ContentslistsavailableatScienceDirect
The
Quarterly
Review
of
Economics
and
Finance
j o u r n al ho me p a g e :w w w . e l s e v i e r . c o m / l o c a t e / q r e f
The
impact
of
liquidity
crises
on
cash
flow
sensitivities
Wolfgang
Drobetz
a,
Rebekka
Haller
a,
Iwan
Meier
b,
Vefa
Tarhan
c,d,∗aFacultyofBusiness,HamburgUniversity,Moorweidenstraße18,20148Hamburg,Germany bHECMontréal,3000,chemindelaCôte-Sainte-Catherine,Montréal(Québec)H3T2A7,Canada cQuinlanSchoolofBusiness,LoyolaUniversity,820N.MichiganAvenue,Chicago,IL60611,USA dBilkentUniversity,Ankara,Turkey
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received1June2016
Receivedinrevisedform9February2017 Accepted6March2017
Availableonline20March2017 JELclassification:
G01 G31 G32 Keywords: Cashflowsensitivity Financialconstraints Liquiditycrises Investmentspending Supplysideshock
a
b
s
t
r
a
c
t
Weexaminetherelationshipbetweenliquiditycrisesandfrictionsinraisingfunds,andfindthatboth
thegapbetweenthecashflowsensitivitiesoffinanciallyhealthyandweakfirmsandthecashflow
sensitivitiesofhealthyandweakfirmsthemselvesarepositivelycorrelatedwiththeseverityofliquidity crises.Usingamulti-equationmodelofcashflowsensitivities,wefindthatmoderateliquiditycrises mostlyaffectfirms’financingactivities.Therecentfinancialcrisiswasespeciallysevereforfinancially weakfirmsandcurtailedboththeirinvestmentandfinancingdecisions.Financiallyhealthyfirmswere abletoprotecttheirinvestmentsbymaintainingfinancialflexibility.
©2017BoardofTrusteesoftheUniversityofIllinois.PublishedbyElsevierInc.Allrightsreserved.
1. Introduction
Intheirseminalstudy,Fazzari,Hubbard,andPetersen(1988) use investment-cashflow sensitivities to show that financially weakfirmsexperiencemoredifficultyraisingexternalfundsthan financially healthy firms.1 They interpret this as evidence that
financiallyweakfirmsfacefrictionsinaccessingcapitalmarkets. Werefertothisapproachfordetectingcapitalmarketconstraints as the “traditional” test. All subsequent studies essentially use Fazzarietal.’s(1988)single-equationmodel,whichregresses capi-talexpendituresoncashflows(alongwithsomecontrolvariables). Astatisticallysignificantand positivecoefficientoncashflowis interpretedasevidenceofconstrainedaccesstocapitalmarkets. Thesingle-equationapproachpositsthat,ifafirmcutsits
invest-∗ Correspondingauthorat:QuinlanSchoolofBusiness,LoyolaUniversity,820N. MichiganAvenue,Chicago,IL60611,USA.
E-mailaddress:vtarhan@luc.edu(V.Tarhan).
1Inthisstudy,weusetheterms“financiallyweak”and“financiallyhealthy”rather
than“constrained”and“unconstrained,”whichareusedintheearlierliterature. Thisisbecause“constraint”wouldthenbeusedinthreedifferentcontexts,which webelieveisoverlyconfusing:(1)asabove,todenoteafirm’soverallhealth,(2)to denotethelevelofaccesstocapitalmarkets,and(3)thesensethatwealsoestimate ourmodelsubjecttosomeconstraints.
mentsinresponsetoacashflowshortfall,itmustbebecauseit isunabletoraisesufficientexternalfundstocompensateforthe shortfall,andisthusfinanciallyconstrained.
Ourprimary aiminthis articleistoexaminewhetherfirms’ constraintsinaccessingcapitalmarketsvarywiththe macroeco-nomicenvironment.Inparticular,weexpectliquiditycrisestohave adverseeffectsonfirms’abilitiestoraisefunds.Againstthe back-dropthatpreviousstudiesusedpooleddatathatcovereddifferent liquidityenvironments,wemakeseveralcontributionstothe liter-atureoncapitalmarketaccessconstraints.
First,becausepooleddatamaymaskdifferencesinhowfirms behaveunderdifferentliquidity-relatedeconomicstates,we esti-mateourmodelseparatelyduringnon-crisisperiods,duringthe relativelymoderateliquiditycrisespriorto2007,andduringthe recent2007–2009financialcrisis(orsubprimemortgagecrisis). Wethenexplorewhetherthetraditionalmeasureofcapitalmarket frictions–thecashflowsensitivitydifferencesbetweenfinancially weakandhealthyfirms–increaseswiththeseverityofliquidity crises.
Second, to get a complete understanding of the correlation betweenliquidityconditionsandthesizeofthehurdlesfirmsface inraisingfunds,weestimatecashflowsensitivitiesseparately dur-ingnon-crisisandliquiditycrisisperiods,andthencomparethetwo setsofestimatestoassesswhethercapitalmarketfrictionsincrease http://dx.doi.org/10.1016/j.qref.2017.03.004
withtheseverityofliquiditycrises.Wehypothesizethatfinancially weakfirmswillbeaffectedmorethanhealthyfirmsduring liq-uiditycrises,andevaluatethedifferencesincashflowsensitivities betweennon-crisisandliquiditycrisisperiodsseparatelyfor finan-ciallyweakandhealthyfirms.Theseextensionsofthetraditional testenableustoassesshowcrisisseverityimpactstheconstraints inaccessingcapitalmarkets.
Third,capital marketfrictions imposecostsonfirmsby cre-atingdistortionsintheirfinancialdecisions,e.g., intheformof underinvestmentproblems.2 Thetraditionalmetricfordetecting
capitalmarketconstraintsonlyindicatestowhatextentweakfirms operatesuboptimallyrelativetohealthyfirms.Ifthecashflow sen-sitivitiesofbothfinanciallyweakandhealthyfirmsincreaseduring liquiditycrises, but thegapbetweenthecash flow sensitivities remainsthesame,thetraditionalmeasureforfinancialconstraints indicatesnochangeinfrictions.Incontrast,weinterpretthesum ofthechangesinthecashflowsensitivitiesfromnormaltimesto crisisperiodsfacedbyfinanciallyweakandhealthyfirmsasthe totalcostsofliquiditycrisesfortheeconomy.
Fourth, both survey (Campello, Graham, & Harvey, 2010; Campello,Giambona,Graham,&Harvey,2011)andempirical stud-ies(Duchin,Ozbas,&Sensoy,2010;Ivashina&Scharfstein,2010; Almeida,Campello,Laranjeira,&Weisbenner,2011;Bliss,Cheng,& Denis,2013)showevidencethatthesubprimemortgagecrisiswas unusuallysevere.Ifcashflowsensitivitiesarealegitimate mea-sureoftheseverityofliquiditycrises,theyshouldreflectfirms’ increaseddifficultiesinraisingfundsviadebtandequityofferings duringtherecentfinancialcrisis.Havingthiscrisisinoursample offersanidealcasefortestingwhetheritwasunusuallyseverein termsofmeasuresofcapitalmarketaccessconstraints.
Finally,whilealmostallpriorstudiesintheliteratureuseFazzari etal.’s (1988)single-equationmodel,weimplementthe multi-equation cash flow sensitivity model of Gatchev, Pulvino, and Tarhan(2010).Duringaseverecrisis,adverseeffectsareunlikely tobeconfinedtofirms’investmentdecisions.Theymayalsoaffect theirfinancingandshareholder distributiondecisions.Sincethe Gatchevetal.(2010)modelincludesalltheimportantinvestment, financing,andshareholderpayoutvariables,wecanprovidea com-prehensivedescriptionofhowfirmsparticularlycopedwiththe recentfinancialcrisis.
AlargestrandofliteraturethatfollowsFazzarietal.’s(1988) approach hasconfirmedtheir findings,and concurs that finan-ciallyweakfirmsexhibithigherinvestment-cashflowsensitivities thanhealthyfirms(Gilchrist&Himmelberg,1995;Allayannis& Mozumdar,2004;amongothers).However,otherstudiesfind evi-dencetothe contrary,and showa non-monotonic relationship between investment cash flow sensitivities and financial con-straints(Kaplan&Zingales,1997;Kadapakkam,Kumar,&Riddick, 1998;Cleary,1999).3Oneexplanationfortheseconflictingresults
maybethatthepresenceandextentoffinancialconstraintsare notdirectlyobservable(Moyen,2004;Becchetti,Castelli,&Hasan,
2 Iffirmsareunabletofullyoffsetdecreasesincashflowsbyapproachingcapital
markets(orbydrawingdowntheircashbalances),thennotonlyaretheir invest-mentactivitieslikelytobecomesuboptimal,buttheiroperating,financing,and shareholderdistributionactivitiesmayaswell.Forthesakeofsimplicity,mostofour discussionsfocusontheunderinvestmentproblemasanexampleofthedistortions causedbycapitalmarketfrictions.
3 Moyen(2004)positsthatdebtfinancingmayexplainthenon-monotonic
rela-tionshipbetweeninvestment-cashflowsensitivitiesandfinancialconstraints.The effectofdebtfinancingoninvestmentsisnotcapturedbythesingle-equationmodel, thusitmagnifiestheinvestment-cashflowsensitivityofhealthyfirms.Otherstudies focusonfirmswithloworevennegativecashflowsandoperatinglosses(Allayannis &Mozumdar,2004;Bhagat,Moyen,&Suh,2005;Cleary,Povel,&Raith,2007)or life-cycleeffects(Hovakimian,2009).Internationalevidenceofinvestment-cashflow sensitivitiesisshowninLove(2003),Lins,Strickland,andZenner(2005),andFrancis, Hasan,Song,andWaisman(2013).
2010).Different studies usedifferentproxies for financial con-straints,andpotentiallyestimatecashflowsensitivitiesdifferently. AnotherpossibleexplanationisthatTobin’sqmaybeestimated witherror,implyingthatinvestmentsaresensitivetocashflows becausecashflowsreflectgrowthopportunitiesratherthan finan-cialconstraints(Erickson&Whited,2000;Roberts&Whited,2012). BrownandPetersen(2009)extendthesingle-equation frame-workalongtwodimensions.First,theyassumeadjustmentcostsin investments,andincludelaggedinvestmentsintheirmodel. Sec-ond,theyaddavariablethatcapturestheamountoffirms’external financing.4BondandSöderbom(2013)usesimulateddatatoshow
thatinaconstrainedregimethesensitivityofinvestmenttocash flow conditionalon measuresof Tobin’sqincreases monotoni-callywiththecostforexternalfinancing.McLeanandZhao(2014) findthattheinvestment-cashflowsensitivitydeclineswhen eco-nomicconditionsimprove(andexternalfinanceislesscostly).They directlytesttherelationshipbetweencashflowsandfinancing vari-ables,andfindthatdebtandequityofferingsensitivitiestocash flowsbecomemorepronouncedaseconomicconditionsimprove. Ourmainresultsareasfollows:first,intermsofthetraditional metric,wefindthatfirmsdonotfacefrictionsinraisingfunds dur-ingnormaltimes.However,evenduringtherelativelymoderate crisisperiodspriorto2007,wefindevidencethatcapitalmarket frictionsinthetraditionalsenseexist.Thesefrictionsaredetected infirms’financingdecisions.Incontrast,duringtherecent finan-cialcrisisof2007–2009,frictionsaredetectedinboththe financing-andinvestment-cashflowsensitivitiesoffinanciallyweakfirms.
Second,wedetectthepresenceofcapital marketconstraints inthefinancing-cashflowsensitivitiesoffinanciallyweakfirms evenduringmoderatecrises.Thisevidenceisstrongerduringthe recentfinancialcrisis,whenweagainfindcapitalmarketaccess constraintsinbothinvestment-andfinancing-cashflow sensitiv-itiesoffinanciallyweakfirms.Thecapitalbudgetsofthesefirms werecurtailedsharplyduringthesubprimemortgagecrisis:the sizeoftheinvestment-cashflowsensitivityestimateis$0.554fora $1declineincashflows(thehighestestimatefortheother subsam-plesisonly$0.041).Thisfindingconfirmsthattherecentfinancial crisiswasanunusuallysevereliquiditycrisis.
Third,weshowevidencethatcapitalmarketfrictionsof finan-ciallyhealthyfirmsareprimarilyconfinedtotheirfinancing-cash flowsensitivities.Thesefirmswereabletoinsulatetheircapital expendituresfromtheeffectsofthepre-2007liquiditycrises.Even duringtherecent financialcrisis,theywereabletoprotect96% of theircapital expenditures. It seems that thekey to protect-inginvestmentsishavingfinancialflexibilityintermsofunused short-termborrowingcapacityandsufficient excesscash. How-ever,althoughthesefirmsareconsideredfinanciallyhealthy,they wereunabletoissuelong-termdebtduringtherecentfinancial cri-sis,whichsupportsthewidelyassertedobservationthatthecapital markets“froze.”
Finally,withtheexceptionoffinanciallyweakfirmsduringthe recentfinancialcrisis,thefrictionsinraisingfundsarenotobserved inthe investment-cashflow sensitivities thattheliterature has focusedon.Instead,wedetecttheminthecashflowsensitivitiesof thefinancingvariables,primarilyintheleveragevariables (short-andlong-termborrowingsandreductionincashholdings).Overall, weconcludethatthereisastrongandpositivecorrelationbetween thedifficultiesfirmsfacein raisingfundsand thepresenceand severityofliquiditycrises.
4Dasgupta,Noe,andWang(2011)investigatetheallocationofcashflowsto
alternativeusesbasedonthecashflowidentity.Theyfindthatfirms’behavioris consistentwiththepeckingorderhypothesisofMyersandMajluf(1984).In partic-ular,firmsusea$1cashinflowtoaddtocashbalancesandreduceexternalfinancing ratherthanpayingoutdividendsorsubstantiallyincreasinginvestments.
Theremainderofthepaperis structuredasfollows:Section 2 describes our methodology and presents the data. Section 3 presentstheresultsofourtestsforthetimevariationofinvestment andfinancingcashflowsensitivities,conditionalonfirms’ finan-cialhealth.Section4documentsthecorrelationbetweenaggregate (economy-wide)capitalmarketaccessconstraintsandtheseverity ofliquiditycrises.Section5concludes.
2. Methodologyanddata
2.1. Traditionalmodelofinvestment-cashflowsensitivities
Fazzarietal.(1988)proposeasingle-equationmodelto deter-minewhetherfirmsfaceconstraintsinaccessingcapitalmarkets. Intheirspecification,thecapitalexpendituresvariableisregressed oncashflows(bothdividedbybeginning-of-periodcapitalstock) andonTobin’sq,whichcontrolsforfirms’investment opportu-nities.Wereplicatethetraditionalsingle-equationframeworkby estimatingthefollowingmodel:
CAPXi,t
PPENTi,t−1 =at+ai+b1×Qi,t−1+b2×
CFLi,t
PPENTi,t−1+εi,t, (1) whereCAPXiscapitalexpenditures,PPENT isnetproperty,plant, andequipment,QisTobin’sq,andCFLiscashflows.Theintercept termsatandaicontrolfortimeandfirmfixedeffects,respectively.5
Thecoefficientb2oncashflowsistheestimateofthe
investment-cashflowsensitivity.Apositiveandhighestimateforthisvariable indicatesthatfirmsareforcedtoreducecapitalexpendituresin responsetocashflowshortfalls.Theimplicationintheliterature isthatthesefirmsareunabletoraisesufficientexternalfundsto offsetcashflowshortfalls.
2.2. Gatchevetal.’s(2010)systemofequationsmodel
Gatchevetal.(2010)suggestthatstaticsingle-equation mod-elssuffer from anomitted variables problem, especially in the caseofvariableswithinertia.Supportingtheirconjecture,wefind thatcapitalexpenditures,assetsales,sharerepurchases,and div-idendsexhibitthehighestinertia.Adjustmentcostsofstopping andrestartinginvestmentsinfixedassetsandthefactthatfirms haveannualreplacementprojectsmayexplainthisinertiain capi-talexpenditures(intertemporaldependenceoffinancialdecisions). Incontrast,financingvariablesdonotexhibitverystronginertia becauseadjustmentcostsarelower.It issurprising thatearlier studiesfocusoncapitalexpenditures,thevariablewiththehighest degreeofpersistencethatlikelysuffersfromtheomittedvariables problemthemost.6
Gatchevetal.(2010)alsoarguethatfirmsmayrespondtocash flowshortfallsbyadjustingothervariables besidesinvestments. Single-equationmodelsdonotacknowledgethatsourcesoffunds equalusesoffundsforallfirmsandatalltimes(interdependence of financial decisions).Ignoring thesources-equal-uses identity producesinefficientcoefficientestimates.Conclusionsaboutthe presenceorabsenceofcapitalmarketfrictionsarereachedonthe basisof differencesinestimatedcoefficients acrosssubsamples.
5ThevariableconstructionswiththeCompustatitemsaredetailedinAppendix
AinTableA1.
6Ourresultsstronglyconfirmthisargument.ThevectorKinEq.(3)below
containsthecorrespondingcoefficientsforlaggedmodelvariables.Inresultsnot reported,wefindthattheestimatesforlaggedcapitalexpenditures,assetsales, sharerepurchases,anddividendsexhibitthehighestinertia,withlagged coeffi-cientsof0.969,0.848,0.706,and0.591,respectively,forthefull1972–2011sample period.Incontrast,theestimatesforlaggedlong-andshort-termborrowingaswell asforinternalborrowing(i.e.,useofcashbalances)fromourmodelexhibitnoinertia (estimatesof0.059,0.058,and0.018,respectively).
Therefore, precisionof theestimates isimportantin this litera-ture.Theinterdependenceproblemmayalsoberesponsiblefor misclassifyingfirmsas“unconstrained”whentheyareeffectively “constrained”(andviceversa).7
FollowingGatchevetal.’s(2010)methodology,weestimatea systemofnineequationssubjecttothesources-equal-uses con-straint.Thefollowingaccountingidentityalwaysholdsexpost:
CAPXt+ACQUISt−ASALESt+RPt+DIVt−EQUISSt−LTDt
−STDt+CASHt≡CFt, (2)
whereCAPXiscapitalexpenditures,ACQUISisacquisitions,ASALE isassetsales,RPissharerepurchases,DIVisdividends,EQUISSis equityissues,andCASHischangesincashbalancesfromyear t−1 toyear t. LTDand STDare long- and short-term debt issues,respectively,andCF iscashflows.AllvariablesinEq.(2) areassumedtobeendogenous,exceptCF.TableA1inAppendixA givesthedefinitionsweuseinconstructingthesevariables.
Inestimating oursystemofequations model,wecontrolfor firms’investmentopportunitiesbyusingthemarket-to-bookratio ofequity(MB).Weincludefirmsize(SIZE),measuredbythe loga-rithmofthebookvalueoftotalassets,toaccountforthepossibility thatlargerfirmshaveeasieraccesstoexternalcapital.Weestimate thefollowingsystemofnineequations,subjecttotheconstraints inEq.(3)below8:
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
−CAPXt −ACQUISt ASALESt −RPt −DIVt EQUISSt LTDt STDt −CASHt⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=L [CFt]+K⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
−CAPXt−1 −ACQUISt−1 ASALESt−1 −RPt−1 −DIVt−1 EQUISSt−1 LTDt−1 STDt−1 −CASHt−1⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
+M MBt SIZEt+
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
−eCAPX,t −eACQUIS,t eASALES,t −eRP,t −eDIV,t eEQUISS,t e LTD,t e STD,t −eCASH,t⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
, (3)7Forexample,iffinanciallyweakfirmsarenotabletoaccesscapitalmarkets,they
mayprotecttheircapitalexpendituresbysellingassets.Infact,surveyevidencein Campelloetal.(2011)showsthatmostfinanciallyweakfirmsfundedtheirongoing operationsbysellingassetsduringtherecentfinancialcrisis(inadditiontodeepcuts incapitalexpenditures).Thiseffectweakensthecorrelationbetweencashflowsand capitalexpenditures,andthereforethesefirmsappearlessconstrained.
8ThesystemofequationsinEq.(3)andtheconstraintsinEq.(4)arederivedfrom
amodelwheremanagersminimizeadjustmentcosts.Adjustmentcostsincrease withhowfarthedecisionvariablesdeviatefromtheirtargetlevels.Moreover,the fasteristhemovetothedesiredtargetlevel,thehigheraretheadjustmentcosts.See Gatchevetal.(2010)andSpindtandTarhan(1980)forthederivationofthemodel.
whereeCAPX,...,eCASH aretheerrortermsassociated withthe
ninefinancing,investment,andpayoutvariables.InEq.(3),theuse variablesarecapitalexpenditures,acquisitions,sharerepurchases, dividends,andchangesincashholdings.9WeestimateEq.(3)
sub-jecttothesources-equal-usesidentityof Eq.(2),which implies imposingthefollowingsetofconstraintsfortheparameter matri-ces:
iL=−1, iK=01×9, iM=01×2. (4)
ThefirstrestrictioninEq.(4)statesthat,whenthereisashockin cashflows,thesumoftheestimatesforsourcesandusesvariables willaddupto$1,andwillhavetheoppositesignofthechange incashflows.If,forexample,thesourcevariableCFincreasesby $1,theneither(1)theothersourcevariablesmustdeclineby$1, (2)theusevariablesmustincreaseby$1,or(3)theresponseof somecombinationofsourceandusevariablesmustsumto$1.If theshockoriginatesfromtheexogenouscontrols(MBandSIZE) orfromanylaggeddependentvariable,thetotalresponseacross equationsmustsumtozero(secondandthirdsetofrestrictionsin Eq.(4))asthesevariablesdonotrepresenteithersourcesoruses offundsinthecurrentperiod.10
2.3. Identifyingtheliquiditycrises
We divide our sample years into liquidity crisis and non-liquidity crisisperiods to examine financing constraints under different liquidity-related regimes. As per Reinhart and Rogoff (2011), we identify the three pre-2007 liquidity crises as fol-lows:(1)theoilcrisisof1975,(2)thesavingsandloancrisisof 1984–1991,and(3)theinformationtechnologybubbleandenergy crisisof2002–2003.11Incontrasttotherelativelymoderate
liq-uiditycrises,thefinancialsystemsufferedmuchmorepronounced adverseliquidityconditionsduringthemostrecentcrisis,when bank, bond, and equity financing sources were simultaneously affected(Gorton&Metrick,2012;Duca,2013).
Severalstudiesmaintainthattherecentfinancialcrisis (sub-primemortgagecrisis)encompassed2007through2009(Campello etal.,2010;Duchinetal.,2010;Almeidaetal.,2011;Kahle&Stulz, 2013).Wealsousethistimeperiodtodatethecrisis.Ouranalysis usesannualdataevenifacrisisbeginsandendsduringthecourse ofayear.Therefore,weassumethatitseffectsarereflected,atleast partially,inend-of-yeardata.From2010onward,firmshadrecord levelsofprofitsandcashholdings(Sanchez&Yurdagul,2013),and theFederalReservecontinueditsaccommodativemonetary pol-icywithextremelylowinterestrates.Therefore,thereasonbehind thelackof corporate investmentswasnot highcostsorscarce availabilityoffunds,butratherademandshockcausedbyfirms’ perceptionsthatglobaluncertaintieswereexceedinglyhigh(Mian &Sufi,2014).12
9 Wecanconsiderassetsalesasanegativeusevariableorasasourcevariable.In
bothcases,itsexpectedsignisnegative,whichindicatesthatadeclineincashflows isexpectedtotriggerassetsales.
10 Thesources-equal-usesconstraintinEq.(4)caneitherbedirectlyinthedata,
orimposedwhenestimatingthesystem.Chang,Dasgupta,Wong,andYao(2014) differentiatebetweenwhattheyrefertoas“explicitconstraints”(asusedinGatchev etal.,2010)and“implicitconstraints,”i.e.,wherethedataisconstructedtosatisfy thesources-equal-usesidentity.
11 See Carmen Reinhart’s webpage: http://www.carmenreinhart.com/data/
browse-by-topic/topics/7/.
12 Intheirthirdquarter2012DukeUniversityCFOBusinessOutlooksurvey,Graham
andHarvey(2012)showthat97%ofrespondingCFOsindicatedthata50-basispoint reductioninborrowingcostswouldnotcausethemtoinitiate,accelerate,orincrease theirinvestments.Evenforaone-percentage-pointreduction,whichissubstantial giventhealreadyverylowinterestrates,over91%oftheparticipantsresponded inthesamemanner.Whenthequestionwasrepeatedforatwo-percentage-point reduction,84%oftherespondentsgavethesameanswer.
2.4. Data
OuranalysisuseslistedU.S.firmsfromCompustatNorth Amer-ica over the 1971–2011 period. We report our results for the 1972–2011 sample period because the model in Eq. (3) uses one-yearlaggedvariables.Thesampleincludesallfirmswith con-solidatedbalancesheetdataandpositivevaluesfortotalbookand marketvaluesofassets.AsinGatchevetal.(2010),weexclude financialfirms(SICcodes6000–6999)andregulatedutilities(SIC codes4900–4999).TableA1summarizesthedefinitionsand Com-pustatcodesofthesource,use,andcontrolvariablesthatenterour multi-equationmodel.FollowingGatchevetal.(2010),wereplace missingvalueswithzero.Finally,wetrimthemarket-to-bookratio atthe95%tail.Thefinalfullsampleconsistsof217,900firm-year observations.
Table1showsthemeansandstandarddeviationsforall vari-ables in our model as a proportion of total assets (except for firmsizeandmarket-to-book ratio).PanelAreports descriptive statisticsforthefullsample;PanelsBandCfurthersubdividethe sampleaccordingtofirms’financialhealth.Allpanelsshow descrip-tivestatisticsfor thefullsample period,thenon-crisisyears of 1972–2007and2010–2011,thethreeliquiditycrisisperiodsprior to2007,andtherecentfinancialcrisisof2007–2009.PanelA indi-catesthatthemeancashflowsduringthefullsampleperiodis slightlynegativebutnot significantlydifferentfromzero. How-ever,cashflowsdeclineastheliquidityconditionsintheeconomy becomemoresevere.Whileaverageannualcashflowsare−0.1% duringnormaltimes,theysharplydecreaseto−5.4%duringthe recentfinancialcrisis.Averageannualshort-andlong-term bor-rowings,aswellaschangesincashholdings,alsodeclineduring therecentfinancialcrisisperiodrelativetonon-crisisyears.Capital expendituresarearound6.5%oftotalassetsduringnormaltimes andduringthepre-2007crisisepisodes,buttheydeclineto5.1% duringtherecentfinancialcrisis.
To classify firms as financially weak or healthy, we apply HadlockandPierce’s(2010)criteria.Theirindexconsidersfirmsize andageaspredictorsoftheextentfirmsfacefinancialconstraints. Thesize–age-index(SA-index)iscalculatedasfollows:
SA-index=−0.737×Size+0.043×Size2−0.040×Age, (5)
whereSizeisthenaturallogarithmoftotalassets,andAgedenotes thenumberofyearsa firmhasnon-missingstockpricedatain Compustat North America. A higher (lower) SA-index value is consistentwithgreater (smaller) financingobstacles. Firm-year observationswithahighSA-indexvalue(abovethe60%percentile) areclassifiedasfinanciallyweak;firm-yearobservationswithalow SA-indexvalue(belowthe40%percentile)areconsidered finan-ciallyhealthy.13
PanelsBandCofTable1reportdescriptivestatisticsforthe subsamplesoffinanciallyweakandhealthyfirms,respectively.The negativeaveragecashflows(−0.5%)andthehighstandard devia-tion(1.2%)forthefullsampleinPanelAcomefromweakfirms (PanelB),whichhavemeancashflowsthatarenegative(−10.6%) withahighstandarddeviation(1.8%).Incontrast,healthyfirms (PanelC)exhibitmeancashflowsof7.1%,withalowstandard devi-ation(0.1%).Whilecashflowsdeclineforweakfirmswhenmoving fromnormaltimes(−9.4%)tothepre-2007crises(−10.5%)andto therecentfinancialcrisis(−24.4%),thecashflowsofhealthyfirms remainfairlystablethroughtheseeconomicstates(between7.0% and8.6%).Weakfirmsalsoexhibithighermarket-to-bookratios. Moreover,comparedtohealthyfirms,theydistributefewerfunds toshareholdersandissuemoreequity.Theyalsoborrowmoreand
13Asarobustnesscheck,wealsouseAltman’s(1968)Z-scoretoassessthe
Table1
Descriptivestatistics. PanelA:fullsamples
Fullsample 1972–2011 Normaltimes 1972–2011 Pre-2007crises 1972–2006
Recentfinancialcrisis 2007–2009
N Mean SD N Mean SD N Mean SD N Mean SD
Cashflow 217,900 −0.005 1.163 143,501 −0.001 0.627 62,086 −0.006 1.923 12,313 −0.054 0.828 Capitalexpenditures 217,900 0.065 0.086 143,501 0.066 0.086 62,086 0.066 0.088 12,313 0.051 0.078 Acquisitions 217,900 0.016 0.062 143,501 0.016 0.062 62,086 0.014 0.062 12,313 0.023 0.073 Assetsales 217,900 0.010 0.092 143,501 0.009 0.082 62,086 0.013 0.118 12,313 0.005 0.053 Sharerepurchases 217,900 0.010 0.055 143,501 0.009 0.057 62,086 0.009 0.051 12,313 0.018 0.059 Dividends 217,900 0.011 0.097 143,501 0.011 0.103 62,086 0.011 0.087 12,313 0.011 0.077 Equityissues 217,900 0.061 0.234 143,501 0.065 0.230 62,086 0.052 0.248 12,313 0.059 0.194 Long-termdebt 217,900 0.008 0.921 143,501 0.013 0.316 62,086 −0.003 1.653 12,313 −0.003 0.253 Short-termdebt 217,900 0.000 0.473 143,501 0.002 0.353 62,086 −0.004 0.674 12,313 −0.004 0.454 Cashbalances 217,900 0.004 0.511 143,501 0.008 0.529 62,086 −0.002 0.417 12,313 −0.022 0.689 Market-to-book 217,900 1.442 0.886 143,501 1.445 0.897 62,086 1.399 0.848 12,313 1.619 0.917 Firmsize 217,900 4.447 2.338 143,501 4.465 2.288 62,086 4.199 2.400 12,313 5.486 2.306 PanelB:financiallyweakfirms
Fullsample 1972–2011 Normaltimes 1972–2011 Pre-2007crises 1972–2006
Recentfinancialcrisis 2007–2009
N Mean SD N Mean SD N Mean SD N Mean SD
Cashflow 87,155 −0.106 1.825 57,397 −0.094 0.967 24,834 −0.105 3.033 4924 −0.244 1.276 Capitalexpenditures 87,155 0.067 0.104 57,397 0.068 0.103 24,834 0.069 0.107 4924 0.048 0.088 Acquisitions 87,155 0.012 0.058 57,397 0.012 0.059 24,834 0.010 0.055 4924 0.016 0.069 Assetsales 87,155 0.015 0.141 57,397 0.013 0.125 24,834 0.020 0.180 4924 0.006 0.077 Sharerepurchases 87,155 0.008 0.066 57,397 0.008 0.069 24,834 0.008 0.062 4924 0.009 0.052 Dividends 87,155 0.009 0.148 57,397 0.009 0.159 24,834 0.008 0.128 4924 0.009 0.107 Equityissues 87,155 0.115 0.348 57,397 0.122 0.339 24,834 0.098 0.377 4924 0.118 0.287 Long-termdebt 87,155 −0.006 1.444 57,397 0.004 0.468 24,834 −0.028 2.605 4924 −0.001 0.352 Short-termdebt 87,155 −0.005 0.733 57,397 −0.002 0.542 24,834 −0.012 1.053 4924 −0.014 0.707 Cashbalances 87,155 −0.006 0.795 57,397 0.004 0.823 24,834 −0.017 0.648 4924 −0.066 1.080 Market-to-book 87,155 1.560 1.066 57,397 1.549 1.069 24,834 1.541 1.047 4924 1.789 1.097 Firmsize 87,155 2.397 1.453 57,397 2.463 1.418 24,834 2.039 1.400 4924 3.436 1.502 PanelC:financiallyhealthyfirms
Fullsample 1972–2011 Normaltimes 1972–2011 Pre-2007crises 1972–2006
Recentfinancialcrisis 2007–2009
N Mean SD N Mean SD N Mean SD N Mean SD
Cashflow 87,159 0.071 0.121 57,402 0.070 0.118 24,833 0.070 0.129 4924 0.086 0.118 Capitalexpenditures 87,159 0.064 0.067 57,402 0.066 0.068 24,833 0.063 0.063 4924 0.053 0.065 Acquisitions 87,159 0.019 0.062 57,402 0.019 0.059 24,833 0.017 0.065 4924 0.027 0.072 Assetsales 87,159 0.006 0.029 57,402 0.005 0.024 24,833 0.007 0.039 4924 0.004 0.027 Sharerepurchases 87,159 0.012 0.046 57,402 0.011 0.047 24,833 0.010 0.041 4924 0.026 0.060 Dividends 87,159 0.014 0.034 57,402 0.014 0.029 24,833 0.016 0.042 4924 0.013 0.035 Equityissues 87,159 0.017 0.056 57,402 0.018 0.059 24,833 0.015 0.053 4924 0.015 0.045 Long-termdebt 87,159 0.016 0.127 57,402 0.018 0.122 24,833 0.013 0.137 4924 0.004 0.130 Short-termdebt 87,159 0.003 0.093 57,402 0.004 0.083 24,833 0.001 0.112 4924 0.000 0.095 Cashbalances 87,159 0.008 0.077 57,402 0.008 0.076 24,833 0.007 0.077 4924 0.008 0.089 Market-to-book 87,159 1.354 0.698 57,402 1.367 0.719 24,833 1.292 0.634 4924 1.511 0.726 Firmsize 87,159 6.407 1.641 57,402 6.378 1.623 24,833 6.299 1.637 4924 7.298 1.606 Thistableshowsthenumberoffirm-yearobservations(N),themean,andthestandarddeviation(SD)ofallthesource,use,andcontrolvariablesofthemulti-equation modelinEq.(3).Allnumbers,exceptformarket-to-bookandfirmsize,areexpressedasfractionsoffirmtotalassets.Market-to-bookistrimmedatthe95%level.Thefull sampleconsistsof217,900firm-yearsfromCompustatNorthAmericabetween1972and2011.PanelAsummarizesthedescriptivestatisticsforthefullsample,andPanels BandCshowthedescriptivestatisticswhenthefullsampleissubdividedintofinanciallyhealthyandweakfirmsusingthesize–ageindex(SA-index)ofHadlockandPierce (2010).TheSA-indexiscomputedas(−0.737×Size)+(0.043×Size2)−(0.040×Age),whereSizeisthelogoftotalassets,andAgeisthenumberofyearsthefirmiscontained
inCompustatwithanon-missingstockprice.Firm-yearswithanSA-indexabovethesixtiethpercentile(belowthefortiethpercentile)areconsideredfinanciallyweak (financiallyhealthy).Weexcludeanyfirmsthatfallinbetween.AllvariablesaredefinedasinTableA1.
relylessoninternalfinancinginallsubperiods.Thisbehaviorof weakfirmssupportsthefindingsintheliterature(Fama&French, 2005;Gatchev,Spindt,&Tarhan,2009).
2.5. Estimatingtheinvestment-cashflowsensitivity: single-versusmulti-equationmodels
Beforewepresentourmainfindings,wedocumentthe superi-orityofourmulti-equationmodelbycomparingestimatesofthe investment-cashflow sensitivitiesbased onthesingle- andthe
multi-equationmethodology.PanelAofFig.1comparesthe esti-matesofinvestment-cashflowsensitivitiesforthepooledsample offinanciallyweakandhealthyfirmsovernon-crisisandliquidity crisisperiodsusingboththesingleandmulti-equationapproach. AsinChenandChen(2012),wefindthatthesingle-equation esti-matesofinvestment-cashflowsensitivitieswereunaffectedduring therecentfinancialcrisis;infact,theyactuallydeclined.Incontrast, theestimatesobtainedfromourmulti-equationmodelshowthat investment-cashflow sensitivities increased duringthis period.
Panel A: All Firms Panel B: Financially Weak versus Healthy Firms 0.00 0.10 0.20 0.30 0.40 0.50 0.60 1972-1974 1975 1976-19831984-19911992-20012002-20032004-20062007-20092010-2011 In vest ment -cash fl ow se n si ti vi ty
Crisis and non-crisis periods Multi-equation model
Single-equation model
0.00 0.10 0.20 0.30 0.40 0.50 0.60 1972-1974 1975 1976-19831984-19911992-20012002-20032004-2006 2007-20092010-2011 In vest men t-cash fl ow se n si ti vi ty
Crisis and non-crisis periods Multi-equation model - financially weak firms Multi-equation model - financially healthy firms Single-equation model - financially weak firms Single-equation model - financially healthy firms
Fig.1.Comparisonofsingle-andmulti-equationestimates.
Thisfigureplotstheinvestment-cashflowsensitivitycoefficientsusingtwodifferentmodels.First,thesingle-equationmodelisdefinedasinFazzarietal.(1988)andChen andChen(2012):PPENTCAPXi,t−1i,t =at+ai+b1×Qi,t−1+b2×
CFLi,t
PPENTi,t−1+εi,t,where
CAPXi,t
PPENTi,t−1 isthefirm’scapitalexpenditures,deflatedbyitsbeginning-of-periodnetproperty,
plant,andequipment,Qi,t−1isthepreviousyear’sTobin’sq,andPPENTCFLi,ti,t−1 isthefirm’sinternalcashflow(depreciationandamortizationplusincomebeforeextraordinary
items),deflatedbyitsbeginning-of-periodnetproperty,plant,andequipment.b2isthesingle-equationinvestment-cashflowsensitivitycoefficient.Themodelincludes
withinfirmandyearfixedeffects.Second,themulti-equationmodelisspecifiedasinEq.(3),andestimatedsubjecttotheconstraintsinEq.(4).Allthevariablesofthismodel aredefinedasinTableA1.Bothmodelsareestimatedovertime,usingninesubsamplesthatreflectthecrisisandnon-crisisperiodsinoursample,whicharedescribedin Section2.3.InPanelA,themodelisestimatedusingallfirmsinthesample;inPanelB,wedifferentiatebetweenfinanciallyweakandhealthyfirmsbyusingthesize–age-index (SA-index)ofHadlockandPierce(2010).Theindexiscomputedas(−0.737×Size)+(0.043×Size2)−(0.040×Age),whereSizeisthelogoftotalassets,andAgeisthenumber
ofyearsthefirmiscontainedinCompustatNorthAmericawithanon-missingstockprice.Firm-yearswithanSA-indexabovethesixtiethpercentile(belowthefortieth percentile)areconsideredfinanciallyweak(financiallyhealthy).
Therefore,thesingle-equationestimatesfailtoverifythefindings oftherecentliteratureonthesubprimemortgagecrisis.
Whenweexaminefinanciallyhealthyandweakfirmsseparately inPanelBofFig.1,theestimatesfromthesingle-equationmodel suggestthat,evenforfinanciallyweakfirms,theinvestment-cash flowsensitivitywasnotaffectedduringtherecentfinancialcrisis. Incontrast,theinvestment-cashflowsensitivitiesfromthe multi-equationmodelconfirmthattheimpactoftherecentfinancialcrisis oncapitalexpenditureswasparticularlysevereforfinanciallyweak firms.Mostimportant,themulti-equationestimateforthecapital expenditure-cashflowsensitivitycoefficientsoarsto0.554, imply-ingthatfinanciallyweakfirmshadtocutcapitalexpendituresby $0.554inresponsetoa$1shortfallincashflowsduringtherecent financialcrisis.14 Theevidenceof thesingle-equationestimates
thatevenfinanciallyweakfirmswereabletoraisecapitalduring
14 The$0.55cutincapitalexpendituresoffinanciallyweakfirmsduringthisperiod
isabouteighteentimeslargerthantheestimateforthesamefirmsduringthethree previousliquiditycrisesthatweconsiderhere($0.03versus$0.55;thedifference issignificantatthe1%level).Whenweestimateamodifiedversionofthe single-equationmodelthatincludeslaggedcapitalexpendituresasanexplanatoryvariable, weobservethatinvestment-cashflowsensitivitiesdecline.However,theyremain muchhigherthantheestimatesfromthemulti-equationmodel,whereeach equa-tionincludesnotonlyitsownlaggeddependentvariable,butalllaggeddependent variablesinthesystem.Thepresenceofapositiveomittedvariablesbiasmayexplain thepuzzlereportedinearlierstudiesthatestimatesofinvestment-cashflow sen-sitivitiesarelargeandsignificantevenforsubsamplesoffinanciallyhealthyfirms. Previousstudiesprovidenoexplanationforthiscontradictoryresult.Wepositthatit isanartifactoftheomittedvariablesbias,whichinflatesthesizeofinvestment-cash flowsensitivitiesforbothtypesoffirms.
suchasevereliquiditycrisisissurprising,anditraisesconcerns abouttheabilityofasingle-equationmodeltoadequatelycapture capitalmarketaccessconstraints.
3. Behavioroffirmsduringcrisisandnon-crisisperiods
Thequestionof how investment-cashflowsensitivities vary withliquidityconditionsoftheeconomyhasreceivedlittle atten-tion in theliterature. Anecdotal evidence suggeststhat, during severeliquiditycrises,firmsfacemoredifficultiesinraisingfunds tofinancetheirinvestmentsand operations. In Section3.1,we examine whetherincludingtherecent financial crisis,which is consideredtobethemostsevereliquiditycrisissincetheGreat Depression(Brunnermeier,2009), affects cash flow sensitivities estimates.Section3.2proceedsbyestimatingourmulti-equation modelfordifferentsubsamplesofcrisisandnon-crisisperiodsto analyzehowcapitalmarketaccessconstraintsincreasewiththe severityofa crisis.Section 3.3analyzesthesymmetry offirms’ reactionstocashflowchanges.Sincethepooledsampleoffirms couldmaskimportantdifferencesbetweenfinanciallyweakand healthyfirms,inSection3.4weestimateourmodelseparatelyfor financiallyweakandhealthyfirms.
3.1. Theeffectoftherecentfinancialcrisisoncashflow sensitivities
Inthissection,weassesswhethertherecentfinancialcrisiswas indeedsevereby comparingthecash flow sensitivityestimates
thatresultfromourmulti-equationmodelforsampleperiodsthat excludeandincludethe2007–2009crisis.Wefirstestimatethe systemofequationsmodelinEq.(3),subjecttotheconditionsin Eq.(4),forthefullsampleperiodandforthefullsampleperiod excludingthe2007–2009crisis.Wethencomparethetwosetsof estimatestoassesswhethertheinclusionofsuchaseverecrisis producesestimatesthatsuggestfirmsfacefrictionsinraisingfunds. Weimplementourtestsseparatelyforboththepooledsampleof firmsandforfinanciallyweakandhealthyfirms.
PanelAofTable2displaystheestimatesofthecashflow sen-sitivitiesfortheninevariablesinourmulti-equationmodel(the elementsofvectorL).Allvariablesaremeasuredinlevels,andto accountforyearfixedeffects,wesubtractannualmeansfromeach variable.15Forallsubsampleestimations,webasetheweightsand
annualmeansontherespectivesubsamples.Columns(1)and(4) showtheresultsfor thepooled sampleoffinanciallyweakand healthyfirms.Theresultsforthefinanciallyweakfirm subsam-plearepresentedincolumns(2)and(5),whilecolumns(3)and(6) displaytheresultsforthefinanciallyhealthyfirmsubsample.Panel Bshowsthestratifiedbootstraptestsfordifferencesbetweenthe estimates.16
Ifincludingtherecentfinancialcrisisgeneratescashflow sen-sitivitiesthatindicatefirmsfacehigherhurdlesinraisingfunds, thentheestimatesofthetraditionalcapitalaccessconstraint mea-surecouldalsobeaffected.Afirstresultthatisreassuringforour modelisthat48ofthe54estimatedcoefficientsinTable2havethe expectedsigns.Forexample,column(1)showsthatwhenfirms experiencea $1decline in cash flows(a sourcevariable), they reduceinvestments(both capital expenditures and acquisitions decline),increaseassetsales,issueequity,reduceshareholder pay-outsintheformofbothdividendsandsharerepurchases,increase theirshort-andlong-termborrowings,anddrawdowntheircash holdings.
Column(1)ofPanelAshowsourresultsfromthefullsample periodexcludingthe2007–2009financialcrisisforthepooled sam-pleoffinanciallyweakandhealthyfirms,whilecolumn(4)contains theresultsforthepooledsampleoffirmswhentherecentfinancial crisisisincluded.Stratifiedbootstrapdifferencetestsofcolumn(1) minuscolumn(4)areshownincolumn(1)ofPanelB.Withthe exceptionofcashbalances(significantatthe10%level),allpaired differencesintheestimateslackstatisticalsignificance.Inclusion orexclusionoftherecentfinancialcrisisdoesnotaffectthe esti-matesforthepooledsampleoffinanciallyweakandhealthyfirms. However,theseresultsmaymaskdifferencesbetweenfinancially weakandhealthyfirms,andwenextestimatethemodelforthese twotypesoffirmsseparately.
Theestimatesforfinanciallyhealthyfirmsareincolumns(3) and (6)of PanelA. Thepatternsdisplayed byhealthy firmsfor thefullsampleperiodwithandwithouttherecentfinancial cri-sisaresimilartothebehaviorofthepooledsampleofweakand healthyfirms.Infact,thestratifiedbootstrap differencetestsin column(5)ofPanelBshowthatagainonlycashbalancesexhibit marginalstatisticalsignificance.Incontrast,theestimatesfor finan-ciallyweakfirmsincolumns(2)and(5)ofPanelAaresensitive totheinclusion/exclusionoftherecentfinancialcrisisinthefull sampleperiod.17Weakfirmsareadverselyaffectedbytherecent
15 Asarobustnesstest,werepeattheanalysismeasuringvariablesinfirst
differ-encesusingCleary’s(1999)methodology.Ourresults(notreported)arerobustwith respecttohowvariablesaremeasured.
16 Webootstrapthep-valuesofthedifferencetestusingstratifiedsamplingwith
1000iterations.Forthefullsample,wedividefirmsintothreegroupsofequal size(financiallyweak,neutral,andfinanciallyhealthy),usingHadlockandPierce’s (2010)size–ageindexasameasureoffinancialconstraints.
17 Whenweexcludetherecentfinancialcrisisfromthesampleperiod,column(2)
showsthatsixofthenineestimatedcashflowsensitivitiesoffinanciallyweakfirms
financialcrisis,andincludingthe2007–2009crisisinthesample resultsinabiggercutincapitalexpendituresandinlower long-termborrowingswhenfirmsfacecashflowshortfalls(column(4) ofPanelB).
Thetraditionalmetricofconstrainedaccesstoexternal capi-talmarketstestswhetherthedifferencesbetweentheestimates forfinanciallyweakandhealthyfirmsaresignificantlydifferent, asreportedincolumns(2)and(3)of PanelB.Whentherecent financialcrisisisexcludedfromthefullsampleperiod (column (2)),weobservethatonlythedifferencesinsharerepurchasesand inchangesincashholdingsshowmarginalstatisticalsignificance. However,whenthiscrisisisincludedinthefullsampleperiod (col-umn(3)),wefindthatweakfirmscuttheircapitalexpenditures significantlymorethanhealthyfirms,theyborrowlessshort-and long-termfunds,issuemoreequity,andrelylessoncashbalances (lessinternalborrowing).
To summarize, neitherthe estimates for financially healthy firmsnorthepooledsampleoffinanciallyhealthyandweakfirms areaffectedbytheinclusion/exclusionoftherecentfinancialcrisis inthesampleperiod.However,financiallyweakfirmsareaffected adverselywhenthiscrisisisincludedinthesampleperiod.When thedataincludestherecentfinancialcrisis,thetraditionallydefined capitalmarketfrictionsbecometighter,suggestingthatthe occur-renceofliquiditycrisesispositivelycorrelatedwiththetraditional measureofcapitalmarketfrictions.
3.2. Capitalmarketaccessconstraintsandcrisisseverity:the pooledsampleoffirms
We next examine whetherthe link betweencapital market accessconstraintsofthepooled sampleoffinanciallyweakand healthyfirmsincreaseswithliquiditycrisisseverity.Inparticular, weestimateourmodelforthepooledsampleoffinanciallyweak andhealthyfirms,andcomparethecoefficientestimatesforthe fullsampleperiodandthethreesubsampleperiods:(1)non-crisis years,(2)thethreemoderatepre-2007liquiditycrises,and(3)the 2007–2009crisis.Again,only theestimatesfor theelementsof vectorLarereportedinTable3.18Thetableincludestwo
addi-tionalrowsfortotalborrowings:Totaldebt,definedasthesumof theestimatesforshort-andlong-termdebtissues,andLeverage, definedastotaldebtissuespluschangesincashholdings.
TherearesixmainfindingsinTable3.First,PanelAshowsthat thedirectionsoftheresponsesarerobusttothechoiceofsample period.Infact,exceptforequityissues(whichareeconomically small),theestimatesforthecashflowsensitivitieshavethesame signsinallfoursampleperiods.Second,33ofthe36estimated coef-ficientshavetheexpectedsign.Inresponsetoacashflowshortfall,
havetheexpectedsigns.Onlyoneofthefivestatisticallysignificantcoefficientshas an“incorrect”sign.Thecoefficientforfinanciallyweakfirms’assetsalesispositive andstatisticallysignificantatthe5%level,indicatingthatweakfirmsrespondto cashflowshortfallsbyreducingassetsales(althoughthe$0.01changeis economi-callysmall).Whenweincludethesubprimemortgagecrisis,eveneightofthenine coefficientsfortheweakfirmsubsampleincolumn(5)havethe“correct”sign,and sevenofthesearestatisticallysignificant.
18BothTables2and3donotreporttheestimatesfortheown-laggeddependent
variables(estimatesforthediagonalelementsofmatrixK)andthetwoexogenous variablesMBandSIZE(estimatesfortheelementsofmatrixM).Becausethese vari-ablesrepresentneithersourcesnorusesoffundsinthecurrentperiod,theestimates shouldsumtozeroacrossthesystemofequations.Theestimatesforlagged depen-dentvariables(notreported)areconsistentwiththenotionthatadjustingrealassets entailshighcosts,whiletheadjustmentcostsoffinancingvariablesisrelativelylow. Thereismorepersistenceinrealassets,andignoringtheirlagsinthespecifications ofcashflowsensitivityregressionscancausesignificantomittedvariablebiases. Moredetailedresultsareavailableuponrequest.
Table2
Fullsampleestimates. PanelA:cashflowsensitivities
Fullsample1972–2011(excluding2007–2009) Fullsample1972–2011
Allfirms(1) Weakfirms(2) Healthyfirms(3) Allfirms(4) Weakfirms(5) Healthyfirms(6)
Capitalexpenditurest 0.017 0.021 0.020 0.031 0.316** 0.030 Acquisitionst 0.021 0.017*** 0.017 0.033 0.021** 0.028 Assetsalest −0.006 0.010** −0.007 −0.005 0.023* −0.005 Sharerepurchasest 0.041* −0.006 0.038 0.044* 0.004 0.042 Dividendst 0.028 0.026*** 0.028 0.030* 0.022** 0.028 Equityissuest −0.001 −0.040 −0.001 0.002 −0.087** 0.002 Long-termdebtt −0.417*** −0.461*** −0.416*** −0.307*** −0.118** −0.310*** Short-termdebtt −0.400*** −0.476*** −0.408*** −0.395*** −0.346*** −0.404*** Cashbalancest 0.068 −0.025 0.065 0.157*** 0.109*** 0.154*** Usest+Sourcest 1.000 1.000 1.000 1.000 1.000 1.000 Numberofobservations 205,587 82,230 82,235 217,900 87,155 87,159
PanelB:bootstrapdifferencetests
Differenceincashflowt (1):(1)−(4)
Differenceincashflowt (2):(2)−(3)
Differenceincashflowt (3):(5)−(6)
Differenceincashflowt (4):(2)−(5)
Differenceincashflowt (5):(3)−(6) Capitalexpenditurest −0.014 0.001 0.285** −0.295** −0.010 Acquisitionst −0.012 0.000 −0.007 −0.004 −0.011 Assetsalest −0.002 0.017 0.029* −0.013 −0.001 Sharerepurchasest −0.003 −0.044* −0.038** −0.010 −0.004 Dividendst −0.002 −0.002 −0.006 0.004 0.000 Equityissuest −0.003 −0.039 −0.089** 0.047 −0.003 Long-termdebtt −0.111 −0.045 0.192** −0.343** −0.106 Short-termdebtt −0.005 −0.067 0.058* −0.129 −0.004 Cashbalancest −0.090* −0.089* −0.046* −0.134 −0.090*
ThistablereportstheestimatesforcashflowsensitivitiesfromthesystemofequationsinEq.(3),subjecttotheconstraintsspecifiedinEq.(4).Allvariablesaredefinedasin TableA1.Thefullsampleconsistsof217,900firm-yearsfromCompustatNorthAmericabetween1972and2011.Theregressionsareestimatedusinglevels.PanelApresents theestimatesforthesampleexcludingtherecent2007–2009financialcrisisincolumn(1);thefullsampleresultsareincolumn(4).Bothsamplesarefurthersubdivided intofinanciallyweak(columns(2)and(5))andfinanciallyhealthyfirms(columns(3)and(6)),usingtheSA-indexofHadlockandPierce(2010).TheSA-indexiscomputed as(−0.737×Size)+(0.043×Size2)−(0.040×Age),whereSizeisthelogoftotalassets,andAgeisthenumberofyearsthefirmiscontainedinCompustatwithanon-missing
stockprice.Firm-yearswithanSA-indexabovethesixtiethpercentile(belowthefortiethpercentile)areconsideredfinanciallyweak(financiallyhealthy).PanelBreports pairwisedifferencetestsforthecoefficientsinPanelA.Thep-valuesoftheestimatedcoefficientscontrolforfirm-levelclusteringusingRogers’(1993)method.Thep-values ofthedifferencetestarebootstrappedusingstratifiedsamplingand1000iterations.***,**,and*denotesignificanceatthe1%,5%,and10%levels,respectively.
investmentandpayoutvariablesdecline,whileassetsales, borrow-ings,anduseofcashbalancesincrease.19
Third,PanelAshowsthattheestimatesforthecashflow sen-sitivityofcapitalexpendituresincreasemonotonicallyasliquidity conditionsintheeconomybecometighter.However,theestimates areeconomicallysmall–rangingfrom0.007to0.041–andtheir statisticalsignificanceduring themoderate andsevereliquidity crisisperiodsisonlymarginal(columns(3)and(4)).Wefindno statisticalsignificanceforanyofthestratifiedbootstrapdifference testsinPanelBthatcomparethecashflowsensitivityofcapital expendituresinthethreeliquidityregimes.
Wenote thata recurring findingin ouranalysesis thatthe biggestreactionsto cash flow shocks, and mostof the statisti-callysignificantcoefficients,areobservedintheleverageequations. In fact, cash flow sensitivities of short-term borrowings, long-termdebt issues, and changes in cash balances dominate both investmentand distributionresponses.20 In other words,while
single-equationmodelsfocus on capital expenditures todetect capitalmarketfrictions, ourresultsindicatethatcapital market frictionsarereflectedprimarilyinborrowingvariablesandcash balancesratherthanincapitalexpenditures.
19 Onlythecashflowsensitivityestimatesforequityissuesinthefullsampleperiod
andthetwoliquiditycrisissubsampleperiodshave“incorrect”signs;however,none ofthethreeequityissuecoefficientswiththe“incorrect”signisstatistically signif-icant.Fortheothervariablesinthemodel,20ofthecoefficientsarestatistically significant.
20 Becausethecoefficientestimatesforequityissuesarebotheconomicallysmall
andstatisticallyinsignificantforallthreesubperiods,wedonotcommentonthem here.Cashflowsensitivitiesofshareholderdistributions(dividendsandshare repur-chases)arealsorelativelysmall.
Fourth,themoreseveretheliquiditycrisis,thetighterarethe restrictionsfirmsfaceindebtmarkets.Whenfirmsexperiencea $1shortfallincash flowsduringnon-crisis years,theyincrease theirtotal borrowingsby $0.843 and theirleverage by $0.905. Whentheshortfallincashflowsoccursduringthethreepre-2007 crises,weobservethatfirmsuseslightly lessdebt($0.776)and leverage($0.873),suggestingthatevenmoderateliquiditycrises havesomeadverseeffectsonfirms’financingactivities.Totaldebt declinesevenmoredramaticallyduringtherecentfinancialcrisis ($0.475).Onecouldarguethatusing$0.067lessborrowingsand having$0.032worthoflowerleverageduringthemoderatecrisis episodesrelativetonon-crisisyearsisnoteconomically meaning-ful.However,attimesthepresenceofcapitalmarketfrictionsmay beobservedinthecompositionofthefinancingresponse,rather thaninitssize.
Fifth,duringmoderateliquiditycrises,firmssubstitute short-termborrowingsforlong-termdebt;however,wefindnoevidence for sucha substitution effectduring the recent financial crisis. Whenanegative$1cashflowshockoccursduringnon-crisisyears, firmsborrow$0.478oflong-termfunds(57%oftotaldebt).In con-trast,duringthepre-2007liquiditycrises,long-termdebtissues dropto$0.187(24%oftotaldebt),suggestingthatfirmsface dif-ficultiesinissuinglong-termdebt(orlong-termdebtmaybetoo costly)evenduringmoderateliquiditycrises.
The decline in long-term borrowings is consistent withthe viewthattheeffectsof financingconstraintsaremorelikelyto bedetectedinfirms’long-termdebtissues thanin their short-termborrowings(Gatchevetal., 2009).Thestratifiedbootstrap differencetestsincolumns(1)of PanelBindicatethat, relative tonon-crisisyears,the$0.292decline inlong-termdebtis
sta-Table3
Investmentandfinancingcashflowsensitivities:crisisversusnon-crisisperiods. PanelA:cashflowsensitivitiesusinglevels
Fullsample1972–2011 Normaltimes1972–2011 Pre-2007crises1972–2006 Recentfinancialcrisis2007–2009
(1) (2) (3) (4) Capitalexpenditurest 0.031 0.007 0.029* 0.041* Acquisitionst 0.033 0.010 0.046 0.039 Assetsalest −0.005 −0.006* −0.001 −0.009 Sharerepurchasest 0.044* 0.042** 0.034** 0.070 Dividendst 0.030* 0.028* 0.022** 0.024** Equityissuest 0.002 −0.001 0.004 0.005 Long-termdebtt −0.307*** −0.478*** −0.187*** −0.093 Short-termdebtt −0.395*** −0.365*** −0.589*** −0.382*** Cashbalancest 0.157*** 0.062 0.097*** 0.348*** Usest+Sourcest 1.000 1.000 1.000 1.000 Totaldebtt 0.702 0.843 0.776 0.475 Leveraget 0.859 0.905 0.873 0.822 Numberofobservations 217,900 143,501 62,086 12,313
PanelB:bootstrapdifferencetests
Differenceincashflowt Differenceincashflowt Differenceincashflowt
(1):(2)−(3) (2):(2)−(4) (3):(3)−(4) Capitalexpenditurest −0.022 −0.034 −0.012 Acquisitionst −0.036 −0.029 0.007 Assetsalest −0.006 0.002 0.008 Equityissuest 0.008 −0.028 −0.036 Sharerepurchasest 0.007 0.004 −0.003 Dividendst −0.005 −0.006 −0.001 Long-termdebtt −0.292*** −0.386*** −0.094*** Short-termdebtt 0.224* 0.017 −0.207** Cashbalancest −0.035 −0.286*** −0.251*** Totaldebtt 0.068 0.369* 0.301* Leveraget 0.033 0.083 0.050
ThistablereportstheestimatesforcashflowsensitivitiesfromthesystemofequationsinEq.(3),subjecttotheconstraintsspecifiedinEq.(4),fordifferentsampleperiods. AllvariablesaredefinedasinTableA1.Twoadditionalvariablesareincluded:(1)Totaldebtt,whichrepresentstheabsolutesumoftheestimatesforLong-termdebtt
andShort-termdebtt,and(2)Leveraget,whichaddsCashbalancesttoTotaldebtt .Thefullsampleconsistsof217,900firm-yearsfromCompustatNorthAmerica
between1972and2011.Allregressionsareestimatedusinglevels.PanelAshowsthecoefficientestimatesforthefullsampleincolumn(1),thenon-crisisyearsofthe 1972–2011periodincolumn(2),theliquiditycrisisperiodspriorto2007incolumn(3),andtherecent2007–2009financialcrisisincolumn(4).Section3.2providesa detailedexplanationofthecrisisandnon-crisisperiods.PanelBreportspairwisedifferencetestsforthecoefficientsfromPanelA.Thep-valuesoftheestimatedcoefficients controlforfirm-levelclusteringusingRogers’(1993)method.Thep-valuesofthedifferencetestarebootstrappedusingstratifiedsamplingand1000iterations.Firmsare dividedintothreestrata(financiallyweak,neutral,andhealthy)ofequalsizeusingtheSA-indexofHadlockandPierce(2010)asameasureoffinancialhealth.Theindexis computedas(−0.737×Size)+(0.043×Size2)−(0.040×Age),whereSizeisthelogoftotalassets,andAgeisthenumberofyearsthefirmiscontainedinCompustatwitha
non-missingstockprice.Firm-yearswithanSA-indexabovethesixtiethpercentile(belowthefortiethpercentile)areconsideredfinanciallyweak(financiallyhealthy).***, **,and*denotesignificanceatthe1%,5%,and10%levels,respectively.
tisticallysignificant(atthe1%level)duringthepre-2007liquidity crises.Atthesametime,short-termborrowingsincreasefromtheir non-crisislevelof$0.365to$0.589;thischangeissignificantatthe 10%level.Thesubstantialshiftinthematuritycompositionofdebt towardshort-termborrowingsappearstobeabetterindicatorof thecapitalmarketfrictionsfirmsfaceduringthemoderateliquidity crisesthanthesmalldeclineintotaldebt($0.068).
Long-termdebtissuesdeclinefurtherduringtherecentfinancial crisisto$0.093(around20%oftotaldebt),whichisnotsignificantly differentthanzero.Whilefirmssubstituteshort-forlong-termdebt duringtheearliercrisesrelativetonon-crisisperiods,wedonot observethissubstitutioneffectduringtherecentfinancialcrisis.In fact,short-termdebtdeclinesrelativetothepre-2007crises(from $0.589to$0.382),implyingthatfirmsborrowessentiallythesame amountofshort-termfundsduringtherecentfinancialcrisisas theydoduringnon-crisisyears($0.365versus$0.384).Thereare severalpossibleexplanationsforthedifferencesbetweenthecash flowsensitivitiesofshort-termdebtduringtherecentfinancial cri-sisandearlierliquiditycrises.Onepossibilityisthatmorefirms violatedthecovenantsoftheirlinesof creditduringtherecent financialcrisis.Banksusedtheescapeclausestriggeredbythese technicaldefaults and renounced theirlines ofcredit contracts withasubstantialnumberoffirms,whichwerelikelyfinancially weakfirms.Whenweexaminethebehavioroffinanciallyweak
andhealthyfirms(seeTable5below),wewillbeableto deter-minewhetherthedifferenceincashflowsensitivitiesofshort-term debtbetweenfinanciallyweakandhealthyfirmsincreasesduring the2007–2009crisis.Anotherpossibilityisthat,asnotespayable matured,bankssimplydidnotrenewthem,eveniffirmswerenot intechnicaldefault(Ivashina&Scharfstein,2010;Almeidaetal., 2011).21
Sixth,whileleveragealsodecreasesmonotonically,reductions inleveragearemoremutedthandeclinesintotaldebt.Infact,the bootstrapdifferencetestsshowthatthedeclinesinleveragedonot reachstatisticalsignificance.Becausethedifferencebetweentotal debtandleverageisuseofcashbalances,theuseofcashbalances mustoffsetthedeclineintotaldebt.PanelAconfirmsthattheuse ofcashbalancesincreasesmonotonically.While,duringnon-crisis years,firmsdrawdowncashbalancesby$0.062aftera$1decline incashflows,thisestimateincreasesto$0.097duringthe moder-atepre-2007crises,andto$0.348duringtherecentfinancialcrisis (thetwolatterestimatesaresignificantatthe1%level).Moreover, becausecashrepresentsnegative(internal) debt,thedifficulties
21Thedefinitionofshort-termdebtincludesthecurrentportionoflong-termdebt.
Firmsmayhavebeenunabletooffsetthematuringcurrentportionofdebtwithnew long-termborrowings.
Table4
Testingforsymmetryofpositiveandnegativecashflowshocks. PanelA:fullsample
Positivecashflowshockt Negativecashflowshockt Differenceincashflowt
(1) (2) (3):(1)−(2) Capitalexpenditurest 0.044 0.003 0.040 Acquisitionst 0.038 0.024 0.014 Assetsalest −0.002 −0.010 0.008 Sharerepurchasest 0.045 0.041** 0.004 Dividendst 0.042 0.005 0.037 Equityissuest 0.008 −0.009** 0.016** Long-termdebtt −0.268* −0.400*** 0.132 Short-termdebtt −0.375*** −0.437*** 0.062 Cashbalancest 0.195** 0.071* 0.124 Usest+Sourcest 1.000 1.000 Numberofobservations 217,900 217,900
PanelB:financiallyweaksampleduringtherecentfinancialcrisis(2007–2009)
Positivecashflowshockt Negativecashflowshockt Differenceincashflowt
(1) (2) (3):(1)−(2) Capitalexpenditurest 0.090** 0.551*** −0.461** Acquisitionst 0.037 0.007 0.030 Assetsalest −0.003 0.041*** −0.045** Sharerepurchasest 0.062 0.004 0.059* Dividendst 0.052* 0.008 0.044 Equityissuest −0.116* −0.036 −0.080* Long-termdebtt −0.152 −0.157 0.005 Short-termdebtt −0.175*** −0.112 −0.063* Cashbalancest 0.313*** 0.168*** 0.145** Usest+Sourcest 1.000 1.000 Numberofobservations 4924 4924
ThistablereportstheestimatesforcashflowsensitivitiesfromthesystemofequationsinEq.(3),subjecttotheconstraintsspecifiedinEq.(4).Allvariablesaredefinedas inTableA1.Thefullsampleconsistsof217,900firm-yearsfromCompustatNorthAmericabetween1972and2011.InPanelA,themulti-equationsystemisestimatedfor thefullsample,withtheright-handsideincludinginteractionvariablesequaltothechangeincashflowsmultipliedbyadummyvariablethatisequaltoonewhenthecash flowvariableispositive,andzerootherwise.InPanelB,weestimatethemulti-equationsystemforthefinanciallyweakfirmsduringthe2007–2009financialcrisis,with theright-handsideincludinginteractionvariablesequaltothechangeincashflowsmultipliedbyadummyvariablethatisequaltoonewhenthechangeincashflowis positive,andzerootherwise.Thep-valuesoftheestimatedcoefficientscontrolforfirm-levelclusteringusingRogers’(1993)method.Thep-valuesofthedifferencetests (column(3))arebootstrappedusingstratifiedsamplingand1000iterations.Firmsaredividedintothreestrata(financiallyweak,neutral,andhealthy)ofequalsizesusing theSA-indexofHadlockandPierce(2010)asameasureoffinancialhealth.TheSA-indexiscomputedas(−0.737×Size)+(0.043×Size2)−(0.040×Age),whereSizeisthe
logoftotalassets,andAgeisthenumberofyearsthefirmiscontainedinCompustatwithanon-missingstockprice.***,**,and*denotesignificanceatthe1%,5%,and10% levels,respectively.
inraisingexternalfundscanthusalsobedetectedbycomparing therelativecontributionsofexternalandinternalborrowingsto leverage.Theuseofcashbalancesaccountsforonly7%ofleverage duringnon-crisisyears,butitincreasesto13%and42%duringthe moderatecrisesandtherecentfinancialcrisis,respectively.The lat-terpercentageprovidesadditionalevidenceoftheseverityofthe crisis.
Anotherobservationis that ourresultsconfirmthe findings in Bliss et al. (2013),who document a significantreduction in dividends and share repurchases during thefinancial crisis. In particular,a$1shortfallincashflowsinducesfirmstocut distri-butions(sharerepurchasesplusdividends)by$0.070,$0.056,and $0.094,duringnon-crisis,moderatecrisis,andthesubprime mort-gagecrisisperiods,respectively.However,thestratifiedbootstrap differencetestsreportedinPanelBshowthatnoneofthe differ-encesindividendsandsharerepurchaseestimatesisstatistically significant.
Overall, the results based on the pooled sample of finan-cially weakand healthy firms indicate that firms face frictions inraisingfunds duringliquiditycrises,and thatthesize ofthe obstaclestoraisingfundsincreaseswiththeseverityofliquidity crises.Wealsofindthatfirmsrespondtocashflowshortfallsby rearrangingtheirfinancing activities,especiallytheirborrowing decisions, rather than by changing their investments. Adjust-ing investment and distribution variables involves substantial costs,whichlikelyexplainsfirms’reluctancetouse
investment-and/or distribution-related instruments to counter cash flow shocks.22
3.3. Testingthesymmetryofthereactionstocashflowchanges Ourtestsassumethat afirm’sreactiontopositivecash flow changesisequalinmagnitudebutoppositeinsigntoitsreactionto negativecashflowshocks.However,thepresenceofcapital mar-ketfrictionsismoreaboutbeingabletoraiseexternalfundswhen facedwithnegativecashflowchangesthanitisaboutretiring capi-talinresponsetopositivecashflowchanges.Wethusinterpretour
22Weimplementtworobustnesstestsforthefullsampleandallsubsample
peri-ods.First,overtheyears,theU.S.economytransitionedfrombeingdominatedby industrialfirmswithlargecapitalexpenditurestobeingdrivenbytechnologyfirms, wheretheimportanceofcapitalexpendituresdeclinedandtheimportanceofR&D activitiesincreased.Wethusmodifytheinvestmentvariablebysummingcapital expendituresandR&Dexpenses.Ourresults(notreported)arerobusttoincluding R&Dexpenses.Second,weusequarterlyinsteadofannualdata.Thebenefitofthis approachisthatitallowsforamoreprecisedatingofthecrisisperiods.In par-ticular,fortherecentfinancialcrisis,wechoosethefourthquarterof2007asthe beginningandthefirstquarterof2009astheend(Kahle&Stulz,2013).Tocapture theintertemporalnatureoftheGatchevetal.(2010)modelwithquarterlydata,we expandthesystemofequationsinEq.(3)andtheconstraintsinEq.(4)to incorpo-ratefourquarterlytimelags,whichleadstoasystemofequationsmodelwith36 equationsandthirty-nineconstraints.Wefindthattheresults(notreported) pro-ducedbyquarterlydataarequalitativelysimilartothoseforannualdata,andthus concludethatthemodelisalsorobustwithrespecttofrequencyoftheobservations.
Table5
Investmentandfinancingcashflowsensitivities:effectsoffinancialhealth. PanelA:cashflowsensitivities
Normaltimes(1972–2011) Pre-2007crises(1972–2006) Recentfinancialcrisis(2007–2009)
Weakfirms Healthyfirms Weakfirms Healthyfirms Weakfirms Healthyfirms
(1) (2) (3) (4) (5) (6) Capitalexpenditurest 0.034 0.007 0.033*** 0.030 0.554*** 0.041* Acquisitionst 0.013** 0.005 0.014* 0.046 0.007 0.025 Assetsalest −0.001 −0.007 0.002 −0.001 0.042*** −0.010* Sharerepurchasest −0.006 0.039* 0.005 0.033** 0.004 0.076 Dividendst 0.028*** 0.028 0.015 0.022** 0.002 0.018* Equityissuest −0.035 −0.001 −0.092*** 0.003 −0.038* 0.005 Long-termdebtt −0.483*** −0.479*** −0.357*** −0.185*** −0.138 −0.100 Short-termdebtt −0.469*** −0.377*** −0.396*** −0.588*** −0.118 −0.393*** Cashbalancest −0.056 0.057 0.090*** 0.098*** 0.181*** 0.342*** Usest+Sourcest 1.000 1.000 1.000 1.000 1.000 1.000 Totaldebtt 0.952 0.856 0.753 0.774 0.256 0.494 Leveraget 0.896 0.913 0.843 0.872 0.437 0.836 Numberofobservations 57,397 57,402 24,834 24,833 4924 4924
PanelB:bootstrapdifferencetests Differencein cashflowt(1): (1)−(2) Differencein cashflowt(2): (3)−(4) Differencein cashflowt(3): (5)−(6) Differencein cashflowt(4): (1)−(3) Differencein cashflowt(5): (1)−(5) Differencein cashflowt(6): (3)−(5) Differencein cashflowt(7): (2)−(4) Differencein cashflowt(8): (2)−(6) Differencein cashflowt(9): (4)−(6) Capitalexpenditurest 0.027 0.002 0.513** 0.001 −0.520** −0.521** −0.023 −0.034 −0.011 Acquisitionst 0.008 −0.031 −0.017 −0.001 0.006 0.007 −0.040 −0.019 0.021 Assetsalest 0.007 0.003 0.052* −0.003 −0.042* −0.040* −0.007 0.003 0.010 Sharerepurchasest −0.045 −0.028* −0.072 −0.011** −0.010 0.002 0.006 −0.036 −0.042 Dividendst 0.000 −0.007 −0.016 0.013 0.026** 0.013 0.006 0.010 0.004 Equityissuest −0.035 −0.095*** −0.043 0.057 0.003 −0.054 −0.004 −0.006 −0.002 Long-termdebtt −0.004 −0.171** −0.038 −0.126** −0.345*** −0.219** −0.294** −0.379*** −0.085 Short-termdebtt −0.092 0.192* 0.275** −0.073*** −0.351* −0.278** 0.212* 0.017 −0.195** Cashbalancest −0.113* −0.008 −0.161** −0.146*** −0.237*** −0.091* −0.041 −0.285*** −0.244*** Totaldebtt 0.096 -0.021 -0.238 0.199** 0.696*** 0.497*** 0.082 0.362* 0.280* Leveraget -0.017 -0.029 -0.398** 0.053 0.458*** 0.406*** 0.041 0.077 0.036
ThistablereportstheestimatesforcashflowsensitivitiesforfinanciallyweakandhealthyfirmsfromthesystemofequationsinEq.(3),subjecttotheconstraintsspecifiedin Eq.(4),fordifferentsampleperiods.AllvariablesaredefinedasinTableA1.Twoadditionalvariablesareincluded:(i)Totaldebtt,whichrepresentstheabsolutesumofthe estimatesforLong-termdebttandShort-termdebtt,and(ii)Leveraget,whichaddsCashbalancesttoTotaldebtt .Thefullsampleconsistsof217,900firm-yearsfrom CompustatNorthAmericabetween1972and2011.PanelAshowsthecoefficientestimatesforthefullsampleincolumn(1),thenon-crisisyearsofthe1972–2011period incolumn(2),theliquiditycrisisperiodspriorto2007incolumn(3),andtherecent2007–2009financialcrisisincolumn(4).Section2.3providesadetailedexplanationof thecrisisandnon-crisisperiods.EachcolumninPanelAisfurthersubdividedintofinanciallyweakandhealthyfirmsusingthesize–age-index(SA-index)ofHadlockand Pierce(2010).TheSA-indexiscomputedas(−0.737×Size)+(0.043×Size2)−(0.040×Age),whereSizeisthelogoftotalassets,andAgeisthenumberofyearsthefirmis
containedinCompustatwithanon-missingstockprice.Firm-yearswithanSA-indexabovethesixtiethpercentile(belowthefortiethpercentile)areconsideredfinancially weak(financiallyhealthy).Weexcludefirmsinbetweenfromthisanalysis.PanelBreportspairwisedifferencetestsforthecoefficientsfromPanelA.Thep-valuesofthe estimatedcoefficientscontrolforfirm-levelclusteringusingRogers’(1993)method.Thep-valuesofthedifferencetestarebootstrappedusing1000iterations.***,**,and* denotesignificanceatthe1%,5%,and10%levels,respectively.
resultsinthecontextoffirms’responsestocashflowshortfalls. Totestwhetherfirmsreactsymmetricallytopositiveandnegative cashflowchanges,weestimateEq.(3),subjecttotherestrictionsin Eq.(4),byincludinganinteractionterm.Theinteractionterm repre-sentstheproductofcashflowsandadummyvariablethatisequal toonewhencashflowchangesarepositive,andzerootherwise. Table4showstheresultsofthissymmetrytest.
InPanelA,thesymmetrytestsareconductedforthefullsample period,whilePanelBshowstheresultsofsymmetrytestsforthe subsamplethatcoverstherecentfinancialcrisis.Inbothpanels, column(1)showsfirms’reactionstoa$1positivecashflowshock. Firms’reactiontoa$1shortfallincashflowsisshownincolumn (2).Tocorrectlyinterprettheresultsincolumn(2)ofbothpanels, thecoefficientsmustbemultipliedby−1.Finally,column(3)tests fordifferencesbetweenpositiveandnegativecashflowshocks.
In bothpanels, 17 of the18estimated coefficients havethe expectedsign.InPanelA,theunexpectedsignisobservedinthe equityissuesequation,whenfirmsexperienceapositivecashflow shock.Theestimate,whichissmallandstatisticallyinsignificant, indicatesthatfirmsincreasetheirequityissuesinresponseto pos-itivecashflow shocks.InPanelB, theunexpectedsignisinthe
assetsalesequation;firmsreduceassetsaleswhenthey experi-enceacashflowshortfall.Mostimportantly,ineightofthenine equationsforthefullsampleperiod,firmsrespondtopositiveand negativecashflowshockssymmetrically(onlyequityissueslack symmetry).Forexample,PanelAshowsthatfirmsborrow$0.437of short-termfundswhentheyexperiencea$1shortfallincashflows (column(2)).Conversely,whentheyexperiencea$1increasein cashflows,theyretire$0.375ofshort-termfunds(column(1));the resultingdifferenceisnotstatisticallydifferentfromzero(column (3)).
PanelBofTable4showstheresultsfromsymmetrytestsforthe recentfinancialcrisis.Theestimatesforcapitalexpendituresand cashbalancesaresignificantlydifferentinresponsetopositiveand negativecashflowshocks.Inparticular,a$1declineincashflows leadstoasubstantialcut($0.551)incapitalexpenditures(column (2)).Asymmetricresponsewouldrequirethatcapitalexpenditures increasebyasimilaramountwhenfirms’facea$1increaseincash flows.However,asshownincolumn(1)ofPanelB,a$1increase leadstoonlya$0.090increaseincapitalexpendituresduringthis crisis.