1  

DoInstitutionalInvestorsInfluenceCapitalStructureDecisions?

RoniMichaelyandChristopherVincent*

January2013

ABSTRACTThispaperempirically investigates therelationshipbetween institutionalholdingsandcapital structure. Institutionsmayaffect capital structure through theirmoni‐toringandinformation‐gatheringroles.Atthesametime,institutionsmaygravitatetoward firms with specific capital structures, forming leverage‐based investmentclienteles.Usingimpliedtradesgeneratedfrommutualfundoutflowsasaninstru‐mentforinstitutionalholdings,andasemi‐naturalexperimentinwhichadditiontotheS&P500 Indexprovidesanexogenous shock to institutionalholdings,we findthatinstitutionalholdingsareasignificantdeterminantoffirms’capitalstructures:Achangeininstitutionalholdingscausesanoppositechangeinleverage.Moreover,usingdynamicpanelestimation,wefindthatwhileinstitutionsaffectcapitalstruc‐turedecisions,changesinleveragedonotaffectinstitutionalholdings,andthereisnoevidenceofaclienteleeffect.Finally,we find that firms lower their leverage inresponsetoincreasedinstitutionalholdingsbybecomingmorelikelytoissueequi‐ty,andlesslikelytoincreasedebt.Whileourfindingsareconsistentwithmodelsinwhich institutions substitute for debt by monitoring and reducing informationasymmetry problems, further evidence suggests that the effect on asymmetric in‐formationdominates.

*RoniMichaelyisfromtheJohnsonSchoolofManagement,CornellUniversityandtheIDC.ChristopherVincentisfromtheJohnsonSchoolofManagement,CornellUniversity.Emails:RoniMichaelyrm34@cornell.edu;ChristopherVincentcjv34@cornell.edu.Wegreatlyap‐preciatethecommentsofJacobBoudoukh,DavidDeAngelis,MarkFlannery,ItayGoldstein,VidhanGoyal,YanivGrinstein,OhadKadan,AndrewKarolyi,ShimonKogan,YelenaLarkin,Mark Leary, Kai Li, Michelle Lowry, Jillian Popadak, Gideon Saar, Denis Sosyura, LukasRoth,FengZhang,andseminarparticipantsatCornellUniversity,GeorgetownUniversity,Nanyang Technological University, National University of Singapore, Singapore Manage‐mentUniversity,UniversityofAlberta,UniversityofBritishColumbia,UniversityofMichi‐gan, University of Syracuse, University of Toronto, University of Utah, the Corporate Fi‐nanceConferenceatHKUST,theCorporateFinanceConferenceattheUniversityofWash‐ingtonatSt.Louis,andtheSEIsummerconferenceatIDC.

2  

Institutionsarearguablythemostimportantandpowerfulclassofinvestors. Theiraver‐

ageequityownershipinU.S.firmshasincreasedeight‐foldoverthepastthirtyyears,and

by theendof2009theyheld70%of theaggregateU.S.marketcap.1 Forboth largeand

smallfirms,institutionsarenow,moreoftenthannot,themajorityinvestorgroup.Inthis

paperweexaminehowinstitutionalinvestorsaffectfirms’capitalstructure,andhowfirms’

financialleverageaffectsinstitutionalinvestors’decisionstoholdtheirequity.

Through the effects of institutional holdings on agency costs, asymmetric infor‐

mation,andtaxes,institutionsmayaffectthewayinwhichfirmsstructuretheircapital.As

weelaborateinthenextsection,thereareseveralreasonswhythisislikelytooccur.First,

bymonitoringmanagementthroughvoiceor(threatof)exit,institutionsasequityholders

reduce conflicts of interests between management and shareholders (e.g., Shleifer and

Vishny(1986),AdmatiandPfleiderer(2009),andLevit(2012)).Second,institutionsgath‐

er informationandmaketradesbasedonthat information,and indoingsoreduce infor‐

mation asymmetry problems associated with equity (e.g., Sias (2004), and Bushee and

Goodman(2007)). Finally,asequityholders, institutionshave,onaverage,arelative tax

advantageoverindividualinvestors.

While institutional investors interactwiththeaforementionedfrictionsunderlying

many capital structuremodels, the resulting nature of the relationship between institu‐

tionalholdingsandcapitalstructureisnotclear. Ononehand, institutionalholdingsand

debtcouldbesubstitutes.If,forexample,organizationalinefficienciesmustbecontrolled

asinJensen(1986),threateningtosellsharesorunseatmanagersiftheyengagein“empire

building”maybeaseffectiveascommittingmanagerstopledgefundstocreditors. Or,in

thecontextofMyersandMajluf(1984), if institutionalinformation‐gatheringandtrading

producesinformation(Sias(2004)),theadverseselectioncostsofequitymaydecline,lead‐

ingfirmstotilttowardahigherpercentageofequityfinancingintheircapitalstructures.

Inbothcases,institutionalholdingsanddebtwouldbesubstitutes.

Ontheotherhand,institutionalholdingsanddebtcouldactascomplements.Inan

agencysetting,itmaybethatbyincreasinginvestorprotectioninstitutionalinvestorsena‐

bleoutsideshareholderstoimplementdevicessuchasdebtthatlimitmanagementdiscre‐

1Thisfigureiscalculatedusing13FdataasfiledwiththeSEC,includesonlypubliclytradedequity,andex‐cludesADR’sandforeignincorporatedfirms.

3  

tionandbetteraligntheobjectivesofmanagersandshareholders(e.g.the“outcomemod‐

el” in La Porta, Lopez‐de‐Silanes, Shleifer, and Vishny (2000)). Or in an information‐

asymmetry setting, itmightbe that institutionshelp solve futureunderinvestmentprob‐

lems caused by the adverse selection costs of equity (again, through their role as infor‐

mationproducers), thusallowingfirmstoforgoissuingequitytoday(Viswanath(1993)).

In either case, the relationshipbetween institutionalholdingsanddebtwouldbea com‐

plementaryone.

While it isperhapsmostnatural toconjecture thatmanagers takeall information,

includingtheextentofinstitutionalequityholdings,intoconsiderationbeforemakingcapi‐

talstructuredecisions,institutionsmayalsotakefirms’leverageintoaccountwhenmaking

theirowninvestmentdecisions. Assuch,theeffectmaygointheoppositedirection,and

capitalstructuredecisionsmayaffect the levelof institutionalholdings ina firm. Recent

workhasshowninstitutionalinvestorstoforminvestmentclientelesbasedonfirmcharac‐

teristicssuchassizeandliquidity(GompersandMetrick(2001)),payoutpolicy(Grinstein

andMichaely(2005)),andcorporategovernance(McCahery,Sautner,andStarks(2010)),

whichtogethersuggestthatinstitutionsmayalsoformcapitalstructureinvestmentclien‐

teles.

Predicatedon the testable implicationsofextantcapital structure theories,weex‐

aminetheinteractionbetweeninstitutionalholdingsandcapitalstructureusingannualda‐

tafromtheperiod1979to2009.Ourfirstfindingisastrongandrobustnegativerelation‐

ship between institutional holdings and leverage:more institutional holdings in period t

areassociatedwithlessleverageinperiod(t+1).Theeconomiceffectofinstitutionalhold‐

ingson leverage is comparable inmagnitude toknowndeterminantsof leverage suchas

profitabilityandindustrymedianleverage,andtherelationshipholdsaftercontrollingfor

importantfactorsthatcovarywithinstitutionalholdings,suchasinsiderholdingsandana‐

lystcoverage,aswellasforknowndeterminantsofleverage,includinginitialfirmleverage

(Lemmon,Roberts,andZender(2008)).

Althoughwemakenopriorclaimsregardingthedirectionofcausality,weareinter‐

ested in whether the effect of institutional holdings on leverage is of a causal nature.

Hence,weemploythreetechniquestoremovethepotentialfeedbackeffectofleverageon

holdings:twoseparateinstrumentalvariablesapproachesandasemi‐naturalexperiment.

4  

OurfirsttechniqueisanArellano‐Bonddynamicpanelestimationinwhichweinstrument

forchangesininstitutionalholdingswithlaggedholdings.

Oursecondtechniqueisalineartwo‐stageleastsquaresestimationinwhichweuse

impliedmutualfundtradinginducedbyinvestoroutflowsasaninstrumentforinstitution‐

alholdings(Edmans,Goldstein,andJiang(2011)). Impliedchanges inmutual fundhold‐

ingsaffectaggregateinstitutionalholdings,butnotforreasonsassociatedwithfirmcharac‐

teristicssuchasleverage.Thatis,whenamutualfundexperienceslargeoutflowsitmust

sellaportionofitsholdingstorepayinvestors.Byusingimpliedtradesbasedonprevious‐

lydisclosedmutualfundholdings,wehaveaninstrumentthatcorrelateswithinstitutional

holdingsformechanicalreasonsratherthaninformationalreasons,andhencesatisfiesthe

exclusionrestriction.

OurfinaltechniquelooksatthechangeinleveragesurroundingadditiontotheS&P

500Index.Standard&Poor’sexplicitlyaimstomakeindexinclusionaninformation‐free

event,andassuchweuseitastreatmentintwodifference‐in‐differencesestimationsthat

testwhetherleveragechangesdifferentiallyfortreatedfirms–thoseaddedtotheS&P500

–relativetoamatched‐sampleofcontrolfirms–thosethatdonotexperiencetheshockto

institutionalholdingscausedbyindexaddition.

Even though the sample sizes vary dramatically across techniques, with all three

approacheswefindthatleveragedecreasesinresponsetoincreasesininstitutionalhold‐

ings.Withthe2SLSestimation,wefindthatthedecreaseinleveragestemmingfromanin‐

creaseininstitutionalholdingsisabout2.8timeslargerthanwhatwefindwithoutinstru‐

mentation, and represents an8.0% change in leverageper standarddeviation change in

holdings. Correspondingly, inournaturalexperiment,weshowthatafterbeingaddedto

theS&P500Index,firmsdecreasetheir leverageby1.5%(7.7%ofthetreatmentgroup’s

average leverage) relative to a groupof similar, propensity‐score‐matched control firms.

Takentogether,theseresultsareconsistentwithawiderangeofagency‐andasymmetric

information‐basedmodelsof capital structure thatpredict a relationshipadvancing from

agencycostsorinformationasymmetrytoleverage.

Atthesametime,wefindtheeffectofcapitalstructureoninstitutionalholdingsto

bemoreambiguous. Inboth level and first‐difference regressions,we find that leverage

hasanegativeeffecton institutionalholdings. However, inanArellano‐Bondestimation

5  

featuring (internal) instrumental variables,we find no effect of leverage on institutional

holdings. In accordance with most capital structure theories, our findings point to a

strongereffectadvancingfrominstitutionalholdingstocapitalstructurethanfromcapital

structuretoinstitutionalholdings.

Next, we investigate how firms change their capital structure in response to a

changeininstitutionalholdings. Wefindthatanincreaseininstitutionalholdingssignifi‐

cantlyraisesthesubsequentprobabilityofafirmbeinganetequityissuer,withoutaffect‐

ingtheprobabilityofdebtissuance.Furthermore,thesizeofequityissuancesalsoincreas‐

eswith positive changes in institutional holdings. Consistentwith direction of causality

frominstitutionstoleverage,andnottheotherwayaround,wedonotfindthatanincrease

inequity issuances inyear t results inan increase in institutionalholdings inyear (t+1).

Together,theseresultssuggestthatequityissuancesarethemainchannelthroughwhich

firmsadjusttheircapitalstructureinresponsetoincreasedinstitutionalholdings.

Althoughourresultssupportbothagency‐andinformation‐basedmodelsthatimply

institutional investorshaveanegative influenceonfirms’ leverage,wefindevidencethat

information asymmetry is the friction through which they most strongly impact capital

structure. Firmsparticularlyprone tohighagencycostsare, ingeneral, large,profitable,

cash‐richfirmswithrelativelyfewgrowthopportunities.Ontheotherhand,firmsparticu‐

larlypronetoinformationasymmetryandhighadverseselectioncostsaregenerallysmall,

young,cash‐strappedfirmswithrelativelymanygrowthopportunities.Wefindtheeffect

ofinstitutionalholdingsonleveragetobestrongerinsmallfirmswithmanygrowthoppor‐

tunities(“asymmetricinformationfirms”)thaninlargefirmswithfewgrowthopportuni‐

ties(“agencyfirms),althoughinpracticeinformationasymmetryandagencycostsarenot

mutuallyexclusivefrictions.

Furthermore,wefindnoevidencethattaxconsiderationsplayaroleintherelation‐

shipbetweeninstitutionalholdingsandleverage.WeutilizetheJobsandGrowthTaxRe‐

liefReconciliationActof2003(whichlowerstherelativetaxadvantageonequitybyinsti‐

tutions) to test whether the relationship between institutional holdings and leverage

changed.Wefindnodifferenceintheeffectofinstitutionalholdingsonleverageinthepe‐

riodbeforethetaxchangecomparedtotheperiodafter.

6  

Finally,wefindtheeffectofinstitutionalholdingsonleveragetobequiterobust.In

additiontoestimatingtheeffectduringvarioussubperiods,weredefineleverageinanum‐

berofdifferentways–usinga logit transformation,usingonly long termdebt,using the

marketvalueofassetsratherthanthebookvalue,usingdebtplusequityinthedenomina‐

torratherthantotalassets,andbydeductingexcesscashfromtotaldebt–andourresults

holdregardlessofthespecification.Inaddition,wecontrolforpotentialcovariatessuchas

market timing considerations, liquidity, and the corporate governance environment, and

findthatinstitutionalholdingsmaintainsitsstrongeffectonfirms’leverage.Lastly,wefind

ourresultstoberobusttousingquarterlydataratherthanannualdata.

Whilepriorliteraturehasnotfocusedontherelationshipbetweencapitalstructure

and institutional investors, the ideathat institutionsmayaffectcorporate financialpolicy

hasnotbeenignored.Specifically,recentworkhasshowninstitutionalinvestorstoplaya

role inmergersandacquisitions (Gaspar,Massa, andMatos (2005)), analyst activityand

accuracy(Ljungqvist,Marston,Starks,Wei,andYan(2007)),payoutpolicy(Grinsteinand

Michaely (2005)), executive compensation (Hartzell and Starks (2003)), CEO turnover

(Huson,Parrino,andStarks(2001)),andcorporategovernance(GillanandStarks(2000)).

Along these lines, Cronqvist and Fahlenbach (2009) find strong evidence of blockholder

fixed effects in firms’ investment and financial policies, as well as in firm performance

measures.Ourpapercontributestothisliterature.

Therestofthepaperisorganizedasfollows.InSectionIwereviewthetheoretical

determinantsforcapitalstructureanddiscusshowinstitutionalinvestorsimpactthesede‐

terminants,andbyextension,capitalstructure.InSectionIIwedescribeourdataandcon‐

structseveralimportantvariables.InSectionIIIwetesttherelationbetweeninstitutional

holdingsandleverage.InSectionIVweconclude.

I.Hypothesisdevelopmentandrelatedliterature

Manycapitalstructuretheoriessuggestthatagencycosts,asymmetricinformation,

andtaxesareimportantfrictionsthatenablethechoiceofleveragetoaffectfirmvalue.In‐

stitutional investors,however,oftenpossesscommoncharacteristicsthatmayimpactthe

severitytowhichfirmsareaffectedbythesefrictions,therebyalteringtheprocessthrough

whichcapitalstructureisdetermined.Inthissection,webrieflyreviewseveralofthemost

7  

relevant theories of capital structure, discuss some common institutional characteristics

thatsuggestinstitutionalinvestorsinteractwiththesefrictions,anddetailtheimplications

oftheseinteractionsforfirms’capitalstructures.

Onesetofcapitalstructuremodelsposits that theseverityofconflictsof interests

between managers and equity holders leads to a particular leverage ratio. Jensen and

Meckling(1976)arguethatthisconflictiscostlybecauseitincentivizesmanagerstoover‐

indulgeinactivitiesbeneficialonlytothem,astheybeartheentirecostofrefrainingfrom

theseactivitieswhilereapingonlyaportionofthegains.Debtalleviatesthisinefficiencyby

increasing themanager’s (relative) share of equity, thus increasinghis ownership of the

residual claim. In a similar vein, Jensen (1986) argues that debtmitigates shareholder‐

manager agency conflictsby committingmanagers topledge funds to repay creditors, in

turnlesseningtheamountof“freecash”availablefororganizationalinefficienciessuchas

consumingperquisitesandempirebuilding.

Institutions,however,canaffecttheseverityoffirms’agencycosts,becauseinstitu‐

tions engage in monitoring (e.g., Shleifer and Vishny (1986), Maug (1998), and Huson,

Parrino,andStarks(2001)). Inthepresenceofineffectivecorporatecontrolmechanisms

(Jensen (1993)), “monitoring” activities are value‐enhancing undertakings ranging from

minimalinterventioninacompany’saffairstomoreaggressivetechniquesofshareholder

activism,includingthe“WallStreetWalk,”initiatingpressurecampaigns,andjockeyingfor

a seat on the board of directors (e.g., Parrino, Sias, and Starks (2003), Gillan and Starks

(2007),AdmatiandPfleiderer(2009),andMcCaheryetal. (2010)). Onestrandof litera‐

ture suggests that blockholders and investorswith large holdings and long horizons are

particularlyeffectivemonitors (Edmans (2009)andChen,Harford, andLi (2007)),while

anotherstrandsuggeststhatentirefundfamiliesvoteinagreementduetocentralizedvot‐

ing departments or ISS guidelines for proxy voting (e.g., Matvos and Ostrovsky (2010),

Morgan,Poulsen,Wolf,andYang(2011),andIliev,Lins,Miller,andRoth(2012)).Thus,the

implicationisthatallinstitutions,tosomeextent,canbeeffectivemonitors.Forexample,

FennandRobinson (2009) find that even sponsorsof indexedETFsdevote timeand re‐

sources tomonitoring firms’management,andareactivebyvotingstrategically inproxy

contests.Thegoalofsuchactivitiesistoreducetheconflictsofinterestthatexistbetween

8  

shareholdersandmanagers, and indoing so, institutional investors canpotentially influ‐

encecapitalstructuredecisions.

Whilemanymodelssuggestthatshareholder‐manageragencyconflictsdrivecapital

structuredecisions(e.g.,Stulz(1990),Zwiebel(1996),andGomes(2000)),themannerin

which institutionsaffect this relationship isnot clear. Thedivergent theoretical implica‐

tionsarewell illustratedintwomodelsputforthbyLaPortaetal.(2000):thesubstitute

modelandtheoutcomemodel. In thecontextof thesubstitutemodel, institutionalhold‐

ings provide investor protection while acting as an alternative bonding mechanism for

debt,suggestingthat,allelseequal, firmswithahighproportionof institutionalholdings

needlessdebtinequilibrium.Ontheotherhand,intheoutcomemodeloutsideinvestors

musthaveadequatemechanismsinplace,perhapsintheformoflawsandregulations,be‐

foretheycanforcemanagementtoincreasedebtandtherebylimitpotentialwealthexpro‐

priation. Because institutional investors increase investorempowermentandprotection,

theoutcomemodelimpliesthatfirmswithastronginstitutionalshareholderbasewillhave

moredebt.

A second set of capital structure models suggests that firms choose their capital

structure in response to the information environment in which they operate. In these

models, insiders (managers) possess information regarding their firm’s future return

streamor investmentopportunities (i.e. their firm’s intrinsicvalue) that isnotknown to

outsideinvestors.Ingeneral,debtispreferabletoequityinfirmswithhigherlevelsofin‐

formationasymmetrymainlybecauseitcarrieslessadverseselectioncoststhanequity.

Institutionalinvestorscanplayanimportantroleinasymmetricinformationmodels

ofcapitalstructurebecausetheydevoteconsiderableresourcestocollecting information.

Manyinstitutionsaresubjecttostateorfederalfiduciaryandprudencystandards(seeDel

Guercio(1996))incentivizingthemtocollectinformationaboutthefirmstheyinvestinin

ordertodecreasethelikelihoodofbeingsuedforviolatingthesestandards(Allen,Bernar‐

do,andWelch(2000)). Thisprocessleadsthemtobemoreinformedthanothertypesof

investors, but also affects the adverse selection costs of equity by decreasing the infor‐

mationgapbetweenoutsideandinsideshareholders,becauseatleastaportionofthein‐

formation theycollect is reflected in their tradingpatterns (Sias (2004), andBusheeand

Goodman(2007)).

9  

Aswithagencymodels,asymmetricinformationmodelsofcapitalstructuredonot

unilaterally imply thesamerelationshipbetween institutionalholdingsand leverage. On

onehand,thetwocanbeinverselyrelated.InMyers(1984)andMyersandMajluf(1984),

due to information asymmetry andmispriced equity, firmsmay under‐invest in positive

NPVprojects.Asaresult,firmsfundinvestmentsbychoosingfinancingsourcesthatmini‐

mizeadverseselectioncosts,preferringretainedearningsoverriskydebtoverequity(the

peckingorder).Institutionalinvestorsmitigatetheadverseselectioncostsofequity,soas

equity financing becomes comparatively cheaper, firms with high institutional holdings

shouldhavelowerleveragethanthosewithlowinstitutionalholdings.2

On theotherhand, institutionalholdings canhaveapositive effect on leverage in

asymmetricinformationmodelsofcapitalstructure. Viswanath(1993)buildsontheMy‐

ersandMajluf(1984)intuitionthatthedilutivecostsofissuingequitymaycausefirmsto

rejectworthwhileprojects,andconstructsamulti‐periodmodel inwhich firmsoptimally

issuestocktoday,wheninformationasymmetryislow,toavoidthepotentiallossofaposi‐

tiveNPVprojectinthefutureduetoinformationasymmetryproblemsatthattime.How‐

ever, as institutions narrow the information gap betweenmanagers and (future) share‐

holders,theydecreasethedilutivecostsoffutureequityissuances.Assuch,astronginsti‐

tutionalshareholderbasemayallowfirmstoavoid issuingequitytoday, implyingacom‐

plementaryrelationshipbetweeninstitutionalholdingsandleverage.

Thetaxstatusofinstitutionalinvestorsisathirdcharacteristicthatsuggestsinstitu‐

tionsmayaffectcapitalstructure.Asequityholders,institutionsgenerallyhavearelative

tax advantage over individual investors. Many institutions are tax‐exempt, while many

otherspay taxesononlya smallportionof thedividends they receive from their invest‐

ments.Thus,allelseequal,afirmwithalargerbaseofinstitutionalequityholdersshould

issue comparatively more equity than a firm owned by (higher‐tax‐paying) individuals.

This implication is consistentwith theAllen,Bernardo,andWelch (2000)argument that

managerscanappealtothetaxadvantageonequityincomeenjoyedbyinstitutions,leading

institutional investors to formtaxclientelesconditionalonthedividendpoliciesof firms.

2InsignalingmodelssuchasRoss(1977),institutionalholdingscansubstitutefordebtinnarrowingthein‐formationgapbetweenoutsideandinsideshareholders, leaving firmswithlessneedtosignaltheirqualitywithcostlydebt.Thesemodels,however,havenotreceivedmuchempiricalsupport.

10  

Thus, tax considerations suggest that higher institutional holdings should be associated

withahigherportionofequityinfirms’capitalstructure.

Inadditiontotheoriesrelatedtoagency,information,ortaxes,arecentstrandofthe

capital structure literature examines the dynamics of firms’ capital structures as deter‐

mined by the costs of making adjustments (e.g. Leary and Roberts (2005), Strebulaev

(2007),andByoun(2008)),andtherelationshipbetweentheseadjustmentsandfirmchar‐

acteristicssuchascashflowsandinvestmentpolicies(e.g.,Faulkender,Flannery,Hankins,

and Smith (2012)). If institutions, as a group, have amitigating effect on firms’ agency

costsandinformationasymmetryproblems,costlyadjustmentmodels implythat, ingen‐

eral,firmswithgreaterinstitutionalholdingswilladjusttheircapitalstructuremorequick‐

ly. Whilethis is interestinginitsownright,thefocusofthispaperisontherelationship

betweeninstitutionalinvestorsandthedocumentedheterogeneityin leverageratios,and

not on the relationbetweenownership structure and the speedof adjustment of capital

structure.

Throughoutthissectionwehavediscussedcapitalstructuretheoriesinconjunction

with the premise that institutional investors can affect the capital structure policies of

firms.Thesetheories,however,donotexcludethepossibilitythatinstitutionsmayalsobe

affectedbyfirms’capitalstructures.Andindeed,priorstudieshaveshowninstitutionalin‐

vestors to exhibit preferences for certain firm characteristics. In particular, Del Guercio

(1996)findsthatinstitutionstendtoinvestinlowerriskfirms,whileGompersandMetrick

(2001) find that institutions prefer to hold large, liquid firms. Grinstein and Michaely

(2005)andDesaiandJin(2011)studyinstitutionalpreferencesforpayoutpolicy,withthe

former finding that institutions prefer dividend‐paying stocks over non‐dividend‐paying

stocks,andthelatterfindingthatinstitutionsformtax‐basedclientelesinrelationtopay‐

outpolicy.Intermsofcorporategovernance,McCaheryetal.(2010)findthat,ingeneral,

corporategovernanceisimportanttotheinstitutionalinvestmentdecision,whileHartzell

andStarks(2003)findthat,morespecifically,institutionsrevealapreferenceforfirmsin

whichexecutivecompensationexhibitshighpay‐for‐performancesensitivity. Thus,while

itmay bemost intuitive to think of the relationship between institutional investors and

capitalstructuretoadvancefrominstitutionstofirms,wealsoconsiderthealternativethat

institutionschoosetheirinvestmentsatleastinpartbecauseoffirms’leveragepolicies.

11  

II.Data,constructionofvariables,andsummarystatistics

Our data consists of all firm‐year observations for U.S. incorporated firms in the

CRSP–Compustatmergeddatabasebetween1979and2009, excludingAmericanDeposi‐

tary Receipts, utility companies (SIC codes 4900 – 4999), and financial firms (SIC codes

6000–6999).3 Thesample iscomprisedof firms listedontheNYSE,Amex,andNasdaq,

withtherequirementthatallfirm‐yearshavenon‐missingandnon‐negativedatafortotal

assets, total debt, and market capitalization.4 Following Lemmon et al. (2008), we

winsorizeratiovariablesattheonepercentlevelinanattempttolessentheeffectsofout‐

liers,andwerestrictleverageratiostotheclosedunitinterval.

InstitutionalholdingsdatacomesfromtheThomson‐ReutersInstitutionalHoldings

Database, as reported on Form 13F filed with the Securities and Exchange Commission

(SEC).Allinstitutionalmanagerswithmorethan$100millioninassetsundermanagement

arerequiredtofile,detailingallequityholdingsofmorethan$200,000or10,000shares.

Following Grinstein and Michaely (2005) we set institutional holdings equal to 0% for

firmsnotappearinginthe13Fdatabase(potentiallybecausetheinstitutionalmanagersdo

notmeettheentire13Ffilingrequirements),andwerestrict institutionalholdingstothe

closedunit interval. Ourresultsareunchanged,however, ifweremovetheobservations

notappearinginthe13Fdatabaseorallowinstitutionalholdingstoexceed100%(asocca‐

sionallyhappensduetofiscal/calendaryeardifferences).

After merging the institutional holdings data with the CRSP‐Compustat database

andremovingobservationsnotmeetingtheaforementioneddatarequirements,weareleft

with130,202firmyearobservationsacross31years. Oursamplecontainsanaverageof

4,200firmsperyear,andfluctuatesbetween3,228firmsin1979and5,741firmsin1997.

Becauseweonlyrequirepositiveassets,apositivemarketcap,andnon‐missingdebt,near‐

lyallofthefirmsintheNYSEarerepresentedeachyear.

3ThistimeframewaschosenbecausetheThomson‐Reutersinstitutionalholdingsdatabeginsin1979.4Ourresultsarerobusttoalternatespecificationsinwhichwerequirestockpricetobeatleast$5andassetstobeatleast$1million.Furthermore,multivariableanalysesimplicitlyrequirenon‐missingdataforallrele‐vantvariables,sothenumberoffirm‐yearsusedinthevarioustestsfluctuatesdependingonthespecificationchosen.

12  

WeimposeoneadditionaldatarequirementforourArellano‐Bond(1991)estima‐

tions. Since we use lagged levels to instrument for each contemporaneous, differenced

equation,werequireallfirmstoappearconsecutivelyforatleastfouryearsatsomepoint

duringoursampleperiod.Thisexcludesfirmsleavingoursamplebetween1979and1981,

andfirmsenteringoursamplebetween2007and2009.Afterimposingthisadditionaldata

requirement,weareleftwith122,859firmyearobservationsforourArellano‐Bondtests,

foranaverageof3,963firmsperyear.

Weconstructourvariablesinaccordancewithseveralimportantpriorstudies.The

fullsetofvariabledefinitionscanbefoundintheAppendix,butwehighlightourkeyvaria‐

bleshere.Wedefineinstitutionalholdings(Hold)astheratiooftotalsharesheldbyinstitu‐

tionsattheendofeachcalendaryeartothetotalnumberofsharesoutstandingattheend

ofthelastfiscalyear.Wesimilarlydefineinstitutionalownershipbytheinstitutionshold‐

ingthefivelargestpositions(Hold5)asaproxyforownershipconcentration.Whilecoor‐

dinationmechanismsexistforinstitutionstoactcollaboratively,itstillmightbethatlarger,

moreconcentrated institutional investorsaremoreeffectiveatmitigatingagencyand in‐

formationasymmetryconcerns(ShleiferandVishny(1986)).Inmostcasesourresultsdo

notchangewhenswitchingbetweenHoldandHold5,butwhentheeffectisnon‐negligible,

wespecificallyaddressit.

Ourotherprimary variable of interest is firm leverage (TDA),whichwedefine as

perFrankandGoyal(2009)astheratiooftotaldebttothebookvalueofassets. Forro‐

bustnesswerepeatour testswithseveralalternativemeasuresof leverage: totaldebt to

themarketvalueofassets,long‐termdebttothebookvalueofassets,long‐termdebttothe

marketvalueofassets,totaldebtdividedbytotaldebtplusstockholder’sequity,andtotal

debtlessexcesscashdividedbytotalassets.5 Ourresultsholdwhenusingthesealterna‐

tivemeasures,andarealsorobusttotheinclusionofoffbalancesheetoperatingleasesinto

ourdefinitionofleverage(TDLA,asperGraham,Lemmon,andSchallheim(1998)).

TheremainingvariablesaredefinedintheAppendix.There,wedefinenetdebtand

equityissuancesandissuersasperFamaandFrench(2005);adjustedreturn,beta,sales,

andadividendpayerdummyasperGrinsteinandMichaely(2005);cashcowasperBrav,

5Wedefineexcesscashascashholdingsminusrevenue*.02ifcashholdingsexceed2%ofrevenue,andzerootherwise.

13  

Graham,Harvey,andMichaely (2005); initial firm leverageasperLemmonetal. (2008);

and assets, change in assets, market cap, profitability, market‐to‐book, capital expendi‐

tures, tangibility, intangible assets, collateral, total payout, a lossmaking dummy, and a

uniquenessdummyasperFrankandGoyal(2009).

Table Ipresentssummarystatistics forboth theentiresampleperiodand for two

nearlyequallengthsubperiods(theearliersubperiodcontainssixteenyearsofdatarather

than fifteen). Non‐ratiomeasures (market cap, log(assets), and log(sales)) are inflation‐

adjustedto2009dollars,andincreaseovertime.Table1showsthattheaveragepublicly

tradedfirmissignificantlylargernowthanitwas30yearsago.

Theratiomeasuresinthetablealsochangesubstantiallyovertime,includingaggre‐

gate institutionalholdings(Hold)andtop‐5 institutionalholdings(Hold5). Averagehold‐

ingsmorethandoublefromthefirsthalfofthesampletothesecond(19%to40%),while

top‐5institutionalholdingsnearlydouble(11%to20%).Eventhesefigures,however,un‐

derstatethecurrentimportanceofinstitutionalequityinvestments. In2009,meanhold‐

ings increased to 55% overall, but jumped to 76% for firms in the fourth size quintile

($1,169millionaveragemarketcap),and77%forfirmsinthefifthandlargestsizequintile

($15,279millionaveragemarketcap).Invalue‐weightedterms,byyear‐end2009,institu‐

tionsheld71%ofthe$10.6trilliondollarmarketcapforallfirmsinoursample.

Averageleverage,ontheotherhand,fallsfrom25%inthefirsthalfofthesampleto

22%inthesecondhalf,whileleverageincludingoff‐balance‐sheet(OBS)operatingleases

falls from 34% to 29%. This slightly smaller drop (in percentage terms) illustrates the

growthinoperatingleasesfoundinFranzenetal.(2009).

III.Results

A.Theeffectofinstitutionalholdingsonleverage

The implication of traditional capital structuremodels is that by interactingwith

frictionssuchasagencycostsandasymmetric information, institutional investors’equity

holdingscanplayaroleinthedeterminationofleverageratios.

To characterize the relationship between institutional holdings and leverage, we

mustcontrolforotherfirmcharacteristicsthataffectleverageandmaybecorrelatedwith

institutional holdings. To this end,webeginour empirical studywith amultivariate re‐

14  

gressionwithleverage(timet)asthedependentvariable.Sinceoneofourgoalsistoex‐

amine thecausaleffectof institutionalholdingson leverage,weexaminehowthe lagged

valueofinstitutionalholdingsaffectsleverage(whilethisisonlyapartialsolution,welater

examinetheissuemorethoroughly). Ouradditionalindependentvariablesaremeasured

at time (t – 1), and, aside from institutional holdings, are commonly used in the capital

structureliterature(see,e.g.,FrankandGoyal(2009)):market‐to‐book,logofassets,prof‐

itability, tangibility,median industry leverage, adividendpayerdummy, capital expendi‐

tures,collateral,intangibleassets,totalpayout,changeinassets,andindustryuniqueness.

Wealsocontrolforinitialfirmleverage,asLemmonetal.(2008)findleverageratiostobe

surprisinglystableovertime. Lastly,weincludeyeardummyvariablesinallofourtests,

andincertainspecifications,weincludefirmfixedeffects.Weoptnottoincludeindustry

fixed effects in our regressions because Lemmon et al. (2008) find that industrymedian

leveragevitiatesthepoweroftheindustryfixedeffecttoexplainleverageratios,andsub‐

stantiallyincreasestheadjustedR‐squaredintheprocess.Noneoftheresultsinthispaper

change,however,ifweincludeindustryfixedeffects.TableIIpresentstheresultsofthese

panel regressions,with t‐statisticsandsignificance levels reflectingstandarderrorsclus‐

teredatthefirmleveltoallowtheerrortermtobeheteroskedasticandcorrelatedwithin

firms.

Thedata incolumn1showthat firmswithhighinstitutionalholdings,onaverage,

havelowleverageevenaftercontrollingforotherfirmcharacteristics.Withonlyyearfixed

effectsascontrols,thecoefficientonHoldis‐0.028(unreported),andafteraddingcontrol

variables known to explain the variation in leverage, such as initial firm leverage (Ini‐

tial_TDA)andthemedianwithin‐industryleverageratio(Industry_TDA),theabsolutemag‐

nitudeincreasesto‐0.100(column1). Theresultissignificantatthe1%level,andisro‐

busttochangesinempiricalspecification,suchasusingtop‐5institutionalholdingsorre‐

strictingthesampletofirmswithleverage>10%.Infacttheresultisevenstrongerwhen

weusemarketleverageratherthanbookleverage,orifweexcludefirmswith0%institu‐

tionalholdings(generallyfirmsthatdonotappearonany13Fforms).

Incolumn2weincludeafirmfixedeffecttocontrolforpotentialfirm‐specificomit‐

tedvariables. Whiletheeffectof institutionalholdingsonleveragesomewhatabates,the

15  

coefficientof‐0.068remainssignificantatthe1%level.6Ineconomicterms,theregression

estimateincolumn1(2)suggeststhatanincreaseininstitutionalholdingsbyonestandard

deviationisassociatedwithaleverageratio2.9%(2.0%)lower,onaverage,inthefollow‐

ingyear. Toput thatnumber inperspective,2.9%representsapproximately12%of the

averageleverageratio(23.6%)duringoursampleperiod.Infact,afterfirststandardizing

allvariables tohaveameanof0andastandarddeviationof1, theeffectof institutional

holdings(‐0.133)iscomparableinmagnitudetothatofprofitability(‐0.147)andindustry

medianleverage(0.127),andistwiceasstrongastangibility(0.068). Theonlyvariables

with a considerably stronger effect are assets (0.207) and initial leverage (0.316). Fur‐

thermore,wefindthiseffecttobeapermanentone.Thatis,wedonotfindevidenceofa

reversal after regressing leverageon the second lagor third lag of institutional holdings

(thecoefficientsremainnegative),nordowefindthatthesecond/thirdlagsaffectleverage

atallwhenthefirstlagisincluded(notreported).

Giventhatpriorresearchhasfoundbothinstitutionalholdingsandleveragetovary

strongly by firm size (e.g. Gompers and Metrick (2001) and Bennett, Sias, and Starks

(2003)),inournextspecificationweinteractholdingswithadummyvariablerepresenting

each firm’s annualmarket capquintile to seewhether the relationbetween institutional

holdingsandleveragevariesbyfirmsize.Asreportedincolumn3,wefindthisinteraction

tobemeaningfuleveninthepresenceofafirmfixedeffect.Thenegativerelationshipper‐

sistsinallmarketcapquintiles,althoughtherelationshipisonlymarginallysignificantin

the largestsizequintile. Intermsofeconomicmagnitude, institutionalholdingshavethe

largesteffecton leverageamongstmedium‐sizedfirms:those inquintiles2,3and4. We

findthat theaverageeffecton leverageofa (quintile‐specific)onestandarddeviation in‐

crease in institutionalholdingsvariesbetween ‐0.5%for the largest firms,and ‐1.1%for

the firms in quintile 3. By allowing the relationship between leverage and institutional

6Lemmonetal.(2008)notethatwhilemostcoefficientsremainsignificantinaleverageregressionwithfirmfixedeffects,themagnitudesofthecoefficientsshrinkconsiderably(TableVintheirpaper).We,however,donotfindsuchstarkdifferencesbetweencoefficientsinourspecificationwithfixedeffectsandourspecifica‐tionwithout.Thedifferencebetweenourfindingsandtheirsstemsfromtreatmentoftheerrorterm.Lem‐monetal.imposeafirst‐orderautoregressivestructureontheirerrorterm,andwedonot.Wedo,however,testforserialcorrelationintheerrorsusingavariantoftheDurbin‐Watsontestderivedforusewithanun‐balancedpanel(Bhargava,Franzini,andNarendranathan(1982)),anddonotfindconclusiveevidenceinfa‐vorofanautoregressiveerrorstructure.

16  

holdingstovarybyfirmsizeweexplainalargerportionofthevariationinleverageratios,

increasing our adjusted R2 from 32.3% to 35.0% without a firm fixed effect, and from

63.9%to64.6%withafixedeffect.Furthermore,ourresultscomparefavorablytosimilar

testsencounteredintheliterature,whereadjustedR2sintheabsenceofafirmfixedeffect

peakataround30%.7

Inunreportedanalysis,were‐estimatecolumns1through3aftersubjectingourde‐

pendentvariableTDAtoalogittransformation.Sinceleverageisaproportionboundedbe‐

tween zero and one, the regressions in columns 1 through 3might bemisspecified, and

performing a logit transformation alleviates any concerns this may cause. Here, and

throughoutthepaper,ourresultsareunchangedwhenusingthetransformed,unbounded

dependentvariable. Consequently, intheremainingtableswereportresultsonlyforour

boundedleverageandinstitutionalholdingsvariables,asthecoefficientsontheindepend‐

ent variables have straightforward interpretations. Finally, althoughwe cluster standard

errorsatthefirmlevel,were‐runthetestsincolumns1through3accordingtothetech‐

niqueoutlinedinFamaandMacBeth(1973).Ourconclusionsdonotchange.

One factor thatmay affect capital structure ismarket timing (Baker andWurgler

(2002)).Inaddition,equitymarkettimingmayinteractwithinstitutionalholdingsaswell.

For example,Alti andSulaeman (2012) show that firms issue equity afterhigh stock re‐

turnsonlywheninstitutionaldemandisstrong.Theirfindingmightimplythattherelation‐

shipwefindbetweenleverageandinstitutionalholdingsisnotonlybecauseoftheeffectof

institutionson the frictionsunderlying capital structuremodels, but ratheranartifact of

firmstimingthemarketwithequityissuances.

Wecontrolforthispossibilityintwoways.First,weincludetheBakerandWurgler

(2002) external finance weighted average (EFWA)market‐to‐book ratio (defined in the

Appendix)inaregressionmirroringtheoneinColumn1,andourresultholds(notreport‐

ed). Second, tospecificallyaddress the findingsofAltiandSulaeman(2012),we include

laggedabnormalstockreturn(realizedreturnrelativetotheCAPMprediction–definedin

theAppendix)ontherighthandsideoftheestimationinColumn4.Bycontrollingforab‐

7Insimilarregressions,Lemmonetal.(2008)produceanadjustedR2of30%withdatafrom1965to2003,while Frank and Goyal (2009) produce adjustedR2s of 30%, 14%, 14%, and 21% for the 1970’s, 1980’s,1990’s,and2000to2003,respectively.

17  

normalstockreturn,weestimatetheeffectofinstitutionalholdingsonleverageoutsideof

thecontextstudiedinAltiandSulaeman(2012),andwecontinuetofindthatinstitutional

holdingshaveanegativeeffectonleverage.Infact,themagnitudeofthecoefficientactual‐

lyincreasesfrom‐0.068inColumn2to‐0.070inColumn4,althoughthesampleisslightly

smaller(laggedreturnrequirestwoprioryearsofpricingdata,asopposedtotheoneprior

yearrequiredbythevariablesinColumn2).Furthermore,ifwerestrictthesampletoin‐

cludeonlyfirmsthatdidnothavealargeabnormalreturnintheprioryear(e.g.ifwein‐

cludeonlyfirmswithreturnsinthe25thto75thpercentile),ourresultshold.Basedonthe‐

seresults,weconcludethatourfindingsarenotamanifestationoffirmstimingthemarket

withequityissuances.

Aprimarymotivationforthispaperisthatinstitutionalinvestorsaffecttheconflict

ofinterestbetweenoutsideequityholdersandmanagers. However,asecondconflictex‐

ists between equity holders and debt holders (e.g., Jensen and Meckling (1976), Myers

(1977),andDiamondandHe(2011)),anditisunclearwhetherinstitutionalequityinves‐

torshaveaneffectonthisconflict.Ononehand,greaterinformation‐gatheringandeffec‐

tivemonitoringbyinstitutionsreducestheopacityoffirm’assets.Indeed,Edmans(2009)

shows that blockholders attenuatemanagers’myopic behavior,making bonds less risky,

whileJiang,LiandShao(2010)findthatwheninstitutionsarebothequityholdersanddebt

holders,thecostofborrowingfalls.Ontheotherhand,greateralignmentbetweenmanag‐

ers and shareholders can lead to more managerial risk‐taking and higher costs of debt

(Ortiz‐Molina(2006)),particularly in theabsenceoftakeoverprotections(Cremers,Nair,

andWei(2007)).

Inourempiricalspecificationswecontrolforthesepotentialeffects inavarietyof

differentways.Mostbroadly,wecontrolforfirms’creditratingstoproxyfortheeffectof

institutionalholdingsonthecostofdebt.Inaddition,wecontrolforthelevelofmanagerial

equityholdingsbecauseDatta,Iskandar‐Datta,andRaman(2005)showthatmanagersare

more likely tomakecapital structuredecisions inalignmentwith thepreferencesofout‐

sideequityholdersiftheirownshareholdingsarehigh.Andfinally,giventhattheconflict

of interestbetweenshareholdersandbondholdersdependson thecorporategovernance

environmentofa firm(Cremersetal. (2007)),wecontrol for theGompers, Ishii,Metrick

(2003)GIMIndexandthenumberofanalyststhatcovereachfirm.

18  

Wefindthattheeffectofinstitutionalholdingsonleveragepersistsaftercontrolling

foreachofthesemeasures. Controllingforcreditratingsdoesnotmateriallychangeany

result,andcontrollingfortheGIMIndexdoesnotchangeourresultsandsignificantlyre‐

ducesthesamplesize,soweonlyreporttheestimatesaftercontrollingforinsiderholdings

(column 5) and analyst coverage (column 6). We find that after controlling for insider

holdings theeffectof institutionalholdingson leverageremainsnegativeandsignificant,

eveninaregressionwithfirmfixedeffects. Thecoefficientof‐0.044issmallerinmagni‐

tude than the ‐0.068 found in column 2, but the samples are very different since

ExecuCompinsiderholdingsdataisonlyavailableafter1991,andonlyforS&P1500firms.

Overall,wefindtheeffectofinstitutionalholdingsonleveragetobequiterobusttothepo‐

tentialeffectofinstitutionalholdingsontheconflictofinterestbetweenequityholdersand

debtholders.

Next,weexaminethepossibilitythattherelationshipbetweeninstitutionalholdings

andleveragestemsfromimportantcovariatesofinstitutionalholdings–analystcoverage

andequityliquidity–ratherthanfromtheholdingsthemselves.Chang,Dasgputa,andHil‐

ary(2006) find that firmswithmoreanalystcoveragehave lower targetdebtratios,and

thatfirmswithlessanalystcoveragearelesslikelytoissueequityandmorelikelytotime

themarketwhen they do issue. Furthermore,Hennessey andWhited (2005) show that

leverageispersistentandnegativelycorrelatedwithlaggedliquidity.

Toaccountforthecorrelationbetweeninstitutionalholdingsandanalystcoverage,

wefollowChangetal.(2006)andusedatafromI/B/E/StodefineAnalystCoverageasthe

maximumnumberofanalystswhomakeannualearningsforecastsinanymonthduringthe

twelvemonthspriortofiscalyearend,whileassumingthatfirmsnotcoveredbyI/B/E/S

have no analyst coverage. And to account for equity liquidity we include the Amihud

(2002)illiquiditymeasure(definedintheAppendix). Wefindthatinstitutionalholdings

remainasignificantdeterminantofleverageaftercontrollingforthenegativerelationship

betweenanalystcoverageand leverage(column6),andthat therelationship isrobust to

theinclusionofAmihud’sliquiditymeasure(unreported).

Tofurtherextendouranalysis,wemodifyourdefinitionofleveragetoincludeoff‐

balance‐sheet(OBS)operatingleasesincolumn7.Franzenetal.(2009)provideevidence

ofadramaticincreaseintheuseofOBSoperatingleasesoverthelastpastdecades,advo‐

19  

catingthatamorecompletemeasureofleveragewouldincludeOBSleasesonbothsidesof

the balance sheet. Furthermore, Rampini and Viswanathan (2010) and Eisfeldt and

Rampini (2009) suggest, respectively, that leasing increases firms’debt capacitybecause

“leasingtangibleassetsrequireslessnetworthperunitofcapital”and“repossessionofa

leasedassetiseasierthanforeclosureonthecollateralofasecuredloan.”Assuch,operat‐

ingleasesmayplayaroleintherelationbetweeninstitutionalholdingsandleverage.

As per Franzen et al. (2009) andGrahamet al. (1998),we calculate leases as the

capitalizedvalueofnon‐cancellableOBSoperatingleases,andincorporatethatvalueinto

bothtotalassetsandtotaldebt.Wefindthatwhiletherelationbetweenholdingsandlev‐

eragebecomesslightlylessnegativeineachsizequintile(e.g.from‐0.060forfirmsinthe

thirdsizequintileincolumn3to‐0.050forthesamefirmsincolumn7),ourresultsarero‐

busttotheinclusionofOBSleases.8Infact,ouradjustedR2sincreasesignificantlywhenwe

includeleases.Withoutleases,thespecificationincolumn1yieldsanadjustedR2of32.3%,

butwhenweaddOBSleasesintodebtandassetstheadjustedR2increasesto37.1%(not

reported). Similarly, comparing columns 3 and 7,we see the adjustedR2 increase from

64.6%to69.0%whenleasesareincluded.

Finally,priorliteraturesuggeststhattherelationbetweeninstitutionsandfrictions

suchasagencycostsandasymmetricinformationmaydifferbyinstitutiontypeorinvest‐

menthorizon(Changetal.(2007)),althoughitisnotclearapriorihowthisshouldaffect

firms’leverageintheaggregate.Toexaminewhethertherelationshipbetweeninstitution‐

al holdings and leverage differs by these institutional characteristicswe utilize type and

horizonclassificationsascalculatedandprovidedbyBrianBusheeonhiswebsite.9While

wedonotreporttheresultsofthesetests,wefindthattherelationbetweeninstitutional

holdingsandleveragedoesnotparticularlyvarybyinstitutiontype. Investmentadvisers

(mutual funds, hedge funds, etc.) have the strongest effect, and the largest holdings, but

pensionfunds,insurancecompanies,andbanksallhaveanegativeandsignificanteffectas

well.Universityandfoundationendowments,aswellas“miscellaneous”institutions,have

aninsignificanteffect.Asforinvestmenthorizon,wefindthatlong‐terminvestors(“dedi‐ 8Thesampleincolumn6isdifferentfromthatincolumn3,sowecannotdirectlycomparecoefficients.How‐ever,whenwere‐run theTDA regressionsusing theTDLAsample, theHold coefficients in theTDA regres‐sionsarestillmeaningfullylargerthantheHoldcoefficientsintheTDLAregressions.9http://acct3.wharton.upenn.edu/faculty/bushee/

20  

cated” institutions)andquasi‐indexershaveanegativeandsignificanteffecton leverage,

whileshort‐terminvestors(“transient”institutions)haveanegativeeffectthatisonlysig‐

nificantinaregressionwithfirmfixedeffects. Quasi‐indexershavethelargesteffect,and

alsohavethelargestholdings.

Sofarwehavecontrolledfordifferencesinfirmcharacteristicsbyincludinglagged

levelsofimportantvariablesinourregressions,andhavealleviatedconcernsoveromitted

variablesby including firm fixedeffects. Estimating theeffectof changes in institutional

holdings on changes in leverage is anotherway to control for unobservedheterogeneity

and omitted variables, but also allows us to estimate the effect of holdings on leverage

whilecontrollingforlaggedchangesinleverage.Tothisend,TableIIIpresentstheeffects

oflaggedchangesininstitutionalholdingsonchangesinleverage.Theoutputincolumn1

showsthatachangeininstitutionalholdingsresultsinasubsequentchangeinleverageto

theoppositedirection. The coefficientof ‐0.014 is significantat the1% level, evenafter

controllingforlaggedchangesinleverage.

ThesecondcolumnofTableIIIpresentstheeffectofchangesinstitutionalholdings

on subsequent changes in leverage,withOBS leasesadded tobothdebtandassets. The

magnitudeoftheeffectisover50%largerthaninthecasewithoutOBSleases,andtheco‐

efficientremainssignificantatthe1%levelevenaftercontrollingforlaggedchangesinlev‐

erage.Aswasthecaseinlevels(TableII),theeffectofchangesininstitutionalholdingson

changesinleverageisrobusttotheinclusionofOBSleases.

Overall,ourfindingsthusfarlendsupporttothehypothesisthatinstitutionalhold‐

ings and debt exist as partial substitutes, rather than complements, in themitigation of

problemsstemmingfromagencycostsorinformationasymmetry.Firmsinwhichinstitu‐

tions have a strong presence tend toward low leverage, and firms in which institutions

strengthentheirpresenceareassociatedwithlowerleverage.What’smore,ourresultsin

TablesIIandIIIsuggestthattherelationshipmaybeofacausalnatureratherthansimply

anassociation, aswe find that firmsdecrease (increase) their leverage inresponse to in‐

creases(decreases)ininstitutionalholdings.Wetestthisindetaillaterinthepaper.

B.Theeffectofleverageoninstitutionalholdings

21  

Whilemostcapital structure theories implicitlysuggestcausality from institutions

tofirms’leverage,onecanthinkofscenarios,notinconsistentwiththesetheories,inwhich

causalitywouldgointhereversedirection.Infact,severalrecentpapersshowinstitution‐

alinvestorstoexhibitpreferencesoverfirmcharacteristics,establishingaprecedentsug‐

gesting that firms’ capital structuremay influence institutional investors’ equity holding

decisions.Inparticular,institutionshaveshownpreferencesoverfirmcharacteristicssuch

asvisibility,transactioncosts,andvolatility(Falkenstein(1996)),payoutpolicy(Grinstein

andMichaely(2005);DesaiandJin(2011)),andcorporategovernance(FerreiraandMatos

(2008),andMcCaheryetal. (2010)). Furthermore,DelGuercio(1996) findsthat institu‐

tionstilttheirportfoliostoward“prudent”stocks,whichinthecontextofthispapermight

implylow‐leveragefirms.

Totesttheeffectof leverageoninstitutionalholdingswestartwithalevelregres‐

sionwherethedependentvariableisinstitutionalholdings,measuredattimet.Theinde‐

pendentvariables,measuredat time(t–1),are leverage, logsales,beta,annualadjusted

return,totalpayout,market‐to‐book,adividendpayerdummy(equalto1intheeventthe

firm pays a dividend), and year dummy variables. Following Grinstein and Michaely

(2005),weselectlogsalesandbetatoaccountforrisk,themarket‐to‐bookratiotocontrol

forgrowthopportunitiesandasymmetricinformation,andourtwopayoutpolicyvariables

becausetheauthorsfindarelationshipbetweendividendpayersandholdings,aswellas

betweenrepurchaseactivityandinstitutionalholdings.TableIVpresentstheeffectoflev‐

erageoninstitutionalholdingsinPanelA,andtheeffectofchangesinleverageonchanges

ininstitutionalholdingsinPanelB.

In Panel A we find that, on average, institutions have higher holdings in low‐

leveragefirmsthaninhigh‐leveragefirms.Withoutafirmfixedeffect(column1),there‐

sultissignificantandsuggeststhataonestandarddeviationincreaseinlaggedleverageis

associatedwitha2.5%decreaseininstitutionalholdings,afigurecorrespondingto10%of

theaverage full‐period institutionalholdings (24%). Witha firmfixedeffect (column2),

theeffectabatesbyaround30%,althoughitremainssignificant.Inaddition,weverifythat

ourresultsarerobusttotheinclusionofOBSoperatingleasesintoourdefinitionoflever‐

age,arenotdrivenbyatimetrend(werepeatourtestsforseveralsubperiods),andarenot

22  

duetothepossiblemisspecificationofinstitutionalholdings(were‐runthetestsaftersub‐

jectingHoldtoalogittransformation).

Asanalternativetoafixedeffectsregression,wetesttherelationbetweeninstitu‐

tional holdings and leverage in first differences rather than in levels, and include lagged

changesininstitutionalholdingsasacontrol.TheresultsarereportedinPanelB.Wefind

that,onaverage,institutionalholdingsdecreaseafterfirmsincreasetheirleverage,regard‐

less if leverage is calculatedwithorwithoutOBSoperating leases (wedonot report the

testthatincludesOBSleases).Evenaftercontrollingforthelaggedchangeininstitutional

holdings,theΔTDA coefficientof‐0.037issignificantatthe1%level.

C.Addressingtheissueofendogeneity

Sofarwehavefoundthatinstitutionalholdingsnegativelyaffectleverage,andthat

leveragenegativelyaffectsinstitutionalholdings.Whileregressionsonlagged,differenced

variables arean important step towardaddressing the issueofendogeneity, theydonot

solve the issue. In this subsectionwe employ three techniques to remove the potential

feedback effect of leverage on holdings: two separate instrumental variables (IV) ap‐

proachesandasemi‐naturalexperiment.OurfirstIVapproachisadynamicpanelestima‐

tionsimilar to theonedevelopedbyHoltz‐Eakin,Newey,andRosen(1988)andArellano

andBond(1991),whileoursecondIVapproachisaclassicaltwo‐stageleastsquaresesti‐

mationusingmutual fundflowinducedtradingasan instrumentfor firm‐specific institu‐

tionalholdings.Finally,ourthirdapproachisasemi‐naturalexperimentinwhichaddition

totheS&P500Indexprovidesanexogenousshocktoinstitutionalholdings.

C.1.DifferenceGMM

TheArellanoandBond(1991)“difference”generalizedmethodofmomentsestima‐

torfitslinearmodelswithonedynamicvariableandadditionalcontrolsonshort,wideNx

Tpanels.Itisdesignedtohandlefixedeffectsandendogeneityofregressorsbyusingpast

valuesofendogenousvariablestoinstrumentforcurrentchangesinthosevariables. The

procedure avoids dynamic panel bias (Nickell (1981)), and deals with potential biases

causedbythecorrelationofinstitutionalholdingsandleverageovertime.

23  

Weruntwoseparateestimations,onewithleverageasthedependent,dynamicvar‐

iable,andonewithchanges in institutionalholdingsas thedependent,dynamicvariable.

Themodelsarespecifiedasfollows:

Leverageit , InstHoldingsi,t‐1 xi,t‐1' εit

InstHoldingsit , Leveragei,t‐1 xi,t‐1' πit

and

E E E 0andE E E 0

(1)

whereiindexesthefirm,tindexestime,xisavectorofcontrols,µiandφiarethefirmfixed

effects,andνitandωitare the idiosyncraticshocks,whichareorthogonal toµiandφi, re‐

spectively. The vectorx can contain variables that are endogenous, predetermined, and

exogenous,aswellasdeeperlagsofthedependent,dynamicvariable.Theestimationisa

two‐step procedure: first, variables are differenced to remove firm fixed effects, and se‐

cond,aGMMestimatorinstrumentsendogenousvariableswiththeiravailablelagsinlev‐

els.

TableVpresentstheestimatesfromourArellano‐Bondestimationswithleverageas

thedependent,dynamicvariable inPanelA, and institutionalholdingsas thedependent,

dynamicvariableinPanelB.Inbothpanels,theinternalinstrumentsbeginatthethirdlag

toavoidproblemsstemmingfromautocorrelatederrorterms,andwecollapseourinstru‐

mentmatricestoeliminateproblemsassociatedwithexcessiveinstruments.

TheresultsinPanelAshowastrong,negativeeffectofinstitutionalholdingsonlev‐

erage.ThecoefficientonHoldissignificantatthe1%level(column1),evenaftercontrol‐

lingforotherimportantdeterminantsofleverage,autocorrelationininstitutionalholdings

andleverage,timevaryingidiosyncraticshocks,andtimetrends.Inaddition,theresultis

robusttotheinclusionofOBSoperatingleasesintoourdefinitionofleverage(notreported

inthetable).

Inthespecificationestimatedincolumn2,wesubstituteHold5,theholdingsofthe

fivelargestinstitutionalinvestors,fortotalholdings(Hold).Theresultisagainsignificant

andisevenstrongerthantheonepresentedincolumn1,withtheabsolutemagnitudeof

24  

thecoefficientincreasingfrom‐0.148to‐0.183.Becausewearetestingacausalrelation‐

ship, itmakes intuitivesense that the top five institutional investorshaveamoresignifi‐

cant impactonleveragethanall institutionscombined,astheylikelyhavethehighest in‐

centivestomonitor,collectinformation,andengageinactivism.

We test thestrengthofour instrumentswith theprocedureoutlined inStockand

Yogo(2005)andreturnaKleibergen‐PaapF‐statisticof68.0,whichexceedsallStockand

Yogo(2005)instrumentalvariablecriticalvaluesforweakidentification. Bothspecifica‐

tionspresentedinPanelApasstheArellano‐Bondtestforsecond(third)orderautocorre‐

lationinlevels(differences).Insum,ourinstrumentspasstestsforbothinstrumentvalidi‐

tyandstrength.Finally,whilethespecificationpresentedincolumn1failstheHansentest

for joint validity of the full instrument set, the J‐test, the specification in column2using

top‐5holdingspasseseasily.

PanelB,however,tellsadifferentstory.Whileourpreviousregressiontestsinlev‐

elsandindifferencesshowleveragetohaveanegativeeffectoninstitutionalholdings,this

effectdisappearsonceweremovetheeffectofholdingsonleverage.Withtheinstitutional

holdings as the dependent, dynamic variable in column 1, the coefficient onTDA is, at ‐

0.012, insignificantly different from zero. Similarly, after substituting top‐5 institutional

holdingsfortotalholdings,wefindincolumn2thattheeffectofleverageisstillinsignifi‐

cant.BothspecificationspasstheHansenJ‐test,suggestingthevalidityofthefullsetofin‐

struments,aswellastheStockandYogotestsforinstrumentweakness.10

Alltold,theresultsinTableVsuggestthatafterremovingtheendogenouscompo‐

nents of the relationship between institutional holdings and leverage, institutions affect

firms’ leveragethroughtheirholdings,butchangesin leveragedonot leadinstitutionsto

altertheirholdings.

C.2.Two‐StageLeastSquares

Next,weestimatetheextenttowhichinstitutionalholdingsaffectcapitalstructure

usinganinstrumentalvariableandlineartwo‐stageleastsquares.Toremovethepotential

10InanalternativespecificationtoPanelB,column1,werestrictourinstrumentmatrixtoincludeonlyonelaggedchangeininstitutionalholdings,andalsoincorporateourinstrumentalvariablefromthenextsection,MFFlow,intotheestimation.Leveragecontinuestohaveaninsignificanteffectoninstitutionalholdings.

25  

feedbackeffectofleverageoninstitutionalholdingsweneedaninstrumentforinstitutional

holdings that is unrelated to the unobserved factors that affect firms’ financial leverage.

Our instrument, implied trades induced bymutual fund outflows (MFFlow), is borrowed

fromEdmans,Goldstein,andJiang(2011).11

The ideabehind the instrument is that outflowsbymutual fund investors lead to

changesinmutualfundportfolioholdings,asthefundsmustsellaportionoftheirholdings

torepayinvestorswhowishtoredeemtheirinvestment.If,however,inresponsetoinves‐

toroutflows,mutualfundstradebasedoninformationrelatedtofirms’capitalstructures,

actualtradesmaynotbeavalidinstrumentforinstitutionalholdings.Toalleviatethiscon‐

cern,weuseimpliedtradesconstructedfrominvestorflowsandpreviouslydisclosedport‐

folioholdings.Impliedtradesarelikelytosatisfytheexclusionrestrictionbecausemutual

fundinvestors’decisionstoaccumulateordivestsharesinafundareunrelatedtothecapi‐

tal structure policies of the individual firms held by the fund. After all, if the investors

caredto,theycouldtradethestocksoftheindividualfirmsratherthantrademutualfund

shares. Thus, impliedtradesconstructedusingmutual fundflowswill leadtochanges in

institutionalholdingsthatmayaffectcapitalstructure,butthatarenotmotivatedbyit.

Onepotentialconcernwithusingfundflowinducedtradingasaninstrumentforin‐

stitutionalholdingsisthatthepricepressureresultingfromexcessivefundflowsmaylead

firmstotimethemarketwithdebtandequityissuances,thuscreatingaspuriousrelation‐

shipbetweeninstitutionalholdingsandleverage.GaoandLou(2011)investigatethisissue

andfindsomesupport formarket timingconsiderations,predominately in firmsmore fi‐

nanciallyconstrained.Themagnitudeoftheeffecttheyfind,however,isverysmall.Specif‐

ically,GaoandLoufindthataonestandarddeviationincreaseinflowinducedpricepres‐

sureisassociatedwithadecreaseinthesubsequentdebttoequityratioofapproximately

5.6basispoints. We,ontheotherhand,findinTableIIthataonestandarddeviationin‐

crease in institutionalholdings isassociatedwitha228basispointdecrease in leverage,

andifweruntheregressionusingthedebttoequityratioinplaceofleverage,wefindthe

negativeassociationtobegreaterthan1,100basispoints.Withtheeconomiceffectofin‐

stitutionalholdingsoncapital structure tobenearly200 times larger than theeconomic

11Edmansetal.(2011)studytheimpactofstockpricesontakeoverprobabilities,andnotethatpricemaybeendogenous,necessitatinganinstrumentalvariablesapproach.

26  

effectofflowinducedpricepressure,theconcernthatmarkettimingconsiderationsgener‐

ateaspuriousrelationshipbetweeninstitutionalholdingsandleverageappearstobemi‐

nor.

Theinstrumentweuse,MFFlow,isthefirm‐specificannualdollarchangeinholdings

impliedbymutualfundinvestoroutflowsandpreviouslydisclosedmutualfundholdings.

Specifically,forfirmiinquartert,

MFFlowi,t, x , , x ,

, x ,, (2)

whereFj,tisthetotaloutflowexperiencedbyfundjinquartert,TAj,t‐1isfundj’stotalassets

attheendofthepreviousquarter,SHARESi,j,txPRCi,t‐1isthedollarvalueoffundj’sholdings

ofstocki,andVOLi,tisthetotaldollartradingvolumeofstockiinquartert.Thesummation

isover funds j forwhichquarterly investoroutflowsequalorexceed5%of fund j’s total

assets(i.e.‐Fj,t/TAj,t‐1≥5%).Edmansetal.(2011)arguethattheimpactofoutflowonpric‐

eswilldependonastock’sliquidity,andassuchscaleMFFlowbytradingvolume.Although

thisconcernislessrelevantforourpaper,weproceedinkind.Finally,tocompletethecal‐

culation,wesumMFFlowacrossthefourquartersinagivencalendaryear.

TableVIpresentstheeffectofinstitutionalholdingsonleverage,estimatedviatwo‐

stageleastsquares.12Wefindthatafterremovingthefeedbackeffectofleverageoninsti‐

tutional holdings, the effect of holdings on leverage increases substantially. Our results

showthataonestandarddeviationincreaseininstitutionalholdingsleadstoastatistically

andeconomically significantdecrease in leverageof8.0%, as compared to2.9%without

instrumentation(column1,TableII).TheresultissupportedbyafirststageF‐statisticof

358.21,whichhiseasilylargeenoughtorejectthehypothesisthattheinstrumentisweak

(StockandYogo(2005)).

12Duetodataconstraints,thesampleinTableVIdiffersslightlyfromthatinTableII.MFFlow,obtainedfromAlexEdmans’swebsite,iscalculatedannuallybetween1980and2007,inclusive,andassuch,2008and2009areexcludedfromthetestsinTableVII.Repeatingcolumn1inTableIIIusingthissample,however,yieldsaverysimilarcoefficientforinstitutionalholdingsof‐0.080.

27  

C.3.S&P500IndexInclusion

Asourfinalmethodofcontrollingforthepotentialfeedbackeffectofleverageonin‐

stitutional holdings, we conduct a semi‐natural experiment that utilizes the exogenous

shocktoinstitutionalholdingsassociatedwiththeinclusionofafirmintotheS&P500In‐

dex(PruittandWei(1989)).Aghion,VanReenen,andZingales(2008),whoalsouseaddi‐

tion to the S&P 500 Index as an instrument for institutional holdings, mention three

sourcesforthisincrease:indexfunds,non‐indexersthatbenchmarkagainsttheIndex,and

institutionsappealingtotheimpliedendorsementofbroadindexingtosatisfyprudentman

restrictions.

S&P500inclusionisavalidinstrumentforinstitutionalholdingsbecauseitiscon‐

sideredbyStandardandPoor’stobeaninformation‐freeevent.Whenachangeismadeto

anS&Pequityindex,Standard&Poor’sstatesinitspressreleasethat“companyadditions

toanddeletionsfromanS&Pequityindexdonotinanywayreflectanopiniononthein‐

vestmentmeritsofthecompany.”Furthermore,Neubert(1985)describesthecriteriafor

S&P 500 selection to be: “size, industry classification, capitalization, trading vol‐

ume/turnover, emerging companies/industries, and responsivenessof themovementsof

stockpricetochangesinindustryaffairs,”allofwhicharepublicinformationalreadyfac‐

toredintofirmvaluationsandmetrics.

Aswithmutualfundflowinducedtrading,markettimingconsiderationscanpoten‐

tiallyrenderS&P500inclusionaninvalidinstrument.Thatis,asStandard&Poor’saimsto

constructanindexthatisbothrepresentativeandstable,itoftenchooseslargefirmswith

goodpastperformance.Giventhis,firmsthatareaddedtotheindexoftenexperiencelarge

stockreturnsinthequartersprecedinginclusion.Itcouldbethecase,then,thatfirmstake

advantageofthelargereturnsandissueequitywhilestockpricesarehigh,inturnlowering

leverageandcreatingaspuriousnegativerelationshipbetweeninstitutionalholdingsand

leverage.Infact,AltiandSulaeman(2012)findthatfirmstimethemarkettoissueequity

onlywhentheyhavebotha favorablestockpriceandhighinstitutionaldemandfortheir

shares.

Whileweareunabletocompletelyeliminatethispossibility,weemployapropensi‐

tyscorematchingtechnique(RosenbaumandRubin(1983,1985),andHeckman,Ichimura,

andTodd(1997))inwhichwematcheachfirmthatenterstheS&P500toacontrolfirm

28  

withtheclosestpropensityscorebasedonthecumulativeCAPM‐adjustedstockreturnfor

thequarterbeforeinclusionandthequarterofaddition.Wealsomatchonsize,market‐to‐

book,leverage,institutionalholdings,assetgrowth,andliquidity,andweforcematchesto

beinthesameindustryatthesametimeperiod(year‐quarter).

For our S&P500natural experiment,weuse quarterlydata fromCompustat, and

count 526 first‐time additions to the Index between 1979 and 2009 (excluding financial

firmsandutilities).Aswerequirenon‐missingleverageforfourquartersafteraddition,we

lose75observationsofthe526(including24fromfirmsthatwereaddedin2009–thefi‐

nalyear inoursample). Furthermore, sincewematchon laggedreturn,werequire two

quartersofnon‐missingpricespriortoaddition,decreasingoursamplebythe8firmsthat

wereaddedinthefirsttwoquartersof1979–thefirstyearinoursample.Finally,welose

28observationsdue tomissingdata required forourmatchingprocedure (inparticular,

thecomponentsofAmihud’s illiquiditymeasure), leavinguswith415S&P500additions

betweenthethirdquarterof1979and2008.

TobeeligibleforthecontrolgroupwerequirethatafirmhasneverbeenintheS&P

500Index,that ithasbeendroppedfromtheIndexat leastoneyearpriortobecominga

match,orthatithasbeenintheIndexforatleastoneyearbeforebecomingamatch.Ex‐

cluding firms in the S&P 500 from the control group would make finding appropriate

matchesmoredifficult,butitwouldalsomakeiteasiertogenerateresultsinadifference‐

in‐difference framework. Because finding qualitymatches is important,we include S&P

500firmsinthecontrolgroupandindoingsosubjectourselvestoadifficulttestingenvi‐

ronment.Thismatchingprocedureyieldsacontrolgroupinwhich209firmsaremembers

oftheS&P500Index,and206firmsarenot.However,inanalternateprocedure,weforce

allmatchestobemembersoftheS&P500IndextocontrolforapossibleS&P500fixedef‐

fect,andourresultshold. Withour treatmentandcontrol samplesnowconstructed,we

conductbothadifference‐in‐differencepairedt‐testandadifference‐in‐differenceregres‐

siontest.TheresultsarepresentedinTableVII.

PanelAofTableVIIpresentssummarystatisticsonequarterbeforeS&P500addi‐

tionforboththetreatmentandcontrolgroups.Thegroupsareverysimilar.Weseethat

thetreatmentgroupisslightlylargerintermsofmarketcapitalizationandtotalassets,alt‐

houghthegroupshaveequalmarket‐to‐bookratios.Furthermore,thetreatmentgrouphas

29  

marginallyhigher institutionalholdingsandleveragebeforeadditiontotheIndex. While

notreported,wefindthatbetweenthequarterbeforeadditionandthequarterofaddition,

institutionalholdingsincreaseby3.4%,onaverage,forthetreatedfirms,andonly0.6%,on

average,forthecontrolfirms.Thedifferencebetweenthesechangesissignificantlydiffer‐

entfromzero. Importantly,thecumulativeabnormalreturninthequarterpriortoaddi‐

tionandthequarterofadditionisactuallyhigherforthecontrolgroupthanforthetreat‐

mentgroup,helpingtoassuageconcernsthattheleveragedecreasewefindstemsfromthe

markettimingconsiderationsstudiedinAltiandSulaeman(2012).

In Panel B we present the results of a difference‐in‐difference paired t‐test. We

comparethemeandifferenceinleveragebetweenthetreatmentandcontrolgroupsinthe

quarterpriortoadditiontothemeandifferenceoneyearafteraddition,whichshouldpro‐

videfirmswithenoughtimetoaltertheircapitalstructurepolicies inresponsetothein‐

creaseininstitutionalholdingscausedbyS&P500addition.Wefindthatthemeandiffer‐

ence between treatment and control groups shrinks by (a statistically significant) 0.015,

from 0.017 before addition to 0.002 after addition. The mean difference produces a t‐

statisticof ‐2.061, indicatinga change in leveragestemming fromthe increase in institu‐

tionalholdingsthatresultsfromindexinclusion.13

Finally,inPanelCwepresenttheresultsofourregressionanalysis.Forthistestwe

create two new dummy variables, one indicating whether the firm joins the S&P 500

(Treatment=1inthiscase),andtheotherindicatingwhetherthetimeperiodisbeforeor

afterinclusion(Post=1afterinclusion,forboththetreatmentfirmsandthematchedsam‐

ple). The difference‐in‐difference estimator of interest is the interaction of these two

dummyvariables,Treatment*Post. Inorder to isolate theeffectofS&P500addition,we

restrictoursample to theperiodoneyearbefore inclusiontooneyearafter inclusion(9

quartersperfirm)forboththetreatmentandcontrolgroups.Intheregressionwecontrol

forthedeterminantsof leverageusedinTableII,aswellaslaggedandcontemporaneous

adjustedreturnand(Amihud’s)illiquiditytomitigateconcernsrelatedtomarkettiming. 13Tofurtherensurethatthedifferentialchangeinleveragebetweenthetreatmentandcontrolgroupisin‐deedduetotheinclusionintotheS&P500,weusethefollowingplacebotest:WelookbacktwoyearsbeforeadditiontotheS&P500,andrepeatourdifference‐in‐differencepairedt‐testfromPanelB(usingthesamefirmsinthetreatmentandcontrolgroups).Wefindthemeandifferenceinleverage(andinstitutionalhold‐ings)betweenthetreatmentandcontrolgroupsbeforeandafterthisparticularquartertobeindistinguisha‐blefromzero.

30  

WefindtheeffectofTreatment*Postonleveragetobenegativeandsignificant,with

acoefficientof‐0.015andt‐statisticof‐2.00.TheresultcomplementsourfindingsinPanel

B, andpoints toa significantdifference in the change in leveragebetween the treatment

andcontrolgroupsafter index inclusion,evenaftercontrolling forotherknowndetermi‐

nantsofleverageandmarkettimingforces.The1.5%changeinleverageiseconomically,as

wellasstatisticallysignificant,as it represents7.7%of theaverage treatment firms’pre‐

event leverage (19.4%), and is already beingmeasured above‐and‐beyond the “natural”

changeinleverageinthecontrolfirms.

Takentogether,wehavepresentedthreedifferenttests:Arellano‐Bond,2SLS,anda

semi‐naturalexperimentusinginclusiontotheS&P500Index.Thosetestsdifferdramati‐

callyintheirsamplesize,andthenatureoftheteststhemselvesdiffersacrossexperiments.

Nonetheless,allthethreetestssupportthesameconclusion:institutionalholdingshavea

causalnegativeinfluenceoncapitalstructure.Inaddition,ourdynamicpanelresultssug‐

gestthatinstitutionsdonotunequivocallychangetheirholdingsinresponsetochangesin

leverage.Weinterpretthisasevidencethatinstitutionscarryoutactivitiesthateitherdi‐

rectlyorindirectlymitigateagencyorinformationalfrictionsregardlessoftheexistingin‐

ternalmechanismsfordoingdo.Institutionsmonitor,forexample,fortheirownpurposes,

anddonotdecreasetheirinvestmentinafirmjustbecausethefirmincreasesitsleverage

andhencefurtherbondsmanagerialactivities.Itisuptofirms,then,todecreasecostlyin‐

ternalinvestorprotectionmechanismswhenlargeinstitutionalinvestorsprovideanalter‐

nativeexternalinvestorprotectionmechanism.

D.AgencyCosts,AsymmetricInformation,orTaxes?

Wepositedthatinstitutionalinvestorsmayhaveaneffectonfirms’capitalstructure

because theyareuniquelyqualified to interactwith frictions that, at least inpart, deter‐

minefirms’ leverageratios. Ourresultsthusfarsupportbothagencyandasymmetricin‐

formation models that predict changes in institutional holdings will results in opposite

changesinleverage.Thisevidence,however,isalsoconsistentwithtaxconsiderationsbe‐

causeoftherelativetaxadvantageenjoyedbyinstitutionalinvestorsoverindividualinves‐

torsonequityincome.Inthissection,wetrytoshedsomelightontherelativeimportance

31  

ofthesefrictions(asymmetricinformation,agency,andrelativetaxation)forthenegative

relationbetweeninstitutionalholdingsandleverage.

Tothisend,weidentifyfirmslikelytofacehighlevelsofagencycostsandfirmslike‐

ly tosuffer fromhighdegreesof informationasymmetry. We thencompare theeffectof

institutional holdings on leverage between the groups. Firms particularly prone to high

agencycostsare,ingeneral,large,profitablefirms,withfewgrowthopportunitiesrelative

to their cash holdings. On the other hand, firms commonly prone to information asym‐

metry problems are generally small, young, cash‐strapped firms with relatively many

growthopportunities.

Wesort firmsannually intomarket capquintilesandmarket‐to‐bookquintiles, as

themarket‐to‐bookratioproxiesforgrowthopportunities.Wethenclassify“Small”firms

asfirmsinmarketcapquintiles1and2,and“Large”firmsasfirmsinmarketcapquintiles

4and5.Similarly,weclassify“Low”growthopportunityfirmsasfirmsinmarket‐to‐book

quintiles1and2,and“High”growthopportunityfirmsasfirmsinmarket‐to‐bookquintiles

4and5.Usingtheseclassifications,wedefineagencyfirmsasfirmsinboththe“Large”and

“Low”categories,andasymmetricinformationfirmsasfirmsinboththe“Small”and“High”

categories.Separatingfirmsinthismannerallowsustotestwhethertherelationshipbe‐

tweeninstitutionalholdingsandleverageisdrivenbyonefrictioninparticular.

PanelAofTableVIIIreportstheresultsofthistest.Whilewefindinstitutionalhold‐

ingstobeanegativeandsignificantdeterminantofleverageforbothtypesoffirms,there‐

lation isconsiderablystronger forasymmetric information firms(column1). Thecoeffi‐

cientsof‐0.147and‐0.050forasymmetricinformationandagencyfirms,respectively,rep‐

resentchangesof‐2.1%and‐1.4%relativetoaonestandarddeviationincreaseinlagged

institutionalholdings. Togivethosenumberssomeperspective,2.1%represents9.0%of

average leverage forasymmetric information firms(23.3%),while1.2%represents5.1%

ofaverageleverageforagencyfirms(27.4%).

Thespecificationsincolumns1and2includeafirmfixedeffect,andtheresultsare

robusttoalternatedefinitionsofLarge‐LowandSmall‐High,aswellastoalternatecatego‐

rizationsinwhichwereplaceSmall/Largewitha“cashcow”indicator(notreportedinthe

table). FollowingBravetal. (2005),wedefinecashcow firmsasprofitable firmswitha

32  

strongcreditrating(Aorbetter)andaP/EratiolowerthanthemedianP/Eratioamong

profitablefirmswithastrongcreditrating.

Next,weextend this analysisbyestimating theeffectofholdingson leverage ina

dynamicpanelseparatelyforagencyandasymmetricinformationfirms.Theresultsincol‐

umns3and4supportandextendthoseincolumns1and2. Thatis, inadynamicpanel,

institutionalholdingsonlyaffectfirmspronetoasymmetricinformationproblems(column

3),astheeffectofinstitutionalholdingsonleverageisnegativebutinsignificantforfirms

prone to agency costs (column4). For robustness,wealso run the specifications in col‐

umns1and2 in first‐differences,andalternativelycombinethetests intooneregression

with a dummy variable designating agency or asymmetric information firms, and obtain

complementaryresults(notreportedinthetable).

TheresultsinPanelAsuggestthatinformationasymmetryisthedominantfriction

behindthenegativerelationshipbetweeninstitutionalholdingsandleverage.Thetaxad‐

vantage of institutional investors, however, implies a relationship in the same direction.

Consequently, we employ a semi‐natural experiment to address whether tax considera‐

tionsplayaroleintherelationshipbetweeninstitutionalholdingsandleverage.Specifical‐

ly,wetestwhetherthetaxadvantageoverequity incomefor institutional investorscom‐

pared to individual investors drives the negative relation between institutional holdings

andleverage.TheJobsandGrowthTaxReliefReconciliationActof2003dramaticallylow‐

eredthetaxrateonequityincomeforindividualinvestors,anddiminishedtherelativetax

advantageenjoyedbyinstitutionalinvestors.Iftaxconsiderationsarethedrivingforcebe‐

hindourresultsthusfar,weshouldseeaweakereffectofinstitutionalholdingsonleverage

after2003thanbefore.

Totestthishypothesiswerestrictoursampletotheyears2000to2006,excluding

2003,andconstructtwonewinstitutionalholdingsvariables,Hold_00‐02andHold_04‐06,

to test the relationship between institutional holdings and leverage pre‐ and post‐tax

break.TheresultsofthistestarereportedinPanelB.Bothcoefficientsarenegativeand

significant,‐0.077forthepre‐2003periodand‐0.086forthepost‐2003period,andanF‐

testwithap‐valueof0.401revealsthattheyareinsignificantlydifferentfromoneanother.

TheresultsinPanelBsuggestthattaxconsiderationsarenotbehindthenegativerelation‐

ship between institutional holdings and leverage. In unreported analysis,we re‐run the

33  

testinPanelBusingfirstdifferencesasopposedtolevels,andweconfirmthatwecannot

rejecttheequalityofthecoefficientsbeforeandafterthe2003taxbreakswereintroduced.

Furthermore,theresultisrobusttotheinclusionofleasesintothedefinitionofleverage.

E.Howdofirmschangetheirleverage?

Wehaveestablishedthatchangesininstitutionalownershipnegativelyaffectlever‐

ageratios. Next,we investigatethenatureof thisprocess. That is,howdofirmschange

theircapitalstructureinresponseto,say,anincreaseininstitutionalownership?Isitdone

byissuingmoreequity,repurchasingfewershares,orbydecreasingnetdebt?Weexamine

therelationintermsoftheprobabilityofanissuanceaswellasthesizeofissuances.

Wedefineequity issuanceanddebt issuance inaccordancewithFamaandFrench

(2005)andLemmonetal.(2008),amongothers.Equityissuance(%ChangeEquity)isthe

productof thesplit‐adjustedchange insharesoutstandingandthesplit‐adjustedaverage

stockprice,normalizedbytotalassetsattime(t–1).Consequentially,afirmisanequity

issuer(EquityIssue=1)if%ChangeEquity>1%,andanequityrepurchaser(EquityRepo=

1)if%ChangeEquity<‐1%.Similarly,debtissuance(%ChangeDebt)isthechangeintotal

debt (long‐term+short‐term)normalizedby totalassetsat time (t–1). Finally, thenet

changeincapitalstructure(%ChangeNetDebt)issimply%ChangeDebt‐%ChangeEqui‐

ty,andafirmisanetdebtissuer(NetDebtIssue=1)if%ChangeNetDebt>0.

Totesttheeffectofinstitutionalholdingsonissuancedecisions,werunlogisticre‐

gressionsseparatelyforequityissuances,equityrepurchases,andnetdebtissuances,with

lagged changes in institutional holdings as our primary independent variable. In these

testsweincludethecontrolvariables,infirstdifferences,fromTableIV,aswellaslagged

annualadjustedreturnandbetatocapturepossiblemarkettimingeffects. Theresultsof

ourlogisticregressionsarepresentedinthefirstthreecolumnsofTableIX.

Wefindthatanincreaseininstitutionalholdingsisassociatedwithahigherproba‐

bilityofafirmbeinganequityissuerinthefollowingyear(column1),andalowerproba‐

bilityofafirmbeinganequityrepurchaserinthefollowingyear(column2).Bothresults

are statistically significant at the 1% level after clustering standard errors by firm. The

logitcoefficientof1.409incolumn1impliesthataonestandarddeviationincreaseinthe

change in institutionalholdings isassociatedwitha14.8%increase intheoddsofa firm

34  

beinganequity issuer in the followingyear. In termsofprobabilities,nearly39%of the

firm‐yearobservationsincolumn1haveequityissuances.Thus,increasingtheoddsofeq‐

uityissuanceby14.8%translatestoa3.3%increaseintheprobabilityoftheaveragefirm

inoursamplebeinganetequityissuer.Similarly,thelogitcoefficientof‐0.368incolumn2

impliesthataonestandarddeviationincreaseinthechangeininstitutionalholdingleads

toa0.4%decreaseintheprobabilityoftheaveragefirminoursamplebeinganequityre‐

purchaserduring the followingyear. Our results complement thoseofChang,Chen, and

Dasgupta(2011),whoreportthatthelikelihoodofequityissuancesincreaseswiththeper‐

centageofsharesheldbyshort‐terminstitutionalinvestors.

Column3presentstheeffectofchangesininstitutionalholdingsontheprobability

ofafirmchangingitscapitalstructurebyissuingmoredebtthanequityinagivenyear(Net

Debt Issue =1). 40%of the firm‐years inour sample experienceanet change in capital

structure in thisdirection. Incolumn3wefindthatan increase in institutionalholdings

leadstoalowerprobabilityofafirmissuingmoredebtthanequity.Specifically,thecoeffi‐

cientof‐0.353suggeststhataonestandarddeviationincreaseinthechangeofinstitutional

holdingsleadstoa0.8%decreaseintheprobabilityoftheaveragefirmchangingitscapital

structurebyissuingdebtaboveandbeyonditsequityissuances.

For robustness,we redefine equity issuances using a 5% cutoff rather than a 1%

cutoff in order to exclude minor fluctuations due to stock option plans, conversion of

bonds,etc.Althoughtheseresultsarenotreported,wefindanevenlargereffectofinstitu‐

tional holdings on equity issuances and repurchases, and the effect remains statistically

significant.Inaddition,were‐estimatetheregressionsincolumns1through3usingtop‐5

institutionalholdingsrather thanaggregateholdings,andourresultsareunchanged. Fi‐

nally,we estimate an alternate specification inwhichwe define equity issuance using a

deal‐baseddummyvariableconstructedusingissuancedatafromSDC,andourresultsare

qualitativelyunchanged.

Next,weexaminetheeffectofchangesininstitutionalholdingsonthesizeofequity

andnetdebtissuances(columns4and5,respectively).Thereisnocolumnspecificallyfor

repurchasesbecausethedependentvariablesinthispanelaresimplynetchanges,andas

suchcanbepositiveornegative.)Column4complimentsourfindingsincolumn1,aswe

findthatchangesininstitutionalholdingsarerelatednotonlytotheprobabilityofanequi‐

35  

tyissuance,butalsotothesizeofequityissuancesasapercentageoftotalassets.Thecoef‐

ficientof0.061issignificantatthe1%level.

Whilewetesttheeffectofchangesininstitutionalholdingsonchangesinleverage

inTableII(andfindanegativerelation),thetestsinTableIXaredifferent.Here,wefixthe

denominator(totalassets)andlookexplicitlyatchangesinequityandnetdebtoutstanding

ratherthanatthechangeintheleverageratio.Whereasbeforeourtestsallowedustosee

whetherornotfirmschangetheirleverageinresponsetochangesininstitutionalholdings,

thetestsinTableIXallowustoseehowfirmschangetheirleverage.Thus,incolumn5we

estimatetheimpactofachangeininstitutionalholdingsonthenetchangeincapitalstruc‐

ture(%ChangeDebt‐%ChangeEquity).Wefindinstitutionalholdingstobestronglyneg‐

atively related to ourmeasure of net debt issuance, indicating thatwhen firmsdecrease

theirleverageinresponsetoanincreaseininstitutionalholdings,theydosoprimarilyby

issuingequityratherthanbydecreasingdebt.

Wealsofindthatincreasesanddecreasesininstitutionalholdingshavesymmetric

effectsonequityandnetdebtissuances.Forexample,ourresultsindicatethatfirmsboth

decrease theirnetdebt issuancesafteran increase in institutionalholdings,and increase

their net debt issuances after a decrease in institutional holdings. (These results are not

tabulated).

Insum,inthissectionweshowthatchangesininstitutionalholdingshaveapositive

and significant effect on theprobability and size of subsequent equity issuances, both in

gross termsandnetofdebt issuances. However, inunreportedresults,wealsoexamine

whether changes indebtorequity issuances inyear t affect institutionalholding inyear

t+1.Wefindnoevidencethatinstitutionalholdingsareaffectedbychangesinequity,debt,

ornetdebt,reinforcingthedirectionofcausalityfrominstitutionstofirmsasfarascapital

structureisconcerned.

IV.Conclusion

In thispaperweanalyzewhether institutionalholdings influencecapitalstructure

decisions,andwhetherfirms’financialleverageaffectsinstitutionalinvestors’decisionsto

hold theirequity. Theoriesof capital structure imply that firmschoose their leverage in

responsetomarketfrictionssuchasagencycostsandasymmetricinformation.Meanwhile,

36  

institutional monitoring and information‐gathering affect firms’ agency costs and infor‐

mation environment, opening channels through which institutions may influence firms’

choicesof leverage. Intheagencyframework,institutionalinvestorsserveasanexternal

disciplinary mechanism for management, lessening the need for internal disciplinary

mechanismssuchasdebt. Intheasymmetric informationframework, institutional inves‐

tors decrease information asymmetry and reduce the adverse selection costs of equity,

servingtodecreasetheamountofdebtneededforsignalingequilibriums,andtolowerthe

costofequityrelativetodebt.

Wefindanegativerelationshipbetweenleverageandinstitutionalholdings.Thatis,

firmswithalargepercentageoftheirsharesheldbyinstitutions,onaverage,haverelative‐

ly low leverage ratios. Wealso find that changes in institutionalholdingsareassociated

withchangesinleverageintheoppositedirection(andviceversa).Toestablishcausality,

weusethreetechniquestoadjustfortheendogenouscomponentsoftherelationshipbe‐

tweeninstitutionalholdingsandleverage.InanArellano‐Bonddynamicpanelestimation

thatutilizesinternalinstruments(laggedlevels)forbothchangesinleverageandchanges

inholdings,wefindthatanincreaseininstitutionalholdingsleadstoadecreaseinlever‐

age,butthatachangeinleveragedoesnotleadtoachangeinholdings.Inacomplemen‐

tarytestusingtwo‐stageleastsquaresandimpliedtradesinducedbymutualfundoutflows

asaninstrumentforinstitutionalholdings(Edmansetal.(2011)),weisolatetheeffectof

institutionalholdingsonleverageratiosandagainfindasignificantlynegativerelationship.

Finally,weconfirmthenegativeeffectofholdingsonleverageusingthepositiveshockto

institutional holdings that results from S&P 500 Index inclusion as treatment in a semi‐

naturalexperiment.Whilethetechniquesaredifferent,andthesamplesizesvarydramati‐

callyacrossexperiments,theresultsconsistentlysuggestthatinstitutionalholdingsnega‐

tivelyaffectfirms’financialleverage,butthatfirms’leveragedoesnotunequivocallyaffect

institutionalholdings.

Withteststhatallowustocharacterizehowfirmslowertheirleverageinresponse

tochangesininstitutionalinvestment,wefindthatasinstitutionalholdingsincrease,firms

becomemore likely to issue equity,while the probability of debt issuance is unchanged.

Furthermore,wefindthatthesizeofequityissuancesarerelatedtochangesininstitution‐

alholdings,both ingross termsandnetofdebt issuances,suggesting that inresponseto

37  

changesininstitutionalholdings,firmsaltertheircapitalstructurepredominantlythrough

equityissuances.

Ourresultsarestronglysupportiveofcapitalstructuretheoriesthatpredictasub‐

stitutiverelationshipbetweenleverageandinstitutionalholdings. Usingteststhatdiffer‐

entiatebetween firmsprone toagencycostsand firmsprone toasymmetric information

problems,weprovideevidence that institutionsmoststrongly influencecapital structure

through their effect on information asymmetry. Furthermore, the preferential tax treat‐

mentofinstitutionalinvestorsasequityholdersalsosuggestsanincreaseinequityrelative

todebtasinstitutionalholdingsincrease.However,wefindnoevidencethattaxconsidera‐

tionsplayaroleinthisrelationship.

Finally,ourresultssuggestthatsignificantlydifferentmotivesunderlietheinterac‐

tionbetweeninstitutionsanddividends(GrinsteinandMichaely(2005))andthe interac‐

tionbetweeninstitutionsandleverage(thispaper). GrinsteinandMichaely(2005)show

thatinstitutionsdonotaffectfirms’payoutpolicies,butthatchangesinpayoutpoliciesaf‐

fectinstitutionalholdings.Moreover,theyshowthattheinteractionismostpronouncedin

largefirmsandfirmsthatarecashcows,consistentwiththemonitoringroleofinstitutions.

We,ontheotherhand,findthatinstitutionsdoaffectleverage,butthatchangesinleverage

donotunambiguouslyaffectinstitutionalholdings.Furthermore,wefindthattheinterac‐

tionismostpronouncedinsmall,growthfirms,inlinewiththenotionthatinstitutionalin‐

vestorsreduceasymmetric information, inturninducingfirmstoreduceleverage. These

contrastingfindingsmay,perhaps,pointnotonlytodifferentinteractionsbetweeninstitu‐

tionalholdingsandvariousfinancialpolicies,butalsotodifferingrolesplayedbydividends

and leverage inresolvingasymmetric informationandagencyconflictsbetweenmanage‐

mentandoutsideequityholders.

38  

Appendix

A.VariableDefinitions

Leverage(TDA)istheratiooftotaldebt(DLTT+DLC)toassets(AT)

Leverageincludingleases(TDLA)istheratiototaldebt(DLTT+DLC)plusthepresentvalue

of off balance sheet (OBS) non‐cancelable operating leases (Oplease) to assets

(AT+Oplease).Opleaseiscalculatedas:

RentExp0 1

5

1

,

whereRentExp0 is thecurrentyearrentalexpense(XRENT),MLPt is theminimum

leasepaymentsforyearst=1,…,5(MRC1,MRC2,MRC3,MRC4,MRC5),andKdisthe

costofdebtcapital(setto10%followingGraham,Lemmon,andSchallhem(1998)).

Institutional holdings (Hold) is the ratio of total shares held by institutions to the total

numberofsharesoutstanding(CSHO)

Institutionalholdings,top‐5(Hold5)istheratioofthetotalsharesheldbytheinstitutional

investorswith the five largestpositions to the totalnumberofsharesoutstanding

(CSHO)

Logofassets(Assets)isthenaturallogarithmofassets(AT)

Changeinlogassets(ChgAssets)istheannualchangeinlogassets(AT)

Marketcap (MktCap) is theprice (PRCC_F) times the totalnumberof sharesoutstanding

(CSHO)

Logofsales(Sales)isthenaturallogarithmofsales(SALE)

Profitability(Profit)istheratioofoperatingincomebeforedepreciation(OIBDP)toassets

(AT)

Market‐to‐bookratio (MktBk) is the ratioofmarketvalueof assets (MVA) tobookassets

(AT). MVA isthesumofthemarketvalueofequity(PRCC_F*CSHPRI)plusdebtin

currentliabilities(DLC)pluslong‐termdebt(DLTT)pluspreferred‐liquidationval‐

ue(PSTKL)minusdeferredtaxesandinvestmenttaxcredit(TXDITC).

Capitalexpenditure(Capex)istheratioofcapitalexpenditure(CAPX)toassets(AT)

Tangibility(Tang)istheratioofnetproperty,plant,andequipment(PPENT)toassets(AT)

Intangibleassets(Intang)istheratioofintangibles(INTAN)toassets(AT)

39  

Collateral(Colltrl)istheratioofinventory(INVT)plusnetproperty,plant,andequipment

(PPENT)toassets(AT)

Dividendpayingdummy(Dividend)isadummyvariabletakingthevalueofoneifcommon

dividends(DVC)ispositive,andzerootherwise

Payout (Payout) is the ratio of commondividends (DVC)plus purchases of common and

preferredstock(PRSTKC)toassets(AT)

Lossmakingdummy(Losses)isadummyvariabletakingthevalueofoneifProfit isnega‐

tive,andzerootherwise

Uniquenessdummy(Unique)isadummyvariabletakingthevalueofoneiftheSICcodeof

thefirmisbetween3400and4000,andzerootherwise

Cashcow(CashCow)isadummyvariabletakingthevalueofoneifafirmisprofitable(Prof‐

it>0),hasa credit ratingofAorbetter (SPLTICRMbetween2and8), andaP/E

lowerthanthemedianP/EamongprofitablefirmswithacreditratingofAorhigh‐

er,andzerootherwise

Medianindustryleverage(Industry_TDA)isthemedianratiooftotaldebttoassetsbyindus‐

trybyyear,withthe49industriesdefinedasperFamaandFrench

Initialfirmleverage(Initial_TDA)isthefirstnon‐missingvalueofLeverage

Insiderholdings(Hold_Insider) is thetotalpercentageofsharesownedby insiders,asper

Execucomp

Amihud’sIlliquidityisthemonthlyratioofabsolutestockreturntoitsdollarvolume,aver‐

agedoverthepriortwelvemonths

“EFWA”Market‐to‐BookisdefinedasperBakerandWurgler(2002)as:

∑∗ ,

wheresummationsbeginateithertheIPOyearorthefirstyearofCompustatdata,

andeandddenotenetequity(SSTK‐PRSTKC)andnetdebt(DLTIS‐DLTR+DLCCH)

issues

Equity issuance (%ChangeEquity) is the ratioof thesplit‐adjustedchange in sharesout‐

standing (CSHOt – CSHOt‐1*(AJEXt‐1/AJEXt)) times the split‐adjusted average stock

price(PRCC_Ft+PRCC_Ft‐1*(AJEXt/AJEXt‐1))toendofyeart‐1assets(AT)

40  

Equityissuer(EquityIssue)isadummyvariabletakingthevalueofoneif%ChangeEquity>

1.0%,andzerootherwise

EquityRepurchaser(EquityRepo)isadummyvariabletakingthevalueofoneif%Change

Equity<‐1.0%,andzerootherwise

Debtissuance(%ChangeDebt)isratioofthechangeintotaldebt(DLTT+DLC)fromyeart‐

1toyearttoendofyeart‐1assets(AT)

Netdebtissuance(%ChangeNetDebt)isthe%ChangeDebt‐%ChangeEquity

Netdebtissuer(NetDebtIssue)isadummyvariabletakingthevalueofoneif%ChangeNet

Debt>0,andzerootherwise

Beta(Beta)istheannualCAPMbetaofeachfirm,asreportedinCRSP

Pastabnormalreturn(AdjRet)iseachfirm’sannualreturnadjustedbytheCAPMreturnus‐

ingthefirm’sbeta,the10‐yearTreasurybondyield,andtherealizedreturnonthe

S&P500index

41  

REFERENCESAdmati,AnatR.,andPaulPfleiderer,2009,The‘WallStreetWalk’andShareholderActivism:Exitas

aFormofVoice,ReviewofFinancialStudies22,2445‐2485.Aghion,Philippe, JohnVanReenen,andLuigiZingales,2009, InnovationandInstitutionalOwner‐

ship,NBERWorkingPaperSeries14769.Allen,Franklin,AntonioE.Bernardo,andIvoWelch,2000,ATheoryofDividendsBasedonTaxCli‐

enteles,JournalofFinance55,2499‐2536.Alti, Aydogan, and Johan Sulaeman, 2012, When Do High Stock Returns Trigger Equity Issues?,

JournalofFinancialEconomics103,61–87.Amihud,Yakov,2002,Illiquidityandstockreturns:cross‐sectionandtime‐serieseffects,Journalof

FinancialMarkets5,31‐56.Arellano,Manuel,andStephenBond,1991,SomeTestsofSpecificationforPanelData:MonteCarlo

EvidenceandanApplicationtoEmploymentEquations,ReviewofEconomicStudies58,277‐297.

Baker,Malcolm,andJeffreyWurgler,2002,MarketTimingandCapitalStructure,JournalofFinance57,1‐32.

Bennett, JamesA.,RichardW.Sias,andLauraT.Starks,2003,GreenerPasturesandtheImpactofDynamicInstitutionalPreferences,ReviewofFinancialStudies16,1203‐1238.

Bhargava,Alok,L.FranziniandWijiNarendranathan,1982,SerialCorrelationandtheFixedEffectsModel,ReviewofEconomicStudies49,533‐549.

Brav,Alon,JohnR.Graham,CampbellR.Harvey,andRoniMichaely,2005,Payoutpolicyinthe21stcentury,JournalofFinancialEconomics77,483‐527.

Bushee,Brian J., andTheodoreH.Goodman, 2007,Which Institutional InvestorsTradeBasedonPrivateInformationAboutEarningsandReturns?,JournalofAccountingResearch45,289‐321.

Byoun,Soku,2008,HowandWhenDoFirmsAdjustTheirCapitalStructurestowardTargets?,Jour‐nalofFinance63,3069‐3096.

Chang, Xin, Yangyang Chen, and Sudipto Dasgupta, 2011, Institutional Investor Horizons, Infor‐mationEnvironment,andFinancing,WorkingPaper,http://ssrn.com/abstract=2022173.

Chang,Xin, SudiptoDasgupta, andGillesHilary,2006,AnalystCoverageandFinancingDecisions,JournalofFinance61,3009‐3048.

Chen,Xia, JarradHarford,andKaiLi,2007,Monitoring:Which InstitutionsMatter?, JournalofFi‐nancialEconomics86,279–305.

Cremers,K. J.Martijn,VinayB.Nair,andChenyangWei,2007,GovernanceMechanismsandBondPrices,ReviewofFinancialStudies20,1359–1388.

Cronqvist,Henrik,andRüdigerFahlenbrach,2009,LargeShareholdersandCorporatePolicies,Re‐viewofFinancialStudies22,3941–3976.

Datta,Sudip,MaiIskandar‐Datta,andKatrikRaman,ManagerialStockOwnershipandtheMaturityStructureofCorporateDebt,JournalofFinance60,2333‐2350.

DelGuercio,Diane.1996,Thedistortingeffectoftheprudent‐manlawsoninstitutionalequityin‐vestments,JournalofFinancialEconomics40,31‐62.

Desai,MihirA.,andLi Jin,2011,Institutionaltaxclientelesandpayoutpolicy, JournalofFinancialEconomics100,68‐84.

Diamond,DouglasW.,andZHe,2011,ATheoryofDebtMaturity:TheLongandShortofDebtOver‐hang,WorkingPaper,http://ssrn.com/paper=1539650.

Edmans,Alex,2009,BlockholderTrading,MarketEfficiency,andManagerialMyopia,JournalofFi‐nance64,2481–2513.

42  

Edmans,Alex,ItayGoldstein,andWeiJiang,2011,TheRealEffectsofFinancialMarkets:TheImpactofPricesonTakeovers(June6,2011).AFA2009SanFranciscoMeetingsPaper;JournalofFinance,Forthcoming.AvailableatSSRN:http://ssrn.com/abstract=1102201.

Eisfeldt,AndreaL.,andAdrianoA.Rampini,2009,Leasing,AbilitytoRepossess,andDebtCapacity,ReviewofFinancialStudies22,1621‐1657.

Falkenstein,EricG,1996,PreferencesforStockCharacteristicsasRevealedbyMutualFundPortfo‐lioHoldings,JournalofFinance51,111‐135.

Faulkender,Michael,Mark J. Flannery, KristineWatsonHankins, and JasonM. Smith, 2012, CashFlowsandLeverageAdjustments,JournalofFinancialEconomics103,632–646.

Fama,EugeneF.,andKennethR.French,2005,Financingdecisions:who issuesstock?, JournalofFinancialEconomics76,549‐582.

Fama,EugeneF.,andJamesD.MacBeth,1973,Risk,return,andequilibrium:Empiricaltests,JournalofPoliticalEconomy81,607‐636.

Fenn,Scott,andBradleyRobinson,2009,ProxyVotingbyExchange‐TradedFunds:AnAnalysisofETFVotingPolicies,Practices,andPatterns,PROXYGovernance Inc. for the IRRC InstituteforCorporateResponsibility.

Ferreira,MiguelA.,andPedroMatos,2008,Thecolorsofinvestors’money:Theroleofinstitutionalinvestorsaroundtheworld,JournalofFinancialEconomics88,499‐533.

Frank,MurrayZ.,andVidhanK.Goyal,2009,CapitalStructureDecisions:WhichFactorsAreRelia‐blyImportant?,FinancialManagement38,1‐37.

Franzen, Laurel,KimberlyRodgersCornaggia, andTimothyT. Simin, 2009, Capital Structure andtheChangingRoleofOff‐Balance‐SheetLeaseFinancing,WorkingPaper,http://ssrn.com/abstract=1452971.

Gao, Pengjie and Dong Lou, 2011, Cross‐Market Timing in Security Issuance, AFA 2012 ChicagoMeetingsPaper,AvailableatSSRN:http://ssrn.com/abstract=1787187

Gaspar,José‐Miguel,MassimoMassa,andPedroMatos,2005,Shareholderinvestmenthorizonsandthemarketforcorporatecontrol,JournalofFinancialEconomics76,135‐165.

Gillan,StuartL.,andLauraT.Starks,2000,Corporategovernanceproposalsandshareholderactiv‐ism:theroleofinstitutionalinvestors,JournalofFinancialEconomics57,275‐305.

———,2007,TheEvolutionofShareholderActivismintheUnitedStates,JournalofAppliedCorpo‐rateFinance19,55‐73.

Gomes,Armando,2000,GoingPublicwithoutGovernance:ManagerialReputationEffects,JournalofFinance55,615‐646.

Gompers, Paul A., Joy Ishii, and AndrewMetrick, 2003, Corporate governance and equity prices,QuarterlyJournalofEconomics118,107‐155.

Gompers,PaulA.,andAndrewMetrick,2001,InstitutionalInvestorsandEquityPrices,TheQuarter‐lyJournalofEconomics116,229‐259.

Graham, JohnR.,MichaelL.Lemmon,and JamesS.Schallheim,1998,Debt,Leases,Taxes,andtheEndogeneityofCorporateTaxStatus,JournalofFinance53,131‐162.

Grinstein,Yaniv,andRoniMichaely,2005, InstitutionalHoldingsandPayoutPolicy, JournalofFi‐nance60,1389‐1426.

Hartzell, Jay C., and Laura T. Starks, 2003, Institutional Investors and Executive Compensation,JournalofFinance58,2351‐2374.

Heckman,JamesJ.,HidehikoIchimura,andPetraE.Todd,1997,MatchingasanEconometricEvalu‐ationEstimator:EvidencefromEvaluatinga JobTrainingProgramme,ReviewofEconomicStudies64,605‐654.

Hennessy,ChristopherA,andToniM.Whited,2005,DebtDynamics,JournalofFinance60,1129–1165.

Holtz‐Eakin, Douglas, Whitney Newey, and Harvey S. Rosen, 1988, Estimating VectorAutoregressionswithPanelData,Econometrica56,1371‐1395.

43  

Huson,MarkR., RobertParrino, andLauraT. Starks, 2001, InternalMonitoringMechanisms andCEOTurnover:ALong‐TermPerspective,JournalofFinance56,2265‐2297.

Iliev,Peter,KarlLins,DariusMiller,andLukasRoth,2012,ShareholderVotingandCorporateGov‐ernanceAroundtheWorld,WorkingPaper,http://ssrn.com/abstract=1702546.

Jensen,MichaelC.1986,AgencyCostsofFreeCashFlow,CorporateFinance,andTakeovers,Ameri‐canEconomicReview76,323‐329.

———,1993,TheModernIndustrialRevolution,Exit,andtheFailureofInternalControlSystems,JournalofFinance48,831‐880.

Jensen,MichaelC.,andWilliamH.Meckling,1976,Theoryofthefirm:Managerialbehavior,agencycostsandownershipstructure,JournalofFinancialEconomics3,305‐360.

Jiang,Wei,KaiLi,andPeiShao,2010,WhenShareholdersAreCreditors:EffectsoftheSimultaneousHolding of Equity and Debt by Noncommercial‐Banking Institutions, Working Paper,http://ssrn.com/abstract=1619927.

Kleibergen,Frank,andRichardPaap,2006,Generalizedreducedranktestsusingthesingularvaluedecomposition,JournalofEconometrics133,97‐126.

LaPorta,Rafael,FlorencioLopez‐De‐Silanes,AndreiShleifer,andRobertW.Vishny,2000,AgencyProblemsandDividendPoliciesaroundtheWorld,JournalofFinance55,1‐33.

Lemmon,MichaelL.,MichaelR.Roberts,andJaimeF.Zender,2008,BacktotheBeginning:Persis‐tence and the Cross‐Section of Corporate Capital Structure, Journal of Finance 63, 1575‐1608.

Leary,MarkT.,andMichaelR.Roberts,2005,DoFirmsRebalanceTheirCapitalStructures?,JournalofFinance60,2575‐2619.

Ljungqvist,Alexander,FeliciaMarston,LauraT.Starks,KelseyD.Wei,andHongYan,2007,Conflictsofinterestinsell‐sideresearchandthemoderatingroleofinstitutionalinvestors,JournalofFinancialEconomics85,420‐456.

Levit,Doron,2012,SoftShareholderActivism,Workingpaper,universityofPennsylvania.Maug, Ernst, 1998, Large Shareholders asMonitors: Is There a Trade‐Off between Liquidity and

Control?,JournalofFinance53,65‐98.McCahery,JosephA.,ZachariasSautner,andLauraT.Starks,2010,BehindtheScenes:TheCorpo‐

rate Governance Preferences of Institutional Investors, Working Paper,http://ssrn.com/abstract=1571046.

Morgan,Angela,AnnettePoulsen,JackWolf,andTinaYang,2011,Mutual fundsasmonitors: evi‐dencefrommutualfundvoting,JournalofCorporateFinance17,914‐928.

Myers, Stewart C., 1977,Determinants of corporate borrowing, JournalofFinancialEconomics 5,147‐175.

Myers,StewartC.,andNicholasS.Majluf,1984,Corporatefinancingandinvestmentdecisionswhenfirmshaveinformationthatinvestorsdonothave,JournalofFinancialEconomics13,187‐221.

Neubert,AlbertS.,1985,TheInsandOutsoftheS&P500,MarketPerspectives3.Nickell,Stephen,1981,BiasesinDynamicModelswithFixedEffects,Econometrica49,1417‐1426.Ortiz‐Molina,Hernan,2006,TopManagementIncentivesandthePricingofCorporatePublicDebt,

JournalofFinancialandQuantitativeAnalysis41,317–340.Parrino, Robert, RichardW. Sias, and Laura T. Starks, 2003, Votingwith their feet: institutional

ownershipchangesaroundforcedCEOturnover,JournalofFinancialEconomics68,3‐46.Pruitt,StephenW.,andK.C.JohnWei,1989,InstitutionalOwnershipandChangesintheS&P500,

JournalofFinance44,509‐513.Rampini, AdrianoA., and S. Viswanathan, 2010, Collateral and Capital Structure,Working Paper,

http://ssrn.com/abstract=1356940.Rosenbaum,PaulR.,andDonaldR.Rubin,1983,TheCentralRoleofthePropensityScoreinObser‐

vationStudiesforCausalEffects,Biometrika70,41‐55.

44  

———, 1985, Constructing a Control GroupUsingMultivariateMatched SamplingMethods ThatIncorporatethePropensityScore,AmericanStatistician39,33‐38.

Ross, Stephen A., 1977, The Determination of Financial Structure: The Incentive‐Signalling Ap‐proach,BellJournalofEconomics8,23‐40.

Shleifer,Andrei,andRobertW.Vishny,1986,LargeShareholdersandCorporateControl,JournalofPoliticalEconomy94,461‐488.

Sias,RichardW.,2004,InstitutionalHerding,ReviewofFinancialStudies17,165‐206.Stock,James,andMotohiroYogo,2005,Asymptoticdistributionsofinstrumentalvariablesstatistics

withmanyweakinstruments,IdentificationandInferenceforEconometricModels:EssaysinHonor of Thomas Rothenberg, D.W.K. Andrews and J.H. Stock eds., Cambridge UniversityPress,Cambridge,U.K.

Strebulaev,IlyaA,2007,DoTestsofCapitalStructureTheoryMeanWhatTheySay?,JournalofFi‐nance62,1747‐1787.

Stulz,RenéM.,1990,Managerialdiscretionandoptimalfinancingpolicies,JournalofFinancialEco‐nomics26,3‐27.

Viswanath,P.V.,1993,StrategicConsiderations, thePeckingOrderHypothesis,andMarketReac‐tionstoEquityFinancing,JournalofFinancialandQuantitativeAnalysis28,213‐234.

Zwiebel, Jeffery,1996,DynamicCapitalStructureunderManagerialEntrenchment,AmericanEco‐nomicReview86,1197‐1215.

45  

TableISummaryStatistics

This table reports simplemeans foranumberofcorporate finance. MarketCap isdefinedas InstitutionalHoldings(Hold)isdefinedastotalinstitutionalholdingsrelativetototalsharesoutstanding.Similarly,Top5institutionalholdingsisthepercentoftotalsharesoutstandingheldbythefivelargestinstitutionalinvestors.Leverage(TDA)isdefinedastotaldebtscaledbybookassets.Log(Assets)(Assets)isthenaturallogarithmoftotalassets.MarkettoBook(Mkt/Bk)isdefinedasthemarketvalueofassetsscaledbythebookvalueofas‐sets. Profitability (Profit) is defined as operating income before depreciation scaled by book assets.Log(Sales) (Sales) is thenatural logarithmof total sales. CapitalExpenditures (Capex) isdefinedascapitalexpenditurescaledbybookassets. IntangibleAssets(Intang)isdefinedasintangibleassetsscaledbybookassets. Tangibility (Tang) is defined as the tangible assets (net PP&E) scaled by book assets. Collateral(Colltrl)isdefinedastotalcollateral(thesumofinventoryandnetproperty,plant,andequipment)scaledbybookassets.ThesampleconsistsofallU.S.incorporatedfirms,excludingfinancialfirmsandutilities,withapositive market cap, positive assets, and non‐missing total debt. Accounting data is obtained from theCompustat‐CRSPmergeddatabase,andinstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSECfilings.

FullSample SubsamplesVariable 1979to2009 1979to1994 1995to2009MarketCap($MM) 1,757.99 872.47 2,654.70InstitutionalHoldings 0.29 0.19 0.40InstitutionalHoldings,Top5 0.15 0.11 0.20Leverage 0.24 0.25 0.22Log(Assets) 5.03 4.68 5.38MarkettoBook 1.72 1.57 1.89Profitability 0.04 0.07 0.02Log(Sales) 4.95 4.70 5.20CapitalExpenditures 0.07 0.08 0.06IntangibleAssets 0.09 0.05 0.13Tangibility 0.29 0.32 0.25Collateral 0.44 0.51 0.38FirmYears 130,202 65,510 64,692

46  

TableIITheEffectofInstitutionalHoldingsonLeverage

Thistablereportsestimatesfrompanelregressionsof leverageoninstitutionalholdings.Thedependentvar‐iablesaremeasuredattheendoffiscalyeart,andtheindependentvariables,describedinTableIanddefinedintheAppendix,aremeasuredat(t‐1). TDLA isdefinedastotaldebt includingoff‐balance‐sheetoperatingleases scaled by (total assets plus off‐balance‐sheet operating leases). Cap1‐Cap5 are annualmarket capquintiledummyvariablescorrespondingtothesmallestandlargestfirms,ascending.Industry_TDAisdefinedas theannualmedian leverageperFama‐French industrygroup. Initial_TDA isdefinedaseach firm’s firstnon‐missingleverageratio. Dividend isadummyvariableequaltooneifa firmpaysadividendandzerootherwise.Allspecificationscontainanintercept,andthefollowingindependentvariables,omittedforbrevi‐tyanddefinedintheAppendix:Intang,Payout,Unique,Losses,andChgAssets. Columns3and6alsocontaindummyvariablesCap1–Cap4.Column4includesannualstockreturnadjustedforthepredictedCAPMre‐turn. Column5 includes insider holdings (fromExecuComp) scaled by total shares outstanding,with andwithoutafixedeffect,respectively,andreflectdataforS&P1500firmsfrom1992to2009.Incolumn6,Ana‐lystCoverageiscalculatedusingdatafromtheI/B/E/SHistoricalSummaryFileasthemaximumnumberofanalystswhomakeannualearningsforecastsinanymonthduringthetwelvemonthspriortofiscalyearend,whileassumingthatfirmsnotcoveredbyI/B/E/Shavenoanalystcoverage.Incolumn7independentvaria‐blesarenormalizedby(assets+OBSoperatingleases)ratherthan(assets),asourmeasurefor leverageinthiscolumn,TDLA,includesOBSoperatingleases.ThedataconsistofallU.S.incorporatedfirmswithCRSP,Compustat,andinstitutionalholdingsdatabetween1979and2009(withtheexceptionofcolumns4and5),excludingfinancialfirmsandutilities.InstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSEC filings. ***, **, and * represent statistical significanceat the1%,5%,and10% level, respectively, andstandarderrors,reportedinparentheses,areclusteredatthefirmlevel.

47  

TableII–Continued

(1) (2) (3) (4) (5) (6) (7)TDA (t ) TDA (t ) TDA (t ) TDA (t ) TDA (t ) TDA (t ) TDLA (t )

Hold (t ‐1) ‐0.100*** ‐0.068*** ‐0.070*** ‐0.044*** ‐0.057***(0.0069) (0.0080) (0.0085) (0.0158) (0.0079)

HoldxCap1 (t ‐1) ‐0.108*** ‐0.085***(0.0184) (0.0191)

HoldxCap2 (t ‐1) ‐0.082*** ‐0.073***(0.0120) (0.0123)

HoldxCap3 (t ‐1) ‐0.060*** ‐0.050***(0.0099) (0.0099)

HoldxCap4 (t ‐1) ‐0.040*** ‐0.035***(0.0094) (0.0097)

HoldxCap5 (t ‐1) ‐0.020* ‐0.019*(0.0105) (0.0106)

AdjustedRet. (t ‐1) ‐0.012***(0.0014)

InsiderHold. (t ‐1) 0.009(0.0378)

AnalystCov. (t ‐1) ‐0.003***(0.0004)

Assets (t ‐1) 0.022*** 0.031*** 0.053*** 0.030*** 0.020*** 0.037*** 0.039***(0.0011) (0.0027) (0.0030) (0.0029) (0.0065) (0.0028) (0.0031)

Profit (t ‐1) ‐0.138*** ‐0.120*** ‐0.103*** ‐0.109*** ‐0.203*** ‐0.123*** ‐0.106***(0.0079) (0.0091) (0.0091) (0.0101) (0.0465) (0.0091) (0.0113)

Mkt/Bk(t ‐1) ‐0.001 0.001 0.009*** 0.003** 0.000 0.001 0.007***(0.0008) (0.0009) (0.0009) (0.0011) (0.0032) (0.0009) (0.0012)

Industry_TDA (t ‐1) 0.282*** 0.315*** 0.286*** 0.304*** 0.288*** 0.311*** 0.226***(0.0189) (0.0336) (0.0325) (0.0358) (0.0652) (0.0334) (0.0356)

Initial_TDA (t ‐1) 0.346***(0.0101)

Dividend (t ‐1) ‐0.060*** ‐0.024*** ‐0.018*** ‐0.023*** ‐0.012 ‐0.023*** ‐0.013***(0.0034) (0.0038) (0.0037) (0.0041) (0.0080) (0.0038) (0.0039)

Capex (t ‐1) 0.035* 0.021 0.062*** 0.022 0.028 0.022 ‐0.010(0.0177) (0.0154) (0.0152) (0.0170) (0.0481) (0.0153) (0.0202)

Tang (t ‐1) 0.065*** 0.094*** 0.079*** 0.091*** ‐0.004 0.099*** 0.295***(0.0119) (0.0225) (0.0219) (0.0245) (0.0773) (0.0224) (0.0196)

Observations 88,444 88,444 88,444 77,250 15,795 88,444 66,285FirmFixedEffect No Yes Yes Yes Yes Yes YesYearFixedEffect Yes Yes Yes Yes Yes Yes YesAdj.R ‐squared 0.323 0.639 0.646 0.648 0.699 0.640 0.690

Leverageinlevels

48  

TableIIITheEffectofChangesinInstitutionalHoldingsonChangesinLeverage

Thistablereportsestimatesfrompanelregressionsof changes inleverageonchangesininstitutionalhold‐ings.Thefirst‐differenceddependentvariablesarecalculatedasYt‐Yt‐1,andthefirst‐differencedindepend‐entvariables,describedinTableIanddefinedintheAppendix,arecalculatedasXt‐1‐Xt‐2.Incolumn2,inde‐pendentvariablesarenormalizedby(assets+operatingleases)ratherthan(assets),asthemeasureoflever‐ageusedincludesleases. ThedataconsistofallU.S. incorporatedfirmswithCRSP,Compustat,andinstitu‐tionalholdingsdatabetween1979and2009, excluding financial firmsandutilities. InstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSECfilings.***,**,and*representstatisticalsignificanceatthe1%,5%,and10%level,respectively,andstandarderrors,reportedinparentheses,areclusteredatthefirmlevel.

(1) (2)ΔTDA (t ) ΔTDLA (t )

ΔHold (t ‐1) ‐0.014*** ‐0.022***(0.0047) (0.0047)

ΔTDA (t ‐1) ‐0.153***(0.0070)

ΔTDLA (t ‐1) ‐0.143***(0.0079)

ΔAssets (t ‐1) 0.026*** 0.027***(0.0024) (0.0025)

ΔProfit (t ‐1) ‐0.045*** ‐0.039***(0.0064) (0.0083)

ΔMkt/Bk(t ‐1) ‐0.002*** ‐0.004***(0.0006) (0.0007)

ΔIndustry_TDA (t ‐1) 0.065*** 0.069***(0.0204) (0.0229)

ΔDividend (t ‐1) 0.002 0.003(0.0021) (0.0022)

ΔCapex (t ‐1) 0.059*** 0.036**(0.0111) (0.0149)

ΔTang (t ‐1) 0.020 0.051***(0.0172) (0.0165)

ΔColltrl (t ‐1) 0.058*** 0.045***(0.0133) (0.0134)

ΔIntang (t ‐1) 0.021** 0.008(0.0102) (0.0104)

ΔPayout (t ‐1) 0.033*** 0.039***(0.0103) (0.0113)

Observations 72,208 52,127YearFixedEffect Yes YesAdj.R ‐squared 0.033 0.036

49  

TableIVTheEffectofLeverageonInstitutionalHoldings–LevelsandDifferences

Thistablereportsestimatesfrompanelregressionsofchangesininstitutionalholdingsonchangesinlever‐age.InPanelA,thedependentvariablesaremeasuredattheendoffiscalyeart,andtheindependentvaria‐blesaremeasuredat(t‐1). InPanelB,thefirst‐differenceddependentvariableiscalculatedasYt‐Yt‐1,andthefirst‐differencedindependentvariablesarecalculatedasXt‐1‐Xt‐2.Payoutisdefinedasannualdividendsplus repurchases scaledbybookassets. AdjustedRet. is definedasa firms’ annual returnadjustedby theCAPMreturn. Beta isa firm’sannualCAPMbeta. TDA,TDLA,Sales,Dividend,andMkt/Bkaredescribed inTable I, and all variables are defined in theAppendix. The data consist of all U.S. incorporated firmswithCRSP,Compustat,andinstitutionalholdingsdatabetween1979and2009,excludingfinancialfirmsandutili‐ties. InstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSECfilings. ***,**,and*repre‐sentstatisticalsignificanceatthe1%,5%,and10%level,respectively,andstandarderrors,reportedinpa‐rentheses,areclusteredatthefirmlevel.

(1) (2) (1)Hold (t ) Hold (t ) ΔHold (t )

TDA (t ‐1) ‐0.123*** ‐0.087*** ΔTDA (t ‐1) ‐0.037***(0.0080) (0.0079) (0.0042)

ΔHold (t ‐1) ‐0.125***(0.0071)

Mkt/Bk(t ‐1) 0.016*** 0.011*** ΔMkt/Bk(t ‐1) 0.003***(0.0010) (0.0008) (0.0004)

Sales (t ‐1) 0.067*** 0.039*** ΔSales (t ‐1) 0.007***(0.0011) (0.0019) (0.0009)

Dividend (t ‐1) 0.010** 0.011** ΔDividend (t ‐1) 0.004*(0.0047) (0.0047) (0.0021)

Payout (t ‐1) 0.192*** 0.049** ΔPayout (t ‐1) ‐0.005(0.0361) (0.0217) (0.0096)

Beta (t ‐1) 0.049*** 0.013*** ΔBeta (t ‐1) ‐0.001(0.0020) (0.0013) (0.0006)

AdjustedRet. (t ‐1) 0.006*** 0.002*** ΔAdjustedRet (t ‐1) 0.001***(0.0004) (0.0003) (0.0001)

Observations 84,764 84,764 Observations 71,093FirmFixedEffect No Yes FirmFixedEffect NoYearFixedEffect Yes Yes YearFixedEffect YesAdj.R ‐squared 0.521 0.816 Adj.R ‐squared 0.036

PanelA‐Inst.HoldingsinLevels PanelB‐Inst.HoldingsinDifferences

50  

TableVArellano‐Bond–InstitutionalHoldingsandLeverage

ThistablereportsestimatesfromArellano‐Bondautoregressions ofleverageoninstitutionalholdings(PanelA),andinstitutionalholdingsonleverage(PanelB).Thedependentvariablesaremeasuredattheendofyeart,andtheindependentvariables,describedinTablesIandIVanddefinedintheAppendix,aremeasuredasoftime(t‐1).InPanelA,wealsocontrolforColltrl,Intang,Payout,Unique,andLosses.Inbothpanels,internalinstrumentsbeginwithlag3. Accordingly,theArellano‐BondAR(3)test ispresentedinfirstdifferencestoruleout secondorder serial correlationbetween the first‐differencederror termat time t and the level attimet‐3.Thenullhypothesisisnoserialcorrelation.ThedataconsistofallU.S.incorporatedfirmswithCRSP,Compustat,andinstitutionalholdingsdataforaminimumoffourconsecutiveyearsbetween1979and2009,excludingfinancialfirmsandutilities.InstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSEC filings. ***, **, and * represent statistical significanceat the1%,5%,and10% level, respectively, andstandarderrors,reportedinparentheses,arerobusttoanypatternofheteroskedasticityandautocorrelationwithinpanels.

PanelA‐LeverageRegression PanelB‐InstitutionalHoldingsRegression(1) (2) (1) (2)

TDA (t ) TDA (t ) Hold (t ) Hold5 (t )

Hold (t ‐1) ‐0.148*** Hold (t ‐1) 0.924***(0.0260) (0.0335)

Hold5 (t ‐1) ‐0.183*** Hold5 (t ‐1) 0.872***(0.0363) (0.0221)

TDA (t ‐1) 0.635*** 0.650*** TDA (t ‐1) 0.012 0.003(0.0222) (0.0258) (0.0174) (0.0107)

Assets (t ‐1) 0.021*** 0.015*** Mkt/Bk(t ‐1) ‐0.007*** 0.000(0.0027) (0.0018) (0.0008) (0.0004)

Profit (t ‐1) ‐0.031*** 0.055 Sales (t ‐1) ‐0.011*** 0.000(0.0072) (0.0580) (0.0013) (0.0008)

Mkt/Bk(t ‐1) 0.001 0.000 Dividend (t ‐1) ‐0.005 0.001(0.0007) (0.0009) (0.0031) (0.0020)

Industry_TDA (t ‐1) 0.068*** 0.072*** Payout (t ‐1) ‐0.020 ‐0.031***(0.0190) (0.0187) (0.0175) (0.0104)

Dividend (t ‐1) 0.004* 0.004 Beta (t ‐1) ‐0.006*** ‐0.001(0.0022) (0.0076) (0.0010) (0.0006)

Capex (t ‐1) 0.127*** 0.117*** AdjustedRet. (t ‐1) 0.000* 0.000(0.0114) (0.0119) (0.0002) (0.0002)

Tang (t ‐1) ‐0.008 ‐0.012(0.0134) (0.0135)

Observations 71,493 71,493 Observations 71,096 71,093Numberoffirms 9,244 9,244 Numberoffirms 9,396 9,395AR(3)Pr>z 0.844 0.985 AR(3)Pr>z 0.603 0.486HansenJ ‐statistic 75.16 52.28 HansenJ ‐statistic 16.24 9.239Hansenp‐value 0.030 0.463 Hansenp‐value 0.181 0.682

51  

TableVIInstrumentalVariables–TheEffectofInstitutionalHoldingsonLeverage

Thistablereportsestimatesfromlineartwo‐stageleastsquarespanelregressionsofleverageoninstitutionalholdings.Thedependentvariable,TDA,ismeasuredattheendoffiscalyeart,andtheindependentvariables,describedinTableIandIIanddefinedintheAppendix,aremeasuredasoftime(t–1). TheinstrumentforHold incolumn1,MFFlow, isthehypotheticalannualchangeinmutualfundholdingsimpliedbypreviouslydisclosedholdingsandquarterlymutualfundoutflows≥5%oftotalfundassets.Allspecificationscontainanintercept, and the following independent variables omitted for brevity:Unique,Losses andChgAssets. Thedata consist of all U.S. incorporated firmswith CRSP, Compustat, and institutional holdings data between1980and2007(duetotheavailabilityofMFFlowdata),excludingfinancialfirmsandutilities. InstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSECfilings,whileMFFlowdataisobtainedfromAlexEdmans’swebsite.***,**,and*representstatisticalsignificanceatthe1%,5%,and10%level,respectively,andstandarderrors,reportedinparentheses,areclusteredatthefirmlevel.

(1)2SLS:TDA (t )

Instrument MFFlow

Hold (t ‐1) ‐0.292***(0.0378)

Assets (t ‐1) 0.037***(0.0033)

Profit (t ‐1) ‐0.123***(0.0086)

Mkt/Bk(t ‐1) 0.002(0.0010)

Industry_TDA (t ‐1) 0.258***(0.0204)

Initial_TDA (t ‐1) 0.332***(0.0106)

Dividend (t ‐1) ‐0.061***(0.0036)

Capex (t ‐1) 0.061***(0.0192)

Tang (t ‐1) 0.051***(0.0125)

Colltrl (t ‐1) 0.158***(0.0104)

Intang (t ‐1) 0.301***(0.0121)

Payout (t ‐1) ‐0.017(0.0268)

FirstStageAdj.R ‐squared 0.548FirstStageF‐Statistic 354.63Observations 80,611YearFixedEffect YesAdj.R ‐squared 0.291

52  

TableVIIS&P500Inclusion–TheEffectofaShocktoInstitutional HoldingsonLeverage

This table reports leverage difference‐in‐difference tests surrounding addition to the S&P500 Index. ThetreatmentgroupconsistsofasampleofU.S.incorporatedfirmsthatwereaddedtotheS&P500betweenthethirdquarterof1979and2008,whilethecontrolgroupconsistsofapropensityscorematchedsampleofU.S.incorporatedfirms. Financial firmsandutilitiesareexcluded. PanelApresentssummarystatistics forthetwogroupsof firms in thequarterbefore indexaddition. InPanelBweconductadifference‐in‐differencepairedt‐testcomparingthemeandifferenceinleverage(TDA)inthequarterbeforeadditiontothemeandif‐ference four quarters after addition. Panel C reports estimates from a panel regression of leverage on atreatment/controldummyvariable (Treatment=1 if in the treatmentgroup,0otherwise),aPre/PostS&P500additiondummyvariable (Post =1 if after addition,0otherwise), their interactionTreatment*Post, aswellasothercontrolvariables.Thedependentvariable,TDA,ismeasuredattheendofquartert.TomitigateconcernsovermarkettimingwecontrolforlaggedandcontemporaneousCAPM‐adjustedreturn,aswellaslaggedandcontemporaneousilliquidity(Amihud(2002)).Furthermore,wecontrolfortheindependentvar‐iablesusedinTableIIcolumn1.Thesevariables,describedinTableIanddefinedintheAppendix,aremeas‐uredasofquarter(t–1).ToisolatetheeffectoftheshockininstitutionalholdingsstemmingfromS&P500inclusion,werestrictoursampleto9quartersperfirm:theperiod4quartersbeforeinclusionto4quartersafterinclusion.InstitutionalholdingsdataisobtainedfromThomson‐Reutersvia13FSECfilings.***,**,and*representstatisticalsignificanceatthe1%,5%,and10%level,respectively,andstandarderrors,reportedinparentheses,areclusteredatthefirmlevel.

PanelA–SummaryStatsinQuarter(t‐1) PanelC– Difference‐in‐DifferenceRegression

Treatment ControlMarketCap 5,962.6 5,657.9Ln(Assets) 7.3 7.2Mkt/Bk 2.9 2.9Inst.Hold 44.4% 39.4%Leverage 19.4% 17.7%CAR(‐1,0) 1.9% 2.6%Observations 415 415

(1)

TDA (t )

Treatment*Post ‐0.015**(0.0075)

PostDummy 0.006(0.0058)

TreatmentDummy 0.010(0.0091)

Assets (t ‐1) 0.036***(0.0042)

Profit (t ‐1) ‐0.677***(0.1211)

Mkt/Bk(t ‐1) 0.002(0.0022)

Industry_TDA (t ‐1) 0.379***(0.0599)

Initial_TDA (t ‐1) 0.205***(0.0287)

Dividend (t ‐1) ‐0.011(0.0115)

Tang (t ‐1) 0.114***(0.0363)

Observations 5,980Year/QtrFixedEffect YesAdj.R ‐squared 0.387

PanelB–Difference‐in‐Difference

pairedt‐test

(TDA treat‐TDA cont)(t ‐1) 0.017(TDA treat‐TDA cont)(t +4) 0.002(Diff(t +4)‐Diff(t ‐1)) ‐0.015

t ‐statistic ‐2.061p‐value 0.040

Quarter(t ‐1)vs.Quarter(t +4)

53  

TableVIIITheEffectofInstitutionalHoldingsonLeverage– TestsofUnderlyingFrictions

Thistablereportsestimatesfrompanelregressionsof leverageon institutionalholdings. Thedependentvariable,TDA, ismeasuredattheendoffiscalyeart,andtheindependentvariables,describedinTableIanddefinedintheAppendix,aremeasuredasoftime(t–1). InPanelA,wetestwhetheragencycostsorasymmetricinformationisbehindtherelationshipbetweeninstitutionalholdingsandleverage.Asymmet‐ricInformation(Agency)firmsaresmall(large)firmswithmany(few)growthopportunities.Columns1and2containthefollowingindependentvariables,omittedforbrevity:Intang,Payout,Unique,Losses,andChgAssets.ThedataconsistofallU.S.incorporatedfirmswithCRSP,Compustat,andinstitutionalholdingsdata between1979 and2009, excluding financial firms andutilities,with the additional requirement incolumns3and4ofatleastfourconsecutiveyearsofnon‐missingdata. InPanelB,weusedatabetween2000and2006,excluding2003,totesttheimpactofinstitutionalholdingsonleverageconditionalonthesize of the tax advantage enjoyed by institutions over individuals. The new variablesHold_00‐02 andHold_04‐06reflectthis.TheFtestteststheequalityofHold_00‐02andHold_04‐06,withthenullhypothesisbeingthattheyareequal.Thefollowingindependentvariablesareincludedbutomittedforbrevity:Intang,Payout,Unique,Losses, andChgAssets. Institutionalholdingsdata isobtained fromThomson‐Reutersvia13FSECfilings.***,**,and*representstatisticalsignificanceatthe1%,5%,and10%level,respectively,andstandarderrors,reportedinparentheses,areclusteredatthefirmlevel.

(1) (2) (3) (4) (1)Asym.Info.:TDA (t )

Agency:TDA (t )

Asym.Info.:TDA (t )

Agency:TDA (t ) TDA (t )

Hold (t ‐1) ‐0.147*** ‐0.050*** ‐0.220*** ‐0.026 Hold_00‐02 (t ‐1) ‐0.077***(0.0526) (0.0158) (0.0787) (0.0189) (0.0113)

TDA (t ‐1) 0.592*** 0.639*** Hold_04‐06 (t ‐1) ‐0.086***(0.0372) (0.0310) (0.0121)

Assets (t ‐1) 0.022* 0.037*** 0.033*** 0.005* Assets (t ‐1) 0.022***(0.0119) (0.0076) (0.0058) (0.0029) (0.0019)

Profit (t ‐1) ‐0.089*** ‐0.297*** ‐0.058*** ‐0.029 Profit (t ‐1) ‐0.150***(0.0190) (0.0476) (0.0087) (0.0222) (0.0164)

Mkt/Bk(t ‐1) 0.003 0.015* 0.000 ‐0.002 Mkt/Bk(t ‐1) 0.004***(0.0020) (0.0077) (0.0011) (0.0023) (0.0014)

Industry_TDA (t ‐1) 0.333** 0.293*** 0.097 0.062** Industry_TDA (t ‐1) 0.246***(0.1561) (0.0643) (0.0693) (0.0292) (0.0291)

Dividend (t ‐1) ‐0.029 0.002 ‐0.004 0.010*** Dividend (t ‐1) ‐0.048***(0.0234) (0.0076) (0.0101) (0.0040) (0.0057)

Capex (t ‐1) 0.030 0.048 0.055* 0.146*** Capex (t ‐1) ‐0.190***(0.0554) (0.0394) (0.0335) (0.0230) (0.0456)

Tang (t ‐1) 0.130* 0.092 ‐0.007 ‐0.049** Tang (t ‐1) 0.171***(0.0664) (0.0562) (0.0356) (0.0226) (0.0213)

Colltrl (t ‐1) 0.129*** 0.038 0.043 0.040** Colltrl (t ‐1) 0.139***(0.0486) (0.0513) (0.0275) (0.0186) (0.0174)

Observations 10,120 11,206 9,814 12,368 Observations 17,312FirmFixedEffect Yes Yes Yes Yes FirmFixedEffect NoYearFixedEffect Yes Yes Yes Yes YearFixedEffect YesAdj.R ‐squared 0.598 0.737 Pr>F 0.401HansenJ ‐statistic 62.04 57.79 Adj.R ‐squared 0.331Hansenp‐value 0.270 0.409

PanelA‐AsymmetricInformationorAgency PanelB‐TaxesLeverageinlevels(cols1‐2),anddifferenceGMM(cols3‐4) Leverageinlevels

54  

TableIXTheEffectofChangesinInstitutionalHoldingsonEquityandDebtIssuances

Thistablepresentsestimatesfromlogitandlinearregressionsofequityissuancesonlaggedchangesininsti‐tutionalholdings.EquityIssue(EquityRepo)isadummy=1ifthechangeinsplit‐adjustedsharesoutstandingnormalizedbytotalassets,%ChangeEquity,>1%(<‐1%),and0otherwise.%ChangeDebtisthechangeindebtnormalizedbytotalassets,whileNetDebtIssue=1if%ChangeDebt–%ChangeEquity>0,and0oth‐erwise.Allspecificationscontainanintercept.Thecontrolvariables,datasources,andsamplearedescribedinTablesIandIV.***,**,and*representstatisticalsignificanceatthe1%,5%,and10%level,respectively,andstandarderrors,reportedinparentheses,areclusteredatthefirmlevel.

(1) (2) (3) (4) (5)Equity

Issue (t )=1Equity

Repo (t )=1NetDebtIssue (t )=1

%ChangeEquity (t )

%ChangeNetDebt (t )

ΔHold (t ‐1) 1.409*** ‐0.368*** ‐0.353*** 0.061*** ‐0.028**(0.0947) (0.1173) (0.0906) (0.0097) (0.0129)

ΔTDA (t ‐1) ‐0.265*** ‐0.960*** ‐1.048*** 0.095*** ‐0.318***(0.0917) (0.0959) (0.0808) (0.0177) (0.0207)

ΔAssets (t ‐1) 0.777*** ‐0.122*** 0.368*** 0.006 0.066***(0.0417) (0.0387) (0.0331) (0.0066) (0.0081)

ΔProfit (t ‐1) ‐0.055 0.073 ‐0.114* ‐0.061*** 0.035*(0.0756) (0.0657) (0.0640) (0.0167) (0.0190)

ΔMkt/Bk(t ‐1) 0.128*** ‐0.031*** ‐0.013* 0.031*** ‐0.027***(0.0091) (0.0087) (0.0074) (0.0021) (0.0024)

ΔIndustry_TDA (t ‐1) ‐2.033*** 0.594 1.610*** ‐0.097** 0.181***(0.3525) (0.4901) (0.3623) (0.0391) (0.0496)

ΔDividend (t ‐1) 0.056 0.237*** 0.210*** 0.001 0.016***(0.0416) (0.0613) (0.0458) (0.0029) (0.0046)

ΔCapex (t ‐1) ‐0.030 0.511*** 1.509*** ‐0.035 0.144***(0.1506) (0.1766) (0.1582) (0.0238) (0.0311)

ΔTang (t ‐1) ‐0.304 0.536** 0.412* ‐0.095** 0.151***(0.2452) (0.2636) (0.2243) (0.0410) (0.0493)

ΔColltrl (t ‐1) 0.494*** ‐1.105*** 0.565*** 0.078** ‐0.005(0.1916) (0.2075) (0.1710) (0.0320) (0.0375)

ΔIntang (t ‐1) 0.357** ‐0.265 ‐0.287** 0.089*** ‐0.061*(0.1596) (0.1871) (0.1450) (0.0284) (0.0344)

ΔPayout (t ‐1) ‐0.893*** 4.693*** 1.921*** ‐0.095*** 0.143***(0.1876) (0.3194) (0.2116) (0.0207) (0.0276)

ΔAdjustedRet (t ‐1) 0.007** ‐0.007** ‐0.007** 0.001 ‐0.001(0.0031) (0.0031) (0.0028) (0.0006) (0.0007)

ΔBeta (t ‐1) 0.036*** ‐0.077*** ‐0.018 0.002 ‐0.004*(0.0120) (0.0139) (0.0119) (0.0018) (0.0022)

YearFixedEffect Yes Yes Yes Yes YesObservations 58,885 58,878 58,878 58,878 58,878PseudoR ‐squared 0.045 0.036 0.022Adj.R ‐squared 0.039 0.041

LinearRegressionsLogitRegressions