC.,JonathanE.Ingersoll,Jr.,andStephenA.Ross,1985,Atheoryofthetermstructureofinterestrates,Econometrica53,385{407.Cremers,KJMartijn,andAnttiPetajisto,2009,Howactiveisyourfundmanager?anewmeasurethatpredictsperformance,TheReviewofFinancialStudies22,3329{3365.Cremers,Martijn,JoostDriessen,PascalMaenhout,andDavidWeinbaum,2009,Doesskininthegamematter?directorincentivesandgovernanceinthemutualfundindustry,JournalofFinancialandQuantitativeAnalysis44,1345{1373.Cremers,Martijn,andAnkurPareek,2016,Patientcapitaloutperformance:Theinvestmentskillofhighactivesharemanagerswhotradeinfrequently,JournalofFinancialEconomics122,288{306.Cronqvist,Henrik,AnilK.Makhija,andScottE.Yonker,2012,Behavioralconsistencyincorporatenance:Ceopersonalandcorporateleverage,JournalofFinancialEconomics103,20{40.Culbertson,J.M.,1957,Thetermstructureofinterestrates,QuarterlyJournalofEconomics71,485{517.Davis,S.J.,J.Haltiwanger,andS.Schuh,1996,Smallbusinessandjobcreation:Dissectingthemythandreassessingthefacts,SmallBusinessEconomics8,297{315.Davis,S.J.,andJ.C.Haltiwanger,2019,Dynamismdiminished:Theroleofhousingmarketsandcreditconditions,NationalBureauofEconomicResearch(No.w25466).Decker,R.,J.Haltiwanger,R.Jarmin,andJ.Miranda,2014,Theroleofentrepreneurshipinusjobcreationandeconomicdynamism,JournalofEconomicPerspectives28,3{24.DeFusco,A.,S.Johnson,andJ.Mondragon,2017,Regulatinghouseholdleverage,WorkingPaper.DeFusco,A.A.,2018,Homeownerborrowingandhousingcollateral:Newevidencefromexpiringpricecontrols,TheJournalofFinance73,523{573.Evans,AllisonL,2008,Portfoliomanagerownershipandmutualfundperformance,FinancialManagement37,513{534. BIBLIOGRAPHY127Evans,D.S.,andB.Jovanovic,1989,Anestimatedmodelofentrepreneurialchoiceunderliquidityconstraints,JournalofPoliticalEconomy97,808{827.Fisher,Lawrence,andRoman.L.Weil,1971,Copingwiththeriskofinte
139 restrate uctua-tions:Returnstobondholder
restrate uctua-tions:Returnstobondholdersfromnaiveandoptimalstrategies,JournalofBusiness44,111{118.Gagnon,JosephE.,MatthewRaskin,JulieAnnRemache,andBrianSack,2011,The-nancialmarketeectsoftheFederalReserve'slarge-scaleassetpurchases,InternationalJournalofCentralBanking7,3{43.Gete,P.,andM.Reher,2018,Mortgagesupplyandhousingrents,TheReviewofFinancialStudies31,4884{4911.Goldsmith-Pinkham,Paul,IsaacSorkin,andHenrySwift,2018,Bartikinstruments:What,when,why,andhow,NBERWorkingPaper(No.w24408).Goodman,L.S.,2017,Quantifyingthetightnessofmortgagecreditandassessingpolicyactions,BCJLandSoc.Just.37,235.Gourio,F.,T.Messer,andM.Siemer,2016,Firmentryandmacroeconomicdynamics:astate-levelanalysis,AmericanEconomicReview106,214{18.Greenspan,Alan,andJamesKennedy,2008,Sourcesandusesofequityextractedfromhomes,OxfordReviewofEconomicPolicy24,120{144.Greenstone,M.,A.Mas,andH.L.Nguyen,2014,Docreditmarketshocksaecttherealeconomy?quasi-experimentalevidencefromthegreatrecessionand`normal'economictimes,NationalBureauofEconomicResearch(No.w20704).Greenwood,Robin,andStefanNagel,2009,Inexperiencedinvestorsandbubbles,JournalofFinancialEconomics93,239{258.Greenwood,Robin,andDimitriVayanos,2010,Pricepressureinthegovernmentbondmarket,AmericanEconomicReview100,585{590.Greenwood,Robin,andDimitriVayanos,2014,Bondsupplyandexcessbondreturns,ReviewofFinancialStudies27,663{713.Grinblatt,Mark,andMattiKeloharju,2009,Sensationseeking,overcondence,andtradingactivity.,TheJournalofFinance64,549{578.Gupta,Arpit,andKunalSachdeva,2017,Skinorskim?insideinvestmentandhedgefundperformance,WorkingPaper.Guren,A.M.,A.McKay,E.Nakamura,andJ.Steinsson,2018,Housingwealtheects:Thelongview,NationalBureauofEconomicResearch(No.w24729). BIBLIOGRAPHY128Haltiwanger,J.,R.S.Jarmin,R.B.Kulick,andJ.Miranda,2016,Highgrowthyoungrms:Contributiontojob,outputandproductivitygrowth,USCensusBureauCenterforEco-nomicStudiesPaperNo.CES-WP-16-49.Haltiwanger,J.,R.S.Ja
140 rmin,andJ.Miranda,2013,Whocreatesjobs?sm
rmin,andJ.Miranda,2013,Whocreatesjobs?smallversuslargeversusyoung,ReviewofEconomicsandStatistics95,347{361.Hanson,SamuelG.,2014,Mortgageconvexity,JournalofFinancialEconomics113,270{299.Harvey,CampbellR.,YanLiu,andHeqingZhu,2016,...andthecross-sectionofexpectedreturns,TheReviewofFinancialStudies29,5{68.Hathaway,I.,andR.E.Litan,2014,What'sdrivingthedeclineinthermformationrate?apartialexplanation,TheBrookingsInstitution.Hayre,Lakhbir,ed.,2001,SalomonSmithBarneyGuidetoMortgage-BackedandAsset-BackedSecurities(JohnWiley,NewYork).Hicks,JohnR.,1939,ValueandCapital(ClarendonPress,Oxford,UK).Ho,ThomasS.Y.,1992,Keyratedurations:Measuresofinterestraterisks,JournalofFixedIncome2,29{44.Hong,Harrison,andLeonardKostovetsky,2012,Redandblueinvesting:Valuesandnance,JournalofFinancialEconomics103,1{19.Hurst,E.,G.Li,andB.Pugsley,2014,Arehouseholdsurveysliketaxforms?evidencefromincomeunderreportingoftheself-employed,ReviewofEconomicsandStatistics96,19{33.Hurst,E.,andA.Lusardi,2004,Liquidityconstraints,householdwealth,andentrepreneur-ship,JournalofPoliticalEconomy112,319{347.Iacus,S.M.,G.King,andG.Porro,2019,Atheoryofstatisticalinferenceformatchingmethodsincausalresearch,PoliticalAnalysis27,46{68.Ivkovic,Zoran,andScottWeisbenner,2005,Localdoesaslocalis:Informationcontentofthegeographyofindividualinvestors'commonstockinvestments,JournalofFinance60,267{306.Jarmin,R.S.,andJ.Miranda,2002,Thelongitudinalbusinessdatabase,WorkingPaper.Karahan,F.,B.Pugsley,andA.Sahin,2019,Demographicoriginsofthestartupdecit,NationalBureauofEconomicResearch(No.w25874). BIBLIOGRAPHY129Kerr,S.,W.R.Kerr,andR.Nanda,2015,Housemoneyandentrepreneurship,NationalBureauofEconomicResearch(No.w21458).Khorana,Ajay,HenriServaes,andLeiWedge,2007,Portfoliomanagerownershipandfundperformance,JournalofFinancialEconomics85,179{204.Kiyotaki,N.,andJ.Moore,1997,Creditcycles,JournalofPoliticalEconomy105,211{248.Krishnamurthy,Arvind,andAnnetteVissing-J
141 rgensen,2011,Theeectsofquantitative
rgensen,2011,Theeectsofquantitativeeasingoninterestrates:Channelsandimplicationsforpolicy,BrookingsPapersonEco-nomicActivity215{287.Krishnan,Karthik,DebarshiKNandy,andManjuPuri,2014,Doesnancingspursmallbusinessproductivity?evidencefromanaturalexperiment,TheReviewofFinancialStudies28,1768{1809.Kumar,Alok,2009,Whogamblesinthestockmarket?,JournalofFinance64,1889{1933.Laufer,S.,andA.Paciorek,2018,Theeectsofmortgagecreditavailability:Evidencefromminimumcreditscorelendingrules.,WorkingPaper.Lee,MayerC.J.,D.,andJ.Tracy,2012,Anewlookatsecondliens,NationalBureauofEconomicResearch(No.w18269).Liu,Crocker,andDavidYermack,2012,Wherearetheshareholders'mansions?ceos'homepurchases,stocksales,andsubsequentcompanyperformance,CorporateGovernance3{28.Loutskina,E.,andP.E.Strahan,2015,Financialintegration,housing,andeconomicvolatil-ity.,JournalofFinancialEconomics115,25{41.Lu,Yan,andMelvynTeo,2018,Doalphamalesdeliveralpha?testosteroneandhedgefunds,WorkingPaper.Macaulay,FrederickR.,1938,Sometheoreticalproblemssuggestedbythemovementsofinterestrates,bondyields,andstockpricesintheUnitedStatessince1856,Workingpaper,NationalBureauofEconomicResearch.Malkhozov,Aytek,PhilippeMueller,AndreaVedolin,andGyuriVenter,2016,Mortgageriskandtheyieldcurve,ReviewofFinancialStudies29,1220{1253.Mian,A.,andA.Su,2014,Whatexplainsthe2007{2009dropinemployment?,Economet-rica82,2197{2223.Modigliani,Franco,andRichardSutch,1966,Innovationsininterestratepolicy,AmericanEconomicReview56,178{197. BIBLIOGRAPHY130Mondragon,J.,2018,Householdcreditandemploymentinthegreatrecession,KiltsCenterforMarketingatChicagoBooth{NielsenDatasetPaperSeries1{025.Neumark,D.,J.Zhang,andB.Wall,2007,Employmentdynamicsandbusinessrelocation:Newevidencefromthenationalestablishmenttimeseries,AspectsofWorkerWell-Being39{83.Nguyen,H.L.Q.,2019,Arecreditmarketsstilllocal?evidencefrombankbranchclosings,AmericanEconomicJournal:AppliedEconomics11,1{32.Odean,T.,andB.Barber,1999,T
142 hecourageofmisguidedconvictions:thetradi
hecourageofmisguidedconvictions:thetradingbehaviorofindividualinvestors,FinancialAnalystJournal41{55.Odean,Terrance,1999,Doinvestorstradetoomuch?,AmericanEconomicReview89,262.Palmer,C.,2015,Whydidsomanysubprimeborrowersdefaultduringthecrisis:Loosecreditorplummetingprices?,WorkingPaper.Patnaik,M.,2017,Theimpactofcreditshocks:Microversussmallrms,WorkingPaper.Petersen,M.A.,andR.G.Rajan,1994,Thebenetsoflendingrelationships:Evidencefromsmallbusinessdata,TheJournalofFinance49,3{37.Pool,VeronikaK.,NoahStoman,andScottE.Yonker,2012,Noplacelikehome:Fa-miliarityinmutualfundmanagerportfoliochoice,TheReviewofFinancialStudies25,2563{2599.Pool,VeronikaK.,NoahStoman,andScottE.Yonker,2015,Thepeopleinyourneigh-borhood:Socialinteractionsandmutualfundportfolios,TheJournalofFinance70,2679{2732.Pastor,Lubos,RobertF.Stambaugh,andLucianA.Taylor,2017,Dofundsmakemorewhentheytrademore?,TheJournalofFinance72,1483{1528.Pugsley,B.W.,andA.S,ahin,2018,Grown-upbusinesscycles,TheReviewofFinancialStudies32,1102{1147.Risa,Sefano,RodrigoIbanez-Meier,MinFan,andIdrissMaoui,2008,Primexed-rateprepaymentmodelupdate,Researchreport,BarclaysCapital.Robb,A.M.,andD.T.Robinson,2014,Thecapitalstructuredecisionsofnewrms,TheReviewofFinancialStudies27,153{179.Rosenbaum,P.R.,andD.B.Rubin,1983,Thecentralroleofthepropensityscoreinobser-vationalstudiesforcausaleects,Biometrika70,41{55. BIBLIOGRAPHY131Ross,MyronH.,1966,\OperationTwist":Amistakenpolicy?,JournalofPoliticalEconomy74,195{199.Rubin,D.B.,1973,Matchingtoremovebiasinobservationalstudies,Biometrics159{183.Rubin,D.B.,1974,Estimatingcausaleectsoftreatmentsinrandomizedandnonrandomizedstudies,JournalofeducationalPsychology66,688.Rubin,D.B.,1977,Assignmenttotreatmentgrouponthebasisofacovariate,JournalofEducationalStatistics2,1{26.Sarsons,H.,2017,Interpretingsignalsinthelabormarket:evidencefrommedicalreferrals,JobMarketPaper.Schmalz,M.C.,D.A.Sraer,andD.Thesma
143 r,2017,Housingcollateralandentrepreneurs
r,2017,Housingcollateralandentrepreneurship,TheJournalofFinance72,99{132.Siemer,M.,2016,Firmentryandemploymentdynamicsinthegreatrecession,AvailableatSSRN2172594.Song,Yang,2019,Themismatchbetweenmutualfundscaleandskill,WorkingPaper.Srinivasan,V.S.,andKumarVelayudham,2010,Prepaymentmodelupdate,Researchre-port,BarclaysCapital.Stiglitz,J.E.,andA.Weiss,1981,Creditrationinginmarketswithimperfectinformation,TheAmericanEconomicReview71,393{410.Stoman,Noah,VeronikaKPool,ScottEYonker,andHanjiangZhang,2018,Doshockstopersonalwealthaectrisk-takingindelegatedportfolios?,TheReviewofFinancialStudies32,1457{1493.Tornqvist,L.,P.Vartia,andY.O.Vartia,1985,Howshouldrelativechangesbemeasured?,TheAmericanStatistician39,43{46.Vasicek,OldrichA.,1977,Anequilibriumcharacterizationofthetermstructure,JournalofFinancialEconomics5,177{188.Vayanos,Dimitri,andJean-LucVila,2009,Apreferred-habitatmodelofthetermstructureofinterestrates,WorkingPaper15487,NBER.Wallace,Neil,1967,ThetermstructureofinterestratesandthematuritycompositionoftheFederaldebt,JournalofFinance22,301{312. 132AppendixAAppendixtoChapter1A.1MergesA.1.1ATTOM-NETS(D&B)MergeThemicro-datautilizedbythispaperareauniquemergebetweenfourdierentdatasets.Itisnotatrivialtasktolinkhomeequityextractionactivityofbusinessownerstothetaxrecordsoftheirbusiness,whichnecessitatedanovelapproach.ATTOMisacomprehensivepropertyandtransactionleveldatasetonresidential(andcommercial)realestatepurchasesandrenances.Fromthisdata,priorresearchhasconstructedtheamountofhomeequityextractedforrenancingtransactions(DeFusco,2018)|thisisdescribedinSection2oftheOnlineAppendix.However,linkingATTOMtoinformationonbusinessesownedbyhomeownersisdicult,sinceATTOMdoesnotlistemploymentinformation.Fortunately,ATTOMliststhenameoftheperson/peoplewhoarethelegalownersofthepropertyatthetimeofeachtransaction.Withthisinformation,ATTOMcanbemergedtoNETS,whichformostbusinessesliststhenamesofbusinessowners.T
144 hismergeprovidesadatasetofthehomeequitye
hismergeprovidesadatasetofthehomeequityextractionactivityofhomeownerslinkedtothebusinessesthattheyown.Forcondentialityreasons,namesofpeopleandbusinessesarenotutilizedintheresearchandinformationonindividualrecordsarenotreported.TomergeATTOMtoNETS,thenameofthehomeownersinATTOMaremergedtothenameofrmownersinNETS.Roughly60%ofrmsinNETShavethenameofthermownerlisted.Inbothdatasets,thenamesarerstcleanedandstandardized.Punctuation,prexes,andsuxesareremovedand429commonnamesaremappedtoastandardspelling(e.g.AnneandAnniearebothmappedtoAnn).ATTOMcontainsnamesforuptotwopeopleforeachtransaction.Iftwonamesarelisted,bothnamesareincludedforthemergetoNETS.Forsometransactions,ATTOMliststhenameofatrustorbusiness(usuallyabankforcasesofforeclosuretransactions,whicharenotincludedinthemerge).Inthecaseofatrust,thecleaningalgorithmattemptstoextractanamelistedwiththetrust(i.e.JohnSmithfor"TheJohnSmithFamilyTrust").Themergeisattemptedinsixpasses,withsubsequentpassesbasedonloosercriteria. APPENDIXA.APPENDIXTOCHAPTER1133Foreachpass,mergedDunsnumbers(identierforestablishments)inNETSarecountedassuccessfullymergedifonlyonerecordfromATTOMismerged.IncaseswheremorethanonerecordinATTOMismergedtoabusinessinNETS,therecordedmergeisdropped.Anyrecordsmergedinanearlierpass,evenifthemergewasnotuniqueandconsequentlydropped,arenotincludedinlaterpasses.Intherstpass,therecordsaremergedonfullrstandlastnames.IncaseswherebothATTOMandNETScontainamiddleinitialorname,themiddleinitialisalsoincludedinthemerge.Forthesecondpass,recordsarematchedonrstandmiddleinitialsandfulllastname.RecordsinATTOMandNETSmusthaveaknownmiddleinitialtobeincludedinthispass.NETSallowsforestablishmentsofmulti-unitrmstobelinkedtotheirheadquarters.Insomecases,thenameofthermownerreportedfortheheadquarterisdierentfromthenameoftheownerofanestablishment.Inthethirdpass,thenameoftheheadquarterownerisusedtomergerecordsonfullrst
145 name,middleinitial(ifbothATTOMandNETSlis
name,middleinitial(ifbothATTOMandNETSlistamiddleinitial/name),andfulllastname.Thefourthpassalsousestheheadquarterownernameandmergesonrstnameinitial,middlenameinitial,andfulllastname.Forthefourthpass,bothATTOMandNETSmustcontaininformationonmiddlename.Thesetwopassesreplicatepassesoneandtwowithheadquarterownernameinsteadofestablishmentownername.Mostestablishmentsarealsotheheadquarters,sincemostbusinessesaresingle-unitentities.Inthefthpass,recordsaremergedonfulllastnameandthelatitudeandlongitudeofthehomeinATTOMandtherminNETS.Thispassprimarilyincludeshomebusinesses.Fortheanalysisofcreditconstraints,mostoftheserecordswouldbeexcludedasthesermsarelocatedinthesamezipcodeofthermowner'shome.Latitudeandlongitudeareroundedtothethirddecimalplace,whichrepresents110metersofaccuracy.Forthesixthpass,recordsaremergedontherstthreelettersofrstnameandfulllastname.MiddleinitialsarealsomatchedforcaseswherebothNETSandATTOMcontainmiddleinitial.Additionalpassesinvolvingfuzzymergeswereattemptedbuthandscreeningthesemergesshowedthatthisresultedinanincreaseintherateoffalsemerges.A.1.2NETS(D&B)-SSELMergeTherestricteduseLongitudinalBusinessDatabase(LBD)containsalongitudinalpanelofdataonanestablishment'ssurvival,employmentandpayroll.ThedataincludeeveryemployerestablishmentintheUnitedStatesandareconstructedfromIRStaxrecordsofthebusiness.TheLBDNUMvariabletracksanestablishmentacrosstimeintheLBD.TheLBDdoesnotcontainbusinessnameandaddress,whicharenecessarytomergethedatatoNETS.TheBusinessRegistrar(SSEL)istherawunderlyingdatathattheLBDdataareconstructedfrom.TheSSELcontainsbusinessnameandaddress,whichnecessitatesrsthavingtomergeNETStoSSELpriortothemergingNETStoLBD.BothNETSandSSELcontainbusinessname,address,andindustry.NETScontainsarm'srstandlastyear,whiletheSSELareannuallesthatcontainallemployerestab-lishmentsopenintheyearofthele(withouttheabilitytotrackestablishmentsacrossthe APPENDIXA.APPE
146 NDIXTOCHAPTER1134dierentannual
NDIXTOCHAPTER1134dierentannualles).ForeachannualSSELle,allopenestablishmentsinNETS,basedontherstyear(yearbusinessopens)andlastyear(yearbusinesscloses)variablesinNETSrelativetotheSSELleyear,arekeptforthemergetoSSEL.Toallowforreportingerrorsanddierences,thevalueforrstyearinNETSisallowedtobeuptoveyearsaftertheSSELleyear,andthevalueforlastyearinNETSisallowedtobeuptoveyearspriortotheSSELleyear.OnereasonforreportingdierencesisthattherstyearinNETSistherstyearthebusinessisinoperation,evenifitstartsasanon-employerrm.IntheSSEL,armonlyappearswhenthermhasatleastonepaidemployee.TomergeNETStoSSEL,vepassesareattempted.Priortomerging,companynameandstreetaddressarecleanedtonormalizethestringsandCOMPGEDandSPEDISfunctionsinSASareusedtocalculatedistancescoresonstrings(thisisthebasisofthefuzzymatches).Intherstpass,zipcodesandstreetnumbersareexactmatchedandstreetname,unitnumber,andcompanynamearefuzzymerged.Withinthisrstpass,veseparatemergeswithvaryingdegreesoftightnessforthefuzzypartofthematchareattempted.MatchingontherstvecharactersofcompanynameandtheNAICSsectorofthebusiness(exceptforsector54[professionaloces],whichcontainmanybusinesseswiththesameinitialstringinthecompanyname)isalsoattempted.Inthesecondthroughfthpasses,themaximumscoresfromtheCOMPGEDandSPEDISfunctionsarelowered(thecriteriaaretightened)ascriteriaforotherattributesofthemergeareloosened.Inthesecondpass,zipcodeand2-digitNAICSsectorareexactmatchedandfuzzymatchesonstreetname,unitnumber,andcompanynameareused.Streetnumberandunitnumberareignoredinthispass.Forthethirdpass,zipcodeand4-digitNAICScategoryareexactmatchedandafuzzymatchoncompanynameisattempted.Addressinformationbeyondthezipcodeisnotusedforthispass.Inthefourthpass,county,state,and6-digitNAICScategoryareexactmatchedandafuzzymatchoncompanynameisattempted.Inthefthandnalpass,county,sta
147 te,and4-digitNAICSareexactmatchedandaver
te,and4-digitNAICSareexactmatchedandaverytightfuzzymergeoncompanynameisattempted.Mostmergesarefromtherstpass,witheachsubsequentpassproducingadiminishingnumberofmerges.Duetodisclosurereasons,statisticsonthemergeratefromindividualpassesarenotavailable.A.1.3LBD-SSELMergeTheSSELdoesnotcontainanidentierthatlinksbusinessesthroughtime.OncetheSSELismergedtoNETS,eachrecordintheSSELthatismergedwillhavetheassociatedDunsnumberfromNETS.TheDunsnumberallowsforlongitudinaltrackingofestablishmentsintheSSEL.ForeachDunsnumber,themodalLBDNUMintheLBDisselected.IfthereismorethanonemodalLBDNUM,oneisrandomlyselected.IfthemodalLBDNUMismappedtomorethanoneDunsnumber,theonethatminimizesthedierenceintheestablishment'slastyearbetweenNETSandLBDisselected.Ifthisatie,theonewiththeclosestSICindustrycodesbetweenNETSandLBDisusedasatiebreaker.Ifathirdtiebreakerisneeded,theonethatminimizesthedierencebetweentheestablishment'srst APPENDIXA.APPENDIXTOCHAPTER1135yearinNETSandLBDisselected.Duetodisclosurereasons,statisticsonthemergeratearenotavailable.A.2HomeEquityExtractionDatafromATTOMATTOMcontainsarecordforeveryrealestatepurchaseandrenancetransaction,fromwhichtheamountofhomeequityextractedforeachrenancetransactioncanbeconstructed.ThispaperborrowsfromtheapproachusedbyDeFusco(2018).TheprimarydierencesfromDeFusco(2018)'sapproacharethatthispaperaccountsforre-suborindationofsecondarylienswhentherstlienisbeingrenanced,andseparatelyaccountsforsecond-liensandHELOCs.Thesedierencesarediscussedbelow.ATTOMtrackspropertiesthroughtimebythesr property idvariable,whilehomeownersarenottrackedviaasimilaridvariablethroughtime.Toconstructtheamountofhomeequityextractedforrenancingtransactions,debthistoriesareconstructedforeachpropertybyhomeownertuple.Foreachpurchasetransaction,renancetransactionsthatoccurbeforethenextpurchasetransactionandthatcontainthesamehomeownerlastnameasthepurchasetransactionareusedt
148 oupdatethedebthistories.Therearethreebro
oupdatethedebthistories.Therearethreebroadtypesofrenancing.First,aborrowermaytakeoutahomeeq-uitylineofcredit(HELOC).ATTOMidentiesHELOCmortgagesbythelndr credit linevariable.AHELOCallowsaborrowertodrawuponacreditline,similartoacreditcard.ATTOMreportsthecreditlimitandnottheamountdrawnfromthiscreditline.Tode-terminetheamountofhomeequityextracted,itisassumedthattheentirecreditlineisextractedequity.Inarate-renance,theborrowerdoesnotextracthomeequitybutoriginatesanewmortgageofthesamebalanceastheircurrentoutstandingmortgage(s)toobtainalowerinterestrate.Aborrowermayalsocash-outrenance,wheretheytakeoutanewmortgagethatincreasestheirtotalloanbalance.Withinthistypeofrenance,theborrowereithertakesoutasecondarylien,andgenerallytheentireloanamountishomeequityextracted1,ortakesoutanewrstlienandpaysotheirtotaloutstandingmortgagedebt.Inthelattercase,thenewrstlienbalanceislargerthanthetotaloutstandingmortgagebalance(acrossallliens)andthedierencebetweenthetwoistheamountofhomeequityextracted.Complicatingmatters,ATTOMdoesnotdierentiatebetweenraterenanceorcash-outrenanceandwithincash-outrenancedoesnotstatewhethertherenanceisarstorsecondlien.Threedierentdebthistoriesneedtobetrackedovertimeforeachpropertybyhome-ownertuple.Therstlien,secondlien,andHELOCdebthistoriesareconstructedstartingwitheachpurchaseorigination.Atthetimeofthepurchasemortgageorigination,ATTOMlistsrstandsecondarylienstogether.Duringthemid-2000s,secondliensattimeofpur- 1Theexceptionbeingiftherealreadyisasecondlien,inwhichcasetheamountextractedisthedierencebetweenthenewsecondlienamountandtheoutstandingbalanceontheexistingsecondlien(ifthisispositive-anegativeamountwouldimplyarate-renanceofthesecondlien). APPENDIXA.APPENDIXTOCHAPTER1136chasewerecommonandreferredtoaspiggybackseconds.Thesewereusuallyoriginatedtocircumventprivatemortgageinsurance(PMI)requirementsforhighloanto
149 value(LTV)mortgages.ATTOMonlyrecordstheo
value(LTV)mortgages.ATTOMonlyrecordstheoriginationofeachmortgageanddoesnotprovideadatethemortgageisterminated(paido)ortherenamingbalancedueovertime.Forsimplicity,itisassumedthattherstandsecondliensare30-yearxedratemortgageswithaninterestrateequaltothemarketmortgagerateatthetimeoforigination.2Thedebthistoriesoftherstandsecondlienspaydownovertimeusingtheamortizationscheduleof30-yearxedratemortgages.ThedebthistoryfortheHELOCbalancedoesnotdecreaseovertimesinceHELOCsaregenerallyinterest-onlyfortherstfewyearsofpayments.IfadditionalHELOCsareoriginated,itisassumedthatifthenewcreditlimitishigherthanthepreviouscreditlimitthatthedierencebetweenthetwocreditlimitsisextractedhomeequity.Ifthenewcreditlimitislessthanthepriorcreditlimit,nohomeequityisextracted.InbothcasestheHELOCdebthistoryisupdatedtothenewcreditlimit.Thechallengewithnon-HELOCrenancesisdeterminingwhethertherenanceisaraterenance,acash-outrenancewithanewsecondlien,oracash-outrenancewithanewrstlien.Thedierencebetweentheselasttwoisthatfortheformercase,thecash-outamountistheentirelienamount(lessthecurrentsecondlienbalanceiftherealreadyisasecondlien).Inthelattercase,thecash-outamountisthenewmortgageamountlessthebalanceofalloutstandingliens.Tohelpwiththecategorization,itisnotedthatwithanewrstlien,inmostcasesallexistingliens(includingsecondliensandHELOCs)arepaido.Thisisduetore-subrodination,wherethesecondarylienholderswouldhavetoagreetobere-suborindatedwhentherstlienisbeingrenancedsincetheycontractuallybecometherstlienholderunlesstheywaivethisright(re-subordination).Lendersgenerallydonotwaivethisright,whichnecessitatestheneedtopayothesecondliensandHELOCswhenoriginatinganewrstlien.Tocategorizethemortgage,theabsolutedierencesbetweenthenewloanamountandalloutstandingliens(includingHELOCs)andalsotheoutstandingsecondliens(=0ifnooutstandingsecondlien)arecalculated.I
150 fthenewloanamountiscloserinbalancetothet
fthenewloanamountiscloserinbalancetothetotaloutstandingdebt,itisassumedthattherenanceisreplacingalloutstandingliensandbecomingthenewrstlien.Ifthenewmortgageamountismore(less)thanthetotaloutstandingdebtbalance,itisassumedtobeacash-out(rate-)renance.Inbothcases,thedebthistoriesforexistingsecondliensandHELOCs(ifany)aresetto0andtherstliendebthistoryissettothenewmortgageoriginationbalancewitha30-yearamortizationschedule.Theamountcashedout,ifthemortgageisclassiedasacash-outrenance,issettothedierencebetweenthenewbalancelesstotaloutstandingmortgagedebt.Ifthenewloanamountiscloserinbalancetothesecondlien(whichis=0ifthereisnosecondlien),itisassumedthattherenanceisassociatedwithasecondlien.Ifthereisnooutstandingsecondlien,thentheentireloanamountisextractedhomeequity.Ifthereisanoutstandingsecondlienthentheamountcashedoutisthethedierencebetweenthe 2ThemarketmortgagerateisobtainedfromFreddieMac'sPMMSsurvey. APPENDIXA.APPENDIXTOCHAPTER1137newloanamountandthecurrentbalanceofthesecondlien.Ifthisamountisnegativethennohomeequityisextracted(likelyarate-renanceoftheexistingsecondlien).Inallthreecases,thedebthistoryforthesecondlienissettothenewloanamountwitha30-yearamortizationschedule.Arenancemortgageoriginationcostsapproximately$5,000.Toaccountforthis,$5,000issubtractedfromtheamountofhomeequityextractedcalculatedabove.Iftheamountremainspositiveafterthisdeduction,therenancetransactioniscountedashomeequityextraction.A.3AppendixFiguresforChapter1 FigureA.1:ShareofSmallBusinessesFoundedwithHomeEquityFundingByFirmSizeTheshareofentrantsmallbusinessfundedbypersonalhomeequitybyyearofformation.ThedataareconstructedfromamergeofATTOMandNETS.Smallbusinessesaredenedashaving10orfeweremployeesintheyeartheyarefounded.Abusinessisclassiedasbeingfundedbyhomeequityiftheownerextractsover$5,000ofhomeequityintheyearthatthebusinessiscreatedortheprioryear.RegionsfollowtheCensusregioncl
151 assication.Therawunderlyingdataonly
assication.Therawunderlyingdataonlyincludesbusinessownerswhoownahome.Tocorrectforthis,thetimeseriesareadjustedbythehomeownershiprateofbusinessownersbasedonthepopulationofbusinessownersbyregionandyearfromtheAmericanCommunitySurveymicro-data.Thedataarefurthersegmentedbytheinitialsizeoftherm. APPENDIXA.APPENDIXTOCHAPTER1138 FigureA.2:ShareofEntrantSmallBusinessWithFewerThan10EmployeesTheshareofentrantestablishmentswithfewerthan10employeesatentry.DataarefromthepublicuseCensusBusinessDynamicsStatistics(BDS)andaregroupedbyyearofentry. FigureA.3:ShareofMortgageOriginationbyIndependentMortgageBanksTheshareofpurchasemortgagesoriginatedbyindependentmortgagebanksbetween2005and2015forLosAngeles,Suolk,andBrowardcounties.ThetimeseriesarecalculatedfromHMDAandareweightedbyloanbalance. 139AppendixBAppendixtoChapter2B.1AdditionalRobustnessTestsInthissection,additionalrobustnesstestsareprovidedfortheresultoftheeectofhomepriceshocksonperformance.Inthebaselinespecication,fund(foreachmergednon-disjointperiodoftime)andtimexedeectsareincluded.Asasecondspecication,fundxedeectsarereplacedwithcommutingzonebytimexedeectsandcontrolstocontrolforfundspeciceectsandforzipcodepreferencesfortheirhomeareincluded.Includingcommutingzonebytimexedeects,partialsoutthecommontrendinhomepricegrowthwithincommutingzone,whichwillconrmthattheresultisdrivenbyidiosyncratichomepricegrowthwithinanarrowgeographicarea.Aftercontrollingforzipcodelocationpreferencesitisunlikelythatfundmanagerscanforecasttheirrealizedidiosyncratichomepricegrowth.Underthisframework,theregressionmodelis:bFFCi;t;cz=c+ln(3yrHPIi;t1)+cz;t+Controlsit+i;t;cz(B.1)wherecz;taremonthlytimebycommutingzonexedeectsandControlsitisavectorofcontrolvariablescomprisedof:Morningstarfundcategoryxedeects,numberofmanagerxedeects,anindicatorforifthemanagerownsa
152 secondhome,anindicatorforifthemanagerhas
secondhome,anindicatorforifthemanagerhasanadvanceddegree(aboveundergraduate),andnon-parametricxedeectsforTNA,homevalue,combinedLTV(CLTV),averagezipcodelevelincome(fromthe2010IRSSOI),andthepercentofnon-whitehouseholdsatthezipcodelevel(fromthe2000Census).Standarderrorsareclusteredatthefundlevel(foreachmergednon-disjointperiodoftime).TableB.1presentsresultsusingthissecondspecication.Theresultsaresimilartothebaselinespecication.Aonestandarddeviationpositiveshockinducesadeclineinannualizedalphaof38bps.Next,otheroutcomevariablesareconsideredforthebaselineregressionmodelthatin-cludesfundandtimexedeects.TableB.2,column1estimatestheregressiononthe1-monthaheadrawreturnslessthereturnonthefund'sbenchmarkindex.Thisisestimatedovertheentire2001toJune2018daterangewithoutinteractionsonthehomepricegrowth APPENDIXB.APPENDIXTOCHAPTER2140variable.Overall,amarginallysignicantandnegativecoecientisfound,indicatingthatduringbothnormaltimesandrecessionsthereisanegativeeectonreturnsfromhomepricegrowth.DuringperiodsofeconomicexpansionthisisconsistentwiththeresultsfoundinthispaperandduringperiodsofcontractiontheresultisconsistentwithStomanetal.(2018).Stomanetal.(2018)ndanegativeandmarginallysignicanteectofhomepricesonrawreturnsduringperiodsofnegativehomepricegrowth,whichtheyattributetocareerconcerns.Whilethispaperndsthatduringperiodsofpositivehomepricegrowth,overcon-dencefrompositivehomepricegrowthalsodecreasesperformance.Theyfoundnoeectofhousepricegrowthonrisk-adjustedreturnsduringtheGreatRecession,alsoconsistentwiththeresultsinthispaper.Incolumn2ofTableB.2,thebaselinemodelisestimatedonmonthlyalphaforthesubsequent12-months.ThesubsequentoneyearaheadmonthlyalphaisestimatedfromaregressionofFama-French-Carhartonthesubsequent12monthsusing12monthlyobserva-tions.Theresultisnegativeandsignicantatthe5%level.Column3replacesnetalphawithgrossalpha(alphabeforef
153 ees)andtheresultisunaected.Asafourt
ees)andtheresultisunaected.Asafourthmethodforcalculatingalpha,TableB.3replicatesTable2.2witha1-monthforecastofalphausingex-anteinformation.Foreachmonthlyfundobservation,theprior12monthperiodisusedtoestimateaFama-French-Carhartregression(themonthlyobservationisomittedifthefundmanagergrouphaslessthan12monthsofhistoricaldatamanagingthefund).Usingthesefactorloadings,the1-monthaheadalphaisestimated.Theresultbecomesstrongerwiththisprocedure.Anotherconcernmaybethatformostfunds,housinginformationisonlyobservedforasubsetofmanagers.Ifthesampleisrestrictedtofundsforwhicha100%ofthemanagersweremerged,theresultremainsandslightlyimproves(TableB.2column4).Thechoiceofclusteringatthefundleveldoesnotallowforcorrelationacrossfundswithinamonth.Regressionswithreturnstypicallyclusteratthemonthlevel.However,inthispapertheeectisbeingestimatedoofchangesinhomepricegrowth,whicharecorrelatedacrosstime.Toshowthatthechoiceofclusteringisnotaectingsignicance,incolumn5ofTableB.2standarderrorsareclusteredbytimeandtheresultisunaected.TheremainderofthecolumnsinTable2.2arealsounaectedifclusteredbymonth(availableuponrequest).Anadditionalconcerncouldbethattheresultsareuniquetothemeasureof3-yearlaggedhomepricegrowth.Toproviderobustnessaroundthis,TableB.4replicatesTable2.2withthe3-yearhomepricegrowthmeasurereplacedwith2and4-yearhomepricegrowthmeasures.Theresultsarelargelyunchanged,showingthattheresultisnotdrivenbythechoiceofthehomepricegrowthmeasure.Anotherconcernisthatthechoiceofrestrictingthepopulationtofundmanagergroupsthatlastedforatleast24monthsiscausingasurvivorshiporothertypeofbiasintheresults.TableB.5replicatesTable2.2onsampleswherethefundmanagergroupssurvivedforatleast12monthsandsurvivedforatleast36months.Again,theresultsarelargelyunchanged.Theaboverobustnesschecksruleoutmanypotentialconcernswiththeprimaryresultsofthispaper. APPENDIXB.APPENDIXTOCHAPTER2141B.2AppendixTablesforChapter2 APPENDIXB.APPENDIXTOCHA
154 PTER2142 TableB.1:EectWithinCommuti
PTER2142 TableB.1:EectWithinCommutingZonebyTimeofHomePriceGrowthonFundAlpha (1)(2)(3)(4)AllPeriodsHPI0HPI0AllPeriods ln(3yrHPIi;t1)-0.135-0.264-0.0153(-1.88)(-2.41)(-0.10)ln(3yrHPI+i;t1)-0.284(-2.60)ln(3yrHPIi;t1)-0.0508(-0.35) #Obs87247595792766887247R-squared0.1900.2020.2160.206 tstatisticsinparenthesesp0:10,p0:05,p0:01Robustnessofthebaselinemodelformonthlyforecastsofalpha,withcommutingzonebytimexedeectsandabatteryofcontrols.Estimationoftheregressionmodel:bFFCi;t;cz=c+ln(3yrHPIi;t1)+cz;t+Controlsit+i;t;czwherecz;taremonthlytimebycommutingzonexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative)andControlsitiscomprisedof:Morningstarfundcategoryxedeects,numberofmanagerxedeects,anindicatorforifthemanagerownsasecondhome,anindicatorforifthemanagerhasanadvanceddegree(aboveundergraduate),andnon-parametricxedeectsforTNA,homevalue,combinedLTV(CLTV),averagezipcodelevelincome(fromthe2010IRSSOI),andthepercentofnon-whitehouseholdsatthezipcodelevel(fromthe2000Census).bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfund
155 sareincluded. APPENDIXB.APPENDIXTOCHAPTE
sareincluded. APPENDIXB.APPENDIXTOCHAPTER2143 TableB.2:EectofHomePriceGrowthonVariousFundReturnMeasures (1)(2)(3)(4)(5)Y=RGrossi;t+1Y=i;t+1!t+12mY=GrossbFFCY=NetbFFCClusterbyMonthRBenchmark;t+1Merged100%ofManagers ln(3yrHPIi;t1)-0.113(-1.67)ln(3yrHPI+i;t1)-0.189-0.263-0.302-0.258(-2.10)(-3.46)(-2.43)(-3.38)ln(3yrHPIi;t1)-0.154-0.141-0.00643-0.148(-1.21)(-1.27)(-0.03)(-1.02) #Obs8486473367848642173087719R-squared0.7820.2640.08950.1160.0901 tstatisticsinparenthesesp0:10,p0:05,p0:01Robustnessofthebaselinemodelformonthlyforecastsofalpha,withfundandtimexedeects.Estimationoftheregressionmodel:Yi;t;a=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.Yi;t;aiseither:netbFFC,grossbFFC,Ri;t+1RBenchmark;t+1,ori;t+1!t+12m(estimatedfroma1-yearregressionofFama-French-Carhartwith12monthlyobservations).Column4isrestrictedtofundsforwhich100%ofthemanagersweremergedtoATTOMandcolumn5repeatstheprimaryregressionwithclusteringbymonth(insteadofbyfund).bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromap
156 revioushomethatwasalsomerged).Fundmonths
revioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. APPENDIXB.APPENDIXTOCHAPTER2144 =TableB.3:EectofHomePriceGrowthonFundAlpha,UsingEx-AnteInformation (1)(2)(3)(4)AllPeriodsHPI0HPI0AllPeriods ln(3yrHPIi;t1)-0.190-0.2640.131(-2.70)(-2.56)(0.70)ln(3yrHPI+i;t1)-0.312(-3.40)ln(3yrHPIi;t1)-0.187(-1.20) #Obs73087490102407773087R-squared0.08740.08410.1250.0904 tstatisticsinparenthesesp0:10,p0:05,p0:01Robustnessoftheestimationofalphainthebaselineregressionmodel.Alphaisestimatedwithfactorsestimatedfromex-anteinformation.Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.bFFCi;t;aaremonthlyalphasobtainedfromestimatingafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodelontheprior12monthsofmonthlydata(fromt12tot).Thefactorsfromtheseregressionsarethenusedtocalculatethealphainmontht+1.Ifthefundmanagergrouphaslessthan12monthsofdatathenthemonthisomitted.Standarderrorsareclusteredatthefundlevel.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.OnlyactivelymanagedUSdomesticequityfundsareincluded. APPENDIXB.
157 APPENDIXTOCHAPTER2145 TableB.4:Eect
APPENDIXTOCHAPTER2145 TableB.4:Eectof2and4-YearHomePriceGrowthonFundAlpha (1)(2)(3)(4)(5)(6)(7)(8)2-yearHPIGrowth4-yearHPIGrowth AllHPIHPI0HPI0AllHPI AllHPIHPI0HPI0AllHPI ln(n-yearHPIi;t1)-0.179-0.2700.259 -0.165-0.203-0.121(-2.67)(-2.65)(1.50) (-3.37)(-2.68)(-1.09)ln(n-yearHPI+i;t1)-0.296 -0.212(-3.28) (-3.26)ln(n-yearHPIi;t1)0.0751 -0.147(0.53) (-1.65) #Obs87719589012881887719 81841552832655881841R-squared0.08770.08930.1170.0904 0.08800.09350.1150.0905 tstatisticsinparenthesesp0:10,p0:05,p0:01Robustnessofthebaselinemodelformonthlyforecastsofalpha,withfundandtimexedeects,onthechoiceofthe3-yearhomepricegrowthmeasure.Estimationoftheregressionmodelutilizingeither2yearor4yearhomepricegrowth(inlieuof3yearhomepricegrowth):bFFCi;t;a=c+ln(n-yearHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingatleast24monthsofdata.Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan24monthsaredroppedandfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppediftherearegreaterthan10managersorifTNA,measuredin2000dollars,fallsbelow$15million.Onlyactive
158 lymanagedUSdomesticequityfundsareinclude
lymanagedUSdomesticequityfundsareincluded. APPENDIXB.APPENDIXTOCHAPTER2146 TableB.5:EectofHomePriceGrowthonFundAlphaforFundManagersWhoSurvivedAtLeast12(36)Months (1)(2)(3)(4)(5)(6)(7)(8)Survived12monthsSurvived36months AllHPIHPI0HPI0AllHPI AllHPIHPI0HPI0AllHPI ln(3yrHPIi;t1)-0.163-0.219-0.0383 -0.159-0.229-0.00770(-2.90)(-2.57)(-0.28) (-2.75)(-2.53)(-0.05)ln(3yrHPI+i;t1)-0.242 -0.256(-3.28) (-3.30)ln(3yrHPIi;t1)-0.149 -0.0920(-1.38) (-0.82) #Obs94216650162920094216 78786527972598978786R-squared0.09150.09610.1160.0937 0.08530.08770.1140.0878 tstatisticsinparenthesesp0:10,p0:05,p0:01Robustnessofthebaselinemodelformonthlyforecastsofalpha,withfundandtimexedeects,ontherestrictionoffundmanagershavingtosurviveforatleast24months.Estimationoftheregressionmodel:bFFCi;t;a=c+ln(3yrHPIi;t1)+i+t+a+itawhereiarefundxedeects(foreachmergednon-disjointperiodoftime),taremonthlytimexedeects(interactedwithindicatorsforpositiveandnegativehomepricegrowthinregressionswherehomepricegrowthissegmentedonbeingpositiveornegative),andaarenon-parametricxedeectsforTNA.bFFCi;t;aaremonthlyalphasobtainedfrombackingoutthemonthlypricingerrorsfromafundlevel(foreachmergednon-disjointperiodoftime)regressionoftheFama-French-Carhart4-factormodel.Thepricingmodelisestimatedonmonthlydatausingbetween12and36monthsofdata(dependingonthecolumns).Standarderrorsareclusteredatthefundlevel,foreachmergednon-disjointperiodoftime.Unboundedvariablesarewinsorizedatthe1%and99%level.Fundswithmanagergroupsthatlastedforfewerthan12or36monthsaredropped(dependingonthecolumns)andfundmanagerswhoownedtheirprimaryhomeforfewerthan36monthsaredropped(theexceptionbeingifafundmanagermovedfromaprevioushomethatwasalsomerged).Fundmonthsaredroppedifthere