Do Creditor Rights Increase Employment Risk? Evidence from Loan Covenants

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Do Creditor Rights Increase Employment Risk? Evidence from Loan Covenants Nellie Liang and Antonio Falato 2014-61 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Do Creditor Rights Increase Employment Risk? Evidence from Loan Covenants AntonioFalato NellieLiang1 FederalReserveBoard FederalReserveBoard Thisdraft: October2013 1ViewsexpressedarethoseoftheauthorsanddonotrepresenttheviewsoftheBoardoritsstaff. Contacts: antonio.falato@frb.gov, jnellie.liang@frb.gov. Special thanks to Mark Carey and Greg Nini for their helpwithDealscanandforkindlysharingtheirCompustat-Dealscankey. WethankBillBassett, Sudheer Chava (discussant), Edward Morrison (discussant), Greg Nini, Marco Pagano, Michael Roberts (discussant), Steve Sharpe, Martin Schmalz, Amir Sufi; seminar participants at the Federal Reserve Board; and conferenceparticipantsattheannualmeetingsoftheAmericanEconomicAssociation,AmericanFinance Association,andtheConferenceonEmpiricalLegalStudiesforhelpfulcommentsanddiscussions. BrandonNedwek,NicholasRyan,RichardVerlander,andespeciallySuzanneChangprovidedexcellentresearch assistance. Allremainingerrorsareours.

Abstract Using a regression discontinuity design, we provide evidence that incentive conflicts between firms and their creditors have a large impact on employees. There are sharp and substantial employment cuts following loan covenant violations, when creditors exercise their ex post control rights. Thenegativeimpactofviolationsonemploymentisstrongerforfirmsthatfacemoresevere agency and financing frictions and those whose employees have weaker bargaining power. Employment cuts following violations are much larger during industry and macroeconomic downturns, when employees have fewer alternative job opportunities and reduced bargaining power. Unionelectionsthatcreatenewlaborbargainingunitsleadtohigherloanspreads,consistentwith creditorsrequiringcompensationfortheirreducedcontrolrightswhenlaborisstronger. Overall, these findings enrich our understanding of the consequences of the state contingent transfer of controlrightsbyidentifyingarisk-shiftingchannelfromcreditorstoemployees. Ouranalysisestablishes an endogeneity-free link between financing frictions and employment and offers direct evidencethatbindingfinancialcovenantsareanimportantamplificationmechanismofeconomic downturns.

1 Introduction OnefundamentalcontributionofmoderncorporatefinanceistheinsightbyJensenandMeckling (1976) that firms are a complex nexus of contractual relations.1 Important aspects of this original insight have been developed. In particular, the literature has extensively studied conflicts of interest between shareholders and managers and between shareholders and debtholders (see Stein (2003) for a survey). It is now well understood that these conflicts of interest can potentially be mitigated by contractual features such as, for example, financial covenants that protect lendersbeforenon-paymentordefaultbydefiningastate-contingenttransferofcontrolrights. A growingrecentempiricalliterature(ChavaandRoberts(2008),RobertsandSufi(2009),andNini, SmithandSufi(2009,2012))showsthatloancovenantsareindeedeffectiveatprotectingcreditors’ rights,andthatthetransferofcontrolrightsthataccompaniescovenantviolationshasimportant consequencesforfirminvestmentandfinancialpolicies. Another important insight of the nexus of contracts view has received relatively little attention. Therecanalsobeafundamentalconflictofinterestbetweencreditorsandotherstakeholders stemmingfromthefactthateachofthesegroupshasapriorityclaimonfirmrevenues. InJensen andMeckling(1976),“firmsincurobligationsdailytosuppliers,toemployees,todifferentclasses ofinvestors,etc. Solongasthefirmisprospering,theadjudicationofclaimsisseldomaproblem. When the firm has difficulty meeting some of its obligations, however, the issue of the priority ofthoseclaimscanposeseriousproblems.”Sinceacomplexwebofmultiplecontractsultimately determine the adjudication of claims, the allocation of rights between shareholders and creditorslikelyhasanimpactoncontractualrelationsbetweenthefirmandnonfinancialstakeholders. However, we have virtually no empirical evidence on whether and how creditor rights actually impactnonfinancialstakeholders. 1"Thereisinaveryrealsenseonlyamultitudeofcomplexrelationships(i.e., contracts)betweenthelegalfiction (thefirm)andtheownersoflabor,materialandcapitalinputsandtheconsumersofoutput. Thefirm[...]isalegal fiction,whichservesasafocusforacomplexprocessinwhichtheconflictingobjectivesofindividuals(someofwhom may“represent”otherorganizations)arebroughtintoequilibriumwithinaframeworkofcontractualrelations.Inthis sensethe“behavior”ofthefirmis[...]theoutcomeofacomplexequilibriumprocess."(JensenandMeckling(1976)) 1

Inanattempttofillthegapintheliterature,thispaperexaminesconflictsofinterestbetween creditors and an important class of nonfinancial stakeholders. Specifically, we assess the impact of loan covenant violations on employees. Why would covenant violations affect employees? Loan covenants which are tied to performance indicators protect creditors by defining a transfer of control rights when a covenant is violated (e.g., financial contracting theory of Aghion and Bolton(1992)andDewatripontandTirole(1994)). Suchviolationsprovidecreditorswiththesame rights as would payment defaults, including the ability to accelerate any outstanding principal and to terminate any unused revolving credit facility, and lead to renegotiation of loans on less favorabletermstoborrowers. Inordertoavoidaccelerationandensurecontinuedaccesstocredit, managementmaydecidetoreduceoperatingcostsbycuttingjobsafteracovenantviolationinan attempttoreassurecreditorsaboutthefirm’sabilitytogeneratecashflows. Employeesmayalso be affected directly by creditors’ interventions in the form of “advising” management to reduce headcountandoperatingexpenses,asexemplifiedbythefollowingexcerptfromthefirstquarter 10-QfilingofInterpharmHoldingsin2008:2 Subsequently,onJanuary28,2008,WellsFargoinformedtheCompanythatitwould consider providing the Company with credit availability on the condition that the Company(i)developsandimplementsanewoperatingplanfocusedonincreasingthe amountofeligiblecollateralandreducingcostsand(ii)developanalternativefinancingarrangement. Further,onFebruary5,2008,theCompanyandWellsFargoentered intotheForbearanceAgreement[...]InconnectionwithitsnegotiationoftheForbearanceAgreement,theCompanycompletedarestructuringofitsoperationsonJanuary 25, 2008 and submitted a new operating plan to Wells Fargo, which the Company believes will result in positive cash flow and net profits, and includes [...] reducing payrollandheadcountbyapproximately20%. Consistentwiththisreasoning,wedocumentlarge-sampleevidencethatloancovenantviolationsandtheassociatedtransferofcontrolrightstocreditorsleadtolessjobsecurityforworkers. 2Using keyword searches of SEC filings (with keywords such as employee or headcount or overhead reduction) wefoundseveralcasesofadirectlinkbetweenviolationsandemployementinmanagementdiscussionofviolations. Forinstance, theannual10-KfilingofMeadeInstrumentsCorpin2008readsasfollows: "Weareworkingwithour lender on a potential amendment to our agreement to cure this technical default. Our restructuring plans include implementationofheadcountreductions,corporateoverheadandmanufacturingcosts."Fromthesecondquarter10-Q filingofAdvancedMaterialsin2004: "TheCompanyisintheprocessofattemptingtocureitslineofcreditandterm loan violations. Management has implemented a plan to reduce expenses and improve sales. Selling, general and administrativeexpensesforthefirstquarteroffiscal2004and2003were$397,000and$499,000,respectively,adecrease of $102,000 or 20%. This decrease was due primarily to a reduction in the number of employees as the Company continuestoimproveindividualproductivity." 2

Ours is the first endogeneity-free evidence that financing frictions have a sizable adverse impact onemploymentandthatbindingcontractualcovenantsareanamplificationmechanismofdownturns. This evidence contributes to the classical academic and policy debate on the influence of corporate financing on macroeconomic and financial stability (e.g., Bernanke and Gertler (1989); Sharpe (1994), Hanka (1998), Benmelech, Bergman, and Seru (2011), and Pagano and Pica (2011) focusonemployment),adebatewhichhasbeenrecentlyrevivedintheaftermathofthefinancial crisisandtheensuingGreatRecessionof2008and2009. Specifically, we use a regression discontinuity design pioneered by Chava and Roberts (2008) toachieveidentificationanddocumentsizablejobcutsfollowingaloancovenantviolation,which are concentrated among firms with agency problems and whose employees have weak bargaining power.3 Our sample consists of 11,536 firm-year observations for 2,265 unique US firms that have information on loan covenants in Dealscan and on employment and firm balance-sheet in Compustat between 1994 and 2010, which we complement with 3,129 hand-collected layoff announcements from major news sources to construct measures of employment that do not reflect workers’voluntaryseparations.4 Ourbaselineestimatesindicatethatemploymentfallssharplyin responsetoacovenantviolationbyabout18%peryear,areductionwhichisroughlythreetimes aslargeasthemedianemploymentdropinthesample. Theestimatesarerobusttorestrictingthe sample to include only observations that are "close" to the covenant threshold, where violations canbeplausiblyconsidereda"quasi-random"treatment.5 Employmentcutssubsequenttoviolationsbecomeasdeepas30%peryearforfirmswithmore severe agency problems, and as much as 27% for firms whose employees have weaker bargaining power. Investment cuts are instead bigger when employees have stronger bargaining power, 3Afirmisclassifiedtobeinviolationinanygivenyearwhenthevalueofeitheritscurrentratiooritsnetworthfalls belowthecorrespondingcontractualthreshold.Weexaminebroadercovenantsinourrobustnessanalysis(Table6). 4LayoffannouncementarecollectedfromtheWallStreetJournalandothermajornewssourcesusingFactivaand LexisNexiskeywordsearches. 5Ourbaselineresultsalsopassseveralrobustnesschecks,whichincludeusingaspecificationinchanges,ratherthan levels;probitregressionspecificationsoflayoffs;alternativesamplesanddefinitionsofcovenantviolations;andadding controlvariablestoaddressomittedvariableconcerns. 3

whichisconsistentwithfirmssubstitutingbetweentheirlaborandcapitalmarginsofadjustment. These results are robust across several firm- and industry-level proxies of agency problems and laborbargaining.6 Variationbyagencyproxiessuggeststhatthestate-contingentallocationofcontrolrightstocreditorsiseffectiveasamechanismtomitigateunderlying"quietlife"typefrictions, which make managers reluctant to fire employees (see Bertrand and Mullainathan (2003), Cronqvist et al (2008), and Atanassov and Kim (2009) for evidence). Since labor rights drive a wedge between the impact of creditor rights on employment vs. investment, there is an interplay betweencreditorandlaborrights. Overall,variationbylaborbargainingpowersuggeststhatthere is a "rivalry effect" between creditor and labor rights which is broadly consistent with the nexus ofcontractsperspectiveofJensenandMeckling(1976). Theimpactofviolationsonemploymentisoutsizedinbadtimesandrelativelymutedingood times, a finding that is robust across different proxies based on industry and macroeconomic activity.7 Forexample,violationsleadtoemployeecutsofabout29%inNBERrecessionyears,and their impact was truly outsized at about 42% in the Great Recession of 2008 and 2009. There is even stronger evidence of time-series variation in the employment impact among firms with weaker labor bargaining power and among non-rated firms, which likely have less access to alternative sources of credit. The combination of a higher likelihood of violations and bigger job cutswhenviolationsoccurleadstoanestimatedexpectedimpactonemploymentwhichisabout 5 times larger in recession than in non-recession times, which is consistent with a fundamental tenet of much macroeconomics and finance that financing frictions exacerbate the real impact of downturns. Since employment cuts are concentrated in exactly those times when creditors have arguablythe mostbargaining powerandlabor hasthe least, time-seriesevidence furthercorroboratesourinterpretationthattheimpactofviolationsreflectstherelativestrengthofcreditorand 6Thefirm-levelproxiesofagengyandfinancingfrictionsinclude:cashholdings,Gompers,Ishii,andMetrick(2003) GIM-Index,bookleverage,andthefractionoftotaldebtwithshortmaturity,aswellasadummyforratedstatus. The industry-levellaborbargainingproxiesinclude:degreeofunionrepresentation,laborintensityandflexibility,domestic andforeignproductmarketcompetition. 7WethankMichaelRobertsforsuggestingthesetestsoftime-seriesvariation. 4

laborrights. Ourevidencesofarsuggeststhattheabilityofcreditorstoexercisetheircontrolrightsislimitedbylaborrights. Inourfinalsetoftests,weaskwhetherloantermsimpoundlaborrightsand price in a premium for creditors’ expected loss of effective control when labor is stronger. Our identificationisaregressiondiscontinuitydesignthatexploitstherequirementbyU.S.laborlaws that in order to create a new labor bargaining unit an election is held in which workers vote by majorityrulefororagainstunionrepresentation. WematchadministrativedatafromtheNational Labor Relations Board (NLRB) on all union elections that took place in the US between 1985 and 2010 with loan pricing information from Dealscan and accounting data from Compustat, which resultsinasampleof3,814loansfor1,756uniqueelectionevents. Union wins are reliably associated with higher spreads on loans originated within two years fromtheelection,8 andtheeffectofunionizationisparticularlylargeforlowratedandnon-rated borrowerswhichhavemostscopeforstatecontingenttransferofcontrolrightstocreditors. This resultcontinuestoholdwhenweuseamatchedsamplemethodologyandwhenweconsideronly "close"electionstoaddresspotentialconcernsaboutanticipationofelectionoutcomesandomitted variables issues. This evidence suggests that creditors demand compensation when employees have more bargaining power, further corroborating our interpretation that labor rights limit the impact of creditor rights on employees. The analysis also contributes to the classical literature on the economic effects of unions (e.g., DiNardo and Lee (2004), Lee and Mas (2012)), which has traditionallyabstractedfromcorporatecontrolissues. Overall,ouranalysisindicatesthatcreditorrightsincreaseemploymentrisk. Tothebestofour knowledge,thisisthefirstdirectevidenceconsistentwiththeimportantimplicationofJensenand Meckling(1976)thatthereareconflictsofinterestbetweencreditorsandnonfinancialstakeholders withpriorityclaimsinasmuchascreditcontractsthatmitigateconflictsbetweendebtholdersand 8Forexample,intheoverallsample,themeanloanspreadforunionswinsis189.7basispoints,whichisabout20 basispointshigherthanthemeanloanspreadforunionlosses. This20basispointsdifferentialishighlystatistically andeconomicallysignificantatabout10percentofthe(unconditional)samplemeanofloanspreads. 5

shareholders have spillover effects on employees. We make two main additional contributions to the literature. First, our findings contribute to the literature on the real effects of the state contingenttransferofcontrolrights(ChavaandRoberts(2008),RobertsandSufi(2009),andNini, Smith, and Sufi (2009; 2012)) by identifying a risk-shifting channel from creditors to employees. Our results indicate that the state contingent allocation of control rights to creditors is effective asamechanismtomitigatelabor-related"quietlife"typeagencyfrictions,whichmakemanagers reluctant to fire employees (see Bertrand and Mullainathan (2003), Cronqvist et al (2008), and AtanassovandKim(2009)). Byhighlightingtheinterplaybetweenlaborrightsandcreditorrights andbyexploringthelinkbetweenlaborbargainingrightsandloanspreads,ourresultsalsofillthe gap in the small but fast growing recent literature on labor and finance, which has so far mostly focusedoncapitalstructuredecisions(e.g.,Matsa(2010)andAgrawalandMatsa(2013)). Second, we contribute to the literature on financing and employment by providing identification of financing effects.9 Existing evidence is relatively scant, since the literature has mostly focused on financing and investment, and its interpretation is complicated by identification issues, since financing variables are likely correlated with future growth prospects and firm’s demand for labor. In addition to establishing an endogeneity-free link between financing frictions andemployment,wealsoofferdirectevidencethatbindingfinancialcovenantsareanimportant amplification mechanism of economic downturns,10 which supports theories such as Bernanke andGertler(1989)whereadeteriorationoffirmnetworthamplifiestheeffectofeconomicdownturnsbyexacerbatingfinancingfrictions. Ourevidencehighlightsaspecificchannelforthecredit restriction, namely covenant violations, and indicates that what makes bad times really bad for workers is that creditors are more likely to exercise their control rights at the same time when 9Previous research has focused on the effects of finance and investment (Fazzari, Hubbard, and Petersen (1988); Whited(1992)); theeffectoffinanceonemployment(Ofek(1993), Hanka(1998), Kaplan(1989), MuscarellaandVetsuypens(1990),Davis,Haltiwanger,Jarmin,Lerner,andMiranda(2008),andBenmelech,Bergman,andSeru(2011)). Inaddition,previousstudieshavedocumentedrealcostsofbankruptcy,suchaslostcustomersandemployeerelationships(TitmanandOpler(1994)). 10The influence of corporate financing on macroeconomic and financial stability (e.g., Kyotaki and Moore (1997), BernankeandGertler(1989))and,specifically,onemploymentfluctuations(Sharpe(1994))isaclassictopicinmacroeconomicsandfinance. 6

employees have weaker bargaining power. While there is recent evidence that bankruptcies entail costs for workers (Graham et al (2013)), our evidence shows that the real effects of finance on employment are operative well before bankruptcy, which can help to explain why economic downturnsleadtolargejoblossesevenwhentheydonottriggeralargewaveofbankruptcies,as inthecaseoftheGreatRecessionof2008and2009. 2 Analysis of Loan Covenant Violations and Employment If the transfer of control rights to creditors has adverse consequences for employees, then there shouldbeanegativeeffectofloancovenantviolationsonemployment. Inthissection,weexaminethishypothesisusingtheregressiondiscontinuitydesignapproachpioneeredinthisliterature byChavaandRoberts(2008). Afterformallytestingfortheimpactofviolationsonemployees,we examine which factors are driving the impact and show that the impact of violations varies predictablyinthecross-sectionandinthetime-series,andisconcentratedamongfirmswithgreater agency problems, those with greater bargaining power of creditors relative to employees, and in badtimeswhenlabormarketshavelessslack, whichcorroboratesourinterpretationofthebaselineestimates. 2.1 DataandSampleSelection Our sample consists of all Compustat firms incorporated in the United States that have relevant loan covenant information from Loan Pricing Corporation’s (LPC) Dealscan database for the period1994to2010which,afterapplyingstandarddatafilters,resultsinafinalsetof11,536firm-year observationsfor2,265uniquefirms. Ourloaninformationcomesfroma2011extractofLoanPricingCorporation’s(LPC)Dealscan database. The data consist of dollar-denominated private loans made by bank (e.g., commercial andinvestment)andnonbank(e.g.,insurancecompaniesandpensionfunds)lenderstoU.S.cor- 7

porationsduringtheperiod1981to2010. OursampleconstructionstrategyfollowscloselyChava and Roberts (2008) and Dichev and Skinner (2002). Thus, in this section we summarize the main parts of our sample construction strategy, detail the few parts where it differs from these papers, andrefertoChavaandRoberts(2008)forfurtherdetails. WestartwiththeannualmergedCRSP- Compustat database, excluding financial firms (SIC codes 6000-6999). While Chava and Roberts (2008)primarilyusequarterlydata,weuseannualdatabecausefirmsdonotreportemployment atthequarterlyfrequency. Weacknowledgethatthisdatalimitationislikelytomakeourassessment of when the covenant violation occurs more noisy, although Chava and Roberts document thattheirresultsalsoholdwithannualdata. AllvariablesaredefinedinAppendixA. DatafromCompustataremergedwithloaninformationfromDealscanbymatchingcompany names and loan origination dates from Dealscan to company names and corresponding active dates in the CRSP historical header file. The basic unit of observation in Dealscan is a loan, also referredtoasafacilityoratranche. Loansareoftengroupedtogetherintodealsorpackages. Most of the loans used in this study are senior secured claims, features common to commercial loans. Because information on covenants is limited prior to 1994, we focus our attention on the sample ofloanswithstartdatesbetween1994and2010. Additionally,werequirethateachloancontains a covenant restricting the current ratio, or the net worth or tangible net worth (which we group togetherasnetworthloans)tolieaboveacertainthreshold.11 Sincecovenantsgenerallyapplytoallloansinapackage,wedefinethetimeperiodoverwhich the firm is bound by the covenant as starting with the earliest loan start date in the package and endingwiththelatestmaturitydate. Ineffect,weassumethatthefirmisboundbythecovenant for the longest possible life of all loans in the package. A firm is in violation of a covenant if the valueofitsaccountingvariablebreachesthecovenantthreshold-i.e.,wheneitherthecurrentratioorthenetworthfallsbelowthecorrespondingthreshold.12 Wefocusonnetworthandcurrent 11Wealsorequirethecovenant’scorrespondingaccountingmeasuretobenon-missing. Wealsomanuallyrecover somemissingcovenantinformationbylookingatthepackagenotesprovidedbyDealscan(package_comments). 12Whileconceptuallystraightforward,themeasurementofthecovenantthreshold,andconsequentlythecovenant 8

ratiocovenantsfortworeasons,aselaboratedbyChavaandRoberts(2008)andDichevandSkinner (2002). First, they appear relatively frequently in the Dealscan database.13 Second, and most importantly, the accounting measures used for these two covenants are standardized and unambiguous. As the earlier papers documented, while other restrictions with debt or leverage may oftenbeused,definitionscanvaryacrosscontractswheredebtcanrefertolong-term,short-term, secured,orotherdebt,makingitdifficulttodefineviolations. Inrobustnessanalysis,weconsider theimpactofbroadeningthesetofcovenants. Sinceourfocusdoesnotdiscriminatebetweenthe two covenants, for the purpose of our regression analysis there is a violation if either of the two covenantsisbreachedinanygivenfirm-year. We use two different measures of employment. One is the number of employees from Compustat. Thesecondisadummyvariablewhichisequaltooneforfirm-yearswhenthereiseither one of the 3,129 layoff announcements involving Compustat firms in the press, which we handcollected from the Wall Street Journal and other major news sources obtained from Factiva and Lexis-Nexisnewssearches,orareductioninthenumberofemployeesfromCompustat(seeOfek (1993)forasimilarvariable). Inadditiontoresultsreportedforthisbasiclayoffvariable,wealso report results for a medium-sized layoff dummy which corresponds to labor cuts of more than 5 percent of the workforce (which are those larger than the sample median of the distribution of job cuts), and a large-sized layoff dummy for cuts of more than 10 percent of the workforce (top quartileofthedistributionofjobcuts). Table 1 provides summary statistics for the incidence of loan covenant violations as well as means and medians of our main dependent variable, employment, and standard firm and industry characteristics in the resulting sample of 11,536 firm-year observations for 2,265 unique firms that are bound by either a current ratio or a net worth covenant during the period 1994 to violation,posesseveralchallenges,suchasthepossibilityofmultipleoverlappingdeals,and,importantly,thefactthat covenantthresholdscanchangeoverthelifeofthecontract. WedealwiththesemeasurementissuesfollowingChava andRoberts(2008)(seetheirAppendixBfordetails). 13TableIinChavaandRoberts(2008)showsthatcovenantsrestrictingthecurrentratioornetwortharefoundin 9,294loans(6,386packages)withacombinedfacevalueofoveratrilliondollars. 9

2010. By way of comparison, we also report summary statistics of these variables for other nonfinancial firms in Compustat. Appendix A provides sources and detailed definitions for each of these variables. Overall, our sample is comparable to those used in previous studies (Chava and Roberts (2008), Dichev and Skinner (2002)). As in these studies, our sample of Compustat firms with available loan covenant information contains firms that are somewhat larger, both in terms of assets and number of employees, and have higher cash flow, profits, and leverage relative to other firms in Compustat. The frequency of firm-year observations that are classified to be in violation is 20%, which is in line with the 15% frequency reported in Table 3 of Chava and Roberts (2008),consideringthatthereissometime-aggregationduetothefactthatoursamplefrequency is annual while theirs is quarterly and our longer sample period includes the Great Recession. Anadvantageofhavingalongertime-seriesthanpreviousstudiesisthatwecandocumentsome stylized time-series features of violations. In particular, the frequency of violations is markedly higherinNBERrecessionyears,27%,thaninnon-recessionyears,18%. 2.2 EmpiricalFrameworkandEstimationApproach Our empirical specification follows the approach of Chava and Roberts (2008) and exploits their insight that the "tightness" of loan covenants - i.e., the distance between the covenant threshold and the actual accounting measure - can be used to estimate the causal effect of financing. In particular, we consider covenant violations as the treatment and non-violations as the control, and adopt a regression discontinuity design approach. We can do so since the treatment effect is a discontinuous function of the distance between the underlying accounting variable and the covenant threshold. Specifically, our treatment variable, Bind , is defined as a dummy which it equals one if z z0 < 0, where i and t index firm and year observations, z is the observed it (cid:0) it it currentratio(ornetworth),andz0 isthecorrespondingthresholdspecifiedbythecovenant. it Ourbaselineempiricalmodelis 10

Emp = α+β Bind +γ X +η +λ +ν (1) i,t (cid:2) i,t (cid:0) 1 (cid:2) i,t (cid:0) 1 i t i,t where Emp is (log) employment, Bind = 1 if z z0 < 0 and zero otherwise is the i,t it (cid:0) 1 it (cid:0) 1 (cid:0) it (cid:0) 1 covenant violation dummy, X is a vector of control variables measured at the fiscal year-end i,t 1 (cid:0) prior to the year in which employment is measured, η is a firm fixed effect, λ is a year fixed i t effect, and ν is a random error term assumed to be correlated within firm and potentially heti,t eroskedastic(Petersen(2006)). Controlsincludevariablesthathavebeenpreviouslyemployedin the loan covenants literature, such as firm size, profitability, and operating performance, as well as in employment regressions (Nickell (1984), Nickell and Wadhwani (1991)), such as total labor costs. In robustness analysis, we include additional controls such as market-to-book asset ratio, leverage,Altman’sZ-score,anddiscretionaryaccruals. Theparameterofinterestisβ,whichrepresentstheimpactofacovenantviolationonemployment (i.e., the treatment effect). Because of the inclusion of a firm-specific effect, identification of βcomesonlyfromwithin-firmtime-seriesvariationforthosefirmsthatexperienceacovenantviolation. AsnotedinChavaandRoberts(2008),thenonlinearrelationinequation(1)providesfor identificationofthetreatmenteffectunderverymildconditions. Infact,inorderforthetreatment effect β to not be identified, it must be the case that the unobserved component of employment (ν ) exhibits an identical discontinuity as that defined in equation (1), relating the violation staj,t tus to the underlying accounting variable. That is, even if ν is correlated with the difference, j,t z z0 , our estimate of β is unbiased as long as ν does not exhibit precisely the same disit (cid:0) 1 (cid:0) it (cid:0) 1 j,t continuityas Bind . it 1 (cid:0) Because the discontinuity is the source of identifying information, we also include smooth functions of the distance from the technical default boundary in our baseline specification.14 In- 14Moreprecisely,DefaultDistance(CR)andDefaultDistance(NW)aredefinedasDefaultDistance(CR)=I(Current Ratio ) (CurrentRatio -CurrentRatio0),DefaultDistance(NW)=I(NetWorth ) (NetWorth -NetWorth0),where it (cid:2) it it it (cid:2) it it I(Current Ratio ) and I(Net Worth ) are indicator variables equal to one if the firm-year observation is bound by a it it current ratio or net worth covenant, respectively. The Current Ratio0 and Net Worth0 variables correspond to the it it covenantthresholds. 11

cluding these variables helps to isolate the treatment effect to the point of discontinuity and addresses the concern that the distance to the covenant threshold may contain information about future investment opportunities not captured by the other controls. In addition, we report estimatesofequation(1)usingthesubsampleoffirm-yearobservationsthatareclosetothepointof discontinuity. We follow Chava and Roberts (2008) and formally define the “Discontinuity Sample” as comprising firm-year observations for which the absolute value of the relative distance betweentheaccountingvariableandthecorrespondingcovenantthresholdislessthan0.20. This restriction reduces sample size to 4,469 firm-year observations,which is about 40% of the overall sample. 2.3 TheResponseofEmploymenttoCovenantViolations: BaselineResults Table 2 reports results of estimating equation (1) in the entire sample (Panel A) and in the discontinuitysample(PanelB),respectively. Allthespecificationsincludeyearandfirmfixedeffects, exceptforColumn4whichreferstoaspecificationinchangeswithyearandindustryfixedeffects. First,wereplicatetheresultsofChavaandRoberts(2008)oninvestmentinoursampleevenwith theadditionofseveralrecentyearsofdata(Column0). Inthenextsectionswewilluseinvestment responsesasabenchmarktoassessalternativeexplanationsforourresults. Moving to employment, covenant violations are associated with a sharp decline in employmentofabout19%peryear(Column1). Theeconomicmagnitudeofthejobcutsissubstantial,at aboutthreetimesthe6%medianyearlyemploymentdropintheentiresample. Inthediscontinuitysample,violationsleadtoemploymentdropsofroughlythesamemagnitude(Column1,Panel B).Nonparametricanalysisofaveragepercentageannualchangesinthenumberofemployeesin event time leading to and after the year when a violation occurs confirms that there is a sharp break in average employment in the year of violation (t = 0) and in the one immediately after (t = 1),whichisofroughlythesamemagnitudeastheregressionestimates(Figure1).15 15In the years prior to violation (t = 4, 1), employment changes are close to zero on average. Employment (cid:0) (cid:0) 12

Ourestimatesoftheimpactofviolationsonemploymentarerobusttousingalternativespecifications. Column2addressesomittedvariableconcernsbyincorporatingstandardcontrolvariables (firm size, total wages, cash flows, and ROA), and Column 3 adds smooth functions of the distance from the default boundary to further isolate the discontinuity corresponding to the covenant violation. In this full specification, covenant violations remain associated with a sharp decline in employment, which is about 18% per year. Signs of the coefficient estimates are as expected, and the inclusion of the controls has little effect on the estimated impact of violations. Column 4 reports estimates from the first difference analog to the fixed effects specification in equation (1), whichexaminesthechangeinthenumberofemployeesforagivenfirminagiven year as a function of covenant violations, after controlling for changes in the control variables. Fixed effects and first differences estimators are both consistent under standard exogeneity assumptions (Wooldridge (2002)), thus making the comparison of the two specifications useful to assesswhetherourbaselineequationisproperlyspecified. Thefirstdifferencespecificationyields estimatesthataresimilartothespecificationwithfixedeffects. Finally,Columns5to7addresstheconcernthattheresultsmaybedrivenbyfrictionalvoluntaryseparationsratherthanthefirmdecisiontofireemployees. Wereportestimatesfromaprobit analog to the baseline OLS regression analysis, which examine the probability that layoffs occur for a given firm in a given year as a function of covenant violations and the full set of controls.16 Theimpactonlayoffsisbothqualitativelyandquantitativelyinlinewiththeresultsoftheimpact on the number of employees. Violations lead on average to a 19% higher likelihood of layoff in a given year (Column 5). The impact is nearly identical when we consider only more discrete medium-sizedlayoffeventsinvolvingmorethan5%oftheworkforce(Column6),suggestingthat frictionalseparationsareunlikelytobedrivingourresults. Thecoefficientestimateissmallerfor continues to shrink somewhat in the subsequent years (t = 2,3), with the annual change in number of employees remainingnegativeandbelowitspre-violationaverageatabout-8%peryearonaverage. 16Layoffshavebeenconsideredinpreviouspapersonfinancingandemployment(see,forexample,Hanka(1998)) andareacommonfocusintheempiricallaborliterature. 13

layoffsthatinvolvemorethan10%oftheworkforcebuttheimpactremainseconomicallylargeat about10%,whichisonthesameorderofmagnitudeastheunconditionallikelihoodofoccurrence ofsuchlargelayoffsinoursample(Column7). 2.4 Cross-sectional Variationin the Employment Response: Evidence onAgency and LaborBargainingPower Based on our motivating theory and direct evidence from management discussion of covenant violations, we expect that there should be cross-sectional variation in the impact of violations on employment. First, the impact of violations should be larger for firms with more severe agency frictions,sincecovenantsaredesignedtomitigateagencyandfinancingproblemsandthereisevidencesupportingthe"quietlife"hypothesisthatagencyproblemsmakemanagersmorereluctant to fire employees (see Bertrand and Mullainathan (2003), Cronqvist et al (2008), and Atanassov andKim(2009)). Second,theimpactshouldalsobelargerwheneveremployeesareinarelatively weaker bargaining position with respect to creditors, since financial contracting theory suggests that employment cuts are brought about by a strengthening of creditor rights relative to labor rights. Next,wetestthesetwohypothesesinturn. Table3showsevidenceofvariationbyseveralfirm-levelproxiesofagencyandfinancingfrictions. Ineachyearofthesampleperiod,werankfirmsbasedontheempiricaldistributionofthese proxies, whichincludecashholdings(Column1), Gompers, Ishii, andMetrick(2003)GIM-Index of antitakeover provisions (Column 2), book leverage (Column 3), and the fraction of total debt with short maturity (Column 4), as well as a dummy for rated vs. nonrated status (Column 5). Freecashflows(Jensen(1986))andprotectionfromdisciplinarytakeovers(Manne(1965))arewellknown to exacerbate agency problems. Leverage, especially when mostly short-term and costly to refinance, and lack of credit ratings increase financial constraints risk, thus potentially exacerbating risk-shifting (Jensen and Meckling (1976)). An alternative interpretation is that firms with 14

highleverage,shorterdebtmaturities,andnocreditratingshavelessfinancialslackandfeweralternative borrowing opportunities, which increases their existing lenders’ bargaining power and abilitytoexertinfluenceuponviolation. Weestimatethespecificationofequation(1)withthefullsetofcontrolsandsplines(Column3 ofTable2)separatelyforthetwogroupsoffirmsinthebottomandtopquartilesofthe(year-prior) distribution of each of the four continuous proxies in turn, and for rated vs. nonrated firms. We reportresultsfortheentiresampleinPanelAandforthediscontinuitysampleinPanelB.Forboth samplesandbothoutcomevariables(numberofemployeesandthelayoffdummy),thenegative impact of violations on employment is concentrated among firms that have higher cash-to-asset ratiosandantitakeoverprotection,thosethatarehighlyleveragedandhavemoreshortmaturity debt,andthosewithnocreditrating(Rows2,4,8,and10). WealsoreplicatetheresultsinChava and Roberts (2008) for investment across the different proxies (Rows 6 and 12). Overall, these results suggest that the state-contingent allocation of control rights to creditors helps to mitigate "quiet life" employment distortions, and especially so for firms whose creditors are in a stronger bargainingpositionatthetimeofviolation. Table4examinesvariationbyvariousindustry-levelproxiesofemployees’bargainingpower (Columns1to4)andproductmarketcompetition(Columns5and6).17 Ineachyearofthesample period, we rank firms based on the empirical distribution of these proxies, which include measuresofunionrepresentation(Columns1and2),laborintensityandflexibility(Columns3and4), anddomesticandforeignproductmarketcompetition(Columns5and6). Organizedrepresentation through unions is well-recognized to increase labor bargaining power (Clark (1984), Hirsch (2008)). Highlabortocapitalratiosreflecttechnologicaldifferencesacrossindustriesintheirmix ofproductivefactors,butmayalsoindicategreaterlaboroverhangandweakerbargainingpower. Bargainingpowerisalsoeffectivelyhigherwhenlaborflexibilityislower,reflecting(sunk)costsof 17Usingindustry-levelvariablesreducesthepotentialforsimultaneityandforthecaseoflaborintensityandflexibilityismotivatedbytheintuitionthat,duetotechnologicaldifferences,theextenttowhichfirmsfacedifferentcostsof adjustinglaborvariesacrossindustries. 15

adjustinglabor,whichincludehiring,training,andfiringcostsdueto,forexample,lossofhuman andorganizationalcapitalaswellasfirm-specificemployeesskills(Oi(1962),Hamermesh(1989), Eisfeldt and Papanikolaou (2011)). There is a classical theory literature and recent evidence that product market competition mitigates agency frictions (Hart (1983), Giroud and Mueller (2011)). Thus, we expect that the impact of violations on employment should be strongest for firms in industrieswithhigherdomesticproductmarketconcentrationandlowerimportpenetration. We estimate the full specification with controls and splines (Column 3 of Table 2) of equation (1) separately for the two groups of firms in the bottom and top quartiles of the (year-prior) distribution of each of the industry-level proxies in turn, and report results for the entire sample in Panel A and for the discontinuity sample in Panel B. The employment impact of violations is concentratedamongthosefirmsthatareinindustrieswithlowerunionrepresentation,higherlabor intensity and flexibility, and those in less competitive industries (Rows 1-4, and 7-10). These results are robust for both samples and both outcomes variables (number of employees and the layoff dummy). By contrast, the impact of violations on investment is less negative in industries with lower union representation and those with higher labor intensity and flexibility (Rows 5-6, and 11-12). Thus, labor bargaining power drives a wedge between employment and investment responses, since the impact of violations on employment is smaller while the impact on investmentislargerwhenlaborhasmorebargainingpower. Overall, the cross-sectional variation by firm-level agency proxies suggests that the state contingentallocationofcontrolrightstocreditorsiseffectiveasamechanismtomitigateunderlying "quiet life" type frictions, which make managers reluctant to fire employees (see Bertrand and Mullainathan (2003), Cronqvist et al (2008), and Atanassov and Kim (2009) for evidence). Variationbylaborbargainingsuggeststhatthereisanimportantinterplaybetweencreditorandlabor rights,andthatthisinterplaymatterstounderstandhowstrongercreditorrightsimpactnotonly employment, but also investment. Labor rights drive a wedge between the impact of creditor 16

rights on employment vs. investment, suggesting that there is a "rivalry effect" between creditor andlaborrightswhichisbroadlyconsistentwiththenexusofcontractsperspectiveofJensenand Meckling (1976). Our finding of differential variation between investment and employment also offersanadditionaltestofouridentificationstrategy. Mechanicalexplanationsofouremployment effect as being simply driven by declining assets would predict that employment cuts should be concentratedamongthesamesetoffirmsthatcutinvestment,whichcounterstheevidence. 2.5 Time-series Variation in the Employment Response: Evidence on Industry and BusinessCycleConditions The influence of corporate financing on macroeconomic and financial stability (e.g., Kyotaki and Moore (1997), Bernanke and Gertler (1989)) and, specifically, on employment fluctuations (Ofek (1993),Sharpe(1994),Hanka(1998),andDavis,Haltiwanger,Jarmin,Lerner,andMiranda(2008)) is a classic topic in macroeconomics and finance. The recent financial crisis and the "great recession" in 2008 and 2009 with unemployment rates that peaked at 10 percent and 41 consecutive months of rates above 8 percent have revived the academic and policy interest in understanding the impact of financial frictions in the propagation of the business cycle shocks to employment. However, interpretation of existing evidence based on the relation between measures of financingsuchasleverageratiosorcashflowsandemploymentiscomplicatedbyidentificationissues, since financing variables are likely correlated with future growth prospects and firm’s demand for labor. Thus, we still do not have endogeneity-free evidence on whether financing frictions exacerbate the impact of downturns on employees. In addition, since labor markets have less slack in bad times, by exploiting time-series variation in employee bargaining power we can test whethertheemploymentimpactofviolationsisconcentratedinbadtimes,whenemployeeshave lessbargainingpowerandfeweroutsidejobopportunities. Table5reportsresultsofthistime-seriestests,whereweexaminevariationbyvariousproxies 17

ofbadtimes(Columns1to3)vs. goodtimes(Columns4to6). Foreachyearofthesampleperiod, wegroupfirmsintotwobinsbasedonwhetherornot(denotedby"Yes"or"No")inthatyearthere isanindustrydownturn(Column1),arecessionbasedontheNBERdates(Column2),the"great recession" (Column 3), an industry expansion (Column 4), the high-tech boom (Column 5), and the"greatmoderation"(Column6). Weestimatethefullspecification18 ofequation(1)separately forthetwobinsandreportresultsfortheentiresampleinPanelA.19 The employment impact of violations is outsized in bad times and relatively muted in good times(Rows1and3),aresultthatisrobustacrossthedifferentproxiesofgoodandbadtimesand our two main outcomes variables (number of employees and the layoff dummy). For example, estimatedresponsesimplythatviolationsleadtoemployeecutsofabout29%inNBERrecession years (Column 2, Row 1), and to even bigger cuts of about 42% in the Great Recession (Column 3,Row1). Evidenceoftime-seriesvariationintheinvestmentimpactisweaker. Forexample,the investment response in NBER recession periods is -0.9% (Column 2), which is about the same as theaverageinvestmentimpactinTable2. Therelativelylesspronouncedtime-seriesvariationin theinvestmentimpactwithrespecttotheemploymentoneisconsistentwithtime-seriesvariation inemployees’bargainingpowerbeinganimportantdriveroftheemploymentresponse. InPanelBwereportresultsofthesamesetoftestswhenwefurtherstratifythesamplebased on firm and industry characteristics that were used in the analysis of Tables 3-4 and the number of employees is the outcome. The firm-level characteristic we consider is firm credit rating status (Rows 7 and 8). There is solid evidence that firms that have access to bond markets tend to substitute bonds for loans in bad times (Kashyap, Stein, and Wilcox (1993), Ivashina and Becker (2011)). Based on this evidence as well as our results in Table 3, we expect the time-series effects tobeconcentratedamongnonratedfirms,whichhavelessaccesstopublicdebtmarkets. Wealso presentresultsforoneofourindustry-levelproxiesforunionrepresentation, unionmembership 18Withthefullsetofcontrolsandsplines(Column3ofTable2). 19ResultsforthediscontinuitysamplearequalitativelysimilartothoseinPanelAandareomittedforbrevity. 18

(Rows9and10). Indeed,time-seriesvariationintheemploymentimpactismorepronouncedfor firmsthatdonothaveacreditrating(Row7)andforthoseinindustrieswithlowerunionmembership(Row9),whichsuggeststhattheinterplaybetweencreditorsandlaborbargainingpower isanimportantfactorbehindthepropagationeffectofviolationsindownturns. 2.5.1 Theemploymentresponseinrecessions: acalibration Theory suggests that financial contracting may exacerbate the effect of economic downturns on employmentandouranalysisinTable5offersdirectevidenceinsupportofthisnotion. Buthow large is the overall amplification effect of loan covenants? In order to facilitate a quantitative assessment of the economic magnitude of the employment impact of violations in bad times, we provide a simple calibration of the additional job cuts in recessions associated with a covenant violation. There are two related but distinct sources of amplification. First, as shown in Table 1, covenant violations are more frequent in bad times as measured by NBER recession years. Second, our estimates in Table 5 imply that employee cuts are bigger in response to any given violation. Again, basedontheNBERdefinition, innon-recessionperiodsviolationleadsto8.9%cut in employees (Column 2, Row 2 of Table 5) and the frequency of bind is 18 percent (Table 1); in recessions,violationleadsto29%cut(Column2,Row1ofTable5)andthefrequencyofbindis27 percent. Puttingtheseeffectstogether,theimpliedexpectedimpactofcovenantviolationsonemploymentinrecessiontimesisgivenby-0.078=0.27 -0.29, whichisabout5timeslargerthanthe (cid:2) impact in non-recession periods, -0.016=0.18 -0.089. A similar calculation for investment sug- (cid:2) gests that there is also amplification, but much less than for employment: violations lead to an expectedcutininvestmentinnon-recessiontimesof0.001=0.006 0.18, vs. acutinarecessionof (cid:2) 0.0024=0.009 0.27, which is only twice as large. Thus, the interplay of labor and creditor rights (cid:2) leadstolargeramplificationeffectsofviolationsonemploymentthanoninvestment. 19

2.6 Robustness InTable 6we examinetherobustness oftheemployment impactof violationstofour batteriesof tests,whichcompriseusingalternativespecificationstoaddressoutliersandtimingissues(Panel A), using alternative samples and definitions of covenant violations to address alternative explanations and potential measurement error issues (Panel B), and including additional control variables to address potential omitted variables concerns (Panels C and D). In all the tests, we take the full specification20 of equation (1) as our starting point and report results for both the entire sample (Columns 1 and 3) and the discontinuity sample (Columns 2 and 4). Starting with Panel A, estimates from a median (quantile) regression specification are somewhat larger than OLS estimates(Row[1]),suggestingthatoutliersareunlikelytobedrivingourresults. Addingalagged dependentvariable(Row[2]),21 onemorelagandtwoleadsofBind(Rows[3]and[4])alsoleaves the estimated impact little changed, suggesting that sluggish employment dynamics and related timingissuesarealsonotdrivingtheresults. MovingtoPanelB,estimatesderivedusinganalternativedefinitionofBindbasedontheviolationdummieshand-collectedfromSECfilingsbyNini,Smith,andSufi(2012),whichareshown inRow[5],remainlargeandareofthesameorderofmagnitudeasourbaselineestimates,suggestingthatpotentialmeasurementerrorfromnotusingactualviolationsisnotanimportantconcern. A broader definition of Bind that includes the full set of covenants with threshold information available in Dealscan (Row [6]) also leads to strongly statistically and economically significant estimates, though notably lower than our baseline, consistent with the reasoning in Chava and Roberts (2008) that net worth and current ratio covenants have most bite and are defined most unambiguously, thus giving rise to the least potential attenuation bias from measurement error. Excludingfirm-yearswhenthereisadivestitureofassets(Row[7])orthefinancialcrisis(Row[8]) 20Withthefullsetofcontrolsandsplines(Column3ofTable2). 21WeareawareoftheissuethatOLSestimatesmaybebiasedinsmall-Tunbalancedpanelswithfirmfixedeffects andalaggeddependentvariable. InadditionalrobustnesstestswehaveexperimentedwithanIV-GMMestimation approach(BondandVanReenen(2007)),whichyieldssimilarcoefficientestimatesfortheemplymentimpactofviolations. 20

hasalsolittleimpactonourmainestimates,whichdoesnotsupportalternativemechanicalexplanationsbasedonindirecteffectsfromsimplyshrinkingfirmsorone-timefinancialcircumstances. Panels C and D verify that our results are robust to controlling for several additional factors that might affect employment,22 suggesting that omitted variables are not likely to be an important concern. 2.7 SummaryofResults In sum, the results so far show that covenant violations and the associated transfer of control rights to creditors lead to sizable employment cuts. In the cross-section, the employment cuts are concentrated among firms with more severe agency and financing frictions, as well as those whosecreditorshavemorebargainingpowerandwhoseemployeeshavelessbargainingpower, suggestingthatthestatecontingentallocationofcreditorcontrolrightsisaneffectivemechanism to mitigate underlying "quiet life" type agency frictions. In addition, our evidence suggests that thereisaninterplaybetweencreditorsandlaborrights, sincelaborrightsaffecttheoveralleffectivenessofthetransferofcontrolrightstocreditors,whichisbroadlyconsistentwiththenexusof contractsperspectiveofJensenandMeckling(1976). The evidence of differential variation of the employment and the investment impact by labor power corroborates our interpretation that violations have a direct effect on employees because theyleadtoatransferofcontrolrightstocreditors. Amechanicalexplanationoftheeffectofviolationswouldpredictcutsinbothemploymentandinvestmentwhichwouldnotvarywithlabor bargaining power. Finally, our results on time-series variation of the employment impact of violationsfurthercorroboratesourinterpretationthatthefundamentalrivalrybetweencreditorand labor rights is driving our results, since employment cuts are concentrated in exactly those times when creditors have the most and labor has the least bargaining power. The time-series analysis 22Inparticular, weincludeinvestment(Row[9]); adummyforwhetherthefirmundergoesadivestitureofassets inanygivenyear(Row[10]);2nd-and5th-ordernon-linersplinesofthedistancefromthecovenantthreshold(Rows [11]and[12]);bookleverage(Row[13]);Tobin’sQ(Row[14]);Altman’sZ-score(Row[15]);anddiscretionaryaccruals (Row[16]).Theestimatedimpactofcovenantviolationsonemploymentisstableacrossallthesedifferentcontrols. 21

also offers the first endogeneity-free evidence that, consistent with a fundamental tenet of much macroeconomicsandfinanceliterature,financingfrictionsexacerbatetheimpactofdownturnson employees. 3 Analysis of Union Elections and Loan Pricing Our main evidence so far is that covenant violations lead to substantial employment cuts, but lesssowhenlaborhasbargainingpower,whichsuggeststhatlaborrightslimitcreditors’control rights. In this section’s additional analysis of loan pricing terms, we ask whether creditors anticipate that labor rights may limit their control rights upon violation of a covenant. If this is the case, thenweexpectthatloantermsshouldimpoundthestrengthoflaborbargainingrightsand price in a premium for creditors’ expected reduced effective control. These tests offer subsidiary evidenceofaninterplaybetweenlaborrightsandcreditorrights,whichfurthercorroboratesour interpretation of the employment impact of violations. The analysis also contributes to the classical literature on the economic effects of unions (e.g., DiNardo and Lee (2004)), which has traditionally abstracted from corporate control issues and focused on the impact of unions on labor marketoutcomes,suchaswages(seeLeeandMas(2012)forarecentstudyonunionsandequity prices). While there is solid evidence that unionized workers have stronger bargaining rights (Clark (1984), Hirsch (2008)),23 empirical tests based on labor union representation face a classical endogeneity challenge: cross-sectional comparison of loan pricing between unionized and nonunionizedfirmsiscomplicatedbypotentialomittedvariablebiasifthetwogroupsoffirmsdiffer along other characteristics that may affect loan prices. To overcome this challenge, we assemble anewdatasetthatcombinesinformationonelectionstoestablishunionrepresentationwithloan and firm information from Dealscan and Compustat. Our main identification is a regression dis- 23Whichincludebargainingoverwages,pensions,andavariateyofwork-relatedissueswiththeemployer. 22

continuitydesignthatexploitstherequirementbyUSlaborlawsthatinordertocreateanewlabor bargainingunit,anelectionisheldinwhichworkersvotefororagainstunionrepresentationand a simple majority rule is followed to determine whether or not they become unionized. We look at yield spreads of loans issued in the two years following elections and ask whether there is a significant spread differential depending on whether the result is a win or a loss for unions. Resultswithinacloserangearoundthe50percentmajoritythresholdareour"discontinuitysample," withinwhichelectionoutcomesareplausiblya"quasi-random"experiment. Wealsocomplement theselocalestimateswithamatched-sampleanalysisfortheoverallsample. Intheremainderofthissection,afterdescribingoursampleselectionandconstructioncriteria, wesummarizetheresultsontheimpactofunionizationonloanpricing. 3.1 Data WematchadministrativedatafromtheNationalLaborRelationsBoard(NLRB)onallunionelections that took place in the US between 1985 and 2010 with loan data from our 2011 extract of Dealscan for firms that have balance sheet variables available in Compustat.24 This is a labor intensive task since it involves matching company names and union election dates for a very large number of events from NLRB to company names and corresponding active dates in the CRSP-Compustathistoricalheaderfile. Sincethebulkofoursampleconstructionstrategyfollows closelytheliteratureontheeconomicimpactofunionizationevents(DiNardoandLee(2004),and especially Lee and Mas (2012), whose Data Appendix we refer to for details), we only highlight ourmaininnovations. Availability of loan pricing information from Dealscan restricts our usable NLRB data with respect to previous studies, which generally rely on a longer time series (1961-) and the entire Compustat universe. Due to this constraint and in order to insure that our hypothesis testing 24Sincewearenotusingloancovenantinformationforthispartoftheanalysis,wearenotconstrainedbycovenant dataavailabilityand,hence,wecanusetheentireDealscansample. 23

hasenoughpowerevenforthe"discontinuitysample"definedwithinanarrowbandaroundthe 50 percent majority threshold, we take several steps to increase the sample size. The main step involves implementing a second-pass name match for NLRB firms that were not matched in the CRSP-Compustat historical header file, for which we used: (i) a list of historical company names retrievedfromCapitalIQ;and(ii)alistofhistoricallinksbetweenCRSP-Compustatfirmsandthe companynamesoftheiroperatingsegmentsandsubsidiariesalsofromCapitalIQ. Summary statistics for the final sample of 3,814 loan observations for 1,756 unique election events that have information on the percentage vote for unionization during the period 1985 to 2010aretabulatedinTable7. PanelAreportsmeans(andmedians)forunionelectionvariables: a unionwindummy,whichisequaltooneforanygivenelectionthatresultsinawinfortheunion, andtwoimportantelectioncharacteristics,size,whichisthenumberofemployeesinvolved,and percentage share of votes that were cast in favor of unionization. Overall, these statistics are broadlyinlinewiththoseinpreviousstudies(e.g.,LeeandMas(2012)),indicatingthatloaninformationavailabilityfromDealscandoesnotleadtoissueswithselectionfromtheNLRBuniverse. Firmsinoursamplearelarger,morelikelytobehighlyrated,andhavesomewhatlowerspreads thanotherfirmsinDealscan-Compustat,anotherfeaturewesharewithpreviousstudies. Finally, asimplediagnosticcomparisonofpre-eventfirmcharacteristicsandloanspreadsbetween firms where elections resulted in a win and those that resulted in a loss for the union is tabulatedinPanelBfortheoverallsample,andinPanelCforthe"discontinuitysample"of"close" elections,definedasanarrowrange(avotesharerangeof 5%)aroundthemajority(50%)thresh- (cid:6) old needed for the union to win representation. While there are some residual differences in the overallsample,especiallyintermsofpriorspreads,thesedifferencesgoawayinthediscontinuity sample,whichvalidatesourkeyidentifyingassumption. 24

3.2 TheImpactofUnionizationonLoanSpreads: Results Table 8 reports results of simple t-tests of differences between mean loan spreads in the first and in the second year after union elections (Columns (1) to (4) and Columns (5) to (8), respectively) dependingonwhethertheelectionresultedinawinoralossfortheunion. InPanelA,wereport resultsfortheentiresample(Columns(1),(5))andforvarioussub-samplesthatexcludeelections involving,inturn,operatingsubsidiaries(Columns(2),(6)),fewerthan150employees(Columns (3), (7)), and those involving both fewer than 150 employees and investment grade-rated firms (Columns(4),(8)). Union wins are associated with significantly higher loan spreads, a result which is robust to the different sub-samples and both time windows. In the overall sample, the mean loan spread for borrowers where unions win the election is 189.7 basis points, which is about 20 basis points higher than the mean loan spread for borrowers where unions lose (Column 1). This spread differentialisnotonlystatisticallysignificant,butalsoeconomicallysignificantatabout10percentof the sample mean loan spread. The differential triples in magnitude when we consider relatively largerelectionsinvolvinglowratedandnonratedborrowersanditaboutdoublesforloansissued two years after the election. Combined, these results suggest that the effect of unionization on loan spreads is long lasting and is concentrated among the firms that have most scope for state contingenttransferofcontrolrightstocreditors.25 InPanelB,werepeattheset-testsofdifferencesbutnowsharpenouridentificationbyexploitingtheuniquefeatureoftheNLRBdatathatwecanobservethepercentagevoteforunionization inanygivenelection. Weusethepercentagevotevariabletorestrictthesampleandincludeonly "close" elections, which are defined as a narrow range (a vote share range of 5%) around the (cid:6) majority (50%) threshold needed for the union to win representation ("Discontinuity sample").26 25Usingequityprices,LeeandMas(2012)alsofindlonglastingeffectsofunionizationevents. 26Thisregressiondiscontinuitydesignisstandardintheliteratureandreliesonplausiblyexogenous"local"variation inunionizationaroundthe50%threshold,whichisduetothefactthatunionscannotcontroltheassignmentvariable (votes)nearthethreshold. 25

The impact of unionization on loan spreads is larger than in the overall sample, with a premium forunionwinsnowrangingbetweenabout70and150basispoints. GraphicalanalysisinFigure 2, which plots mean loan spreads for each of ten bins of the data sorted on deciles of the union votesharevariable,offeradditionalnonparametricevidencethatthereisasharpbreakinaverage loanspreadsaroundthe50%threshold. Finally, Panel C shows that, when we further stratify the sample based on the number of employees involved in each election, the loan spread differential between union wins and losses increasesmonotonicallywiththesizeoftheelection. Sincelargerelectionsextendthereachoflaborrights, theylikelyimposegreaterlimitationsonthestatecontingenttransferofcontrolrights to creditors. Stronger results for the "discontinuity sample" and for larger elections suggest that anticipationoftheelectionoutcomesandotherpotentialomittedvariableissueslead,ifanything, to downward biased estimates of the effect of unionization. Thus, the impact of unionization of loanspreadsisunlikelytobeanartifactofendogeneityissues. 3.2.1 Matched-SampleAnalysisandAdditionalRobustness In our last set of tests, we use a matched sample methodology analogous to long-run eventstudies(e.g.,BarberandLyons(1997))andconstructa"benchmark"spreadforaportfolioofloans matched on year, industry, and a variety of firm and loan characteristics. We use this approach to check whether the baseline results in Panel A of Table 8 are robust to controlling for common shocks occurring by chance that affect firms with similar characteristics. Results are reported in Table9,whichshowst-testsforthemeansofexcessloanspreadsinthetwoyearsafterunionelections,whicharedefinedasthedifferencebetweenloanspreadsandtheaverageloanspreadfora portfolioofloansmatchedbasedonyear,industry,and,inturn,(decilesof)firmsizeinColumns (1)to(4)ofPanelA;growthopportunities(Markettobookratio)inColumns(5)to(8)ofPanelB; credit ratings in Columns (1) to (4) of Panel C; and year-prior loan spreads in Columns (5) to (8) 26

of Panel D. Robustly across these four benchmarks, union wins remain reliably associated with higherloanspreads. Thedifferentialinexcessspreadsbetweenunionswinsandunionlossesremainseconomicallysignificant. Forexample,intheoverallsample,themeanloanspreadinexcess ofthebenchmarkbasedonyear-priorspreadisabout17basispoints, whichisstillabout10percent of the sample mean loan spread (Panel D, Column 5). The spread differential is again much higher and ranging between about 60 and 90 basis points for relatively larger elections involvinghigh-yieldandnonratedborrowers. Thus,controllingforcommonshocksleavesourbaseline resultslittlechanged. Weimplementedabatteryofadditionalrobustnesstests,whicharenottabulatedtoconserve space.27 In particular, we confirmed that the results in Table 8 are robust to the following: (i) addressingpotentialoutliersbyeitherrepeatingthet-testanalysisonthelogarithmofspreadsor by using Mann-Whitney (z-statistic) tests; (ii) estimating a full-fledged polynomial regression of loan spreads on union win dummy, while controlling for smooth and higher order polynomials of the union vote share as well as standard firm and loan characteristics, and year and industry fixed effects. An additional advantage of this robustness check is that we have verified that the coefficient of the union win dummy remains statistically significant when we cluster standard errorsatthefirmlevel,whichaddressesthepotentialconcernthatmultipleelectioneventsforthe same borrower firm-year may affect our assessment of statistical significance in Table 8;28 (iii) a series of placebo or falsification tests in which we take arbitrary thresholds for the union share vote and an associated "discontinuity" band and examine if unionization around these artificial thresholdsisrelatedtoborrowers’post-electionloanspreads. Wefoundnostatisticalsignificance around the two placebo thresholds we considered (40% and 60%), suggesting that our baseline resultsarenotspurious. Overall, the analysis in this section indicates that union wins in elections to set up new la- 27Tabulationsoftheresultsareavailableuponrequest. 28Wehavealsoverifiedthattheresultsarerobusttoretainingtheoutcomeofthelargestelectiononlyinthecases whentherearemultipleelectionsforanygivenfirm-year. 27

borbargainingunitsareassociatedwithhigherloansspreads,andespeciallysoforfirmsthatare ratedbelowinvestmentgradewherecreditorshavemorescopeformitigatingrisk-shiftingissues through the state contingent transfer of control rights. This evidence suggests that creditors demandcompensationwhenemployeesgainbargainingpower. Thisevidencefurthercorroborates our interpretation that there is a transfer of control rights to creditors in response to covenant violationsandstrongerlaborrightsmitigatetheimpactofcreditorrightsonemployees. 4 Conclusion Stronger creditor rights increase employment risk. We have provided robust evidence that loan covenant violations and the associated transfer of control rights to creditors have significant adverse effects on employment. In response to a loan covenant violation, employment drops by about 18% per year, with even deeper cuts for firms with more severe agency problems, those whose employees have relatively weaker bargaining power, and in times when industry and macroeconomic conditions are weak. Labor rights not only mitigate the employment impact of creditor rights, but also affect creditors’ loan pricing decisions. Ours is the first direct evidencethatevenawayfrombankruptcystatesthereareconflictsofinterestbetweencreditorsand otherstakeholderswithpriorityclaims. Thus,ourevidencesuggeststhatcreditcontractsbetween debtholdersandshareholdershavespillovereffectsonnonfinancialstakeholders. Inaddition,our evidence shows that there are real effects of financial contracting on employment and that these effects are operative before debt default or bankruptcy, which we have argued can contribute to explain why economic downturns lead to large job losses even when they do not trigger a large waveofbankruptcies,asinthecaseoftheGreatRecessionof2008and2009. 28

References [1] Aghion,P,andPBolton,1992,"AnIncompleteContractsApproachtoFinancialContracting," ReviewofEconomicStudies59,473–494. [2] Agrawal, A. and D. A Matsa, 2013, "Labor Unemployment Risk and Corporate Financing Decisions,"JournalofFinancialEconomics,108(2),449-470 [3] Altman,E.,1984,"AFurtherEmpiricalInvestigationoftheBankruptcyCostQuestion,"JournalofFinance39,1067-1089. [4] Andrade, G. and S. Kaplan, 1998. "How Costly is Financial (not Economic) Distress?" Evidence from Highly Leveraged Transactions that Became Distressed," Journal of Finance, 53, 1443-1494. [5] Atanasov, J and E. H. Kim, 2009, "Labor and Corporate Governance: International Evidence fromRestructuringDecisions,"JournalofFinance,64(1),341-374. [6] Barber, B. M. and J. D. Lyons, 1997, “Detecting Long-run Abnormal Stock Returns: The EmpiricalpowerandSpecificationofTestStatistics,”JournalofFinancialEconomics,43(3),341– 372. [7] Beneish, M. and E. Press, 1993, "Costs of Technical Violation of Accounting-Based Debt Covenants,"TheAccountingReview,68,233-257. [8] Benmelech,E.,N.Bergman,andA.Seru,2011,“FinancingLabor,”NBERWP17144,June. [9] Bertrand,M.andS.Mullainathan,2003,"EnjoyingtheQuietLife? ManagerialBehaviorFollowingAnti-TakeoverLegislation",JournalofPoliticalEconomy,11,1043-1075 [10] Billett,M.T.,D.K.Tao-Hsien,andD.C.Mauer,2007,"GrowthOpportunitiesandtheChoice ofLeverage,DebtMaturity,andCovenants,"forthcoming,JournalofFinance. [11] Bond,S.R.,andJ.VanReenen,2007,"MicroeconometricModelsofInvestmentandEmployment,"inJ.J.HeckmanandE.E.Leamereds.: HandbookofEconometrics,Volume6A(Elsevier,Amsterdam). [12] Chava, S. and M.R. Roberts, 2008, "How Does Financing Impact Investment? The Role of DebtCovenants,"forthcoming,JournalofFinance. [13] Clark, K.B., 1984, "UnionizationandFirmPerformance: TheImpactonProfits, Growthand Productivity,"AmericanEconomicReview,74(5),893-919. [14] Cronqvist, H., F. Heyman, M. Nilsson, H. Svaleryd, and J. Vlachos, 2008, "Do Entrenched ManagersPayTheirWorkersMore?"forthcoming,JournalofFinance. [15] Davis, S. J., J. Haltiwanger, R. S. Jarmin, J. Lerner, and J. Miranda, 2008, "Private Equity and Employment,"mimeo,HBS. [16] Dewatripont,M,andJTirole,1994,"Atheoryofdebtandequity: Diversityofsecuritiesand manager-shareholdercongruence,"QuarterlyJournalofEconomics109,1027–1054. [17] Dichev,I.D.andD.J.Skinner,2002."LargeSampleEvidenceontheDebtCovenantHypothesis,"JournalofAccountingResearch,40,1091–1123. [18] DiNardo, J., and D. S. Lee, 2004, "Economic Impacts of New Unionization on Private Sector Employers: 1984–2001,"QuarterlyJournalofEconomics,119,1383-441. 29

[19] Garleanu,N.andJ.Zwiebel,2007,"DesignandRenegotiationofDebtCovenants,"Journalof Finance,forthcoming. [20] Giroud, X. and H. M. Mueller, 2011, "Corporate Governance, Product Market Competition, andEquityPrices,"JournalofFinance,66(2),563-600. [21] Gompers,P.A.,J.L.Ishii,andA.Metrick,2003,”CorporateGovernanceandEquityPrices”, QuarterlyJournalofEconomics,118,107-155 [22] Graham,J.R.,H.Kim,S.Li,andJ.Qiu,2013,"HumanCapitalLossinCorporateBankruptcy," Workingpaper,DukeUniversity [23] GrossmanandHart,1982,"CorporateFinancialStructureandManagerialIncentives",inJohn J. McCall, ed: The Economics of Information and Uncertainty, University of Chicago Press, Chicago,Ill. [24] Hamermesh, D., 1989, "Labor Demand and the Structure of Adjustment Costs," American EconomicReview,79,674-689. [25] Hanka, G., 1998, "Debt and the Terms of Employment,’ Journal of Financial Economics, 48, 252-282 [26] Hart, O. D., 1983, "The Market Mechanism as an Incentive Scheme," Bell Journal of Economics,14,366–382 [27] Hirsch B. T., 2008, "Sluggish Institutions in a Dynamic World: Can Unions and Industrial CompetitionCoexist?,"JournalofEconomicPerspectives,22(1),153–176. [28] Ivashina, V. and B. Becker, 2011, "Cyclicality of Credit Supply: Firm Level Evidence," NBER WorkingPaperNo.17392. [29] Jensen,M.,1986,"AgencyCostsofFreeCashFlow,CorporateFinance,andTakeovers,"AmericanEconomicReview,76(2),323-329. [30] Jensen,M.,andW.Meckling,1976,"TheoryoftheFirm: ManagerialBehavior,AgencyCosts andCapitalStructure,"JournalofFinancialEconomics,3,11-25. [31] Jensen,M.andJ.Warner,1988,"PowerandGovernanceinCorporations,"JournalofFinancial Economics20,3-24. [32] Johnson, S. A., 2003, "Debt Maturity and the Effects of Growth Opportunities and Liquidity RiskonLeverage,"ReviewofFinancialStudies16,pp.209-236. [33] Kashyap,A.,J.Stein,andDWilcox,1993,“MonetaryPolicyandCreditConditions: Evidence fromtheCompositionofExternalFinance,”AmericanEconomicReview,83(1),221-256. [34] KiyotakiN.andJ.H.Moore,1997,”CreditCycles,”JournalofPoliticalEconomy,105(2):211- 48. [35] Lee, D. S., and A. Mas, 2012, "Long-run Impacts of Unions on Firms: New evidence from FinancialMarkets,1961–1999,"QuarterlyJournalofEconomics,127,333-78. [36] Manne,H.,1965,”MergersandtheMarketforCorporateControl,”JournalofPoliticalEconomy,73,110. [37] Matsa, D. A., 2010, "Capital Structure as a Strategic Variable: Evidence from Collective Bargaining,"JournalofFinance,65(3),1197-1232. 30

[38] Muscarella,C.,andVetsuypens,M.,1990,"EfficiencyandOrganizationalStructure: AStudy ofReverseLBOs,"JournalofFinance45,pp.1389-1413. [39] Nickell, S. J., 1984, "An Investigation of the Determinants of Manufacturing Employment in theUnitedKingdom",ReviewofEconomicStudies,51,pp.529-557. [40] Nickell, S.J., 1986, "Dynamic Models of Labor Demand," in Handbook of Labor Economics (V.1),AshenfelterO.andR.Layard(eds.),Elsevier. [41] Nickell,S.J.andWadhwani,S.,1991,"EmploymentDeterminationinBritishIndustry: InvestigationsUsingMicro-Data,"ReviewofEconomicStudies,58,pp.955-969. [42] Nini,G.,D.C.Smith,andA.Sufi,2009,"CreditorControlRightsandFirmInvestmentPolicy," JournalofFinancialEconomics,92(3),400-420. [43] Nini,G.,D.C.Smith,andA.Sufi,2012,"CreditorControlRights,CorporateGovernance,and FirmValue,"ReviewofFinancialStudies,25,1713-1761. [44] Ofek, E., 1993, "Capital Structure and Firm Response to Poor Performance: An Empirical Analysis,"JournalofFinancialEconomics,34(1),3-30. [45] Oi,W.,1962,"LaborasaQuasi-FixedFactor,"JournalofPoliticalEconomy,70(6),538-55. [46] Opler, T. and S. Titman, 1994, "Financial Distress and Corporate Performance," Journal of Finance49,1015-1040. [47] PaganoM.,andG.Pica,2011,"FinanceandEmployment,"CSEFWP283. [48] Petersen, M., 2006, "Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches,"forthcomingReviewofFinancialStudies. [49] Rajan, R. and A. Winton, 1995. "Covenants and Collateral as Incentives to Monitor," Journal ofFinance47,1367-1400. [50] Roberts,M.andA.Sufi,2009,"ControlRightsandCapitalStructure: AnEmpiricalInvestigation,"JournalofFinance,64(4),1657-1695. [51] Sharpe,S.,1994,"FinancialMarketImperfections,FirmLeverage,andtheCyclicalityofEmployment,"AmericanEconomicReview,84(4),1060-1074. [52] Smith,C.,1993."APerspectiveonViolationsofAccountingBasedDebtCovenants,"AccountingReview,68(2),289-303. [53] Smith, C. and J. Warner, 1979, "On Financial Contracting: An Analysis of Bond Covenants," JournalofFinancialEconomics7,117-161. [54] Stein,J.,2003,"Agency,InformationandCorporateInvestment,"inG.M.Constantinides,M. HarrisandR.Stulz,eds.: HandbookoftheEconomicsofFinance(Elsevier,Amsterdam). [55] Stulz,R.,1990,"ManagerialDiscretionandOptimalFinancingPolicies,"JournalofFinancial Economics26,3-27. [56] Sweeney,A.P.,1994,"DebtCovenantViolationsandManagers’AccountingResponses,"JournalofAccountingandEconomics17,pp.281–308. [57] Wooldridge,Jeffrey,2002,EconometricAnalysisofCrossSectionandPanelData(MITPress, Cambridge,Massachusetts). 31

AppendixA:VariableDefinitions Thevariablesusedinthispaperareextractedfromfourmajordatasources: LoanPricingCorporation’s(LPC)Dealscandatabase,COMPUSTAT,CRSP,andtheNationalLaborRelationsBoard (NLRB).Foreachdataitem,weindicatetherelevantsourceinsquarebrackets. Thevariablesare definedasfollows: LoanCovenants[Dealscan]: Bindisadummythattakesvalueofoneifeithernetworthorcurrentratiofallbelowtheirrespectiveloancovenantthresholdsinanygivenfirm-year. NW isthenetworthcovenantthreshold. CRisthecurrentratiocovenantthreshold. OutcomeMeasures: Log(Employment)isthenaturallogarithmofthetotalnumberofemployees(item29). [Compustat] Layoff(All)isadummythattakesvalueofoneifthereisadeclineinemploymentinanygivenyear fromthepreviousyear. WecomplementCompustatdatawithinformationon3,129hand-collected layoffannouncementsfromWallStreetJournalandothermajornewssources[Compustat,Factiva andLexisNexisnewssearches]. Layoff (Medium) is a dummy that takes value of one if there is a larger than average (5%) decline in employment in any given year from the previous year. We complement Compustat data with information on 3,129 hand-collected layoff announcements from Wall Street Journal and other majornewssources[Compustat,FactivaandLexisNexisnewssearches]. Layoff(Big)isadummythattakesvalueofoneifthereisalargerthan10%declineinemployment in any given year from the previous year. We complement Compustat data with information on 3,129hand-collectedlayoffannouncementsfromWallStreetJournalandothermajornewssources [Compustat,FactivaandLexisNexisnewssearches]. Investmentiscapitalexpenditures(item128)overnetproperty,plantandequipmentatthebeginningofthefiscalyear(item8). [Compustat] Loanspreadistheallinspreadonloans,includingfees[Dealscan]. UnionElections[NLRB]: Union Win is a dummy that takes value of one for any given election that results in a win for the union. UnionElectionSizeisthenumberofemployeesthatareeligibletovoteinanygivenelection. UnionVoteShareisthepercentageofvotescastinfavoroftheunioninanygivenelection. FirmandIndustryVariables: BaselineControls: Log(Assets) is the natural logarithm of the book value of assets (item 6), deflated by CPI in 1990. [Compustat] Total Wages is the natural logarithm of total labor expenses (item 42), deflated by CPI in 1990. [Compustat] Cash Flow is the ratio of income before extraordinary items plus depreciation and amortization over the ratio of net property, plant and equipment at the beginning of the fiscal year to total assets. [Compustat] Return on assets (ROA) is the ratio of operating income after depreciation (item 178) over lagged totalassets(item6). [Compustat] DefaultDistance(NW)isthedifferencebetweenthenetworthcovenantthresholdandtotalassets minustotalliabilities. [Compustat] 32

DefaultDistance(CR)isthedifferencebetweenthecurrentrationcovenantthresholdandtheratio ofcurrentassetstocurrentliabilities. [Compustat] Sample-SplitVariables: CashHoldingsistheratioofcashholdings(item1)tototalassets(item6). [Compustat] GIMIndexistheindexofantitakeoverprovisionsbyGompers,Ishii,andMetrick(2003). Leverage is long term debt (item 9) plus debt in current liabilities (item 34) over the sum of long term debt (item 9) plus debt in current liabilities (item 34) plus market value of equity (item 25*item199). [Compustat] ShortTermDebtisthefractionofafirm’stotaldebtthatmaturesinthreeyearsorless. [Compustat] RatingisadummyvariablethattakesthevalueofoneifthefirmhasanS&PLong-TermDomestic IssuerCreditRating[Compustat] Industry Unionization is the share of employees in any given industry-year that are members of a union(Membership)orcoveredbyacollectivebargainingagreement(Coverage). Industry Labor Intensity & Labor Adjustment Costs are measured as the average ratio of number of employeestototalassets(Labor-CapitalRatio)andtheaverageratioofcapitalizedselling,general, andadministrative(SG&A)expensestototalassets(AdjustmentCosts)[Compustat] Industry Product Market Competition is measured by the Herfindahl-Hirschman Index (HHI) and theshareofimportstothetotalvalueofshipments(ImportPenetration)[Compustat] Bad Times are measured as industry-years in the bottom quartile of sales growth (Industry Downturn),adummythattakesvalueofoneinNBERrecessionyears(NBERRecession),andadummy thattakesvalueofonein2008and2009(TheGreatRecession)[Compustat&NBER] GoodTimesaremeasuredasindustry-yearsinthetopquartileofsalesgrowth(IndustryExpansion), and a dummy that takes value of one for firms in the semiconductors, computer manufacturing, andtelecommunicationssectorsforthe1993-2000period(HighTechBoom)andadummythattakes valueofoneinthe1993-2000period(TheGreatModeration)[Compustat&NBER] AdditionalControls: Tobin’s Q (M/B) is the market value of assets divided by the book value of assets (item 6), where themarketvalueofassetsequalsthebookvalueofassetsplusthemarketvalueofcommonequity lessthesumofthebookvalueofcommonequity(item60)andbalancesheetdeferredtaxes(item 74). [Compustat] R&DistheratioofR&Dexpenditures(item46,or0ismissing)overlaggedsales(item12). [Compustat] Advertisingistheratioofadvertisingexpenditures(item45,or0ifmissing)overlaggedtotalsales (item12). [Compustat] Free Cashflow is the ratio to total assets (item 6) of operating income before depreciation (item 13) less interest expense (item 15) and income taxes (item 16) and capital expenditures (item 128). [Compustat] Altman’sZ-Scoreisthesumof3.3timespre-taxincome,sales,1.4timesretainedearnings,and1.2 timesnetworkingcapitalalldividedbytotalassets. [Compustat] AccrualsTWW andAccrualsDDareasdefinedinChavaandRoberts(2008). [Compustat] 33

Table1: LoanCovenantSample: SummaryStatistics Thistablepresentssummarystatistics(meansandmedians)forourmergedDealscan-Compustatsample, whichconsistsof11,536firm-yearobservationsfornonfinancialfirmsbetween1994and2010correspondingtofirmsthathaveatleastoneprivateloanfoundinDealscanwithacovenantthatrestrictscurrentratio or net worth to lie above a certain threshold (Columns 1 and 2). For the sake of comparison, Columns 3 and4reportsummarystatistics(meansandmedians)fortheOtherCompustatsample, whichconsistsof 110,058firm-yearobservationsfornonfinancialfirmsinthesameperiodthathavenomatchinginformation inDealscan. DefinitionsforallvariablesareinAppendixA. Dealscan-Compustat OtherCompustat Mean Median Mean Median (1) (2) (3) (4) LoanCovenantViolations: Bind 0.20 0 n.a. n.a. ExcludingNBERRecessionYears 0.18 0 n.a. n.a. OnlyinNBERRecessionYears 0.27 0 n.a. n.a. EmploymentOutcomeVariables: Employees(000) 7.75 2.16 7.45 0.75 LayoffDummy: All 0.38 0 0.38 0 Medium 0.22 0 0.20 0 Large 0.09 0 0.10 0 Capex 0.07 0.04 0.07 0.04 FirmCharacteristics: Assets(Log) 5.9 6.0 5.1 4.9 TotalWages(Log) 4.7 4.5 4.3 4.5 CashFlow 0.08 0.08 0.04 0.08 ROA 0.10 0.11 0.06 0.10 NetWorth 476 149 471 43 TangibleNetWorth 179.8 60.3 123.8 29.3 CurrentRatio 2.10 1.76 2.30 1.79 CashHoldings 0.10 0.05 0.20 0.10 GIMIndex 9.30 9.00 9.20 9.00 Leverage 0.29 0.28 0.25 0.20 ShortTermDebt 0.13 0.05 0.12 0.05 RatedDummy 0.32 0 0.18 0 Tobin’sQ 1.94 1.36 1.89 1.41 IndustryCharacteristics: Unionization 0.10 0.08 0.09 0.06 LaborIntensity 0.01 0.01 0.01 0.01 Competition(HHI) 0.23 0.18 0.22 0.17 Competition(ImportPenetration) 0.24 0.21 0.24 0.20 Firm-YearObs 11,536 110,058 Firms 2,265 15,186 34

sisylanAnoissergeRenilesaB:tnemyolpmEdnasnoitaloiVtnanevoCnaoL :2elbaT latipacfooitarehtsielbairavtnednepedehT .slortnocdna)"dniB"(erusaemnoitaloivtnanevocanotnemyolpmefostlusernoissergerstneserpelbatsihT snmuloCnitnemyolpmegol;)8002(streboRdnaavahCotselpmasfoytilibarapmocerusneot,)0(nmuloCnidoirepehtfotratsehttastessaotserutidnepxe nienofoeulavsekattahtymmudadna,)4(nmuloCni)seeyolpmeforebmundeggalotseeyolpmeforebmunfooitareht(htworgtnemyolpme,)3(ot)1( ,wofl hsac tpecxe ,selbairav tnednepedni llA .A xidneppA ni era snoitinfied elbairav llA .)7( ot )5( snmuloC ni ffoyal a si ereht nehw raey-mrfi nevig yna eht gnitaloiv ot esolc era smrfi hcihw ni snoitavresbo raey-mrfi sesu ylno B lenaP .elpmas eritne eht rof stluser eht stneserp A lenaP .raey eno deggal era mrfihtobedulcni)3(ot)0(snmuloCnisnoitacfiicepS .)"elpmasytiunitnocsiD"(dlohserhttnanevocehtdnuora)%02 (egnarworranasadenfied,tnanevoc (cid:6) lairesmrfi-nihtiwdnayticitsadeksoretehottsuborsrorredradnatS .stceffedexfiraeydnayrtsudniedulcni)7(ot)4(snmuloCelihw,stceffedexfiraeydna .ylevitcepser%1dna,%5,%01rof***dna,**,*ybdetacidnieraecnacfiingisfosleveL .setamitsetniopwolebraeppanoitalerroc elpmaSeritnE:AlenaP sffoyaL htworGpmE )tnemyolpmE(gol tnemtsevnI :elbairaVtnednepeD )7( )6( )5( )4( )3( )2( )1( )0( tiborP tiborP tiborP slortnoCSLO slortnoCEF EF EF EF :noitacfiicepSledoM egraL muideM llA senilpS& senilpS& slortnoC slortnoCoN slortnoCoN (cid:3)(cid:3)(cid:3)901.0 (cid:3)(cid:3)(cid:3)291.0 (cid:3)(cid:3)(cid:3)881.0 (cid:3)(cid:3)(cid:3)851.0- (cid:3)(cid:3)(cid:3)671.0- (cid:3)(cid:3)(cid:3)471.0- (cid:3)(cid:3)(cid:3)491.0- (cid:3)(cid:3)(cid:3)010.0dniB )010.0( )510.0( )410.0( )720.0( )220.0( )220.0( )520.0( )300.0( 100.0- (cid:3)(cid:3)(cid:3)620.0- (cid:3)(cid:3)(cid:3)720.0- (cid:3)(cid:3)631.0 (cid:3)(cid:3)(cid:3)852.0 (cid:3)(cid:3)(cid:3)162.0 )stessA(goL )200.0( )400.0( )400.0( )850.0( )520.0( )420.0( 100.0- 100.0- 200.0- 980.0 140.0 240.0 segaWlatoT )200.0( )400.0( )500.0( )750.0( )920.0( )920.0( (cid:3)620.0- (cid:3)(cid:3)680.0- (cid:3)(cid:3)(cid:3)681.0- (cid:3)(cid:3)560.0 860.0 950.0 wolFhsaC )610.0( )330.0( )840.0( )230.0( )511.0( )511.0( (cid:3)(cid:3)(cid:3)812.0- (cid:3)(cid:3)(cid:3)114.0- (cid:3)(cid:3)(cid:3)634.0- (cid:3)(cid:3)(cid:3)541.0 (cid:3)(cid:3)(cid:3)514.0 (cid:3)(cid:3)(cid:3)814.0 AOR )230.0( )950.0( )470.0( )530.0( )221.0( )221.0( 000.0- 100.0 100.0- 400.0- 100.0 )WN(ecnatsiDtluafeD )100.0( )200.0( )300.0( )500.0( )800.0( 200.0 110.0 100.0 430.0 (cid:3)(cid:3)910.0- )RC(ecnatsiDtluafeD )600.0( )900.0( )400.0( )030.0( )800.0( (cid:3)(cid:3)(cid:3)812.0 (cid:3)(cid:3)(cid:3)648.0- (cid:3)(cid:3)(cid:3)768.0- (cid:3)(cid:3)(cid:3)133.0 (cid:3)(cid:3)(cid:3)880.0 tpecretnI )840.0( )931.0( )831.0( )380.0( )500.0( oN oN oN oN seY seY seY seY stceffEdexiFmriF seY seY seY seY oN oN oN oN stceffEdexiFyrtsudnI seY seY seY seY seY seY seY seY stceffEdexiFraeY 159,01 159,01 568,01 569,01 560,11 560,11 624,11 635,11 snoitavresbO elpmaSytiunitnocsiD:BlenaP (cid:3)(cid:3)(cid:3)070.0 (cid:3)(cid:3)(cid:3)111.0 (cid:3)(cid:3)(cid:3)251.0 (cid:3)(cid:3)(cid:3)161.0- (cid:3)(cid:3)(cid:3)431.0- (cid:3)(cid:3)(cid:3)041.0- (cid:3)(cid:3)(cid:3)261.0- (cid:3)(cid:3)(cid:3)800.0dniB )810.0( )620.0( )620.0( )040.0( )030.0( )030.0( )130.0( )300.0( 190,4 190,4 550,4 971,4 391,4 391,4 304,4 964,4 snoitavresbO 35

Table3: LoanCovenantViolationsandEmployment:ByProxiesofAgencyandFinancing Thistablepresentsregressionresultsofemploymentonacovenantviolationmeasure("Bind")andcontrolsfordifferent sub-samplesplitsofthedatabasedonex-anteproxiesfortheseverityofagency(Columns(1)and(2))andfinancing (Columns(3)to(5))frictionsfacedbyfirms. Themodelspecificationistheonewithfirmfixedeffects,controls,and splinesasinColumn(3)ofTable2andthedependentvariableislogemploymentinRows[1]-[2]and[7]-[8],adummy that takes value of one in any given firm-yearn when there is a layoff in Rows [3]-[4] and [9]-[10], and the ratio of capital expenditures to assets at the start of the period in Rows [5]-[6] and [11]-[12]. All variable definitions are in Appendix A. We only report estimates of the Bind coefficient and omit estimates of firm controls from the table for brevity(availableuponrequest). PanelApresentstheresultsfortheentiresample. PanelBonlyusesfirm-yearobservationsinwhichfirmsareclosetoviolatingthecovenant,definedasanarrowrange( 20%)aroundthecovenant (cid:6) threshold("Discontinuitysample").Thesesamplesaresplitbetweenbottomandtopquartilesof(year-prior)valuesof thefollowingex-anteproxiesofagencyfrictions:cashholdings(Column(1)),andGompers,Ishii,andMetrick(2003) GIMindexofantitakeoverprovisions(Column(2));andbetweenbottomandtopquartilesof(year-prior)valuesofthe followingex-anteproxiesoffinancingfrictionsleverage(Column(3))andthefractionoftotaldebtwithshortmaturity (Column(4)),aswellascreditratingstatus(Column(5)).Standarderrorsrobusttoheteroskedasticityandwithin-firm serialcorrelationappearbelowpointestimates. Levelsofsignificanceareindicatedby*,**,and***for10%,5%,and 1%respectively. PanelA:EntireSample Agency Financing (1) (2) (3) (4) (5) Cash GIM Leverage ShortTerm Credit Holdings Index Debt Rating Log(Employment) [1] Q1 -0.049 -0.065 -0.052 -0.086 Yes -0.095(cid:3)(cid:3) (0.042) (0.076) (0.050) (0.068) (0.039) [2] Q4 -0.228(cid:3)(cid:3)(cid:3) -0.303(cid:3)(cid:3)(cid:3) -0.200(cid:3)(cid:3)(cid:3) -0.243(cid:3)(cid:3)(cid:3) No -0.185(cid:3)(cid:3)(cid:3) (0.077) (0.072) (0.044) (0.079) (0.028) Layoffs,Probit [3] Q1 0.053(cid:3)(cid:3)(cid:3) 0.023 0.058(cid:3)(cid:3) 0.032 Yes 0.012 (0.017) (0.012) (0.030) (0.024) (0.012) [4] Q4 0.156(cid:3)(cid:3)(cid:3) 0.128(cid:3)(cid:3)(cid:3) 0.154(cid:3)(cid:3)(cid:3) 0.187(cid:3)(cid:3)(cid:3) No 0.112(cid:3)(cid:3)(cid:3) (0.034) (0.060) (0.017) (0.028) (0.012) Investment [5] Q1 -0.003 -0.003 -0.002 -0.001 Yes -0.006 (0.006) (0.006) (0.007) (0.006) (0.004) [6] Q4 -0.020(cid:3)(cid:3)(cid:3) -0.011(cid:3)(cid:3) -0.013(cid:3)(cid:3) -0.011(cid:3)(cid:3) No -0.014(cid:3)(cid:3)(cid:3) (0.006) (0.006) (0.006) (0.004) (0.003) PanelB:DiscontinuitySample Log(Employment) [7] Q1 -0.035 -0.038 -0.037 -0.064 Yes -0.052 (0.058) (0.140) (0.078) (0.074) (0.054) [8] Q4 -0.233(cid:3)(cid:3)(cid:3) -0.228(cid:3)(cid:3)(cid:3) -0.217(cid:3)(cid:3)(cid:3) -0.267(cid:3)(cid:3) No -0.182(cid:3)(cid:3)(cid:3) (0.078) (0.083) (0.067) (0.117) (0.036) Layoffs,Probit [9] Q1 0.041(cid:3)(cid:3) 0.028 0.021 0.024 Yes 0.013 (0.021) (0.019) (0.028) (0.025) (0.024) [10] Q4 0.148(cid:3)(cid:3)(cid:3) 0.178(cid:3)(cid:3) 0.189(cid:3)(cid:3)(cid:3) 0.196(cid:3)(cid:3)(cid:3) No 0.124(cid:3)(cid:3)(cid:3) (0.036) (0.116) (0.023) (0.035) (0.016) Investment [11] Q1 -0.002 -0.006 -0.002 -0.006 Yes -0.006 (0.005) (0.007) (0.008) (0.006) (0.004) [12] Q4 -0.019(cid:3)(cid:3) -0.022(cid:3)(cid:3) -0.014(cid:3)(cid:3)(cid:3) -0.016(cid:3)(cid:3) No -0.013(cid:3)(cid:3)(cid:3) (0.008) (0.011) (0.006) (0.007) (0.004) 36

Table4: LoanCovenantViolationsandEmployment:ByProxiesofLaborBargainingPower Thistablepresentsregressionresultsofemploymentonacovenantviolationmeasure("Bind")andcontrolsfordifferent sub-samplesplitsofthedatabasedonex-anteproxiesforindustryunionization(Columns(1)and(2)),laborintensity andflexibility(Columns(3)and(4))andproductmarketcompetition(Columns(5)and(6)). Themodelspecification istheonewithfirmfixedeffects,controls,andsplinesasinColumn(3)ofTable2andthedependentvariableislog employment in Rows [1]-[2] and [7]-[8], a dummy that takes value of one in any given firm-yearn when there is a layoff in Rows [3]-[4] and [9]-[10], and the ratio of capital expenditures to assets at the start of the period in Rows [5]-[6]and[11]-[12]. AllvariabledefinitionsareinAppendixA.WeonlyreportestimatesoftheBindcoefficientand omitestimatesoffirmcontrolsfromthetableforbrevity(availableuponrequest). PanelApresentstheresultsforthe entiresample.PanelBonlyusesfirm-yearobservationsinwhichfirmsareclosetoviolatingthecovenant,definedasa narrowrange( 20%)aroundthecovenantthreshold("Discontinuitysample").Thesesamplesaresplitbetweenbottom (cid:6) and top quartiles of (year-prior) values of the following ex-ante industry-level proxies union membership (Column (1)) and coverage (Column (2)); the average industry ratio of employees to assets (Column (3)) and SG&A to assets (Column(4)),theHerfindahl-HirschmanIndex(HHI)(Column(5))andthedegreeofimportpenetration(Column(6)). Standarderrorsrobusttoheteroskedasticityandwithin-firmserialcorrelationappearbelowpointestimates.Levelsof significanceareindicatedby*,**,and***for10%,5%,and1%respectively. PanelA:EntireSample Unionization LaborIntensity&Flexibility Competition (1) (2) (3) (4) (5) (6) Member- Coverage Labor- LaborAdj HHI ImportPeship CapitalRatio Costs netration Log(Employment) [1] Q1 -0.223(cid:3)(cid:3)(cid:3) -0.270(cid:3)(cid:3)(cid:3) -0.066(cid:3) -0.216(cid:3)(cid:3)(cid:3) -0.106(cid:3) -0.212(cid:3)(cid:3)(cid:3) (0.082) (0.094) (0.037) (0.043) (0.063) (0.065) [2] Q4 -0.074 -0.075 -0.269(cid:3)(cid:3)(cid:3) -0.061 -0.257(cid:3)(cid:3)(cid:3) -0.141(cid:3)(cid:3) (0.047) (0.047) (0.044) (0.049) (0.055) (0.065) Layoffs,Probit [3] Q1 0.134(cid:3)(cid:3)(cid:3) 0.166(cid:3)(cid:3)(cid:3) 0.029(cid:3)(cid:3) 0.131(cid:3)(cid:3)(cid:3) 0.035(cid:3)(cid:3) 0.098(cid:3)(cid:3)(cid:3) (0.033) (0.038) (0.016) (0.017) (0.017) (0.036) [4] Q4 0.025 0.012 0.160(cid:3)(cid:3)(cid:3) 0.030(cid:3) 0.138(cid:3)(cid:3)(cid:3) 0.033 (0.018) (0.014) (0.016) (0.016) (0.014) (0.024) Investment [5] Q1 -0.002 -0.003 -0.013(cid:3)(cid:3) -0.003 -0.002 -0.014(cid:3)(cid:3) (0.004) (0.004) (0.006) (0.003) (0.007) (0.006) [6] Q4 -0.010(cid:3)(cid:3) -0.009(cid:3)(cid:3)(cid:3) -0.006(cid:3)(cid:3) -0.017(cid:3)(cid:3)(cid:3) -0.013(cid:3)(cid:3)(cid:3) -0.002 (0.004) (0.004) (0.003) (0.007) (0.004) (0.004) PanelB:DiscontinuitySample Log(Employment) [7] Q1 -0.213(cid:3)(cid:3)(cid:3) -0.265(cid:3)(cid:3)(cid:3) -0.056 -0.244(cid:3)(cid:3)(cid:3) -0.098 -0.213(cid:3)(cid:3) (0.078) (0.095) (0.058) (0.059) (0.131) (0.105) [8] Q4 -0.031 -0.056 -0.265(cid:3)(cid:3)(cid:3) -0.062 -0.223(cid:3)(cid:3)(cid:3) -0.131 (0.069) (0.058) (0.095) (0.080) (0.071) (0.095) Layoffs,Probit [9] Q1 0.139(cid:3)(cid:3)(cid:3) 0.180(cid:3)(cid:3)(cid:3) 0.001 0.140(cid:3)(cid:3)(cid:3) 0.016 0.112(cid:3)(cid:3)(cid:3) (0.037) (0.041) (0.029) (0.021) (0.023) (0.051) [10] Q4 0.014 0.001 0.180(cid:3)(cid:3)(cid:3) 0.045 0.137(cid:3)(cid:3)(cid:3) 0.047 (0.024) (0.029) (0.041) (0.035) (0.019) (0.037) Investment [11] Q1 -0.002 -0.003 -0.012(cid:3)(cid:3) -0.001 -0.005 -0.015(cid:3)(cid:3) (0.004) (0.004) (0.005) (0.005) (0.007) (0.006) [12] Q4 -0.011(cid:3)(cid:3) -0.012(cid:3)(cid:3) -0.003 -0.017(cid:3)(cid:3) -0.013(cid:3)(cid:3)(cid:3) -0.001 (0.005) (0.005) (0.004) (0.008) (0.005) (0.006) 37

Table5: LoanCovenantViolationsandEmploymentinGoodandBadTimes Thistablepresentsregressionresultsofemploymentonacovenantviolationmeasure("Bind")andcontrolsfordifferent sub-samplesplitsofthedatabasedonproxiesformacroeconomicconditions. Themodelspecificationistheonewith firmfixedeffects,controls,andsplinesasinColumn(3)ofTable2andthedependentvariableislogemploymentin Rows [1]-[2] and [7]-[12], a dummy that takes value of one in any given firm-yearn when there is a layoff in Rows [3]-[4],andtheratioofcapitalexpenditurestoassetsatthestartoftheperiodinRows[5]-[6]. Allvariabledefinitions areinAppendixA.WeonlyreportestimatesoftheBindcoefficientandomitestimatesoffirmcontrolsfromthetable forbrevity(availableuponrequest). PanelApresentstheresultsfortheentiresample,whichissplitbasedonseveral proxies between good (Columns (1) to (3) and bad (Columns (4) to (6)) times. Panel B further stratifies the sample basedoncreditratingstatus(Rows[7]and[8]),andindustryunionmembership(Rows[9]and[10]). Standarderrors robusttoheteroskedasticityandwithin-firmserialcorrelationappearbelowpointestimates. Levelsofsignificanceare indicatedby*,**,and***for10%,5%,and1%respectively. PanelA:EntireSample BadTimes GoodTimes (1) (2) (3) (4) (5) (6) Industry NBER TheGreat Industry HighTech TheGreat Downturn Recession Recession Expansion Boom Moderation Log(Employment) [1] Yes -0.246(cid:3)(cid:3)(cid:3) -0.290(cid:3)(cid:3)(cid:3) -0.424(cid:3)(cid:3)(cid:3) -0.057 -0.023 -0.088(cid:3)(cid:3)(cid:3) (0.037) (0.057) (0.099) (0.041) (0.068) (0.028) [2] No -0.109(cid:3)(cid:3)(cid:3) -0.089(cid:3)(cid:3)(cid:3) -0.089(cid:3)(cid:3)(cid:3) -0.239(cid:3)(cid:3)(cid:3) -0.230(cid:3)(cid:3)(cid:3) -0.194(cid:3)(cid:3)(cid:3) (0.024) (0.020) (0.020) (0.025) (0.067) (0.030) Layoffs,Probit [3] Yes 0.198(cid:3)(cid:3)(cid:3) 0.157(cid:3)(cid:3)(cid:3) 0.198(cid:3)(cid:3)(cid:3) 0.019 0.053 0.064(cid:3)(cid:3)(cid:3) (0.024) (0.028) (0.059) (0.024) (0.060) (0.015) [4] No 0.053(cid:3)(cid:3)(cid:3) 0.058(cid:3)(cid:3)(cid:3) 0.095(cid:3)(cid:3)(cid:3) 0.146(cid:3)(cid:3)(cid:3) 0.167(cid:3)(cid:3)(cid:3) 0.132(cid:3)(cid:3)(cid:3) (0.012) (0.010) (0.009) (0.012) (0.045) (0.014) Investment [5] Yes -0.009(cid:3)(cid:3)(cid:3) -0.009(cid:3)(cid:3) -0.012(cid:3)(cid:3) -0.006 -0.009(cid:3)(cid:3) -0.008(cid:3)(cid:3)(cid:3) (0.003) (0.004) (0.005) (0.005) (0.004) (0.002) [6] No -0.007(cid:3)(cid:3)(cid:3) -0.006(cid:3)(cid:3)(cid:3) -0.005(cid:3)(cid:3)(cid:3) -0.010(cid:3)(cid:3)(cid:3) -0.008(cid:3)(cid:3) -0.007(cid:3)(cid:3)(cid:3) (0.002) (0.002) (0.001) (0.002) (0.004) (0.002) PanelB:Row1ByFirmandIndustryCharacteristics ByFirmCreditRatingStatus [7] Rated -0.112(cid:3)(cid:3) -0.159(cid:3)(cid:3)(cid:3) -0.083 -0.046 -0.002 -0.022 (0.048) (0.061) (0.200) (0.093) (0.102) (0.040) [8] NotRated -0.315(cid:3)(cid:3)(cid:3) -0.354(cid:3)(cid:3)(cid:3) -0.709(cid:3)(cid:3)(cid:3) -0.066 -0.036 -0.086(cid:3)(cid:3)(cid:3) (0.086) (0.077) (0.201) (0.048) (0.066) (0.032) ByIndustryUnionization(UnionMembership) [9] Q1 -0.340(cid:3)(cid:3)(cid:3) -0.367(cid:3)(cid:3)(cid:3) -0.612(cid:3)(cid:3)(cid:3) -0.091 -0.105 -0.102(cid:3)(cid:3) (0.094) (0.093) (0.153) (0.064) (0.083) (0.044) [10] Q4 -0.148(cid:3) -0.188(cid:3)(cid:3) 0.203 -0.046 -0.001 -0.023 (0.078) (0.078) (0.123) (0.059) (0.091) (0.031) 38

Table6: LoanCovenantViolationsandEmployment:RobustnessAnalysis In this table, we check for robustness of the impact of violations on employment presented in Table 2 to using alternative specifications (Panel A), alternative samples and definitions of covenant violations (Panel B), and to including additional controls (Panels C and D). In all robustness checks, the starting modelspecificationistheonewithfirmfixedeffects,controls,andsplinesasinColumn(3)ofTable2and thedependentvariableislogemployment. Weonlyreportestimatesofthecovenantviolationcoefficient and omit estimates of firm controls from the table for brevity (available upon request). Columns 1 and 3 present the results for the entire sample. Columns 2 and 4 only use firm-year observations in which firmsareclosetoviolatingthecovenant,definedasanarrowrange( 20%)aroundthecovenantthreshold (cid:6) ("Discontinuity sample"). Panel A reports results from the following specifications: a median (quantile) regressionspecificationinRow[1]; aspecificationthataddsalaggeddependentvariableinRow[2]; and specificationsthataddonemorelagandtwoleadsofBindinRows[3]and[4],respectively. PanelBshows resultsfor:usinganalternativedefinitionofBindbasedontheviolationdummieshand-collectedfromSEC filingsbyNini,Smith,andSufi(2012)inRow[5]; usingabroaderdefinitionofBindthatincludesthefull setofcovenantsthathavethresholdinformationavailableinDealscaninRow[6];usingasmallersample thatexcludesobservationswhenthereisadivestitureofassetsinRow[7];andusingasmallersamplethat excludestheyearsofthefinancialcrisisinRow[8]. PanelsCandDpresentresultsforspecificationsthat includethefollowingadditionalcontrols:investmentinRow[9];adummyforwhetherthefirmundergoes a divestiture of assets in any given year in Row [10]; 2nd- and 5th-order non-liner splines of the distance from the covenant threshold in Rows [11] and [12], respectively; book leverage in Row [13]; Tobin’s Q in Row [14]; Altman’s Z-score in Row [15]; and discretionary accruals in Row [16]. All variable definitions areinAppendixA.Standarderrorsrobusttoheteroskedasticityandwithin-firmserialcorrelationappear belowpointestimates.Levelsofsignificanceareindicatedby*,**,and***for10%,5%,and1%respectively. Robustness EstimatedCoeff, Robustness EstimatedCoeff, Test log(Employment) Test log(Employment) Entire Disconti- Entire Disconti- Sample nuity Sample nuity (1) (2) (3) (4) PanelA:AlternativeSpecifications PanelC:AdditionalControls Controllingfor: [1]Median(quantile)regression -0.206(cid:3)(cid:3)(cid:3) -0.218(cid:3)(cid:3)(cid:3) [9]Investment -0.175(cid:3)(cid:3)(cid:3) -0.126(cid:3)(cid:3)(cid:3) (0.030) (0.048) (0.021) (0.032) [2]Includelaggeddependent -0.133(cid:3)(cid:3)(cid:3) -0.106(cid:3)(cid:3)(cid:3) [10]Divestitures -0.176(cid:3)(cid:3)(cid:3) -0.131(cid:3)(cid:3)(cid:3) (0.014) (0.022) (0.021) (0.032) [3]IncludetwolagsofBind -0.152(cid:3)(cid:3)(cid:3) -0.116(cid:3)(cid:3)(cid:3) [11]2-ndSplines -0.171(cid:3)(cid:3)(cid:3) -0.126(cid:3)(cid:3)(cid:3) (0.039) (0.041) (0.021) (0.030) [4]IncludetwoleadsofBind -0.151(cid:3)(cid:3)(cid:3) -0.115(cid:3)(cid:3)(cid:3) [12]5-thSplines -0.169(cid:3)(cid:3)(cid:3) -0.117(cid:3)(cid:3)(cid:3) (0.028) (0.036) (0.021) (0.031) PanelB:AlternativeCovenantsandSamples PanelD:OtherAdditionalControls Controllingfor: [5]Useviolationsdummyfrom -0.120(cid:3)(cid:3)(cid:3) n.a. [13]Leverage -0.172(cid:3)(cid:3)(cid:3) -0.142(cid:3)(cid:3)(cid:3) Nini,Smith,andSufi(2012) (0.024) n.a. (0.022) (0.030) [6]Includeallothercovenants -0.070(cid:3)(cid:3)(cid:3) -0.092(cid:3)(cid:3)(cid:3) [14]Tobin’sQ -0.175(cid:3)(cid:3)(cid:3) -0.132(cid:3)(cid:3)(cid:3) (0.011) (0.027) (0.023) (0.032) [7]Excludedivestingfirm-years -0.162(cid:3)(cid:3)(cid:3) -0.113(cid:3)(cid:3)(cid:3) [15]Z-Score -0.163(cid:3)(cid:3)(cid:3) -0.102(cid:3)(cid:3)(cid:3) (0.022) (0.028) (0.022) (0.031) [8]Excludefinancialcrisisyears -0.163(cid:3)(cid:3)(cid:3) -0.116(cid:3)(cid:3)(cid:3) [16]Accruals -0.164(cid:3)(cid:3)(cid:3) -0.128(cid:3)(cid:3)(cid:3) (0.023) (0.033) (0.023) (0.032) 39

Table7: UnionElectionSample: SummaryStatistics Thistablepresentssummarystatistics(meansandmedians)forourmergedUnion(NLRB)-Dealscan(DS)- Compustat sample, which consists of 3,814 observations for nonfinancial firms between 1985 and 2010 correspondingtoloansthathaveunionrepresentationelectioninformationinNLRBforfirmsinCompustat and loan pricing information in Dealscan one year after the union election event (Column 1 and 2, Panel A).Forthesakeofcomparison,Columns3and4(PanelA)reportsummarystatistics(meansandmedians) for the Other Dealscan-Compustat sample, which consists of the remaining observations in the merged Dealscan-Compustat sample for the same period that have no matching information in NLRB. Panel B reportssummarystatistics(meansandmedians)fortwosub-samplesoftheUnion(NLRB)-Dealscan(DS)- Compustat sample, based on whether the representation election results in a win or a loss for the union. PanelCreportssummarystatistics(meansandmedians)fortwomoresub-samplesoftheUnion(NLRB)- Dealscan(DS)-Compustatsample,basedonwhethertherepresentationelectionresultsina"close"winor a "close" loss for the union and with "closeness" defined as a narrow range (a vote share range of 5%) (cid:6) aroundthemajority(50%)thresholdneededfortheuniontowinrepresentation("Discontinuitysample"). AllvariabledefinitionsareinAppendixA. PanelA:Union(NLRB)-Dealscan(DS)-CompustatSample NLRB-DS-Compustat OtherDS-Compustat Mean Median Mean Median (1) (2) (3) (4) UnionElectionResultsandOtherElectionCharacteristics: UnionWin 0.36 0 n.a. n.a. UnionElectionSize 268.5 141.0 n.a. n.a. UnionVoteShare 0.46 0.43 n.a. n.a. LoanSpreadsAfterUnionElections: LoanSpread,Year=+1(bps) 180 150 223 212 LoanSpread,Year=+2(bps) 181 150 223 212 LoanSpread,Year=+1,+2(bps) 181 150 223 212 LoanSpreads&FirmCharacteristicsBeforeUnionElections: Log(Assets) 9.0 9.2 6.6 6.5 M/B 1.6 1.3 1.6 1.3 RatedDummy 0.58 1 0.55 0 LoanSpread,Year=-1(bps) 179 125 223 212 PanelB:DifferenceinPre-ElectionCharacteristicsbetweenAllUnionWinsvs. Losses UnionWins UnionLosses Mean Median Mean Median (1) (2) (3) (4) LoanSpreads&FirmCharacteristicsBeforeUnionElections: Log(Assets) 8.9 9.2 9.0 9.3 M/B 1.5 1.3 1.6 1.4 RatedDummy 0.57 1 0.59 1 LoanSpread,Year=-1(bps) 184 150 176 125 PanelC:DifferenceinPre-ElectionCharacteristicsbetweenCloseUnionWinsvs. Losses CloseUnionWins CloseUnionLosses Mean Median Mean Median (1) (2) (3) (4) LoanSpreads&FirmCharacteristicsBeforeUnionElections: Log(Assets) 9.1 9.2 9.1 9.2 M/B 1.6 1.3 1.6 1.3 RatedDummy 0.58 1 0.58 1 LoanSpread,Year=-1(bps) 178 125 175 125 40

Table8: UnionElectionsandLoanPricing:BaselineAnalysis Thistablepresentsresultsfortestsofdifferencesinmeansofloanspreadsafterunionelectioneventsdepending on whether the election outcome was a win or a loss for the union. Columns (1) to (4) refer to loan spreads one year after the election and Columns (5) to (8) are for spreads two years after the election. For each spread, Panels A reports results for the entire sample (Columns (1) and (5)) and various sub-samplesthatexcludeelectionsinvolving, inturn, operatingsubsidiaries(Columns(2)and(6)), fewer than150employees(Columns(3)and(7)),andthoseinvolvingbothfewerthan150employeesandinvestmentgrade-ratedfirms(Columns(4)and(8)). PanelBonlyusesobservationsinvolving"close"elections, definedasanarrowrange(avotesharerangeof 5%)aroundthemajority(50%)thresholdneededforthe (cid:6) union to win representation ("Discontinuity sample"). Panel C stratifies the sample based on the number of employees involved in each election. All variable definitions are in Appendix A. Standard deviations appear in square brackets below mean loan spreads and p-values are below the difference in mean loan spreads. Levelsofsignificanceareindicatedby*,**,and***for10%,5%,and1%respectively. PanelA:EntireSample LoanSpread,Year=+1 LoanSpread,Year=+2 (1) (2) (3) (4)=(3)+ (5) (6) (7) (8)=(7)+ All Exclude Exclude Exclude All Exclude Exclude Exclude Subs Small InvGrade Subs Small InvGrade UnionWin 189.7 204.1 214.4 250.1 188.6 200.7 207.9 264.1 [144.0] [145.5] [155.4] [155.5] [159.3] [138.2] [187.3] [194.9] UnionLoss 168.9 163.0 161.1 191.1 153.5 156.0 149.5 198.4 [128.8] [121.7] [139.1] [150.5] [129.9] [129.5] [139.1] [153.9] Difference 20.8(cid:3)(cid:3)(cid:3) 41.1(cid:3)(cid:3)(cid:3) 53.4(cid:3)(cid:3)(cid:3) 59.0(cid:3)(cid:3)(cid:3) 35.1(cid:3)(cid:3)(cid:3) 44.7(cid:3)(cid:3)(cid:3) 58.4(cid:3)(cid:3)(cid:3) 65.6(cid:3)(cid:3)(cid:3) p-value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observations 3,814 2,382 1,652 734 3,811 2,366 1,635 720 PanelB:DiscontinuitySample UnionWin 236.5 244.4 265.7 295.1 226.7 252.9 213.9 259.5 [164.9] [186.3] [199.9] [204.6] [149.5] [149.3] [142.0] [133.3] UnionLoss 167.2 135.7 125.9 141.8 141.6 151.0 119.5 140.1 [146.5] [114.9] [101.7] [110.8] [106.9] [120.1] [102.7] [140.1] Difference 69.2(cid:3)(cid:3)(cid:3) 108.7(cid:3)(cid:3)(cid:3) 139.7(cid:3)(cid:3)(cid:3) 153.3(cid:3)(cid:3)(cid:3) 85.1(cid:3)(cid:3)(cid:3) 101.9(cid:3)(cid:3)(cid:3) 94.4(cid:3)(cid:3)(cid:3) 119.4(cid:3)(cid:3)(cid:3) p-value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observations 470 276 232 102 511 301 227 100 PanelC:LoanSpread,Year=+1byUnionElectionSize (1) (2) (3) (4) (5) (6) (7) (8) N 100 N 150 N 200 N 250 N 300 N 350 N 400 N 450 (cid:21) (cid:21) (cid:21) (cid:21) (cid:21) (cid:21) (cid:21) (cid:21) UnionWin 190.7 214.4 215.9 226.4 232.1 240.5 240.8 253.0 [146.2] [155.4] [171.4] [182.2] [190.2] [200.2] [201.1] [213.5] UnionLoss 167.9 161.1 162.2 151.5 149.5 155.4 153.7 142.1 [138.9] [139.1] [138.1] [143.6] [147.1] [152.4] [159.8] [119.4] Difference 22.8(cid:3)(cid:3)(cid:3) 53.4(cid:3)(cid:3)(cid:3) 53.7(cid:3)(cid:3)(cid:3) 74.9(cid:3)(cid:3)(cid:3) 82.6(cid:3)(cid:3)(cid:3) 85.0(cid:3)(cid:3)(cid:3) 87.1(cid:3)(cid:3)(cid:3) 110.9(cid:3)(cid:3)(cid:3) p-value (0.002) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observations 2,466 1,652 1,204 913 736 601 509 447 41

Table9: UnionElectionsandLoanPricing:Matched-SampleAnalysis Thistablepresentsresultsfort-testsofdifferencesinmeansofaverageexcessloanspreadsinthetwoyears afterunionelectioneventsdependingonwhethertheelectionoutcomewasawinoralossoffortheunion. Averageexcessloanspreadsaredefinedasthedifferencebetweenloanspreadsandtheaverageloanspread foraportfolioofmatchingloans. Columns(1)to(4)ofPanelArefertoaverageexcessloanspreadsover a portfolio of loans matched based on year, industry, and firm size (deciles), while Columns (5) to (8) of PanelBareforaverageexcessloanspreadsoveraportfolioofloansmatchedbasedonyear,industry,and firmgrowthopportunities(Markettobookratiodeciles). Foreachspread,wereportresultsfortheentire sample(Columns(1)and(5))andvarioussub-samplesthatexcludeelectionsinvolving,inturn,subsidiaries (Columns(2)and(6)),fewerthan150employees(Columns(3)and(7)),andthoseinvolvingbothfewerthan 150employeesandinvestmentgrade-ratedfirms(Columns(4)and(8)). Columns(1)to(4)ofPanelCrefer to average excess loan spreads over a portfolio of loans matched based on year, industry, and firm credit ratings, while Columns (5) to (8) of Panel D are for average excess loan spreads over a portfolio of loans matchedbasedonyear,industry,andloanspreads(deciles)oneyearbeforetheelection. Foreachspread, weagainreportresultsfortheentiresample(Columns(1)and(5))andvarioussub-samplesthatexclude electionsinvolving,inturn,subsidiaries(Columns(2)and(6)),fewerthan150employees(Columns(3)and (7)), and those involving both fewer than 150 employees and investment grade-rated firms (Columns (4) and(8)). AllvariabledefinitionsareinAppendixA.Standarddeviationsappearinsquarebracketsbelow meanloanspreadsandp-valuesarebelowthedifferenceinmeanloanspreads. Levelsofsignificanceare indicatedby*,**,and***for10%,5%,and1%respectively. A.Year,Industry,&SizeMatched B.Year,Industry,&GrowthOpp. Matched (1) (2) (3) (4)=(3)+ (5) (6) (7) (8)=(7)+ All Exclude Exclude Exclude All Exclude Exclude Exclude Subs Small InvGrade Subs Small InvGrade UnionWin 24.0 23.6 38.2 139.4 -26.2 -10.1 -27.1 124.6 [121.3] [120.8] [121.1] [140.4] [134.0] [132.2] [135.0] [131.6] UnionLoss 8.7 -2.1 4.7 76.9 -42.7 -51.0 -58.9 34.1 [111.6] [108.2] [115.4] [155.3] [121.1] [115.7] [126.6] [165.1] Difference 15.3(cid:3)(cid:3)(cid:3) 25.8(cid:3)(cid:3)(cid:3) 33.4(cid:3)(cid:3)(cid:3) 62.5(cid:3)(cid:3)(cid:3) 16.5(cid:3)(cid:3)(cid:3) 41.0(cid:3)(cid:3)(cid:3) 31.8(cid:3)(cid:3)(cid:3) 90.5(cid:3)(cid:3)(cid:3) p-value (0.001) (0.000) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) Observations 3,122 1,845 1,378 606 3,124 1,844 1,376 605 C.Year,Industry,&RatingsMatched D.Year,Industry,&PriorSpreadMatched (1) (2) (3) (4)=(3) (5) (6) (7) (8)=(7) All Exclude Exclude +Below All Exclude Exclude +Below Subs Small InvGrade Subs Small InvGrade UnionWin 10.9 9.1 26.9 39.0 19.4 12.9 26.3 140.3 [97.4] [98.4] [120.9] [151.8] [103.9] [106.0] [111.5] [149.6] UnionLoss -5.1 -10.7 -2.3 -22.1 2.2 -6.7 -2.9 59.1 [93.3] [90.4] [95.0] [135.9] [92.3] [84.4] [96.8] [165.3] Difference 16.0(cid:3)(cid:3)(cid:3) 19.9(cid:3)(cid:3)(cid:3) 29.2(cid:3)(cid:3)(cid:3) 61.1(cid:3)(cid:3)(cid:3) 17.2(cid:3)(cid:3)(cid:3) 19.6(cid:3)(cid:3)(cid:3) 29.3(cid:3)(cid:3)(cid:3) 81.2(cid:3)(cid:3)(cid:3) p-value (0.000) (0.000) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) Observations 3,113 1,840 1,372 604 3,106 1,829 1,376 605 42

Figure1:LoanCovenantViolationsandEmployment Thesampleconsistsof11,536firm-yearobservationsfornonfinancialfirmsbetween1994and2010correspondingtofirmsthathaveatleastoneprivateloanfoundinDealscanwithacovenantthatrestrictscurrent ratioornetworthtolieaboveacertainthreshold. Thisfiguresshowsaveragepercentageannualchanges inthenumberofemployeesineventtimeleadingtoandaftertheyearwhenacovenantviolationoccurs. Figure2:UnionizationElectionsandLoanSpreads Thesampleconsistsof3,814observationsfornonfinancialfirmsbetween1985and2010correspondingto loansthathaveunionrepresentationelectioninformationinNLRBforfirmsinCompustatandloanpricing informationinDealscanoneyearaftertheunionelectionevent. Thisfiguresplotsmeanloanspreadsfor each of ten equally-spaced bins of the data sorted on values of the union vote share variable, with the verticallinedenotingthe50%votethreshold. 43