Business Cy le Sensitivity
Peter Pontu h
∗
This version: November 25,2011
Abstra t
Weanalyzetheintera tionsbetweennan ing onstraintsandprodu tmarket
ompe-tition. Finan ially onstrainedrmsfa erestri teda essto externalnan eduring
e o-nomi downturns,pre iselywhentheirinternalfundsde rease. Thisleadstovi ious ir le
dynami s. Wearguethatin ompetitiveindustries ash owsareparti ularlysensitiveto
aggregatesho ks,andtheadversedynami sareamplied. Wendsigni antsupportfor
thishypothesisin rms'operatingprotabilityandxedinvestment. Theadverseee ts
ofnan ing onstraintsarein reasinginthelevelofprodu tmarket ompetition. Market
valuationsdonottakeintoa ountthesedieren esin fundamental risk. Un onstrained
rmsin ompetitiveindustries earn positive abnormalreturns(on average24-40 bpper
month), espe ially following periods of ma roe onomi distress. Furthermore, nan ing
onstraintsae t ompetitiveme hanismswithinindustries. Theindustry-averagelevelof
nan ing onstraintstendstoredu e theintra-industrymean-reversionofrm
protabil-ity. Again,thisregularityisnotpri ed: highlyprotablermsearnalphasof20-29bpper
monthiftheyoperatein industrieswithmany onstrainedrms,but virtuallynoalphas
iftheirindustries havefew onstrainedrms.
Keywords: nan ing onstraints, produ t market ompetition, industry on entration,
business y le,protability,investment.
JELClassi ations: G32,L11.
∗
UniversitéParis-Dauphine-DRMFinan e-Pla eduMaré haldeLattredeTassigny-75775ParisCedex
16 -pontu hgmail. om. I am deeplygrateful to Gilles Chemla (my advisor) and Gordon Phillips. I also
thankMurilloCampello(dis ussantatFMAEuropeanConferen e2011),SandeepDahiya,Mi haelFaulkender,
DenisGromb,GerardHoberg,JayantKale,GaëlleLeFol,VojislavMaksimovi ,andubo²Pástorforvaluable
suggestions. Ialsothankseminarparti ipantsatUniversitéParis-Dauphine,theUniverstyofMaryland,College
Park, GeorgetownUniversity,ESCP Europe,and INSEADDo toral workshop fortheir ommentsand ideas.
Allerrorsare myown. Igratefully a knowledgenan ialsupportfromtheFBFChairinCorporateFinan e,
It is widely a knowledged that orporate earnings are pro y li al and more volatile than the
overall e onomy due to their junior status relative to other laims. However, there is a high
degreeof heterogeneityinearningspro y li alityat the rmlevel. Some ofthis pro y li ality
isdriven byindustrydemand(Gomes,Kogan,andYogo(2009)). Butrelatively littleisknown
about other rm- and industry-level determinants of this heterogeneity, despite the obvious
relevan eof thisquestion.
Inthispaperwefo usontwosu hdeterminants,namelynan ing onstraintsandprodu t
market ompetition. We show that the ee ts of these two fa tors are interrelated. The
intuition behind this idea is simple. The literature on nan ing onstraints suggests that
(i) onstrained rms' investment is more dependent on internal funds (see Fazzari, Hubbard,
and Petersen (1986) and subsequent studies), and (ii) onstrained rms are more subje t
to a ight to quality in apital markets at the onset of an e onomi downturn (Bernanke,
Gertler, and Gil hrist (1996)). A vi ious ir le dynami s is at play: onstrained rms fa e
restri teda essto external nan eexa tlywheninternal fundstendto fall. Thisfall implies
insu ient nan ing for urrent operations and investment proje ts, whi h further redu es
available internal fundset .
In this paperwe arguethat theseverity ofthese adverse dynami s dependson the initial
sho kto ashows. We proposeto study the level of ompetition intheprodu t marketasa
fa torae tingtheinitialsho ktointernalfunds. Con entratedindustriestendtobemore
in-sulatedfromaggregatesho ks(RotembergandSaloner(1986),HouandRobinson(2006)). The
initialadversesho ktointernalfundsduetoworseningma ro onditionswilllikelybestronger
forrmsin ompetitive industries. Thevi ious ir ledynami sinnan ially onstrainedrms
should thereforebe strongest inhighly ompetitive industries.
It hasbeen re ently do umented thatnan ing onstraints have signi ant real ee tson
rms'poli iesduringadversee onomi onditions. Campello,Graham,andHarvey(2010)
per-forma CFOsurveyduring themost re ent nan ial turmoil andshowthat redit onstrained
sig- onstraints have more real ee ts on rms in highly ompetitive industries ompared to
on- entrated industries.
We rst showthat nan ing onstraints in rease thema ro-sensitivity of rms'operating
performan e. Compared to highly onstrained rms, un onstrained rms areabout 30% less
sensitiveto GDPandabout50to75%lesssensitiveto hangesinthebondspread. Consistent
withourhypothesiswe ndthattheseadverse ee tsofnan ing onstraintsaresigni antly
amplied by produ tmarket ompetition. Stated dierently, thedieren e of business y le
sensitivity between onstrained and un onstrained rms is in reasing inthelevel of
ompeti-tion.
Wealsoexaminetherealee tsofnan ing onstraints. Wendthatthesensitivityofxed
investment to aggregate e onomi onditions is in reasing in nan ing onstraints. Industry
ompetition furtheramplies this sensitivityinthehighly onstrained subsample,whereasno
su hsigni antee t ispresent inthe low onstraints subsample. Theseresultsare onsistent
withtheidea thatindustry ompetition alsoamplies thereal ee tsofnan ing onstraints.
We also analyze investments in net working apital and inventories: theee t of
ompe-tition has the same sign but is not signi ant. Lastly, employment growth's sensitivity to
aggregate sho ks is somewhat in reasing in nan ing onstraints, but the ee t of industry
on entration seemsto be ompletely absent.
Our asset pri ing tests reveal that market valuations do not ompletely in orporate the
above dieren es in fundamental risk. We nd that portfolios of un onstrained ompanies
in ompetitive industries earn abnormal positive returns. A moving window analysisreveals
that this abnormal return is mostly earned following periods of e onomi turmoil. On the
otherhand, onstrainedrmsearnnegativereturnsintheseperiodsandvirtuallynoabnormal
returnsinnormalperiods.
Finally, we study how nan ing onstraints ae t ompetitive me hanisms in industries.
We present a set of results about ross-se tional mean-reversion of rm protability. At the
industry by 50 per entage points redu es the mean reversion speed by about 12 per entage
points. This suggests that nan ially onstrained ompetitors are less aggressive. Abnormal
levelsof protability anbesustained fora longerperiod inindustries witha highper entage
of onstrained rivals. Again, this regularity is not fully pri ed: higly protable rms earn
signi ant alphas of 20-29 bp per month if they operate inindustries withmany onstrained
rms,but virtually no alphasiftheir industries have few onstrained rms.
Ourpaper ontributes totheexistingliteratureontherealee tsofnan ing onstraints.
We show that produ t market ompetition is an important determinant of the severity of
the real ee ts of nan ing onstraints. Companies' operating performan e and investment
behavioraremoreae tedbynan ial onstraintsifthermoperatesinahighly ompetitive
industry.
Ourresultshaveimpli ationsfornan ialanalysisatboththermlevel(e.g. forvaluation
and risk assessment) and the aggregate level (e.g. nan ial a elerator ee ts). Our study
also has impli ations for poli y design. It points to the fa t thatrms deemed as nan ially
onstrained andoperatinginhighly ompetitive industriesarethemostfragile elementof the
orporate universe inan e onomi downturn, espe iallyifitisa ompaniedbyaworseningof
overall redit onditions. Poli ies aimed at improving a ess to nan ing for SMEs during a
redit run h should fo usparti ularly on highly ompetitiveindustries.
Thearti le isorganizedasfollows. Inse tion2we present relatedliteratureandformulate
our resear h hypotheses. We then present our data and dis uss measures of nan ing
on-straintsand industry on entration inse tion 3. Ourresults onthe sensitivityofprotability
levelsarepresentedinse tion4. Weanalyzetherealee tsofnan ing onstraintsinse tion
5. Assetpri ingtestsareperformedinse tion6. Se tion 7 presents our ndingson theee t
The omplex relationships between ma roe onomi u tuations, markups and market
stru -turewerestudied byHall(1986),whosuggests thatalarge numberofU.S. industriesoperate
above ompetitive markup levels. This fa thas impli ations for the ma roe onomi s of
pro-du tivitysho ks: non ompetitive stru tures ouldbe asour e ofunderutilization ofresour es
and aggregate u tuations. Other ontributions in this eld in lude Domowitz, Hubbard,
and Petersen (1986), Bils (1987), Rotemberg and Saloner (1986), Rotemberg and Woodford
(1991), and Haskel, Martin and Small (1995). Oliveira Martinsand S arpetta (2002) give a
brief survey of the related theoreti al and empiri al literature. They note that there is no
onsensusonwhethermarkupsarepro y li alor ounter y li al, thelatterrequiringgenerally
some non- ompetitive elements inthemodel(e.g. monopolisti ompetitionwith pro- y li al
produ t variety, or oligopoly with ollusions). For instan e, Rotemberg and Saloner (1986)
developamodelof ounter y li almarkupsinanoligopolisti marketwithimpli it ollusions.
Thebasi ideaofthe modelisthatthe ollusionagreement ismu hlessstableundere onomi
expansions(wheredemand ishighand punishments hurtless), than ine onomi downturns.
Elements of nan ing fri tions were formalized in the ontext of markups by Chevalier
and S harfstein (1996) whodevelop and test amodel of ounter y li al markupswith apital
market imperfe tions where onsumers fa e a swit hing ost. Finan ially onstrained rms
nan e their operations indownturns by in reasing relative pri esand preservingshort-term
prots,whilesa ri inglong-termmarketsharesandprots. Hen etheirmarkupsare
ounter- y li al, makingtheir prots smoother.
Howshouldonere on ilethisresultwithamoretraditionalandintuitiveviewsummarized
in Bernanke, Gertler and Gil hrist (1996), namely that onstrained rms tend to suer in
re essions earlier and more than un onstrained rms? In fa t, we need to point out that
ChevalierandS harfsteinpresentasettingwithalimitednumberofrmsandwith onsumers
fa ing swit hing osts. These osts generate market power for rms whi h they an use to
survive low demandstates bytemporarily harging relatively higher pri es, at the expenseof
A seminal work by Fazzari, Hubbard and Petersen (1988, FHP) introdu ed the ash ow
sensitivity of investment as indi ator of nan ing onstraints. Hubbard (1998) summarizes
the main reasons why investment may respond more to hanges in net worth of onstrained
rms and surveys subsequent work on the subje t. A ri h body of literature was generated
following the ritique by Kaplan and Zingales (1997, KZ) who obje t to the idea that ash
owsensitivityof investment should be in reasing inthelevelof onstraints.
Inoneofthelaterextensionsintoassetpri ingLamont,PolkandSaá-Requejo(2001)study
a ommon fa torinsto kreturnsrelatedtonan ial onstraints(measured bytheKZindex),
buttwounexpe tedresultsshowup: thisfa torearnsonaverageanegativeriskpremiumand
the onstraints fa tor does not seem to respond to ma roe onomi variables. Moyen(2004)
triesto re on ileFHPandKZandproposesamodelofinvestment/ ashowsensitivitywhere
two spe ial ases are onsistent with either FHP or KZ. Whited and Wu (2006) propose an
alternative indexto theKZindex,usingastru tural modelofinvestment. Theyndeviden e
ofa ommonnan ial onstraintsfa torinsto kreturns,whi hearnsapositivebutstatisti ally
not signi ant premium. Interestingly, ontrolling for nan ing onstraints eliminatesthesize
premium.
Morere ently,Gomes,YaronandZhang(2006)proposeaninvestment assetpri ingmodel
wherenan ing onstraintsappearasa ommon fa torinthe ross-se tionofsto kretursand
wherethenegative onsequen esofnan ing onstraintsarepro- y li al. Livdan,Saprizaand
Zhang (2009) in orporate dynami debt into a produ tion asset pri ing model and provide
eviden e insimulated dataof a pri ed nan ing onstraints fa tor,giving a slight superiority
to theWW index overtheKZindex.
Produ tmarket ompetitionwasre ently studiedinnan einthe ontext ofthe ross
se -tionofsto kmarketreturnsbyHouandRobinson(2006),whoshowthatrmsin on entrated
industries earn lower returns even after ontrolling for standard risk fa tors. They attribute
this result to either a onservative innovation poli y, or to barriers against rival entry and
as opposed to systemati aggregate sho ks studied in our paper. Hoberg and Phillips (2010)
analyzeindustry dynami s related to valuationbooms in ompetitive and non ompetitive
in-dustries. They show that ompetitive pressure following a boom in ompetitive industries
tendsto redu e futureprotability,leading to asharp drop inse torvaluation.
Theintera tionbetweenprodu tmarket ompetitionandnan ing onstraintshasre eived
mu hlessattentionintheliterature. Koveno kandPhillips(1997)arguethatleverage-indu ed
onstraintsen ouragepassiveprodu tmarketbehaviorwhiletherivalsbe omemoreaggressive.
Campello (2003) onrms at the industry level the predi tion by Chevalier and S harfstein
(1996) of ounter y lial markups generated byleverage-indu ed onstraints. Campello (2003)
alsoshows at thermlevelthatduringe onomi downturns nan ial onstraintspenalizethe
marketsharegrowthofrmsoperatinginindustrieswithlowleveragelevels,whileithaslittle
ee t in industries with high leverage. Povel and Raith (2004) study the ase of a duopoly
withone rmnan ially onstrainedandshowthatthe onstrained rmtendsto havealower
marketsharedueto ostdisadvantages,whilebothrmsbenetfromahigherpri ethanunder
full ompetition. We should notethat their denitionof onstraints is loserto the notionof
nan ial dependen eon external funding.
Webuildontwomainideasexploredintheliteratureonnan ing onstraints. Therstone
is that internal ash ows signi antly ae t onstrained rms' investment a tivity (Fazzari,
Hubbard,and Petersen(1986)). A ording tothese ondviewdeveloped inma roe onomi s
asthenan ial a eleratorrms fa ing highagen y osts(i.e. nan ially onstrained rms)
aresubje ttoaight toqualityinthe apitalmarketsintheearlystagesofe onomi slumps,
implying an earlier and strongerredu tion of their e onomi a tivity (see Bernanke, Gertler,
and Gil hrist (1996)). The two views taken altogether impli itly point to a vi ious ir le
dynami s. Finan ially onstrained rms have more di ulties with raising external nan e
exa tly when internal ash ows are lowest. These low internal fundsprovide for insu ient
nan ing of urrent operations and investments, whi h implies further redu tion in a tivity
internal ash ows. We proposeto introdu e a fa tor thatae ts theinitial sho kto internal
funds: the level of ompetition in the produ t market. Hou and Robinson (2006) onsider
industry on entrationasapartialinsulationfromma rosho ksduetohigherbarriersofentry.
There are at least two other reasons why on entrated industries should be less exposed to
adversesho ks. First,if ollusion me hanismsworkbetterindownturns(asinRotembergand
Saloner(1986)),then on entratedindustrieswillenjoyhighermarkupsindownturnsthatwill
partially ompensate for the falling demand. Se ond, if onsumers fa ehigh swit hing osts,
rms in on entrated markets will be able to steal some of the onsumer surplus to support
their own ash ows in the short run (as in Chevalier and S harfstein (1996), and Campello
and Flu k (2006)).
Given that the initial fall in internal funds will be stronger in ompetitive industries (an
assumptionstronglysupportedinourdata),thesubsequentadversedynami sfollowingaight
to qualityin reditmarkets shouldbemore severe for onstrained rms inhighly ompetitive
industries. We thereforeaddress thefollowing resear h questions: (i) What aretheee ts of
nan ing onstraintsontheoperatingperforman eandtheinvestment behaviorofrms?,and
(ii)Arethese realee ts ae tedbyindustry on entration? Lastly,we will explorea reverse
relationship between the two fa tors: (iii) Do nan ing onstraints ae t the ompetitive
me hanisms inindustries?
Based on the related literature and e onomi reasoning, we formulate the following
hy-potheses:
H1: Firms in on entrated industries are better insulated fromaggregate sho ks.
Firmsenjoyingmarketpowerin on entrated industries anpasssomeoftheee tsofadverse
aggregatesho ksontotheir ustomers. Hen e,protabilityofrmsin on entrated industries
islessexposedtothebusiness y le,whiletheprotabilityofrmsin ompetitiveindustriesis
moresensitivetotheoveralle onomy. Arationaleforthishypothesisiseitherbarrierstoentry
ofpotential rivalsasinHouand Robinson (2006), better fun tioningof ollusion me hanisms
H2: Firms subje t to nan ing onstraints are more ae ted by aggregate sho ks
than un onstrained rms. Constrained rms aresubje t to a ight to quality in apital
markets at earlystages ofe onomi downturns asinBernanke,Gertler, and Gil hrist (1996).
Giventhatinternal ashowsaredepressedduringre essions,theyhavetoredu etheira tivity
more than un onstrained rmsdue to a vi ious ir le me hanism.
H3: The ee ts of nan ing onstraints in downturns are higher in ompetitive
industries than in on entrated industries. PerH1, rms operatingin ompetitive
in-dustries have their internal ash ows more sensitive to adverse aggregate sho ks. If a
on-strainedrmoperatingina ompetitiveindustry issubje tto aight toquality,asper H2,it
an nan e a lower part of its operations and investment internallythan ifit were operating
ina on entrated industry. Overallit redu es its a tivitymore than a rm ina on entrated
industry.
H4: Finan ing onstraintssigni antlyae tthemean-reversionofrm
protabil-ity. We expe t thata rm an sustain abnormallevels of protability for a longer timeifa
large proportion of its rivals are nan ially onstrained. We use a similar logi as Chevalier
and S harfstein (1996).
We will address H1-H3 by rst studying the business y le sensitivity of operating
prof-itability of rms. Asa se ond step we will analyze real ee ts in terms of xed investment,
networking apitaland inparti ular inventories, andemployment. WeaddressH4inthelast
3.1 Sample rms
We useall dataonU.S. publi ompanies overthe period1977 to 2009 fromtheCRSP
Com-pustat merged database (CCM). We drop nan ial ompanies (SIC between 6000 and 6999)
andutilities (SICbetween 4900 and4999)due toa spe i statusofthese twoindustries. We
drop rm-years with total assets below$10 millionor with ommon equitybelow $5 million,
asinFamaand Fren h (2000). Werequirevalid(and non-negative,where appli able) dataon
totalassets,totalliabilities,property,plant&equipment,sharesoutstanding,sharepri e,sales,
net in ome, in ome before extraordinary items, osts of goods sold, ommon and preferred
dividends.
Wedeneoperatingprotability(ROA)asearningsbeforeextraordinaryitemsandinterest
dividedbytotalassets. Bookequityistotalassetslesstotal liabilitieslesspreferredsto kplus
balan e sheet deferred taxes and tax redits (our denitions follow Fama and Fren h, 2000).
Markettobookratioisthemarketvalueofequity(sharesoutstandingtimesend-of-yearshare
pri e)plustotal bookassetslessbookequitydivided by totalbookassets.
We winsorize all ratio variables at the1% level. We have intotal 78,537 rm-years. Our
average ross-se tion is2,380 observations,thesmallest one being 1,350observations in1977,
thelargestone 3,362 in1998. We present thenumberofobservations inea h yearintable 1.
We present summarystatisti sof our sample intable 2. Themedian assetprotability is
6.7%,while the median ROEis somewhat higherat 8.4%. Themedian M/Bofassets is1.28,
whilethemedian real salesgrowthis about 4.8% forthemedian rm. Mediantotal payout is
only 11.7%. Medianinvestment rateisat 4.7% of assets.
3.2 Business y le variables
We use the following raw ma roe onomi data from the Federal Reserve data base 1
: real
annualGDP,the3-monthTreasurybillrate,10-yearTreasurynoterate,3-monthnon-nan ial
1
Commer ialpaperrate ,and yields onseasonedAaa and Baabonds ratedbyMoody's.
Themain business y levariableis realannualU.S.GDP growth(denoted GDPGR).Our
large sample overs manufa turing and non-manufa turing rms, whi h justies theuse of a
largeaggregateindi ator asopposedto amorenarrowindi atorlike industrialprodu tion. In
additionto our oin ident ma roe onomi indi ator we use three leading indi ators basedon
marketyields ofgovernment and privatedebt instruments.
First,weusethe hangeinthe averageannualspreadbetweenthe3-month ommer ial
pa-perrateandthe 3-month Treasurybillrate(DCPSPR),be auseitis onsideredasapredi tor
of future e onomi u tuations. Spe i ally, downturns are generally pre eded by a spike in
theCPspread andre overies byitstightening (seeFriedmanand Kuttner,1991). Arationale
for this property, besides default risk and monetary onditions, is ompanies' variability in
ash needs over the business y le. A ording to Friedman and Kuttner the CP spread an
beviewedasademand-drivenphenomenon,resultingfrom ompanies'operatingperforman e.
An alternative view isthat spikes inshort therm nan ing osts(related to rising redit risk
premiums) are also a signal of a general tightening of redit onditions, whi h for es
ompa-nies to restrain some of their operations and therefore redu e their protability (as notedby
Bernanke, GertlerandGil hrist,1996).
Foranalogous reasonswe in ludeinourvariablesthe hange inthebondspread(DBSPR)
al ulated asthe hange ofthe average annualspread between seasonedBaaand Aaa bonds.
The third and last variable is the hange inthe average annualterm spread between the
10-yearTreasury noterateandthe 3-month Treasury billrate (DTSPR),following studieson
theinformation ontent of the term spreadwith respe t to future real e onomi a tivity(see
Estrella and Hardouvelis(1991),and morere ently Hamilton andKim(2002)).
All spread variables are taken as hanges in the annual average of weekly values. GDP
growth is al ulated as the rate of hange of the real annual GDP. Given that these ma ro
variables are highly orrelated, we orthogonalize them in a triangular fashion. We start by
takingGDPGRwithout hanges. WethenregressDCPSPRonGDPGRanda onstantandwe
keep theresidualasour nextma ro variable. We thenregress DBSPRonGDPGR,DCPSPR
2
zeromean andunit varian e.
In the remainder of the paper the names GDPGR, DCPSPR, DBSPR, and DTSPR will
refer to these orthogonalized and standardized variables. We plot the GDP growth and the
unadjusted levelsof spreadvariables ongure1.
In gure 2 we plot the dieren ed spread variables beforeand after adjustment for
orre-lation. We notethatthere islittle hange beforeand afterthe adjustment intheCPand the
term spread variable. The bond spreadvariablekeepsthegeneral appearan e,but withsome
dis repan ies duringthe 1980s.
3.3 Measuring nan ing onstraints
We measurenan ial onstraintsusing thesyntheti WhitedandWu(2006)index(WW)and
four fundamentals-based proxies following Almeida, Campello and Weisba h (2004, ACW),
Faulkenderand Wang (2006),and Denisand Sibilkov(2010).
The index by Whited and Wu (2006) is dened as WW
= −0.091 ∗
CF/
TA− 0.062 ∗
Divpos
+ 0.021 ∗
LTD/
TA− 0.044 ∗
LNTA+ 0.102 ∗
ISG− 0.035
SGwhere CFisin omebeforeextraordinary itemsplusdepre iation expense,TAistotalassets,Divposisa dummyvariable
forpositive ashdividend,LTDistotallongtermdebt,LNTAisthenaturallogarithmoftotal
assets,ISGis3-digit level industrysalesgrowth,SG istheindividualrmsales growth. Note
thattheinitial index wasestimated usingquarterly data.
The fundamentals-based proxies ofFC are:
•
P/Oratiomeasuredas(Dividends+Repur hases)/Earningsbeforeextraordinaryitems and interest.•
Size measuredasTotal assets.•
LT rating measured as an indi ator of whether the rm has a S&P long-term issuer rating.•
ST rating measured as an indi ator of whether the rm has a S&P short-term issuer rating.WefollowACW indening rms withhighand lownan ial onstraints. For WW,P/O,
andsize we sortannuallyallrms basedon ea hof thethree variables. Finan ing onstraints
are in reasing in the WW index and de reasing inP/O and size. We dene as highand low
nan ing onstraints theappropriatetop/bottom three de ilesinea h annual ross-se tion.
For the rating variables we dene as onstrained rm-years without a valid non-default
grade rating and with positive debt. Un onstrained rms are those that have a valid
non-defaultrating andpositive debt. Itisnot obviouswhetherrmswithouta ratingandwithout
debt should behave as less onstrained than rms without a rating with positive debt. Not
havingadebt ouldbearesultofstri terinternalpoli ies. Ifthisisthe ase,thenthissituation
is equivalent to not being able to issuedebt, i.e. being onstrained. ACW brieydis uss the
fa tthatthese rmsmaybe either onstrained orun onstrained. Wheneverwedoregressions
on pooled observations we will therefore only usean indi ator variable for theun onstrained
rms, whi h isless disputable. This is equivalent to a more onservative assumption thatall
rms thatdo nothave a ratingare onstrained.
3.4 Measuring market ompetition
Themost ommonlyusedmeasureofthedegreeof produ tmarket ompetition(PMC) isthe
Herndahl-Hirs hmanindex(HHI)dened asthesumofthesquaredmarketsharesofrmsin
theindustry. Themost ommon alternative totheHHIisthesimple on entrationratio C(x)
(the ratio of the x largest rms' market share to the overall market sales). There is a wide
agreement that the HHI is an imperfe t measure of on entration. For example it does not
take into a ount regional on entration ee ts, and it annot apture more omplex market
stru tures su h asSta kelberg-like stru tures(see the dis ussionin Martin 1993, p. 167). In
ee t, intermediate HHI values of about 500 an be asso iated with very dierent levels of
ompetitive stru tures. 3
3
AlsonotethattheDepartmentofJusti e onsidersindustriesmoderately on entratedabovethevalueof
Hoberg and Phillips (2010) from their Web Data Library. 4
These data are onstru ted by
usingtheCensusHHIofmanufa turingindustriesandextendingthemto all3-digitindustries
using the HHI al ulated from publi rms' salesin Compustat. Thisapproa h is a
ompro-misebetween using only Census data(whi h are al ulated only with a5-yearfrequen y and
only overing manufa turingindustries) and al ulatingHHIfromCompustat dataonly,thus
ex luding all private rms (see the ritique by Ali, Klasa and Yeung, 2009). Our measure
of ompetition is PMC
= 1 −
HHI, that we standardize to zero mean and unit varian e. Asa se ondary measure, we dene indi ator variables for the top and bottom 30% observations
basedon thePMCvariable.
We also use as a robustness he k an individual measure of market power as in Peress
(2010),the relative Lerner index (orthe relative pri e- ost margin). The rawLerner index is
denedforea hrmasLerner
it
= (
Salesit
−
COGSit
−
SGAit
)/
Salesit
,whereCOGSistheCostof goods sold and SGA isSelling, general and administrative expenses. We trun ate the raw
Lerner indexesat 0,asnegativevalues ofthe index have little theoreti alfoundationsand we
onsiderthemtoimplyzeromarketpower. Therelativeindexisobtainedbysubtra ting from
the raw value the rm's SIC3 industry sales-weighted average. The relative Lerner variable
hassigni antvariabilitythatmight beanoisymeasure. Asintheprevious asewe onstru t
indi ator variables for the top and bottom 3de iles and usethese inouranalysis.
To better understand the impli ations ofusing the HHI versus thealternative measureof
market power we inspe ted at the 3-digit SIC level the relationship between the tted HHI
valueandtheintraindustrydispersionofindividualrmmarketpower. Wemeasurethelatter
bytheindustrystandarddeviationofrms'rawLernerindexesinagivenyear. Asillustratedin
gure3,thereisrelativelylittledispersionofrmLernerindexesamongthehighly on entrated
industries. Onthe otherhand, the dispersionof rm Lerner indexes variesa lot at thelower
endof theHHI.Wenotethatsome ofthis patternisdue to alower number ofrms inhighly
on entrated industries. The table 4 presents for the year 2005 industries witha HHI below
January2010).
4
industrieswith alow dispersionof Lerner indexes(i.e. relatively equal distribution of market
power) are mostly standardized produ ts and servi es, with the ex eption of Newspapers.
Conversely, industries with dispersed Lerner indexes (uneven distribution of market power)
areindustrieswithlikelygeographi al, te hnologi al or variety-related on entration ee ts.
Given that our alternative measure of market power uses relative Lerner indexes, it will
mostlypi kupthevariabilityofmarketpowerwithintheselower on entration/highdispersion
industries, whereasindustrieswithlowdispersionofLerner indexeswill tendto generate rm
observations with relative Lerner indexes loser to zero. Therefore, when we later use the
relative Lernerindex tomeasuremarketpower, we willmostly he k thatour reasoningholds
inthese intermediate on entration industries.
Lastly,we notethatthedistribution ofnan ing onstraintsa rossthe orporate universe
isnotorthogonaltoindustry on entration. Infa t,mostmeasuresofnan ing onstraintsare
positively orrelatedwithrmsize(seedis ussioninBernanke,Gertler,andGil hrist(1996)).
Smallrmstendtobelesswellknown,lesstransparentandmoresubje ttoagen y osts. But
rmsinmore on entratedindustriestendtobelarger,bydenitionofindustry on entration.
We argue that nan ing onstraints an oexist with on entrated markets (i.e. not all
onstrained rms are in ompetitive industries). We present in table 4 average proportions
of rms withhigh nan ing onstraints inindustries with a Low, Medium and High
on en-tration. As expe ted, the proportion of onstrained rms seems to be negatively orrelated
withindustry on entrationforall onstraintsproxiesex eptSTratings. For example,forthe
WWindex 37%of rm-yearsaredeemedas onstrainedinlow on entrationindustries,while
only about 20%aredeemed as onstrained in on entrated industries. Although theseresults
onrm a lear orrelation between onstraints proxies and on entration, we are reassured
thatthere isstill asigni ant numberof onstrained rmsinhighly on entrated industries.
The most interesting ases of oexisten e of on entration with nan ing onstraints are
in moderate-to-high on entration industries. For example the Air rafts and Parts industry
very small and likely nan ially more onstrained rms su h as EDAC Te hnologies, CADE
Industries,and CPIAerostru tures,all withassetsbelow$35 million.
A se ond example is Household audio&video equipment (SIC3=365) where the HH was
1057 as of2005. The a tors in luded the large Harman International with assetsin ex essof
$2billion,medium-sizedUniversalele troni s($146million),aswellassmallerRo kfordCorp.
($52million) andKoss ($29million), both likely more onstrained.
4 Operating performan e and the business y le
4.1 Estimation method
We estimate a simple model inwhi h we regress rm protability ROA
t
(earnings beforeex-traordinaryitemsandinterestdividedbytotalassets)on(i)asetofma roe onomi variables,
(ii) their intera tion with measures of nan ing onstraints and industry ompetition (where
appli able), (iii)a setof rm ontrols, (iv)a lineartrendand a onstant.
All rm variables are lagged and they ontrol for the expe ted protability level. These
rm-level variables in lude similar variables as those used by Fama and Fren h (2000) for
estimating expe ted protability, namely the market-to-book ratio of assets (M/B Assets),
ommon dividends and repur hases divided by book equity ((D+Rep)/BE), and a dummy
variable indi ating nonzero ommon dividends during the previous year (DDUM). We also
addavariablemeasuring apitalintensity al ulatedasproperty,plantandequipmentdivided
by totalassets (K/A),and aninvestment intensityvariablemeasuredby apitalexpenditures
to totalassets(I/A). Wein lude alineartrendto apture the hanges inaverageprotability
inthe ross-se tionof listedrms.
We estimatethe following equationusing a xed rmee t estimatorwith rm- lustered
standard errors: ROA
it
= α +
4
X
j=1
β
j
Ma roVarj
t
+
5
X
k=1
γ
k
FirmVark
it−1
+
Trendt
+ ǫ
it
(1)setin ludes the M/Bofassets,D/BE,DDUM,K/A andI/A.
This equationis rst estimatedwithout furtherintera tion terms, to serve asbaseline. In
subsequentse tionsweintera ttheMa roVarsetwithlaggedmeasuresofnan ing onstraints
and/or produ t market ompetition, asdened inse tions 3.3 and 3.4. This spe i ation is
similar to the redu ed form spe i ation used by Sharpe (1994) for analyzing rm
employ-ment growth. We will therefore be able to he k whether measures of onstraints and/or
on entration signi antly ae t thesensitivity tothebusiness y levariables.
4.2 Results on market ompetition
In this se tion we rst onrm one of the building blo ks of this paper, namely that that
rms in on entrated industries resist better to aggregate u tuations (hypothesis H1). We
will estimatethe following modi ationof equation (1)in luding thelevels of PMCasof the
previous year,and Ma roVar-PMC intera tions:
(1)
+
PMC+
Ma roVar-PMC intera tions (2)Astwo alternative spe i ations, wewill also estimateaversionthatin ludes lagged
indi- ator variables for high (Hi) andlow(Lo) produ t market ompetition/relative Lernerindex,
and theintera tions ofMa roVar withtheindi ator variables:
(1)
+
PMCindi ators+
Ma roVar-PMC indi ator intera tions (3)Therst olumn oftable5presents oe ientsonthema rovariablesinthebaseline ase
without on entration measures. The oe ients are strongly statisti ally signi ant ex ept
for DCPSPR. The GDP growth variable has a signi antly positive inuen e on operating
protability. Asa ruleofthumb,a onestandard deviation positive sho kinreal GDPgrowth
translatesinto0.36per entagepointsofadditionalprotability. The hangeinthebondspread
Lastly, the term spread also has a signi ant negative ee t on rm protability and is the
strongest in magnitude. A one standard deviation unexpe ted in rease inthe term spread is
asso iatedwith 0.61pointslessprotability.
Column (2)shows that intera ting the ma ro variables with the PMClevels signi antly
alters the pi ture. A one standard deviation in rease in the level of ompetition in reases
thesensitivity to the GDP growth by about one third. Moreover, PMCstrengthens also the
sensitivity to the bond spread (by more than one half per one standard deviation of PMC)
andtothetermspread(bymorethanonethirdperSD).Theseresultssupportourunderlying
hypothesisof insulation againstaggregatesho ks.
WepresentinColumn(3)intera tionswithdummyvariablesforhighandlowPMC.These
result suggest that the ee ts of the PMC levels presented in Column (2) seem to be driven
by the low end of the PMC variable (i.e. on entrated industries), as it is mostly the Lo
dummythathasasigni ant oe ient inthisspe i ation. Con entrated industriesseemto
bemore than one halflesssensitive to theGDP growth andtheterm spread,and seem to be
quasi-insensitive to thebond spread.
Column (4) presents results using intera tions with a dummy variable for high and low
relative Lerner index. As we dis ussed in se tion3.4, this measure of marketpower pi ks up
mostlythe variability inthe intermediately on entrated industries with a highdispersionof
individual rmLerner indexes. Therefore this approa h isa omplement to theprevious two
ases. The results are qualitatively similar to the previous ase. Firms with a high market
power are more than one half less sensitive to the GDP and about one quarter less sensitive
to theterm spread. Firms witha low marketpower aremore than twi eas mu h exposedto
thetermspread. There isasigni ant positive oe ient on theCPspread onbothhighand
lowintera tion oe ients.
The explained variabilityinall spe i ationsisabout 7to 8 per ent. We notethatthese
R-squaredex ludethe variability apturedbythermxedee ts(thexedee trepresents
Wesummarizethisse tionbystatingthatourresultsspeakinfavorofthehypothesisthat
imperfe tly ompetitive industriesbenetfromasigni ant insulation fromaggregate sho ks.
Firmsin on entrated industriesseemtobelessexposedtoboththe oin ident business y le
variable, aswellasto leading indi ators.
4.3 Results on nan ing onstraints
4.3.1 Standalone spe i ation
We rst pro eed to the analysis of the ee t of nan ing onstraints without ontrolling for
industry ompetition. The studied spe i ation will be analogous to the previous ase. We
estimatetheequationusinglaggedindi atorvariablesforhigh(Hi
FC
)andlow(Lo
FC
)nan ing
onstraints basedon theWWindex, thepayout ratio,size,and LTandST ratings, andtheir
intera tions withthe ma ro variables:
(1)
+
FCindi ators+
Ma roVar-FCindi ator intera tions (4)Intable6wesummarizeresultsforvedierentproxiesofnan ing onstraints(notethat
for the two ratings proxies there are only dummies for low onstrained rms due to reasons
dis ussed inthe last paragraphof se tion3.3).
ThesensitivitytoGDPissigni antly lowerforrmsfa inglow onstraints, forallproxies
ex ept the WW index whi h is only signi ant at the 10% level. Depending on the proxy,
un onstrainedrmsareabout30%to70%lesssensitivetoGDPgrowththantheaveragerm.
Constrainedrms have eitheran insigni ant negative intera tion oe ient (WWand size),
or apositive oe ient signi ant at the10%level(P/O ratio).
The results on thebond spread arethe most lear- ut. For all our proxies un onstrained
rms are lessexposedto unexpe ted hanges in thebond spread: on average about 50% less
than average rms. On the other hand, onstrained rms are signi antly more sensitive to
oe ientsareonly signi ant forone proxy,theP/Oratio.
Inall ases,the modelexplains about 6%to7%of thevarian e ofthedependentvariable,
ex luding the rmxed ee ts ontribution. The RMSE, not presented inthetable,isabout
0.09to 0.10 (notethattheSD of the dependent variableis about 0.14).
To sum up, using various proxies of nan ing onstraints we on lude that onstrained
rms'operationsaremore sensitiveto the overallbusiness y le,beingae tedmore by both
the oin ident indi ator as well as one leading indi ator, the bond spread. Our results are
apparently onsistentwiththeighttoqualityee tofBernanke,GertlerandGil hrist(1996)
where onstrained rms tend to suer early in e onomi slumps from a tightening of redit
onditions.
4.3.2 Spe i ation ontrolling for market ompetition
Wewillnowattempttoanalyzethe hypothesisH3,whi h laimsthatthenan ial
a elerator-like ee t do umented inthe previous se tion is strongest in ompetitive industries. We will
estimateanequationthat ontrolsfornan ing onstraints, produ tmarket ompetition, and
their intera tions. Wewill usethe PMClevels asameasure thelevelof ompetition.
(1)
+
FCindi ators+
PMC+
Ma roVar-PMC intera tions+
Ma roVar-FC indi atorintera tions+
Ma roVar-FC indi ator-PMC intera tions (5)Table 7 shows results for this ombined spe i ation. We see that the rst eight rows
of thetable arequalitatively and quantitatively similarto thespe i ation with ompetition
intera tions only (presented intable 5, olumn (2) ): rms in on entrated industrieshave a
betteraggregate sho ks resistan e.
Simple intera tion termswithnan ing onstraints proxiesarelesssigni ant than inthe
survivesinthisspe i ation: allve oe ientsaresigni antlypositiveforthelow onstraints
dummy,andall oe ientsarepositive(onesigni antly)onthehigh onstraintsdummy. The
Term and CPspread oe ients showno lear patterns.
Theparameters ofinterestareon the rossintera tionterms ontainingboththe
ompeti-tionandthe onstraintsvariables. The rossintera tion termofLow onstraintswithPMCis
mostlypositiveandsigni antintwooutofve ases. The rosstermofHigh onstraintsand
PMCis negative and signi ant in two out ofthree ases. Thismeans that onstrained rms
tend to benet relatively more from the ma ro-prote tion ee t of industry on entration.
Conversely, onstrainedrms arethe most exposedto aggregate sho ksin highly ompetitive
industries. This pi ture is even more lear with the ross terms on the Bond spread. Low
onstraints rosstermsareallsigni antly negative at leastatthe10%level, whileHigh
on-straints terms are all positive, one signi ant and two just below the 10% signi an e level.
Again onstrained rms tend to benet more from theinsulation from aggregate sho ks due
to the on entration of their industry. Similar but less straightforward patterns an be seen
on theTermspread ross oe ients.
Inthesespe i ations,themodelexplains6to7%ofthevarian eofthedependentvariable,
ex luding the xed ee t's ontribution. The RMSE(not presented) is inthe interval
0.093-0.098, therefore about one-third of rm protabilty's standard deviation is aptured by the
model.
This se tion provided signi ant eviden e in favor of the hypothesis H3: nan ing
on-straints seem to have stronger real ee ts on the operating performan e of ompanies in the
most ompetitive industries. Put dierently, the ma ro-insulation benet of on entration
is higher for rms deemed as onstrained using a number of ommon proxies of nan ing
In this se tion we will inspe t whether thepreviously dis ussed results on erning operating
performan ehavetheir impa tonrmbehavior,inparti ular intermsof investment and
em-ployment. Inee t,themainimpli ationof thenan ial a eleratoraspresentedinBernanke
et al. (1996) is that nan ially onstrained rms subje t to a ight to quality are for ed to
restraintheir investment morethan otherrms,therebyfurther ontributingto theaggregate
e onomi slowdown andee tively ontributing to thedepthof there ession.
We will estimate an enri hed version of an investment model where we in lude a ma ro
variable and its intera tion with market on entration. As it is standard in the investment
literature we will estimatethis equation onsubsamples basedon various nan ial onstraints
proxies(see for exampleGil hristand Himmelberg(1995)). The estimatedspe i ation is:
Inv
it
= α + β + δ ×
PMC SIC3t−1
×
GDPGRt
+
6
X
k=1
γ
k
FirmVark
it−1
+ γ
7
∗
PMC SIC3t−1
+ ǫ
it
(6)In thefollowing subse tions Inv will be one of three investment rates (inphysi al apital,
innetworking apital, ininventories) or theemployment growth rate. The setof lagged rm
variablesFirmVarin ludestheM/Bratioofassets, ashowmeasuredastheoperatingin ome
beforedepre iationdividedbyassets, apitalintensitymeasuredasPPE/Assets,sizemeasured
asthelogof assets,adividend payer dummy,and previous year'sindustry salesgrowth.
5.1 Fixed investment
Werst present ourresultsfor equation6appliedto investmentinphysi al apital,dened as
the ratio of apitalexpenditures to netproperty,plant andequipment. We ex lude irrelevant
rms operating with verylowphysi al apital, i.e. whose PPE/Assets ratio islowerthan 5%.
The rst observation from table 8 is that the sensitivity of investment to GDP is learly
higher for rms with high onstraints. For un onstrained rms a one standard deviation
0.99-1.50 per entagepointin rease ininvestment.
Ase ondobservationisthattheintera tion ofGDPwithPMCispositiveinall ases,
sug-gestingthatindustry ompetition in reasesthema ro sensitivityofrminvestment. However
the oe ients are only signi ant in the high onstraints subsample, and only for the rst
three proxies. Judging bythe size of the onstrained subsample for the rating-based proxies,
this absen e ofsigni an e ould be due to the rudeness ofthese two onstraints proxies. In
thethree ases wherethe oe ient issigni ant, ithase onomi allysigni ant value. Aone
standard deviation in reaseinPMCin reases the GDPsensitivitybybetween 30%-50%.
We on ludethatwefoundpartialeviden eforanintera tionee tofnan ing onstraints
and industry on entrationinthe y li alsensitivity ofrm investment. Asinthese tionon
operating performan e, we nd that industry on entration benets onstrained rms more
than un onstrained rms.
5.2 Net working apital and inventories
Inthisse tionwe applyequation6to investment innetworking apitaldened asthe hange
innetworking apital dividedbyPPE. Networking apital(NWC)hasbeen analyzedinthe
ontextofrminvestmentbe auseitservesasabuerwhennan ingbe omess ar e(asshown
byFazzari and Petersen (1993)). In ee t, NWC just like anyother omponent of operating
assets has to be nan ed in the balan e sheet of the rm. When available nan e be omes
limited,parti ularlyforthemore onstrainedrms,rmsfa eatrade-owhenallo atingfunds
tovariousasset omponents. FazzariandPetersenarguethatrms anadjustNWCinvestment
at a lower ost than xed investment and therefore NWC is used as a liquidity buer. The
business y le variability of NWC ould thus reveal interesting patterns, omplementary to
those seenonxed investment inthe previous se tion.
We dene net working apital as urrent assets minus ash minus urrent liabilities. We
ex lude ash be ause ash plays an important role for investment poli y, in parti ular for
Table9 onrmsthattheNWCinvestmentsensitivitytoe onomi onditionsvariesstrongly
with the nan ing onstraints status. For low onstraints rms, the sensitivity is relatively
low and signi ant for only two out of ve proxies. For the high onstraints subsample, the
sensitivity to GDP is signi antly positive. A one standard deviation negative sho kon
e o-nomi a tivity redu es NWC investment by3.4-5.5 per entage points. The intera tion ee t
of GDPGR with PMC seems to be mostly positive, but it is only signi ant in one single
asehigh onstraint rms measuredbythe P/Oproxy.
Next, wefo usour attention on asubset ofNWC: inventories. Thisassethasa parti ular
role in ma roe onomi s, be ause hanges in inventories onstitute an important sour e of
aggregate e onomi growth variability. Gertler and Gil hrist (1994) point out dieren es in
the behavior of inventories in small manufa turing rms relative to large rms, the former
a ounting for alarge partof the variabilityof theaggregate inventory hanges.
Wealsofo usspe i allyoninvestmentsininventories,denedasthe hange ininventories
divided by PPE (results not presented, but available upon request). We note that inventory
investment ispro- y li alinallthreesubsamples,butthe ee tisstrongestinthe onstrained
subsample. Asinthe NWC ase, the oe ient on thePMCintera tion term is positive and
strongest inthe onstrained subsample,but issigni ant inonlyone ase.
Our results on NWC and inventories onrm their role of a buer asset: the sensitivity
ofthis itemto aggregatee onomi a tivityis onsiderably higher than thesensitivityof xed
investment. Moreover, onstrained rms'NWCinvestmentsseem to be more sensitive to the
aggregatesho kbyafa tor ofat least2 ompared to un onstrained rms. The intera tionof
onstraints withprodu tmarket ompetition isnot signi ant.
5.3 Employment
The last real ee t that we will analyze is employment growth. Sharpe (1994) shows that
the aggregate sensitivity of rm employment to the business y le is higher for small rms
market ompetition inadditionto onstraints.
We in lude in our analysis only rms whi h have had an asset growth rate within the
(-50%,+100%)range,tomitigatetheee tofrm-yearswithsigni ant orporateeventssu has
mergers andspin-os. Table 10 suggeststhat thereis ama ro-sensitivity dierential between
theHigh and Low onstraints subsamples. Low onstraint rms redu e their employment by
1.2%-2.5%for aone SDde reaseintheGDP,whereas onstrainedrms slashemployment by
2.8%-3.1%.
Intera tions withPMChave neitherasigni ant oe ient, norstablesigns. Constrained
rmsoperatingin on entrated industriesdonotseem have alower employment sensitivityto
theaggregate y le.
Wesummarizethisse tionbynotingthatnan ially onstrainedrms'investmentinxed
assets and working apital and their employment appear more sensitive to the business y le
ompared to un onstrained rms. Industry ompetition strengthens this ee t only on xed
investment.
6 Some asset pri ing results
Previousse tionsshowthatnan ially onstrainedrmsaremoreexposedtoaggregatesho ks.
Thisfundamental risk isparti ularly highifthey operateinhighly ompetitive industries. In
thisse tionweshowthatsto kpri esdonot ompletelyree tthesedieren esinfundamental
risk.
WestartbymergingourfundamentalsdatawithCRSPsto kreturns. Weuseallrmsthat
bothsatisfyour fundamentals datarequirements(see se tion3) andhave valid CRSPreturns
data. At the end of June of ea h year t we take the year t-1 sorts from the fundamentals
dataset and we onstru t portfolios based on PMCand FCtop/bottom ter iles. We usethis
portfolio stru ture to al ulate July t to June t+1 returns. We use three proxies for FC:
Weregress portfolio ex essreturns onthree Fama-Fren h fa torsand themomentum
fa -tor. 5
In table 11 we present time-series regression estimates on the period July 1977 to
De- ember2009.
InPanelA(WWindex)we observethatat lowandmediumlevels ofmarket ompetition,
un onstrained rmshave thelowest marketbeta, asexpe ted. For highlevels of ompetition,
however, marketbetadiers little fordierent levelsof FC. We also notethat market beta is
highest for medium FC rms at all levels of ompetition, although the dieren e is small for
mediumandhighPMClevels. Thislatterresultmightbesurprisingatrst. Butnotethatthe
dieren e inrisk between medium andhigh FCrms is aptured instead bythe SMB fa tor.
Indeed,highFCrms mapstrongly onthe size fa tor.
The most interesting observation inPanel A is that low onstraints rms inmedium and
highly ompetitiveindustriesearnasigni antpositiveabnormalreturn. TheHighPMC/Low
FC portfolio earns about 26bp monthly abnormal return (about 3.1% annualized). This is
surprising given our eviden e that these ompanies bear signi antly less fundamental risk
ompared to highly onstrained rms. Sto k pri es do not seem to ree t this fundamental
riskdierential.
Panel B(payout ratio) presents an even learer pi ture. Marketbetais generally
in reas-inginboth PMCand FC. Moreover, additionalrisk exposure ofhighly onstrained portfolios
is mapped on the SMB fa tor. Again, un onstrained rms in medium and highly
ompeti-tive industries earn a signi ant abnormalreturn of 19bp (2.3% annualized) and 40bp (4.9%
annualized), respe tively.
InPanelC(size)theoverallpi tureremainssimilartoPanelA.Marketbetaisnot
ne essar-ilymonotonous inFC,in parti ular inthe highly ompetitive portfolios. ButtheSMB fa tor
mappings apturealargepartofoverallriskofhighFCportfolios. Mostimportantly,theLow
FC rms earn again signi ant positive abnormalreturns in medium and highly ompetitive
industries(16bpand 24bp,respe tively).
We investigate further the abnormal returns patterns by running moving window
time-5
abnormal returns and their t-stats from 48-month moving window regressions of four orner
portfolios orrespondingto ombinations ofHigh/LowFC andHigh/LowPMC.
First, for all three FC proxies the High FC/High PMC portfolio earned mostly negative
abnormal returns over the se ond half of the 1980s, marked by the beginning of the savings
andloan risis andthesto kmarket rashof 1987. Theperiod 1994-95(Fedtightening, bond
market de line and rising yield volatility) through 1997 (Asian risis) was also a period of
mostly signi ant negative abnormal returns for this portfolio, but only based on the WW
and Size proxies. The rst half of the 2000s is the only period of positive abnormal returns,
signi ant only when using the WW and P/O proxies. The 2007-2009 risis brought again
some negative abnormalreturns.
TheHighFC/LowPMCportfoliohadalsonegativealphasduringtheseperiodsofdistress,
namely the se ond half of the 1990s and the 2007-09 risis. But there were also periods of
signi antly positive returns: therst halfofthe1980s andtheearly 1990s.
The Low FC/High PMC portfolios earned positive alphas during the 1997 Asian risis
period with a peak in early 1998 for all three FC proxies. Similarly, these ompanies earn
positive returns in the post-re ession periods of 2002-03, and 2009. The only instan e of
signi ant negative returnsisthe 1986 period.
Last, the Low FC/Low PMC portfolio abnormal returns are learly lower in magnitude
and only signi ant at times. For example, alphas were signi ant during the 2007-09 risis
period.
Themovingwindowanalysisshedssomelightonthe positiveabnormalreturnsoflow
on-straintsrmsinmediumandhighly ompetitiveindustriesobservedintable11. Un onstrained
rmsinhighly ompetitive industriestendto benetfrompost-downturn onditions when
ar-guably their onstrained rivals are still experien ing di ulties (as do umented in previous
se tions). However,this fa tisnot su iently ree ted invaluations andun onstrained rms
generatepositiveabnormalreturnsinthoseperiods. Thisndingisnewandis omplementary
Also,theabnormalnegativereturnsof onstrainedrms in risisperiodsindi ates thatat
normaltimestheirsto kpri esdonot ree ttheir nan ial`vulnerability'. Thisnding ould
partly explain why previous studies nd that a nan ing onstraints fa tor in sto k returns
earns a negative (Lamont, Polk, and Saá-Requejo 2001) or a positive but insigni ant risk
premium (Whitedand Wu 2006).
7 The ee tof nan ing onstraints on ompetitive me hanisms
7.1 A simple mean reversion model
Up to this point, we have studied how ex ante measures of ompetition intensity ae t the
real ee ts of nan ing onstraints. In this se tion we will look at how ex ante measures of
nan ing onstraintsat therm andindustry levelae t ompetitive me hanisms.
Toaddressthisquestionwewillanalyzehownan ing onstraintsae tthe ross-se tional
mean-reversionme hanismofrmprotability. FollowingFamaandFren h(2000,hereinafter
FF)weestimateasimple partialadjustment modelof mean-reversion ofprotability. Intheir
paperFFproposeatwostepmethod. Intherststeptheyestimateonea hyear's ross-se tion
theexpe tedprotability for ea h rm basedon three variables: M/B of assets,dividends to
bookequityand a dummyindi ating zero dividends. 6
Asase ond step, theauthors estimate
an equation where the hange of protability for a rm in year
t
is explained by the laggedprotability hangeandbythedieren ebetweenobserved andexpe tedprotabilityin
t − 1
.The main FFspe i ation is somewhat disputable, given that expe ted protabilityis linked
to the M/B ratio of assets. In ee t, su h a spe i ation allows the expe ted protability in
subsequentyearsto revert to ee tiveprotabilityinsteadoftheseeing theee tive number
revertingto itsexpe tation(asweimpli itlyexpe tinthenotionofmean-reversion). Ifarm
earnsabnormallevelsofprotabilityinyeart,themarket aninterpretthissurpriseasthenew
normal: M/Bin reasesandthereforenextyear'sexpe tedprotabilityin reases aswell. This
6
Notethatwein ludeverysimilarvariablesasindividualrm ontrolsinourregressionsinse tion4,ex ept
offundamentalinformation bythemarkets. Theauthors nda rateofmean reversioninthis
spe i ation at about 38%a year.
FF in lude also a more robust spe i ation in whi h protability is assumed to
mean-revert to a ommon ross-se tional mean. In this setting the mean-reversion oe ient is
also signi ant, though lower, at about 30% a year. It seems normal that in this ase the
estimatedmean-reversionrateislower,be ausewe apturethepuremean-reversionofee tive
protability toa ommon mean.
We spe ify a model similar to the FF's se ond spe i ation, ex ept that we expli itly
in lude the rossse tional averageprotability inthe previous year. The baselinemodelis
ROA
it
−
ROAit−1
= α + β
ROAit−1
−
ROA agrt−1
+ γ (
ROAit−1
−
ROAit−2
) + ǫ
it
, (7)where ROA
it
is earnings beforeextraordinary items andinterest divided byassets and whereROA agr
t
istheasset-weighted averageROAin the ross-se tionofall rms.We then in lude intera tion mean-reversion terms ontrolling for nan ing onstraints at
the rmand industry level:
ROA
it
−
ROAit−1
= α + β + δ
Lo×
Lo FCt−1
+ δ
Hi×
Hi FCt−1
+ δ
%Hi×
%Hi FCt−1
×
ROAit−1
−
ROA agrt−1
+ γ (
ROAit−1
−
ROAit−2
) + ǫ
it
, (8)whereLoFC(HiFC)isanindi atorvariableofthebottom(top)30%observationsbasedonthe
onstraints proxyinthe previousyear, andwhere %Hi istheproportion ofHiFCrmswithin
the SIC3industry in theprevious year. Equation(8)is slightly modied for therating-based
proxies,where weonlyuseindi atorsfor low onstrained rms(seese tion3.3foradis ussion
of this hoi e).
Weestimatethese equations usingtheFamaandMa Beth (1973) method andwepresent
the results in table 12. The baseline mean-reversion oe ient of 26.9% per year in olumn
being orre ted for in our tables, whi h means that t-statisti s areoverestimated. If we had
a su iently long time series of these slope estimates, we would be able to estimate reliable
auto orrelationestimatesand al ulateadjustedSEsbytheNewey-Westte hnique. Sin ethis
isnot the ase, to infer signi an e we follow Famaand Fren h (2000)and require somewhat
higher t-statisti s for signi an e. FF estimate that with an auto orrelation of 0.5, the SEs
should be in reased by about 40%, therefore the threshold for t-stats should be at about 2.8
insteadof the ommonlyusedvalueof 1.96.
In olumns(2)to (6)weuseve dierentproxiesfornan ing onstraints. Wend partial
eviden ethatrm-level onstraintsae tthemean-reversionrateofrmprotability: fortwo
onstraints proxies (WW index and P/O ratio), the dieren e between onstrained and
un- onstrainedrms issigni antandsuggesting that onstrained rmsrevertto themeanfaster
than un onstrained rms. The eviden e is more lear- ut for the industry-level onstraints:
the per entage of onstrained rms signi antly redu es the mean reversion rate in all three
ases. In reasing the per entage of onstrained rms within an industry by 50 per entage
pointsredu es the mean reversionrate by about 12 per entagepoints. We notehowever that
theper entageofun onstrained rmsinthetworatings-based aseshastheexpe tednegative
oe ient only for LT ratings and the average t statisti is belowour in reased hurdle value
of 2.8inboth ases.
7.2 Two-level mean-reversion
Whenlookingatthespe i ationofequation(7)one ouldarguethattheimpli itassumption
of mean-reversion towards one ommon average protability for all rms is implausible. The
most obviousargument isthat one ommon expe ted protabilitydoesnot take into a ount
dieren es in risk a ross industries. A similar ritique an be made based on dieren es
in te hnology, the level of intangible assets (human apital, ta it knowledge, brand value,
reputation) et .
mean-mean. These ond level isthemean-reversionof theindustry meanto theaggregate mean. If
theabove ritiqueisjustied,weshouldonlyndeviden efortherstlevelofmeanreversion.
Spe i ally,we estimateinthebaseline asetheequation
ROA
it
−
ROAit−1
= α + β
1
ROA
it−1
−
ROA SIC3t−1
+ β
2
ROA SIC3t−1
−
ROA agrt−1
+ γ (
ROAit−1
−
ROAit−2
) + ǫ
it
, (9)where ROA SIC3
t
is the asset-weighted average protability in the SIC3 industry. The hangeinrm protability an therefore be due to eitherthe previous year's dieren e between the
rm and its industry mean, or between the previous year's industry mean and the aggregate
mean.
Analogously,thetwo-level spe i ation ontrolling for nan ing onstraints is
ROA
it
−
ROAit−1
= α + β + δ
Lo×
Lo FCt−1
+ δ
Hi×
Hi FCt−1
+ δ
%Hi×
%Hi FCt−1
×
ROAit−1
−
ROA SIC3t−1
+ λ + κ
Lo×
Lo FCt−1
+ κ
Hi×
Hi FCt−1
+ κ
%Hi×
%Hi FCt−1
×
ROA SIC3t−1
−
ROA agrt−1
+ γ (
ROAit−1
−
ROAit−2
) + ǫ
it
. (10)Table 13 shows results based on the two-level spe i ation. The results at the rst level
(rmvs. industry)arequalitativelysimilartothesimple ase. Individualnan ing onstraints
status seems to be relevant only in some ases (WW and P/O). The signi ant negative
oe ientontheLoFCtermforLTratingsissurprisingasitsuggeststhatlow onstraintsrms
seem torevertfaster than otherrms. Theper entageof onstrained rmshasasigni antly
positive oe ient in all three ases (WW, P/O, size), and the per entage of un onstrained
rms is negative for LT ratings although the average t-stat is still below our onservative
thresholdof 2.8.
The pi ture is ompletelyreversed atthe se ondlevel(industryvs. aggregate): individual
per entage of onstrained rms in the SIC3 industry does not seem to be relevant as no
averaget-stat isabove our threshold of2.8.
Insummary,the two-level spe i ation onrmed thatprotabilitymean-reversionis
hap-peningat both theintra-industry andtheinter-industrylevels. Wefurther onrmedthatthe
proportion of onstrained/un onstrained rms within theindustry seems to be an important
determinant ofintra-industry protability mean-reversion.
Theseregularitiesarenottakenintoa ountbymarketvaluations. Intable14we onstru t
doublesortedportfolioson(i) therelative rmprotabilitywithrespe tto theprotabilityof
its SIC3 industry, and (ii) on theindustry per entage of nan ially onstrained rms within
the SIC3 industry. We that abnormally protable rms earn alphas of about 20-29 bp per
monthifahighper entageoftheirrivalsarenan ially onstrained. Thealphasaresmalland
not signi ant for abnormally protablerms iftheir industries have few onstrainedrms.
8 Robustness
Besides having used ve dierent proxies for nan ing onstraints previously employed in
the literature, we have also tried several robustness he ks for our results. First, we have
employed an alternative spe i ation for the intera tion ee t of nan ing onstraints and
market on entration on the business y le sensitivity of operating protability. Instead of
in luding rm level proxies of lagged protability we use lagged values of protability. We
usethreelaggedlevels ofROA.Dueto knownissueswithlaggeddependent variables inpanel
data, we estimate this equation following Arellano and Bond (1991) with a dieren e GMM
estimator (instrumenting with four lags of endogenous variables, ollapsed). The Ma roVar
set in ludes the same four ma ro variables asin se tion 4. The statisti al properties of this
spe i ation are improved when we in lude rm variables that likely ae t deviations from
a simple autoregressive spe i ation (we uselagged hanges inthe xed investment rate and
in advertising expense). We apture the ee t of nan ing onstraints by estimating the
and DBSPR for the LowFC subsample are relatively lowand sparselysigni ant. The same
intera tiontermsontheHighFCsubsamplearebothstatisti allyande onomi allysigni ant.
Sharpe (1994)notesthatusing anaggregate e onomi variableonrm-leveldata ouldbe
problemati due to dierent industryexposure to ma roe onomi onditions. Asarobustness
he kheproposestouseindustry-yearspe i dummyvariables. We onstru ttwelveindustry
groups based on the 12 Fama-Fren h portfolios and reate a dummy for ea h industry-year. 7
Theresults(notreported)onxedinvestment sensitivityareevenstrongerthan inthegeneral
ase: the intera tion variable ofPMC withGDPGR is signi ant for onstrained rms using
all ve nan ing onstraints proxies, while it is e onomi ally and statisti ally not signi ant
for mediumand low onstrained rms.
The underlying idea of of this paper statingthat nan ing onstraints should ae t more
rms in ompetitive industries relies strongly on the assumption that rm ash ows are
strongly orrelated with thebusiness y lein the rst pla e. Therefore, for industries whose
ashows ovarylittlewiththebusiness y le, theadditionalbenetofa on entrated market
(providing insulation against aggregate sho ks) should be less important. We follow Sharpe
(1994)andCampello(2003)inseparatingindustriesintohighly orrelated(durables),weakly
orrelated (non-durables), and others (non-manufa turing). Our results on thesensitivity
ofoperatingprotabilitytoGDPandbondspreadsho kssurvivequalitativelyinthedurables
subsample, with a loss of signi an e for some proxies. The results survive partially in the
non-manufa turing subsample. The non-durables subsample ompletely losessigni an e on
theGDP ross intera tion terms, as expe ted due to a low orrelation of ash ows with the
e onomi y le in these industries. On the other hand, theBond spread ee t survives even
inthenon-durables industries.
Lastly,we he kwhetherourresultsarenotdrivenbythere ent risis. Ourresultsremain
broadlyun hanged ifwe ex ludethe 2007-2009 period.
7
In this paper we studied the intera tions of produ t market ompetition and nan ing
on-straints. Thetworesear hpathsinthenan ing onstraints literature,namely thatof
invest-ment ash-owsensitivities and thatof nan ial a elerator, point to a vi ious ir leee t of
nan ing onstraints during e onomi downturns. Spe i ally, a onstrained rm must rely
on its internal fundsexa tlywhen ashows aredepressed, for ing itto a stronger redu tion
of operations and investment, further depressing internal fundset . We argue that
ompeti-tive industriesundergo the strongest sho k to ashows at theonset of a downturn, thereby
strengthening the adverse ee t of nan ing onstraintsduring theslowdown.
We nd signi ant support for our hypothesis in the analysis of operating protability
of rms. The ee ts of ma ro variables on the operating performan e of a onstrained rm
are stronger if the rm operates in a ompetitive industry. Stated dierently, nan ially
onstrained rms tend to benetrelatively more from the insulation ee t of a on entrated
produ tmarket.
We onrm that nan ing onstraints have signi ant real ee ts on the business y le
sensitivityof orporate investment andemployment tothebusiness y le. Inxedinvestment
data we nd an amplifying ee t of produ t market ompetition on the ee ts of nan ing
onstraints. On the other hand, investments innet working apital and employment growth
do not seemto besubje tto asimilar ampli ationee t.
In our asset pri ing tests we nd that sto k pri es do not ompletely take into a ount
thesedieren esinfundamentalrisk. Spe i ally,un onstrainedrmsoperatingin ompetitive
industriesearnabnormalpositivereturns,parti ularlyduringperiodsofe onomi andnan ial
distress.
Lastly,weshowthatnan ing onstraintsae t ompetitiveme hanismswithinindustries.
Wendsupportforthehypothesisthattheaveragelevelofnan ing onstraintsinanindustry
tends to redu e the intraindustry mean-reversion of rm protability. Sto k pri es do not
integratethis fa t. Firms withahighprotability relativeto their SIC3industryprotability