Firms' financial policy and labour demand : theory and evidence

Firms’FinancialPolicyandL abourD emand:

T heoryandEvidence ^¤

P ålB ergström andSaraL indberg

A bstract

W e investigate the in‡uence of…nancialleverage on …rms’hiring decisionsinthecontextofahierarchyof…nancemodel. T heanalysisis basedon theEulerequation ofemploymentin thepresenceofconvex adjustmentcosts. W eshowtheempiricalimplications of…rms facing ahierarchyof…nancialcosts, and estimatealinearisedversion ofthe modelon a large panelofSwedish manufacturing …rms. B ootstrap methods areutilised toalleviatesomeoftheestimation problems in- volved. T heempirical…ndings indicatethatthein‡uenceof…nancial leverage on …rms’ hiring decisions di¤ ers signi…cantly between …rms in di¤ erent…nancialregimes.

Keywords: Employmentdetermination, FinancialP olicy, P anelD ata JEL Classi…cation: J23, G 32, D 9 2

¤W earegratefulforcommentsandsuggestionsfrom P er-A ndersEdinG unnarForsling, N ilsG ottfries, B ertilH olmlund, andTorJakobKlette, aswellasfrom seminarparticipants atFIEF and U ppsalaU niversity. A llremainingerrors areofcourseours. SaraL indberg acknowledges …nacialsupportfrom Jacob W allenbergs Stiftelse.

yCorrespondingauthor: SaraL indberg, D epartmentofEconomics, U ppsalaU niversity, B ox513, S-7 51 20 U P P SA L A , e-mail: Sara.L indberg@ nek.uu.se.

1 Introduction

O verthe pastdecade there has been a renewed interestamongeconomists and policy makers in the in‡uence ofcapitalmarketimperfections on the operatingbehaviourof…rms. Firms’investmentand …nancialdecisions will generally notbe independentifthe costoravailability ofinternaland ex- ternal…nance di¤erfrom each other. T his is an importantissue since the presenceofalinkbetween …rms’realand …nancialactivitymaygiveriseto a…nancialpropagation mechanism, amplifyingthee¤ ects ofbusiness cycles and monetarypolicy, andhencehavingimportantpolicyimplications.

T henotionofsuchareal-…nanciallink…ndssupportinempiricalresearch.

T here is alarge body ofmicroeconometricevidence, relatingtocapitalin- vestment, indicatingthat…nancialfactors, suchastheavailabilityofinternal funds orthedebtposition, areimportantforexplainingvariations in …rms’

capitalexpenditures. T hereisalsoevidencethatthisdependencyvariesover di¤ erentgroupsof…rmswithdi¤erent…nancialandinstitutionalcharacteris- tics.¹ A morelimitedliteraturehas directedattention towards …rms’labour demand, reachingsimilarconclusions.²

1See, e.g., Fazzari, H ubbard & P etersen (19 88), W hited (19 9 2), Kashyap, L amont&

Stein (19 9 4), and H ubbard, Kashyap & W hited (19 9 5) forthe U nited States; D evereux

& Schiantarelli (19 9 0), B lundell, B ond, D evereux & Schiantarelli (19 9 2), and B ond &

M eghir(19 9 4) fortheU nited Kingdom;H oshi, Kashyap & Scharfstein (19 9 1) forJapan;

and H ansen & L indberg(19 9 7 ) forSweden.

2See the contributions of, e.g., N ickell& W adhwani (19 9 1), Frisse, Funke & L ankes (19 9 2), Sharpe(19 9 4), and N ickell& N icolitsas (19 9 5).

In this paper, we provide furtherevidence forthe in‡uence of…rms’…- nancialpolicyontheiremploymentdecisions, usingalargepanelofSwedish manufacturing…rms overthe period 19 7 9 -88. W hile in previous studies …- nancialfactors havebeenincludedintotraditionallabourdemandequations on an adhoc basis, theaim ofthecurrentpaperis toderiveamodelwhich is consistentwith the endogenous nature ofboth the employmentand the

…nancialdecision. T o achieve this, we adoptthe hierarchy of…nance ap- proachtocorporate…nance, andderivethesetofEulerequations governing theoptimalpathofemploymentinthepresenceofconvexadjustments costs oflabour.

A hierarchy of…nance may arise forseveralreasons, including tax dis- crimination, agency costs, asymmetric information, and costs of…nancial distress. Itimplies that…rms face a hierarchy of…nancialcosts associated with di¤erentsources of…nance. Forourpurposes, we introduce corporate taxation and agencycosts ofdebtintoastandarddynamicmodeloflabour demand. W hen the …rm solves its intertemporaloptimisation problem, it willacknowledge these distortions and take them intoaccountaccordingly.

T he…rm’s employmentand …nancingdecisions willhencebedetermined si- multaneously, which, as wewillshow, has testableimplications fortheEuler equation. T he analysis is con…ned tothe choice between retained earnings anddebt…nance;thus, newequity…nancingis disregarded. T hisabstraction may bejusti…ed becausedebtis amoreimportantsourceofincremental…- nance— inaverage, only3.7 percentofSwedish corporate…rms issued new

equityovertheperiod19 7 9 -9 5.³

A s noted byB ond & M eghir(19 9 4), thetheoryofahierarchyof…nance predicts thatsome …rms may in any period be in a regime in which oper- ations are constrained by the availability ofinternalfunds. M oreover, the probability to be in eitherregime is both time-varying and endogenous in theinvestmentequation. T heseobservations motivatedB ondandM eghirto testfordi¤erencesintheinvestmentbehaviourof…rmsacrossdi¤ erent…nan- cialregimes. W edevelop asimilartestinthispaper. O urmodelimpliesthat ifthe…rm has exhausted its retained earnings, theemploymentEulerequa- tionwillincludeaterm measuringthedegreeof…nancialleverage. Financial leveragewillhenceonlybeimportantforthose…rms thatdonothaveaccess tointernallygenerated funds, and thehypothesis ofahierarchyof…nancial costs can betested bycomparingthecoe¢cients on thedebtratiobetween theseandother…rms.

T he Eulerequation ofemploymentis estimated using the G eneralized M ethodofM omentestimator(G M M ) proposedbyA rellano& B ond(19 9 1).

U nfortunately, this estimatortends toperform poorly in …nite samples. In particular, theasymptoticstandarderrorsmaybeseverelydownwardbiased.

A saremedytothis…nitesampleproblem, however, severalG M M -bootstrap procedureshavebeensuggested⁴. Consequently, weassistourinferenceusing thebootstrap procedureproposed byB rown & N ewey(19 9 5).

O urempirical…ndings suggestthatthe Eulerequation ofemployment

3T his averageis obtained from theentireCostadata-base, describedbelow.

4See, e.g, B rown & N ewey (19 9 5) and H all & H orowitz (19 9 6). For a comparison between thesetwo, seeB ergström, D ahlberg& Johansson(19 9 7 ).

gives risetoareasonablywellspeci…edmodel. A lthoughthelong-runprop- erties ofthemodelmaybequestionable, we…nd supportforthehypothesis ofahierarchy of…nancialcosts. T here is aclearindication thatthe …nan- cialleveragecapturingtheagencycosts exerts asigni…cantin‡uenceon the labourdemandforthose…rms thatareinaregimeinwhichtheyareliquid- ity constrained in the sense thatinternally generated funds are exhausted.

H owever, because ofthe violation ofthe long-run properties ofthe model (which by nomeans is uncommon feature ofempiricalEulerequations) we thoughtitwouldbedesirabletosupplementtheEulerequationanalysiswith an analysis ofthelabourdemand relation in form ofabehavioralequation.

Failingtoobtain aclosed form solution ofthelabourdemand relation from ourmodel, we did this by estimatinga ”suggestive” labourdemand equa- tion, the properties ofwhich can be intuitively motivated from ourtheory.

T his labourdemand equation is similarin spirittothe modelin N ickell&

N icolitsas (19 9 5), and theresults obtained con…rm ourprevious …ndings on theimplications ofahierarchyof…nancialcosts.

T hepaperis organisedas follows: T henextsectionbrie‡yreviews previ- ous workin this …eld and outlines thetheoreticalfoundationforourmodel.

In section 3, we derive the Eulerequation describing the optimalpath of employmentinthepresenceofcapitalmarketimperfections and convexad- justmentcosts. Section4describes thedataandthesampleselectionproce- dure thatwe use. Italsogives abriefaccountofthe empiricalprocedures, includingthebootstrap tests used. T heeconometricresults arereported in section5, whilesection 6concludes.

2 Financialpolicyandlabourdemand

T here is only a handfulofempiricalstudies investigatingthe link between a…rm’s …nancialpolicy and its employmentdecisions, butthe overallcon- clusionis thatthe…rm’s debtpositionseems toexertasubstantialin‡uence ontolabourdemand. Introducingvarious …nancialvariables intoadynamic labourdemand schedule given by a standard bargaining model, N ickell&

W adhwani (19 9 1) …nd asigni…cantimpactofthe debttoequity ratiofora panelofB ritish …rms. In a related study, N ickell& N icolitsas (19 9 5) …nd thatan increasein interestpayments relativetocash ‡owhas alargenega- tive e¤ecton employment, and thatthis e¤ectis greaterfor…rms thatare relatively smallerorrelatively more highly leveraged. Frisse etal. (19 9 2) obtains similarresults forG erman …rms, usingastaticmodeloflabourde- mand. Finally, Sharpe (19 9 4) …nds a signi…cantrelationship between the

…nancialleverage ofU .S. …rms and the cyclicality oftheirworkforces over the 19 59 -85 period. T his cyclicality was moreoverinversely related tosize ofthe…rm. T heresults suggestthatthemacroeconomybecomes moresen- sitive to disturbances as …nancialleverage increases, and thatsmallerand morehighlyleveraged…rms tend tobeless pronetohoard labour.

T heSharpe(19 9 4) study, in particular, touches onand lends directsup- porttothenotionthat…nancialleveragehasanimportantroleasapropaga- tionmechanism inbusinesscyclesandinthetransmissionofmonetarypolicy totheeconomy. R ecentmacroeconomicliteraturegives twocomplementary viewsontheoperationofthe…nancialpropagationmechanism, bothofwhich

arebased onproblem ofasymmetricinformationin…nancialmarkets.⁵ T he …rst view is the creditview, which emphasises the availability of creditand the role played by banks as intermediaries. Itis based on the presumption thatbanks play a specialrole in the …nancialsystem as they areespeciallywellsuitedtosolveasymmetricinformationproblems incredit markets. B ecauseofthis specialrole ofthe banks, some borrowers willnot haveaccesstocreditmarketsunless theyborrowfrom banks. Soifthefunds availableforbanklending(i.e., deposits less reserveholdings and operating costs) are reduced due to tightermonetary policy orsome macroeconomic disturbance, investmentspendingwillconsequentlydecline. T his e¤ectwill be furtherreinforced ifbanks reactto the reduction ofloanable funds by settingthe interestrate belowthe marketclearinglevel, thereby rationing credit.

T hesecondview, thebalancesheetview, stresses thecreditworthiness of borrowers, suggestingthatswings in the balance sheets willamplify swings in spending. T he logicbehind this is thataweakerbalance sheetposition ofpotentialloan applicants willbe perceived by creditors as agreaterrisk, forwhich theyneed tobecompensated. T hecreditworthiness ofborrowers, commonlymeasuredasthelevelofnetworth, isthuslikelytoa¤ ectthesizeof theagencycostsassociatedwithany…nancialarrangement. Sinceborrowers’

networtharelikelytobeprocyclical, agencycosts tendtodeclineinbooms andriseinrecessions. B ythesamereasoning, agencycosts arealsolikelyto

5Foracomprehensiveoverviewofthe litteratureon …nancialand realinteraction, see G ertler(19 88) and Kashyap & Stein (19 9 3).

behigherwhenevermonetarypolicyis tight.

T heaim ofthispaperisnottostudythe…nancialpropagationmechanism overbusiness cycles and various monetary regimes as such, butratherto investigatethein‡uenceof…nancialleverageontheemploymentdecisionsof individual…rms, in thecontextofthehierarchyof…nancemodel. H owever, since theanalysis is based on creditmarketimperfections ofthesametype crucialtothe operation ofthe balance sheetchannel, ourconclusions may alsoindirectly relate tothe question ofwhetherornotsuch a propagation mechanism indeedis ine¤ect.

O urstudyextends previous researchin this areabyincorporating…nan- cialconsiderations explicitly intothe…rm’s maximisation problem, and not just, ad hoc, adding…nancialvariables tothe labourdemand relation. Ex- ceptfortheinclusionofcapitalmarketimperfections, thebasicfeaturesofthe modelfollowM achin, M anning& M eghir(19 9 3), and theempiricalanalysis is basedontheEulerequationgoverningtheoptimalpathofemploymentin thepresenceofconvexadjustmentcosts oflabour.

W e assume that the …rm can …nance employment and investment ex- penditures by internally generated funds orbydebt, thus disregardingnew equity…nance. D ebt…nanceis assumedtohaveataxadvantagetoretained earnings, buttobeassociated with agency problems duetoasymmetricin- formation. Jensen & M eckling (19 7 6) and M yers (19 7 7 ) showed thatthe limited liability provision ofdebtcontracts creates incentives for…rm man- agers toactin de…ance ofthe interestofcreditors in ordertoincrease the rateofreturn totheshareholder. Sincepotentialcreditors willacknowledge

the incentive problem, as wellas the riskofbankruptcy, they willdemand eitherinterestratepremiums orbondcovenants torestrictthebehaviourof themanagementwith respecttonewdebtissues, dividends, and the main- tenanceofworkingcapital. T hecosts thatariseinthesesituations arecom- monly known as agency costs. T hese costs are likely to increase with the degreeofleverage, sinceagreaterdebttoequityratiomeansagreaterincen- tiveformanagerstodiverge. D ebt…nancewillthereforebecomeincreasingly expensiveas …nancialleveragerises, butstillbethepreferredsourceoffunds atlowlevels ofborrowing, becauseofthetaxshield.

T he…rm willthus befacingahierarchyofcosts associatedwith thetwo sourcesoffunds. T hree…nancialregimeswillemerge: Inthe…rstregime, the

…rm will…nance investmentspendingsolelybydebt, up tothepointwhere themarginalcostofdebtequals themarginalcostofretainedearnings. A ny pro…ts willbeentirelydistributedtotheshareholders. Inthesecondregime, marginalinvestments willbe …nanced by internally generated funds atthe expenseofextradividends. Finally, whenretentions arefullyexhausted, the

…rm willbe in the third regime, and higherlevels ofinvestmentmustbe

…nancedbyadditionaldebt.

In reality, …rms willnever, orjustrarely, be in the …rstregime. L ess thanonepercentofthe…rms intheCoStadata-base⁶distributetheirentire pro…ts as dividends.⁷ W e willtherefore in the empiricalstudy focus only on the second and third regimes. Itturns outthatthe …nancinghierarchy

6Seebelowforadescription.

7 Calculatingthis ratio, wehavecorrected forthefactthat…rms areallowed tofund a certain shareoftheirpro…ts each year, thus postponingtaxation ofthesefunded pro…ts.

implies that…nancialleverage, measured as the debtto capitalratio, will enterinto the employmentEulerequation only forthose …rms thatare in thelatterregime;i.e., …rms whodonothaveaccess tointernally generated funds. T hese …rms can be identi…ed as those notpaying dividends in two consecutiveperiods, andthe…nancialhierarchyhypothesis canbeevaluated bytestingfordi¤ erencesinthedebtcoe¢cientsover…rmsinthetwo…nancial regimes. T heformalmodelis developedinthefollowingsection.

3 T heM odel

3.1 T heoreticalmodel

W e considera competitive …rm operating in a world ofimperfect capital markets. T hemanagers and shareholders ofthe…rm areassumedtoberisk neutral. W ealsoassumethatthemanagersmaketheiroperatingdecisionsin thebeginningofeachperiod, thattheyhaverationalexpectations, andthat theyacton behalfoftheshareholders in ordertomaximisethevalueofthe

…rm. L abourandcapitalareassumedtobequasi-…xedinputfactors;hence, changes in these factors willentailpositive adjustmentcosts. T ofocus on the employmentdecisions and tosimplify the analysis, we assume thatthe adjustmentcostfunctionisadditivelyseparableininvestmentandhires, and wetreatcapitalas predetermined.

In the absence ofprice bubbles, i.e. lim

T! 1 EthQT

k= 0 ¯_{t+ k¡1}VT

i = 0 , the valueofthe…rm atthebeginningofperiodtisgivenbytheexpectedpresent valueofallfuturedividendpayouts:

Vit= E

2 4X¹

j= 0

0

@ Yj k= 0

¯_{t+ k¡1}

1

ADi;t+ jj- t

3

5; (1)

where Ditis the dividends paid in period tand ¯_trepresents the …rm’s one-period discountfactor. T he discountfactoris de…ned as ¯_{t+ k¡1} = 1 fork = 0 and ¯_{t+ k¡1} = (1 + rt+ k¡1)^¡1 fork = 1;2 ;:::;1 , where rtis the shareholders’required nominalrate ofreturn between periods tand t+ 1.

Finally, E [¢j- t] = E_t[¢] denotes the expectations operatorconditionalon information availableinperiodt.⁸

T he dividends are de…ned by the identity ofsources and uses offunds whichstates thatcashin‡ows mustequalcash out‡ows:

Dit = (1¡¿)[F (N ^it;Kit;X it)¡G (H ^it)¡W^itN it¡i^t¡1Bi;t¡1

¡A(Bi;t¡1;K_i;t¡1)]+ B_it¡(1 ¡¼t)B_i;t¡1: (2)

H ereF (N it;K_it;X_it) denotestherealrevenuefunction(denominatedinterms oftheoutputprice), whereN itis labour, Kitis capital, and X itis allother typesofinputfactors. R evenuesareassumedtobeincreasingatadecreasing rate with labour. G (H it) is the realadjustmentcostfunction, assumed to beincreasingand convexinhires, H it, whichis de…nedas

H_it= N _it¡(1 ¡±)N i;t¡1: (3)

8U ncertaintymaycomefrom anumberoffactors, suchas futureinterestrates, input/

outputprices, and technologies.

where 0 ·± ·1 is thevoluntary quitrate. A(Bi;t¡1;Ki;t¡1) is the agency costfunction, whereBi;t¡1 istheamountofone-periodrealdebtissuedinthe previous period and Ki;t¡1 is the capitalstock, hererepresentingcollateral.

T he agency costs are assumed tobe increasingatan increasingrate in the levelofdebt, butdecreasingin the levelofcapital. Finally, Witis the real wagecost, it¡1thenominalinterestratepaidonloans takeninperiodt¡1,

¿ the corporate tax-rate, and ¼ttherateofproductpricein‡ation between periodt¡1 andt.

T hus, equation(2) states thatthedividendpayouts mustbeequaltothe sum ofafter-tax pro…ts and the change in the stock ofdebt. In addition, dividends arerestrictedtobenon-negative:

Dit¸0 : (4)

N ow, the…rm’s objectiveis tochoosethelevels ofemploymentanddebt soas tomaximizeits value, equation(1), subjecttotheconstraints (2), (3), and (4). T osolvethis problem, form theL agrangian function, letting¸itbe theseries of(currentvalue) Kuhn-T uckermultipliersassociatedwith(4) and substitutingforDitandH itusing(2) and(3). T he…rstorderconditions for labourand debtcanthenbederivedas

@Fit

@ N it¡@ Git

@ H it¡W^it= ¡¯tEt

"

1 + ¸i;t+ 1

1 + ¸it

(1¡±)@G i;t+ 1

@H i;t+ 1

#

; (5)

(1 + ¸it) = ¯_tEt

"

(1 + ¸i;t+ 1)

Ã

1 + (1¡¿)

Ã

it+ @Ai;t+ 1

@ Bit

!

¡¼t+ 1

! #

; (6)

forany period t= 0 ;1;:::;1 . Equation (5) is the Eulerequation, charac- terisingthe optimalpath ofhirings alongwhich the …rm willbe indi¤ erent betweenhiringsomeonetodayandpostponingthehireuntiltomorrow. T his equation maybemoreeasilyinterpreted ifwerearrangeitslightly:

@ F_it

@ N it¡Wit= @ G_it

@ Hit¡¯tEt

"

1 + ¸_{i;t+ 1} 1 + ¸it

(1¡±)@G_{i;t+ 1}

@H i;t+ 1

#

: (7 )

Firstconsiderthecasewherethe…rm payspositivedividends, i.e. ¸i;t+ 1 =

¸it= 0 . O nthelefthand sideofequation (7 ) is thenetcontribution tocash

‡owin period tfrom addinganewworkertothe labourforce. Itincludes therealmarginalrevenueproductless therealwagecost. O ntherighthand side is the marginalnethiringcost, comprising the di¤erence between the marginalhiringcostincurredin period tand thediscounted expected value ofthemarginalhiringcosttoberecoveredin periodt+ 1. T helatterarises sincethe…rm has toincurless expenses inperiodt+ 1 inordertoobtainthe samenumberofemployees inthesubsequentperiodandonwards. Equation (7 ) thus tells us thatthe marginalbene…tofanewworkershould be equal tothemarginalnethiringcost.

N extconsiderthe case when eitherofthe non-negativity constraints on dividends is binding. T heKuhn-T uckermultiplierassociated with thebind- ing constraint will then take on a positive value, and the term

(1 + ¸i;t+ 1)= (1 + ¸it) willwork as an additionalweighton the righthand sideexpression. T his means thatiftheconstraintis bindinginperiodt, the hiringcosts tobe recovered in t+ 1 willbe discounted more heavily. T he marginalnethiringcostwillhence be perceived by the…rm as higher, and hirings willbereducedinaccordancewiththat. Consequently, a…rm thatis

…nanciallyconstrainedinthesensethatithas fullyexhausted its retentions willhirefewerworkers thana…rm thathas not, everythingelseequal.

Equation (6) is the …rst-ordercondition fordebt, stating thatthe …rm shouldborrowup tothepointwhereitis indi¤erentbetweenoneadditional unitofdebtand one additionalunitofretentions. A ssumingthatthe …rm pays positivedividends in both periods, themarginalbene…tofissuingone unitofnewdebtto…nancedividendpayments mustbeequaltothepresent valueofrepayingthedebtplus interest. H owever, ifthedividendconstraint isbindinginperiodt, themarginalcostofnewdebtwillexceedthemarginal bene…ts, implyingthatthediscountfactorinequation (7 ) reads

¯_tEt

"

1 + ¸i;t+ 1

1 + ¸it

#

= 1

1 + (1¡¿)(it+ ABit)¡¼^et+ 1

: (8)

Equations (7 ) and (8) form thetheoreticalbasis forourempiricalinves- tigation in section 5. T he empiricalspeci…cation ofthe modelis derived below.

3.2 Empiricalspeci…cation

In ordertogetaspeci…cation thatcan beestimated, weneed afewfurther assumptions on explicitfunctionalforms oftherevenues, as wellas the ad- justmentand agencycosts. FollowingM achin etal. (19 9 3), weassumethat therevenueandadjustmentcostfunctions aregivenby

F (N it;Kit;Xit) = f (Kit;Xit)N _it^®; (9 )

G (H it) = °

2 H _it²; (10)

where ® and °are greaterthan zero⁹. Forthe agency cost function, we assumethefollowingform:

A(B_i;t¡1;K_i;t¡1) = Á 2

B_i;t² _¡1 Ki;t¡1

; (11)

forpositive values ofBi;t¡1, and zerootherwise. Á is the agency costpara- meterassumedtobegreaterthanzero.

D i¤erentiating(9 ), (10), and(11) withrespecttoN it, H it, andBit, wein- serttheresultinto(7 ) and(8). U ndertheassumptionofconstantconditional covariancebetween(µ + ¸i;t+ 1)= (µ + ¸it) and othertimet+ 1 variables, (8 ) can then besubstitutedinto(7 ). U singtheassumption ofrationalexpecta- tions ontheresultingexpressionandrearrangingyields:

9W e use quadraticadjustmentcosts formathematicalconvenience. H owever, we ac- knowledgethatthereislimitedempiricalsupportforthisassumption.(See, e.g.,H amermesh (19 9 3)).

°(1¡±) 1 + (1¡¿)^³it+ Á^³_K^B´

it

´¡¼^et+ 1

(N i;t+ 1¡(1 ¡±)N it) =

°(N it¡(1 ¡±)N i;t¡1) + Wit¡®

µQ N

¶

it+ fi+ st+ 1+ "i;t+ 1; (12) where Q = f (K ;X )N ^® denotes realsales, fiis a…rm-speci…c…xed e¤ ect, st+ 1 is a time-speci…c …xed e¤ ect, and "i;t+ 1 is the forecast error, serially uncorrelatedandorthogonaltoanyinformationinperiodt. T he…rm-speci…c e¤ ectisincludedtoallowforheterogeneityamong…rmsduetodi¤erencesin, e.g., growthopportunities ortheskillofthemanagement. T hetime-speci…c e¤ ectwillcapturemacroeconomicshocks thatarecommontoall…rms.

Finally, T aylorexpanding(12) abouttheindividualmeanofthevariables yields theEulerequation foremploymenttobeestimated:

N i;t+ 1 ' ¯1N it+ ¯₂N i;t¡1+ ¯₃Wit+ ¯₄

µQ N

¶

it

+ ¯₅

µB K

¶

it

+ f_i^¤+ s^¤_{t+ 1}+ "^¤_{i;t+ 1}; (13) wheref_i^¤, s^¤_{t+ 1}, and"_{i;t+ 1}^¤ di¤eronlyproportionallyfrom fi, st+ 1, and"i;t+ 1.¹⁰ T hereduced form parameters ¯_iarerelated tothestructuralparameters as follows:

10N ote thatthe in‡ation rate and the interestrate are subsumed by the time-speci…c e¤ ect.

¯₁ = ^(1¡±)_(1¡±)^{2+ ª} ¯₂ = ¡ª

¯₃ = _°(1^ª_¡±) ¯₄ = ¡_°(1^{® ª}_¡±)

¯₅ = ^{Á±N (1}_ª ^¡¿)

with ª = 1 + (1 ¡¿)^³¹{+ Á^³B =K^´´¡¹¼.

T hemodelis clearlyoveridenti…edinterms ofthestructuralparameters, butwe can stillsay something aboutthe reduced form parameters.. T he modelpredicts that: ¯₁ > 2 , ¯₂ < ¡1, ¯3> 0 , ¯₄ < 0 , and ¯₅> 0 . N ote thatthesum of¯₁ and ¯₂ willbegreaterthan one, implyingthatthesigns ofthe coe¢cients on Q =N , W , and B =K mustbereversed in thelongrun.

T his implies, in particular, thatan increase in the debtratio willhave a negative long-run e¤ecton employment. N ote alsothatthe debttoequity ratiois only expected toentertheequation ifthe…rm has exhausted allof its retentions.

4 EstimationandD ata

4.1 M ethodologicalconsiderations

W eestimateequation(13) usingtheG eneralizedM ethodofM omentsestima- tor(G M M ) developed by A rellano& B ond (19 9 1). T his estimatorexploits the linearmomentconditions implied by the model. T he individuale¤ects are removed by …rst-di¤ erencing the modeland the resulting endogeneity problem is handled byusingappropriateinstruments.

Since the errors in levels are serially uncorrelated, as implied by ratio-

nalexpectations, the errors in …rst-di¤ erences willfollowa…rst-orderM A - process. T his implies thatvalues ofthe dependentvariable (in levels) and otherright-hand side variables mustbe lagged atleasttwotime-periods in ordertobeuncorrelated with thedi¤erenced errorterm and, thereby, valid as instruments in the transformed model. A s the consistency ofthe G M M estimates criticallyhinges onthevalidityoftheinstruments, wetestforthis assertion, using the Sargan-testand the A rellano-B ond m2-criterion. T he Sargan-testtests forthenullofnocorrelationbetween theinstruments and the residuals, while the m2-testtests forthe lack ofsecond-ordercorrela- tion. A corresponding m1-testtests forthe lack of…rst-ordercorrelation.

Estimationis carriedoutusingtheD P D program forO x1.20.¹¹

U nfortunately, thereis atendencyfortheasymptoticstandard errors of the second-step A rellano-B ond estimatorto be downward biased in small samples. T here are two ways ofhandling this. O ne could eitherstick to theconsistent…rst-step estimates, whichhavebetter…nitesampleproperties butare potentially ine¢cient;orone could go forthe e¢cientsecond-step estimates and employabootstrap t-test, which has bettersizeproperties¹², forinference. W e make use ofboth approaches and report the …rst- and second-step asymptotictests togetherwiththebootstrap tests.

T he idea ofthe bootstrap testis simple. Itamounts to estimatingthe distribution ofthe teststatisticby resamplingthe data. W hen performing an asymptotic t-test, the nullhypothesis is rejected ifthe teststatistic is

11Foradescriptionoftheprograms, seeD oornik(19 9 6) andA rellano, B ond& D oornik (19 9 7 ).

12SeetheM onteCarloevidenceinB ergström (19 9 7 ).

unlikely tohave come from at-distribution. W hen performingabootstrap test, we …rstneed toestimate the distribution ofthe teststatistic. T his is done by treating the data as the population and the estimated parameter value, say ^¯ , as thetruevalue. From this ”fake” population, wethen create alarge numberofsamples by randomly drawingobservations with replace- ment, andreestimatethe¯ -parameterforeachofthenewsamples(usingthe samemethodasbefore). T helogicofabootstrap testisthatthereestimated values, say ~¯, willon average be equaltothe ”true” value, ^¯ . Soby con- structingt-tests ofthenull~¯ = ^¯ (which we knowwillbe true in repeated resampling) forallnewsamples, wemayinthiswaysimulatethedistribution ofthe t-statisticunderthe true null. T his distribution willnotonly re‡ect thecharacteristics oftheD G P (datageneratingprocess), butalsothesizeof theoriginalsample.

Forming tests based on the bootstrap distribution has proved to be at leastas reliable as using the asymptotic distribution, and, in mostcases, considerablymorereliable.¹³ G M M -bootstrappinginvolvesafewothertech- nicalobstacles thatmustbeovercome. W edonotwantgointotheseissues here, andreferthereadertoB rown& N ewey(19 9 5) forarigorous treatment oftheissues involved.

4.2 D ataandsample selection

T he data we use is a sample from the CoSta-base which contains annual incomestatements andbalancesheets ofSwedish …rms overtheyears 19 7 9 -

13Seee.g. D avidson & M acKinnon (19 9 6).

19 9 4.¹⁴ T hedatamaterialis collected byStatistics Sweden and includes all manufacturing…rms withmorethan20 employees. A ttimes, …rms leavethe sample, eitherdue to bankruptcies ormergers orbecause they have ”con- tracted” themselves outfrom this class of…rms. Since we cannotidentify thecauseofwhya…rm leaves thesample, weneedtorestrictouranalysis to those…rmsthatremaininthesampleovertheentireperiod, whichwouldgive us abalanced panelof16years. T hereis, however, aselfselection problem with this approach. B y discarding…rms thathave”contracted” themselves outofthe sample, factors with anegative in‡uence on employmentwillbe underestimated. T o cope with this, we use the following sample selection procedure:

1. Includeinthesampleonlythose…rms thathavesu¢cientlymanyem- ployees in the…rstyear(19 7 9 ) toallowthe …rms tocontractoverthe time-periodwithoutleavingthesample.

2. Inspectthe numberofemployees forthe lastobservation ofeach …rm sampled. Ifthis is less thansomecriticalvalue, labelthe…rm ”doubt- ful”.

3. Selectthe length ofthepanelas longas possible, butmakesurethat no”doubtful” …rms arethrown away.

In ordertokeep as manyobservations as possible, wechooseadesign in which the required numberofemployees in the …rstyearis 50, the critical

14Forathorough description oftheCoSta-database, seeH ansen (19 9 7 ).

numberofemployees is 30, and the length ofthe panelis 10 years. T his leaves us with a panelof47 9 …rms overthe period 19 7 9 -19 88. O utofthe

…rms discarded, three…rms werelabelled”doubtful”;i.e., theyhadless than 30 employees as their…nalobservation. T he exclusion ofthese three will hardlycauseanysigni…cantself-selection bias.

T hevariables usedintheempiricalanalysis includelabour, N , realwage costs, W , productivity, Q =N , and thedebttoequityratio, B =E. N otethat weusethedebttoequityratioinsteadofthedebttocapitalratio, as inthe derivationofthemodel. T hisavoidstheproblemsassociatedwithmeasuring thevalueofthecapitalstock. Itmayalsoconstituteabettermeasureofthe

…rm’s …nancialstrenght, ifthe…rm has otherassets thancapital.¹⁵

N is measured as the average numberoffulltime employees, which is roughlythetotalnumberofhours workeddividedbytheworkingtimecon- sideredasstandardhoursateach…rm. W isobtainedbydividingtotalwage costs bythenumberofemployees andQ =N is measuredas realsales divided bythenumberofemployees, both ofwhich arede‡ated usingtheP roducer P riceIndexatatwodigit(SN I) industrylevel. B =E is calculatedusingthe bookvalues ofdebtand equityin accordancewith therecommendations of theSwedish A ssociation ofFinancialA nalysts. T he useofbookvalues was necessary since we have no access to marketinformation. H owever, ifwe wantto make a virtue outofthis necessity, we could argue thatthe book valuesaremoreappropriatetouse, asthosearethe…guresthatbankersgen-

15Fora…rm with nootherassets than capital, B =E is simplybeamonotonictransfor- mation ofB =K: B =E = (B =K)=(1 ¡(B =K)):

erally would audit. A more detailed description ofthe variables and some descriptivestatistics aregiveninthedataappendix.

5 R esults

5.1 Eulerequationestimates

W estartbyestimatingequation(13) withoutallowingfor…nancialregimes.¹⁶ T he results are reported in the …rstthree columns ofT able 1. A s can bee seen, the speci…cation tests do not indicate any problems that would af- fectthe consistency ofthe estimates. Even though the testfor…rst-order serialcorrelation does notindicate asigni…cantly negative correlation as is expected, both the Sargan and the m2 tests suggestthatthe instruments arevalidandthatthespeci…cationis acceptable. T hesetofinstruments in- cludes values ofallvariables inlevels, laggedup tothreeperiods. W hilethe inclusionoffurtherlags as instruments increases thee¢ciencyofestimation, italsoleads theSargantesttoemphaticallyrejectthespeci…cation.

A llcoe¢cients havethesigns thatwewould expectfrom thetheoretical model, with one importantexception: the coe¢cienton the debtratio is negative. B eforeweaddress this issue, we may …rstnote thatthe…rst-step estimates in the …rstcolumn suggestthatonly the lagged dependentvari- ables exertany explanatory powerin the Eulerequation;none ofthe other

16A ll estimations in this section have been performed both with and without time- speci…c …xed e¤ ects with virtually the same results. A s the time e¤ ect in mostcases wereindividuallysigni…cantandalways jointlysigni…cant, wehavechoosentopresentthe estimations includingthesee¤ ects.

estimates aresigni…cantlydi¤erentfrom zero. H owever, consideringthefact thatthe…rst-step estimates areine¢cient, this resultshouldnotbeallthat surprising, and itdoes notpersistin the second-step ofestimation. O n the contrary, in some cases the t-statistics associated with the second-step es- timates are remarkably high, which makes us suspectthatthese standard errors may be estimated with a severe downward bias. T his suspicion is supported by the high bootstrap p-values reported in the third column, al- thoughoneshouldbearinmindthatthebootstrap procedureweuseditself haveatendencytounder-rejectthenullhypothesis. (SeeB ergström (19 9 7 ) forM onte Carlo evidence). T aken together, the evidence seem to suggest thatthevariables included doexertsomein‡uenceupon employment, even though thesigni…cance ofthe wagecosts and the productivity measure are somewhatambiguous.

A s we noted above, the negative coe¢cienton the debttoequity ratio (¡0 :12 ) is notconsistentwith the theoreticalmodel. T herefore, following ourprecedentlineofarguments, wereestimatethemodel, nowallowingfor thepresenceof…nancialregimes, withthedebttoequitycoe¢cientsvarying overthese. T his is donesimplybycreatingtwodummyvariables which are interacted with thedebtratio, the …rstpickingoutthose …rms thatdonot pay positive dividends in twoconsecutive years and the second pickingout the remaining …rms. A s a directimplication ofthe theoreticalmodel, we shouldexpectthedebtratiotobestatisticallysigni…cantonlyforthesubset of…rms thatare notpayingpositive dividends in twoconsecutive periods.

Sincethese…rms are, insomesense, liquidityconstrained, theiremployment

decisions arelikelytobein‡uencedbytheir…nancialstatus. T heresults are presentedinthelastthreecolumns ofT able1.

A scanbeseen, thespeci…cationisnotrejectedbythespeci…cationstests, andthecoe¢cientsonallvariablesotherthanthedebttoequityratioremain practicallythesame. In contrast, thedebtcoe¢cientchanges substantially.

Forthose …rms thatare notpaying dividends, the debtcoe¢cientis now positive (0 :12 ) as implied by theory, whereas the coe¢cient fordividend- paying …rms is no longersigni…cantly di¤ erent from zero. T his supports thehypothesis ofahierarchyof…nanceand suggests that…nancialleverage willonly be importantin a regime in which internally generated funds are exhausted.

Itshouldbenotedthatthebootstrap t-testofthedebtcoe¢cientassoci- atedwiththegroup ofnon-dividendpaying…rms (p = 0 :11) does castsome doubtsonitslevelofsigni…cance. N onetheless, bearinginmindthepotential size distortion ofthe bootstrap testand alsobeing reassured by the small di¤ erence between the …rstand second step estimates, we are inclined to believethatthisdoesnotcontradictthehypothesesofahierarchyof…nance.

5.2 L abourdemandrelation

T he main source ofdisappointmentin ourempiricalmodelis the violation ofthe long run properties suggested by theory: the coe¢cients on lagged employmentdonotsum toa value greaterthan unity. T his result, which bynomeans is uncommonforempiricalEulerequations ofthis type, makes the parameterestimates obtained somewhatless clear-cutto interpret. It

T able1: EulerEquation ofEmployment

I. O riginalSpeci…cation II. Financialregimes V ar 1-step 2-step B ootp 1-step 2-step B ootp

N it 0.49 49 0.49 40 0.49 51 0.49 49

(3.53) (410) 0.000 (3.54) (37 7 ) 0.000

N i;t¡1 0.0821 0.0831 0.0821 0.0832

(2.63) (385) 0.000 (2.63) (329 ) 0.000

W it 0.7 17 0 0.2465 0.7 614 0.2625

(1.56) (5.08) 0.119 (1.54) (4.9 1) 0.100

Q /N it -0.0231 -0.009 5 -0.0236 -0.0101

(-0.48) (-1.7 6) 0.443 (-0.48) (-1.7 8) 0.438

B /Eit -0.1110 -0.1211

(-0.62) (-8.19 ) 0.010

B /E^{nod iv}_it ^{0.117 5 0.1218 0.114}

(0.42) (6.54)

B /E^{d iv}_it -8.6383 -4.8117 0.632

(-0.7 6) (-1.61)

m₁: -1.07 S₁(65): 0.000 m₁:-1.07 S₁(64): 0.000 m₂: -1.23 S₂(65): 0.149 m₂: -1.23 S₂(64): 0.163 W ald (7 ): 0.000 W ald (7 ): 0.000 N otes:

(i) G M M estimates, obtained usingD P D forO x1.20. Foradescription oftheprograms, seeD oornik(19 9 6) and A rellanoetal. (19 9 7 ).

(ii) t-statistics are reported in parentheses. T hey are computed using the asymptotic standard errors, which are obtained usinga heteoscedasticity-robustvariance-covariance matrix. B ootp gives thep-values ofthecorrespondingbootstrap t-test. T hroughout400 bootstrap replications havebeenused.

(iii) A llregressions includetimedummies. W aldgives thep-valueofanasymptoticW ald- testofthejointsigni…canceofthedummies. U nderthenullofnorelationship, theW ald statisticisasymptoticallydistributedas²(k), wherekisthenumberofcoe¢cientstested.

(iv)m₁andm₂ areteststatisticsfor…rst- andsecond-orderserialcorrelationinresidualsin

…rstdi¤ erences. B othareasymptoticallystandardnormalunderthenulls of, respectively, no…rst- and second-ordercorrelation.

(v) S₁(S₂) gives thep-valueoftheasymptotic…rst(second) step Sargan testoftheover- identifyingrestrictions (validity ofinstruments). U nderthenullofvalid restrictions, the Sarganstatisticisasymptoticallydistributedas²(p-k), wherep isthenumberofmoment conditionsandkis thenumberofcoe¢cientsestimated. T hesetofinstrumentsincludesN andQ /N lagged2 and3years, B /E andW lagged1, 2, and3years, andalltimedummies.

would therefore be desirable toestimate the labourdemand relation in the form ofabehaviouralfunction, asacomplementtotheEulerequationanaly- sis. B ecause ofthe involved structure ofthe model, we have notcome up with aclosed form solution forsuch alabourdemand relation. Instead, we haveestimatedabare-bones log-linearlabourdemandequationtowhichwe have added our…nancialvariables ad hoc, in the same manneras in e.g.

N ickell& N icolitsas (19 9 4). Following theirline ofarguments, we should expect…nancialleverageoftheprevious periodtoexertanegativein‡uence on labourdemand, something which also is intuitively consistentwith our previous model.

A llowingfor…nancialregimesinthesamemannerasbefore, theempirical modelweestimateis basedontheequation

nit= ®1nit¡1+ ¯₁wit+ °₁B =E^{nod iv}_i;t¡1 + °₂B =E^{d iv}_i;t¡1+ fi+ st+ "it

wherelowercaseletters indicatelogs¹⁷ . W ethen add su¢cientdynamics to theequation, inordertoobtainavalidspeci…cation, andestimatetheresult- ingequationin…rstdi¤ erencesaspreviously¹⁸. T hemaindi¤ erencebetween this speci…cation and theEulerequation representation, is theintroduction ofcontemporaneous wagecosts.

A s canbeseenfrom theresults inT able2, this speci…cationis approved

17 Foradiscussion on theendogeneity problem which makes itnecessary touselags of the…nancialvariable, seeN ickell& N icolitsas (19 9 4).

18Introducingthe productivity measure orindustry speci…ctrends did notchange the results substantially. W e therefore reporttheresults from themore parsimonious model in which theseregressors areomitted.

T able2: L abourD emandEquation

I. FinancialR egimes V ar 1-step 2-step B ootp

n_i;t¡1 0.7 7 15 0.7 47 0

(11.7 6) (32.3) 0.000

ni;t¡2 -^0.1841 -^0.1652

(-5.34) (-12.4) 0.000

ni;t¡3 0.109 7 0.09 16

(3.55) (8.10) 0.005

wit -0.2036 -0.1584

(-0.02) (-5.53) 0.030

wi;t¡1 0.09 7 8 0.047 0

(2.14) (1.9 6) 0.468

w_i;t¡2 -0.089 2 -0.09 88

(-3.89 ) (-6.29 ) 0.000

B /E^{nod iv}_i;t¡1 -^0.0008 -^0.0009

(-2.7 1) (-15.1) 0.030

B /E^{nod iv}_i;t¡2 -0.0001 -0.0001

(-0.24) (-2.31) 0.562

B /E^{d iv}_i;t¡1 0.0126 0.0112

(3.23) (4.85) 0.09 5

B /E^{d iv}_i;t¡2 ^{0.008 0.009 8}

(2.33) (6.42) 0.085 m₁: -3.9 6 S₁(128): 0.000 m₂: 0.9 3 S₂(128): 0.416

W ald (7 ): 0.000 N otes:

(i) G M M estimates, obtained usingD P D forO x1.20.

(ii) T he setofinstruments includes alllags available from period t-2 forn, w, and q/n, and alllags availablefrom period t-1 forB /E, as wellas alltimedummies.

(iii) Seethenotes in table1 forfurtherexplanations.

byalltests performed. N otethatthem1 nowindicates asigni…cantnegative

…rst-orderserialcorrelation, asonewouldexpectfrom taking…rstdi¤erences.

L ookingattheparameterestimates, theydoresemblethoseofalabourde- mand relation with a fairly lowshortrun employment-wage elasticity and withalongrunelasticityatroughly-0.6,whichseemsratherreasonable. Fur- thermore, the negativesigns ofthedebtcoe¢cients ofnon-dividend paying

…rmsareconsistentwiththe…ndingsofN ickell& N icolitsas(19 9 4), although the magnitude ofthe coe¢cients may appeartobe rathersmall. H owever, thesummarystatisticsinthedataappendixtellusthatthevariablewehave chosentocapturetheagencycosts maypotentiallyincreasequitedrastically forindividual…rms;anincreaseofonestandarddeviation, startingoutfrom the mean, implies a 400 percentincrease ofthis ratio. T he actuale¤ects from changes in this ratioontoemploymentmay hencebe substantialeven thoughthecoe¢cients appearsmall.

T heevidenceonstatisticalsigni…canceofthedebttoequityratioofdiv- idendpaying…rms is ambiguous. O ntheonehand, the…rstandsecondstep asymptotic t-tests suggestthatthe coe¢cients are statistically signi…cant.

O ntheotherhand, thebootstrap t-testsuggeststheyarenot. Inviewofour earlierresultand the potentialsize distortion ofthe bootstrap test, we are in this case inclined tobelieve thatsome e¤ects are actually present. T his resultis, ofcourse, notconsistentwiththetheoryof…nancialregimes, which predictsthat…nancialleverageshouldbeirrelevantfortheemploymentdeci- sionofdividendpaying…rms. B utevenso, thereisacleardi¤ erencebetween the twosubsets of…rms. T he debt-coe¢cientofnon-dividend paying…rms

is negativeincoherencewithourapriori belief, whereas thedebtcoe¢cient ofdividend paying…rms is positive, aresultwhich …ts badlyintothetheo- reticalframeworkofagencycosts. W emakenoattempts tryingtointerpret this result, butjustnotethatthereareimportantdi¤erencesinthein‡uence of…nancialleveragebetween…rms in di¤ erent…nancialregimes.

6 Conclusions

U singthehierarchyof…nanceapproach, wehavederivedan Eulerequation modelofemploymentsuggestinga mechanism through which the …nancial policy of…rms may a¤ ecttheirhiringdecisions and, thus, theirlabourde- mand. T his modelwas estimated usingalarge panelofSwedish manufac- turing…rms overtheyears 19 7 9 -88. A fewpoints areworthstressing:

²Inaccordancewithearlierstudies, we…ndthat…nancialleverageexert an important in‡uence on the hiring decisions of …rms. T he main conclusion is thatthis e¤ectis related to which …nancialregime the

…rm …nds itselfin. Financialleverage only appears to be important forthose…rms thatareinaregimeinwhichretentions havebeenfully exhausted.

²T heEulerequationofemploymentgives risetoareasonablywellspec- i…ed model, although the long-run properties of the model may be questionable. H owever, estimatinga bare-bones log-linearlabourde- mandrelationas acomplementtotheEulerequationanalysis, weget

qualitatively similarresults, givingsome supportforthe hierarchy of

…nancehypothesis.

²T he bootstrap procedure we have used here con…rms thatthe down- wardbiasofthesecond-step estimatesmightbeanimportantproblem.

H owever, as thebootstrap t-testitselfappears tobesomewhatunder- sized, inference may still not be clear-cut. N evertheless, we believe thatbootstrap tests are usefuland essentialcomplements to asymp- totictests in applications ofthis type.

R eferences

A rellano, M ., B ond, S. & D oornik, J. A . (19 9 7 ). D ynamic paneldata es- timation using D P D for O x. N u¢eld College, O xford, available at http://hicks.nuff.ox.ac.uk/Users/Doornik/.

A rellano, M .& B ond, S.R .(19 9 1). Sometestsofspeci…cationforpaneldata:

M onteCarloevidenceandanapplicationtoemploymentequations, R e- viewofEconomicStudies58: 27 7 –29 7 .

B ergström, P. (19 9 7 ). O n bootstrap standard errors in dynamicpaneldata models., W orkingpaper23, D epartmentofEconomics U ppsalaU niver- sity.

B ergström, P., D ahlberg, M . & Johansson, E. (19 9 7 ). G M M bootstrapping and testingin dynamicpanels, W orkingpaper10, D epartmentofEco- nomics U ppsalaU niversity.

B lundell, R ., B ond, S., D evereux, M . & Schiantarelli, F. (19 9 2). Investment andT obin’s Q , JournalofEconometrics51(1-2): 233–57 .

B ond, S. & M eghir, C. (19 9 4). D ynamicinvestmentmodels and the …rm’s

…nancialpolicy., R eviewofEconomicStudies61: 19 7 –222.

B rown, B . & N ewey, W . (19 9 5). B ootstrappingforG M M , M imeo, M IT .

D avidson, R . & M acKinnon, J. G . (19 9 6). T he size distortion ofbootstrap tests, D iscussion P aper9 37 , Q ueen’s InstituteforEconomicR esearch.

D evereux, M . & Schiantarelli, F. (19 9 0). Investment, …nancialfactors, and cash ‡ow: Evidence from U .K. panel data, in G . R . H ubbard (ed.), A symmetric Information, Corporate Finance and Investment, U niver- sityofChicagoP ress, Chicago, pp. 27 9 –306.

D oornik, J. A . (19 9 6). O bject-oriented M atrix P rogramming U sing O x’, T homsonB usiness P ress, L ondon.

Fazzari, S., H ubbard, R . & P etersen, B . (19 88). Financingconstraints and corporateinvestment, B rookings P apers on EconomicA ctivity(1): 141–

9 5.

Frisse, K., Funke, M . & L ankes, F. (19 9 2). Financialstructure and labour demand of W est G erman industrial and commercial companies: A studywithlongitudinaldata, Jahrb. fürN ationalökonomieundStatistik 209 : 106–118.

G ertler, M .(19 88). Financialstructureandaggregateeconomicactivity: A n overview, JournalofM oney, CreditandB anking20(3): 559 –88.

H all, P. & H orowitz, J. L . (19 9 6). B ootstrap criticalvalues fortests based onG eneralized-M ethod-of-M oments estimators, Econometrica64: 89 1–

9 16.

H amermesh, D . S. (19 9 3). L aborD emand, P rincetonU niversityP ress.

H ansen, S. (19 9 7 ). CoSta- Corporate Statistics. U ppsalaU niversity, D ept ofEconomics.

H ansen, S. & L indberg, S. (19 9 7 ). A gencycosts, …nancialderegulation, and corporate investment. A n Eulerequation approach to panel data for Swedish …rms, W orking paper 20, D epartmentofEconomics U ppsala U niversity.

H oshi, T ., Kashyap, A .& Scharfstein, D .(19 9 1). Corporatestructure, liquid- ity and investment: Evidence from Japanese industrialgroups, Q uar- terlyJournalofEconomics106(1): 33–60.

H ubbard, G . R ., Kashyap, A . K. & W hited, T . M . (19 9 5). Internal…nance and…rm investment, JournalofM oney, Credit, andB anking27 (3): 683–

7 01.

Jensen, M .& M eckling, W .(19 7 6). T heoryofthe…rm: M anagerialbehavior, agencycosts, andownership structure, JournalofFinancialEconomics 3(4): 39 5–60.

Kashyap, A . K., L amont, O . A . & Stein, J.C.(19 9 4). Creditconditions and the cyclical behavior of inventories, Q uarterly Journal of Economics 109 (3): 565–9 2.

Kashyap, A . K. & Stein, J. C. (19 9 3). M onetary policy and bank lending, N B ER W orkingP aper4317 .

M achin, S., M anning, A . & M eghir, C. (19 9 3). D ynamic models ofem- ployment based on …rm-level panel data, in J. V an O urs, G . P fann

& G . R idder(eds), L abor D emand and Equilibrium W age Formation, N orth-H olland, A msterdam, pp. 167–19 5.

M yers, S.(19 7 7 ). D eterminantsofcorporateborrowing, JournalofFinancial Economics5(2): 147 –7 5.

N ickell, S. & N icolitsas, D . (19 9 4). W ages, restrictivepractices and produc- tivity, W orkingpaper219 , CentreforEconomicP erformance, L SE.

N ickell, S.& N icolitsas, D . (19 9 5). H owdoes …nancialpressurea¤ect…rms?, D iscussion P aper17 0, InstituteofEconomics andStatistics, U niversity ofO xford.

N ickell, S. & W adhwani, S. (19 9 1). Employmentdetermination in B ritish industry: Investigations usingM icro-D ata, R eviewofEconomicStudies 58(5): 9 55–69.

Sharpe, S. A .(19 9 4). Financialmarketimperfections, …rm leverage, andthe cyclicalityofemployment, A merican EconomicR eview84: 1060–107 4.

W hited, T .M .(19 9 2). D ebt, liquidityconstraints, andcorporateinvestment:

Evidencefrom paneldata, JournalofFinance47 (4): 1425–60.

A D ataA ppendix

A .1 T he Sample

T hesampleweuseconsists of47 9 Swedishcompanies observedovertheyears 19 7 9 – 19 88 and is drawnfrom theCoStabase, whichcontains annualincomestatements and balance sheetforalargenumberof…rms. Itwas selectedaccordingtothefollowingcriteria, upon whichourresults areconditional:

²O nly …rms in the manufacturingindustry is included, i.e., those belongingtoSN I categories 31–38.

²O nly …rms with more than 50 employees are included. T his is due tothe sample selection problem whichis thoroughlydescribed in section 4.3 ofthis paper.

²Firms musthavestrictlypositivewages andsales. T his criterionis appliedinorder toexcludeobviouslyerroneous data.

²Employmentmustnot‡uctuatewith morethan afactorof3 between twoconsec- utiveyears. T his requirementis used toeliminatechanges in employmentwhich is duetorestructuringofcompanies, ratherthantothefactors thatweseektostudy.

T his restriction, however, proved tobenon-restrictivein thesampleselection.

A .2 D e…nitions ofV ariables

T hevariables used in this paperarethefollowing.

²^Employment(N it)is expressed as thenumberof”e¤ ective” workers accordingto thede…nitionoftheCentralSwedishA ccountingB oard(B okföringsnä mnden), B FN

R 4. Itis calculatedas totalhours workeddividedbythe”normal” workingtimeof an employee. CoSta: V ar001 M ean: 424.3 Std dev: 1661

²W agecosts (W it)istotalwagecosts, includingpayrolltaxes, dividedbythenumber ofemployees. Itis expressedinrealterms usingtheP roducerP riceIndexatatwo digit(SN I) industrylevel. CoSta: V ar004/(V ar001£^{P P Ijt}) M ean: 148.9 Std dev:

24.46

²^Sales(Qit)includes both domestic and international sales. It is expressed in realterms using the P roducerP rice Index ata two digitindustry level. CoSta:

V ar005/P P I_jtM ean: 662.3 Std dev: 57 9 .3

²^{D ebt}(Bit)is computedas thesum ofcurrentpayables andlong-term liabilities to corporate group members, currentloan liabilities, and otherlong-term liabilities.

Costa: V ar07 8+ V ar084+ V ar081+ V ar086. M ean: 87 669.16 Std dev: 6689 32.7

²A djusted equity(E_it)is computed as the sum ofbook-value ofequity and tax adjusted value ofuntaxed reserves less the dividends declared. T he corporate tax rateusedincalculations was setto0.5. A fewobservations (eight) quotednegative equity, butratherthan deletingthe entire time series ofthese …rms, we imputed these observations by the value of equity per employee in the lowest percentile observed, times the number ofemployees. Costa: V ar108+ (1-¿)V ar103-V ar110.

M ean: 9 1181.2 Std dev:506053.6

²D ebttoequityrato(B_it=E_it)M ean: 1.57 9 27 6 Std dev: 4.435337

²P roducerP riceIndex(Pjt)atatwodigitindustrylevel. Source: Statistics Sweden