Private Equity and Debt Contract Enforcement: Evidence from Covenant Violations

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Private Equity and Debt Contract Enforcement: Evidence from Covenant Violations Sharjil Haque, Anya Kleymenova 2023-018 Please cite this paper as: Haque, Sharjil, and Anya Kleymenova (2023). “Private Equity and Debt Contract Enforcement: Evidence from Covenant Violations,” Finance and Economics Discussion Series 2023-018r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.018r1. 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.

Private Equity and Debt Contract Enforcement: Evidence from Covenant Violations * SharjilHaque† AnyaKleymenova‡ ThisVersion: September2023 Abstract Using the Shared National Credit supervisory data, we find Private Equity (PE) sponsored firmsviolateloancovenantsmoreoftenthancomparablenon-PEfirms.However,uponcovenant violation, PE-sponsoredborrowersexperiencerelativelysmallerreductionsincreditcommitments, suggesting lenders are more lenient with these borrowers. This limited-punishment effectexistsinbothcovenant-heavyandcovenant-liteloansbutisstrongerforbankswithrelativelyhighercapital. Limitedpunishmentisdrivenbyrepeateddealsandsponsorreputation, aswellasthehigherbargainingpowerofsponsorsinloanrenegotiation. Ourresultsindicate sponsorsgeneratefinancialflexibilitybydampeningdebtcontractenforcementfordistressed borrowers. JELClassification: G21,G23,G32 Keywords: PrivateEquity;Covenants;LoanRenegotiation;SyndicatedLoans *TheviewsexpressedinthispaperarethoseoftheauthorsanddonotnecessarilyrepresenttheviewsoftheFederal Reserve Board or the Federal Reserve System. Preqin’s PE data were obtained by Sharjil Haque, one of the authors prior to employment at the Federal Reserve Board, while he was a Ph.D. candidate at the University of North CarolinaatChapelHill. WewouldliketothankRusAbuzov,ReenaAggarwal,GregBrown,MurilloCampello,Gustavo Cortes,MustafaEmin,OlegGredil,ArunGupta,JesperHaga(discussant),IftekharHasan,IvanIvanov,AnilK.Jain, Stephen Karolyi, Spyridon Lagaras (discussant), Andrey Malenko, Simon Mayer, Greg Nini, Uday Rajan, Doriana Ruffino,FlorinVasvari,PatrickVerwijmeren(discussant),JamesWang,TengWang,MichaelWeisbach,andconference and workshop participants at the Federal Reserve Board, Federal Reserve Bank of Kansas City, LBS Private Capital Symposium2023, EdinburghCorporateFinanceConference, IFABSOxford2023Conference, 4thVaasaBankingResearchWorkshop, theOfficeoftheComptrolleroftheCurrency(OCC),and2023ICFBAfortheirhelpfulcomments andsuggestions. WearegratefultoRobertCoteforguidanceontheSNCdata. JoeYukeandSamuelRossprovided stellarresearchassistance. †FederalReserveBoardofGovernors.Email:sharjil.m.haque@frb.gov ‡FederalReserveBoardofGovernors.Email:anya.kleymenova@frb.gov

Introduction Acentralquestioninfinancialeconomicsishowdebtcontractenforcementaffectsfirmfinancing (Smith Jr and Warner, 1979; Favara, Morellec, Schroth, and Valta, 2017). This question matters particularly for highly leveraged borrowers, such as those backed by private equity (PE) funds. In PE-sponsored leveraged buyout (LBO) deals, lenders typically exercise control rights over the borrowerthroughvariouscovenants.1 PriorresearchhasshownthataPEsponsor’sreputational capitalcanleadtomoregenerouscovenantstructures(DemirogluandJames,2010;Ivashinaand Kovner, 2011).2 These studies have generally focused on covenants observed at loan origination or deal entry, and our understanding of (i) how often PE-sponsored firms violate covenants and (ii)theconsequencesofcovenantviolationsisstilllimited. Inparticular,thereisanimportantgap in our understanding of a lender’s enforcement behavior towards PE-sponsored borrowers after a covenant is violated, which can have adverse consequences for net debt issuance (Roberts and Sufi, 2009a), real investment (Nini, Smith, and Sufi, 2009), and employment (Falato and Liang, 2016). How does the presence of a PE sponsor shape the enforcement of debt contracts following a contractualbreach? Relatedly,howimportantisex-antecontractdesignforex-postrenegotiation and enforcement when a PE sponsor is present? Breach of covenants represents a natural setting tostudyenforcementofdebtcontractsbecausecovenantsappearinnearlyallfinancialloancontracts,andenforceabilityistheirdefiningfeature(BeckerandIvashina,2016). Priorresearchshows lenders punish borrowers by reducing credit availability upon violations. Using an administrativecreditregistryofsyndicatedloans,ourkeyfindingisthatlendersdisplaylimitedpunishment towardsPE-sponsoredfirmsrelativetocomparablenon-PEborrowersuponacovenantviolation. Consequently,PE-backedborrowersretaingreateraccesstocredit(atfavorableterms)evenwhen they are in the early phases of distress. We provide evidence for two related mechanisms potentially explaining limited punishment towards PE-backed borrowers: (i) repeated deals with PE sponsors,whichincentivizeslenderstopreserverelationshiprent,and(ii)relativelyhighbargainingpowerofPEsponsorsvis-a-vislendersinrenegotiatingloancontractsfollowingviolations. 1Covenantsregulatecorporatepoliciesthroughtriggersbasedonfinancialmetrics(ChavaandRoberts,2008). 2Throughoutthetext,weusethefollowingtermsinterchangeably:PEsponsor,financialsponsor,orsimplysponsor. Aloanis“sponsored”or“backed”byaPEfundwhenitprovidestheequitycapitalthatfinancesaleveragedbuyout, whileabankandotherlendersprovidethedebt. 1

To examine this question, we construct a novel database of PE-sponsored loans that contains supervisory information on covenant types and covenant compliance. In particular, we combine confidential loan-level information from the Shared National Credit (SNC) program, which is jointly administered by the Federal Reserve, Federal Deposit Insurance Corporation (FDIC), and Office of the Comptroller of the Currency (OCC), with data from Preqin, which identifies PE-sponsored leveraged buyouts (LBO). The SNC data allows us to directly observe whether a loan covenant is in compliance or not, whether the loan is in compliance only after amendment, andwhetherthelendergrantstheborroweracovenantwaiveratagivenpointintime. TheSNC covenantsampleoffersseveraladvantagesoveralternativedatasetssuchasDealScan,includinga largersampleandgreatercoverageofprivatefirms(Chodorow-ReichandFalato,2022),aswellas detailed syndicate membership structure throughout the life of a given loan (Irani, Iyer, Meisenzahl, and Peydro, 2021). We follow the covenant compliance of over 2,200 PE-sponsored firms that borrowed from the syndicated loan market between 2012 and 2021, covering around USD 2 trillioninsponsoredloansinthecross-section. We begin by presenting several new facts about covenants in leveraged buyouts. First, followingChristensenandNikolaev(2012),weclassifycovenantsintoperformance-basedandnonperformance-based and find over 50 percent of our PE sample observations have covenants directly linked to their current earnings (e.g., maximum leverage, interest coverage, debt service coverage, or fixed charge coverage ratio). Second, loans with performance-based covenants are held by concentrated syndicates (i.e., those with few non-bank institutional investors) consistent withrecentfindingsonsplitcontrolrightsbyBerlin,Nini,andEdison(2020). Third,wefindthat PE-backed firms tend to violate covenants more often than non-PE-backed firms. Their average annual rate of covenant violations is 18 percent when we look at all covenants and 21.9 percent whenweexamineperformance-basedcovenants. Fornon-PEfirms,thisrateis16.1(allcovenants) and 20.4 (performance-based) percent, respectively. We investigate these descriptive findings by estimatingaloan-levellinearprobabilitymodel. WefindthatPE-backedborrowershaveatleasta 4to5percentgreaterprobabilityofviolatingacovenantrelativetonon-PE-backedfirmsbutalso receivemorecovenantwaiversorresets. Establishing a causal impact of PE presence on covenant enforcement presents challenging identification problems as both covenant violations and PE investments are endogenous. The 2

idealempiricalresearchdesignwouldallowforrandomlymatchingPEsponsors,borrowers,and covenant violations. While such a setting is impossible, our research design attempts to address these challenges. In particular, we compare loans of the same type that have observably similar creditrisk,originatedatthesamepointintime,havesimilarcovenants(i.e.,thoselinkedtocurrent performance and those that are not), and are issued by the same bank to borrowers in the same industry-time. This allows us to narrow the only observable dimension that the borrowers differ alongwithwhetherornotaPEfundsponsorsthem. TheidentifyingassumptionisthatabsentPE involvement,bothborrowertypeswouldhaveexperiencedthesameoutcomefollowingcovenant violations. Following Bernstein, Lerner, and Mezzanotti (2019) and Boucly, Sraer, and Thesmar (2011),inourrobustnesstests,wealsore-estimateourbenchmarkspecificationusingamatching procedureforthesetofloansinoursampleforwhichwecanobtainfirm-leveldata. Unobservablefactorscorrelatedwithcovenantviolationsandenforcementbehaviorcouldstill exist. To mitigate this concern, we use an instrumental variable research design and exploit personality or examination style across federal bank examiners, where the endogenous variable is an indicator of covenant violation status. The excluded instrument is the strictness of the bank’s supervisoratthetimeofthebuyoutloanorigination. Supervisorsfrequentlymeetwithbankmanagement to assess bank risk and take corrective actions (Hirtle, Kovner, and Plosser, 2020), but theirassignmenttodifferentlendersisquasi-exogenous(Agarwal,Lucca,Seru,andTrebbi,2014; IvanovandWang,2022). Ourintuitionisthatloansmadeunderstrictersupervisorshavetighter covenants and thus have higher probabilities of covenant violation. Similar to Ivanov and Wang (2022), we exploit personality differences across supervisors, which affect supervisory strictness, hence covenant tightness, faced by lenders within each federal district. Crucially, a supervisor’s historyofstrictnessisconfidentialandunobservedbythePEsponsorandborrower. Across all of our specifications, we find strong evidence of limited punishment towards PEbacked borrowers following covenant violations. Our baseline results show covenant violations lead to credit commitment reductions of around 11-12 percent for all firms. However, this credit reduction is only around 5.0 percent for PE-backed firms. At the extensive margin, the limitedpunishment effect is even stronger. While we focus primarily on the credit amount, we also document limited punishment in terms of loan maturity (i.e., the reduction in loan maturity is lowerforPE-sponsoredborrowers)andinterestratespreads(theincreaseinloanspreadsislower 3

for PE-sponsored borrowers). Further, we examine loan performance and find that loans with a PE sponsor tend to be downgraded more often relative to non-PE, consistent with PE’s higher covenantviolationrate. However,whenwecomparePE-sponsoredloansthatviolatedcovenants with non-PE loans that also violated covenants, we find little evidence that eventual downgrade anddefaultratesaresystematicallydifferentacrossborrowertypes. Next,weconsiderwhetherdifferencesinex-antecontractdesign,specificallycovenantdesign, betweenPEandnon-PEloansaffectex-postrenegotiationandenforcement. BeckerandIvashina (2016)showcovenant-liteloansfeaturelowerenforceabilityduetocreditorcoordinationcosts. To examinethispossibility,were-estimateourbenchmarkregressionsseparatelyforcovenant-heavy loansandcovenant-liteloansand-surprisingly-findthatlimitedpunishmenttowardsPE-backed borrowers is present in both samples. This finding suggests that the creditor coordination cost hypothesis cannot fully explain our benchmark result since loans with traditional maintenance covenants typically feature relatively fewer non-bank institutions (Berlin et al., 2020). Exploring additionalfactorsthatshapeenforcementandrenegotiation,wefindlendersthatdisplaylimited punishmentorleniencytowardsPEtendtohavemoreregulatorycapital,asmeasuredbyabank’s Tier1capitalratioimmediatelybeforeacovenantviolation. Weexploretwo(notmutuallyexclusive)channelsthatcouldpotentiallyexplainlimitedpunishment toward PE. First, we connect our results to models of relationship rent stemming from repeated interactions between PE sponsors and creditors (Malenko and Malenko, 2015; Buccola, 2023). In particular, we posit that reputational capital can mitigate agency costs of lending (Diamond, 1991). Thus, lenders’ willingness to enforce written contracts depends on the expected gains from repeated transactions. Following Demiroglu and James (2010), we construct measures of PE sponsor reputation and find that sponsors with a high reputation in credit markets - measured in terms of LBO deal history - obtain greater leniency from creditors upon covenant violations. Second,weprovideevidenceforabargainingpowerchannelofPEsponsorsinloanrenegotiation upon violation. Survey evidence from Bernstein et al. (2019) shows PE sponsors help their portfoliocompaniesrenegotiatedebtcontractswithbanks,whileLiu(2021)demonstratesthatPE hassuperiorbargainingskills. Weshowthatthelimitedpunishmenteffectispresentevenwhena syndicatedloan’sownershipstructureishighlyconcentrated(i.e.,featuresafewnon-bankinstitu- 4

tionallenderslikeCLOs). Sinceaconcentratedownershipstructurepreserveslenders’bargaining powerandgivesthemhighincentivestorenegotiatethecontractuponcovenantviolation(GiannettiandMeisenzahl,2022),limitedpunishmenttowardsPE-backedborrowersindicatessponsors raise the portfolio company’s bargaining power (relative to non-PE) during loan renegotiations. To dig deeper, we also proxy for a given sponsor’s bargaining power vis-a-vis a given lender by aggregatingalloutstandingloansfromeveryLBOdealbetweenagivensponsorandagivenlead bankintheSNCdata,andfindbankswithhighercumulativecreditexposuretoagivensponsor are more likely to display limited punishment upon covenant breach. As this measure captures a bank’s historical reliance on a given sponsor for deal flow, we expect it to capture a sponsor’s bargainingpower. If sponsors increase bargaining power, we should also see favorable renegotiation outcomes outside of distress because loans are often renegotiated outside of covenant violation or default (Roberts and Sufi, 2009b; Denis and Wang, 2014). Indeed, we confirm that both borrower types frequently renegotiate loan commitments outside of covenant violation. Interestingly, we find such loan amendments outside of distress are associated with higher credit commitment in PEbacked firms relative to non-PE. Because we are able to absorb differences in borrower risk, our interpretationisthatasponsor’sbargainingpowergenerateshigherdebtflexibility. Still, one cannot completely rule out endogeneity concerns related to the non-random selection of PE, even though we match PE and non-PE loans on supervisory risk rating. To alleviate thisconcern,wemergeourSNCsamplewiththeFederalReserve’sFRY-14Qdataoncommercial loans,whichprovidesrichinformationonfirm-levelfinancialvariables. OurbenchmarkresultremainsunchangedwhenwematchPE-backedloanstonon-PEloansbasedonleverage,firmsize, EBITDA,1-yearaheadprobabilityofdefault,andindustryinthepre-buyoutyear. Finally,wealso find evidence that PE-backed borrowers inject equity more often than non-PE to cure covenant violations,butthiseffectisquantitativelysmallerthanthereputationandbargainingpowerchannels. Thissuggestssponsorsprovidemoreoperationalsupportbeforeresortingtoequityinjection asdocumentedinarecentsurveybyGompers,Kaplan,andMukharlyamov(2022),whenportfoliofirmswereindistressduringCovid-19. Overall,ourfindingshighlighthowsponsorsdampen theeffectofstate-contingentallocationofcreditorcontrolrights,allowingPE-backedfirmstohave morefinancialflexibilityindistressandthusdirectlyaffectingthecorporatedebtpolicyofalarge 5

shareofborrowersinthesyndicatedloanmarket. Our paper contributes to several strands of the literature. First, we take a step further in understandingtheroleofcovenantsinshapingthecapitalanddebtstructureofPE-sponsoredfirms. Our paper is closest to Demiroglu and James (2010), Ivashina and Kovner (2011) and Achleitner, Braun, Hinterramskogler, andTappeiner(2012). Thesepaperscomprehensivelyexaminetherole of sponsor reputation in shaping covenant tightness observed at deal origination. More recently, Badoer,Emin,andJames(2021)examinetherelationshipbetweenPEsponsorreputationandthe propensity to use covenant-lite loans. Different from these papers, we are the first to examine outcomes after origination throughout the life of a given loan, focusing on (i) the propensity of PE-backed firms to violate covenants, (ii) the adverse consequences of covenant violations and subsequentresolution,and(iii)ifdifferencesinex-antecovenantdesignorinasyndicatedloan’s dynamicownershipstructurebetweenPEandnon-PEaffectex-postdebtcontractenforcement. Second, we contribute to the broader literature on how debt contract enforcement affects a distressed borrower’s access to credit, using covenant violations as our setting. Several papers showadverseconsequencesofcovenantviolations, includingChavaandRoberts(2008), Roberts and Sufi (2009b), Nini et al. (2009) Nini, Smith, and Sufi (2012), Denis and Wang (2014), Falato and Liang (2016), Berlin et al. (2020), Carey and Gordy (2021), Becher, Griffin, and Nini (2022), Chodorow-Reich and Falato (2022), and Bräuning, Ivashina, and Ozdagli (2022). Different from thesepapers,wearethefirsttoshowhowthepresenceofafinancialsponsordampenstheeffects of covenant violations, thus generating financial flexibility in funding cash flow shortfalls. Since thischanneldirectlyaffectsaccesstocredit,ourpaperalsocontributestotherelatedliteratureon firms’financialconstraints(KiyotakiandMoore,1997;LianandMa,2021). Finally, we contribute to the large literature on the effects of private equity buyouts and offer newinsightsonloanperformance. AssuggestedbyKaplanandStromberg(2009)andrecenttheories(MalenkoandMalenko,2015;GryglewiczandMayer,2022),PEownersaffectfirmoutcomes throughvariouschannels. SeveralpapersstudywhetherandhowPEownersaffectfirmoutcomes andvaluecreation.3 Differentfromthesepapers,ourdataallowsustoexaminetheeffectofPEon loandowngrades,loandefaults,andloanamendmentspost-origination. 3See, for example, Boucly et al. (2011); Axelson, Jenkinson, Strömberg, and Weisbach (2013); Davis, Haltiwanger, Handley,Jarmin,Lerner,andMiranda(2014);Bernsteinetal.(2019);Gornall,Gredil,Howell,Liu,andSockin(2021); Johnston-Ross,Ma,andPuri(2021);Fracassi,Previtero,andSheen(2022);Haque,Jang,andMayer(2022). 6

1 Data and Stylized Facts 1.1 Data We begin by describing our data sources and sample characteristics. We build a large loan-level sample that primarily relies on merging two key datasets containing information on (i) covenant violations and pertinent loan characteristics and (ii) identifying information on private equitysponsoredborrowers. Data on Covenant Compliance: Our data on loan contracts and covenant compliance come from the Shared National Credit Program (SNC). Administered by the Federal Reserve System (FRS), Federal Deposit Insurance Corporation (FDIC), and the Office of the Comptroller of the Currency (OCC), the SNC Program covers all syndicated deals exceeding USD 20 million and held by three or more supervised institutions, which is the SNC inclusion criterion (Ivanov and Wang,2022). Thisincludesloanpackagescontainingtwoormorefacilitiestothesamecompanyat thesameoriginationdate,withatotalloanamountofoverUSD20million.4 Thelendersinclude domestic and foreign institutions, commercial banks, investment banks, insurance companies, investmentcompaniessuchasCLOs,andmutualandhedgefundswhenevertheparentcompany isregulated. Asof2021,SNCcommitmentstotaledUSD5.2trillion.5 Thesyndicatedloanmarket includes both leveraged and non-leveraged loans, and our analysis relies on the entire sample. Recent academic studies using the SNC data to study covenant violations include Chodorow- ReichandFalato(2022)andIvanovandWang(2022). The SNC is a loan-level panel dataset. The reporting frequency is annual before 2015, quarterly in 2015, and semi-annual from 2016 onward. Examiners collect information on covenant compliance for around one-third of the SNC universe of loans. The SNC covenant sample overweightsnoninvestmentgradeandcriticizedloansbutisotherwiserepresentativeofthefullSNC universe (Chodorow-Reich and Falato, 2022). As mentioned earlier, the SNC covenant sample offersseveraladvantagesformeasuringcovenantcomplianceoverpreviousdatasetsconstructed bystartingfromtheDealScandatabaseandhand-collectinginformationonsubsequentloanoutcomes from public filings. In particular, the SNC sample is much larger and contains a large and 4In2018,theminimumcommitmentsizewasraisedtoUSD100million. 5Forfurtherdetails,seethe2022reviewoftheSNCprogram. 7

representative share of nonpublic borrowers. This is particularly important because we are investigating PE-sponsored, typically not publicly traded firms. Moreover, it contains supervisory information on covenant compliance, including when a covenant breach results in a waiver and thelender’sresponsetotheviolation. We observe a number of standard loan-level variables such as loan commitments, utilization rate, maturity, loan type, loan purpose, covenant types, and regulatory classification of loan risk thatwedescribeindetailbelow. Thedataalsobreaksouttheloansyndicatemembership,including non-bank lenders, on a quarterly basis (e.g., CLOs and hedge funds). The lead bank must reportdetailsonagivenloan, eveniftheyarenolongerinthesyndicate. Forasmallersubsetof oursample,wealsoobserveloanspreadsatagivenpointintime. Crucially, we observe Concordance Ratings, which are (time-varying) credit risk ratings that Federal supervisors assign to a loan facility using information related to the borrower provided bytheAgentBank. Theseratingsareprovidedonanumericalscale,wherelowernumbersdenote higher-quality loans. Specifically, a risk rating of 1 denotes an Investment Grade Pass, 2 denotes Non-InvestmentGradePass,3denotesLowestRatedPass,whileratingsof4,5,6and7denoteSpecial Mention,Substandard,Doubtful,andLossrespectively.6 SNCreportsaflagforeachloaninthecovenantsampleforwhethertheloanwasincompliance atagiventime. Moreover,iftheloanremainscompliant,weobservewhetheritwouldhavebeen non-compliant but for a covenant waiver or reset granted by the lender. We follow Chodorow- Reich and Falato (2022) and classify a covenant as breached in either circumstance. However, our resultsarenotsensitivetothisparticulardefinition,whichweshowinourrobustnesstests. Note that a waiver can come with conditions and does not necessarily mean that the violation gets resolvedwithoutadverseconsequencestotheborrower. PEBuyoutListandMatchedSampleInformation: ToidentifyPE-sponsoredLBOs,wecombine the SNC data with information from Preqin.7 Preqin is generally considered a representativedatasourceofPE-sponsoredleveragedbuyoutsandhasbeenutilizedextensivelyintheaca- 6Foranexampleofhowloanqualityismappedfromtheagentbank’sinternalratingtosupervisoryrating,seethis reportingformbytheSNCoffice. 7To match the SNC to our PE dataset, we apply a string matching algorithm following Cohen, Dice, Friedrichs, Gupta, Hayes, Kitschelt, Lee, Marsh, Mislang, Shaton, Sicilian, and Webster (2021) on portfolio company name and industry. Wewenttogreatlengthstoensuretheaccuracyofourdatamerge,whichinvolvedsignificanttimecommitmentsfromseveralresearchassistantsinmanuallycheckingourmatch. 8

demicliterature(see,forexample,BarberandYasuda,2017;Davis,Haltiwanger,Handley,Lipsius, Lerner,andMiranda,2021;ShiveandForster,2022). Preqin’sbuyoutdatacontainsidentifyinginformation on sponsored portfolio companies, industry, the name of the sponsor, and, crucially, deal closing dates, allowing us to distinguish between pre-(post-) PE-ownership samples. Our sample only uses the earliest chronological buyout date if a company is acquired twice or more byaPEfund(secondaryortertiarybuyout). WesupplementourPreqinlistwithdatafromSNC, whichalsocollectsidentifyinginformationonPEsponsors. Our sample period ranges from 2012 to 2021. For most of our sample, loan commitments are observed twice a year, typically once in the first quarter of the calendar year and then again in August of the same year. After filtering out observations for which we do not see covenant compliance and other pertinent loan-contracting information, we begin with a baseline sample covering 43,670 loan-time observations belonging to 11,416 unique credit facilities. These facilitiescover5,660uniqueborrowers,outofwhich2,272arePE-sponsored. Oursamplecontains640 uniquePEsponsors. Aggregatingall(unique)loanstoborrowerswithaPEsponsorinthesample cross-sectionshowsthatsponsoredloansaccountforaroundUSD2.2trillionandUSD1.6trillion incommittedandutilizedloanexposurerespectively. Thesampleincludes6,967covenantviolations,a15.9percentviolationrateinthecross-section,butwithsignificanttime-seriesvariationas weshowsubsequently. OtherData: WealsorelyonCallReportstoextractbank-levelinformation. Forourbenchmark sample, we observe financial variables for participating banks, such as total assets, total equity, total risk-based capital, and total risk-weighted assets. We use these to construct our measures of lender capital ratio. Finally, for part of our analysis, we merge our loan-level sample with firms’ balance sheet data from the Federal Reserve’s FR Y-14Q Corporate Loan Schedule (H1). The FR Y-14 data consists of information on all loan facilities with over USD 1 million in the committedamountheldbyBankHoldingCompanies(BHCs). Thesedataareavailablefrom2012 andrepresentsupervisorydatacollectedaspartoftheFederalReserve’sStressTestingexercise.8 PartAprovidesvariabledefinitionsforoursample. 8WechoosetorestricttheSNCsamplefrom2012onwardinordertooverlapwithFR-Y14Q,whichstartsin2012. 9

1.2 SummaryStatistics Table1reportssummarystatisticsforourloan-timesamplebyPE-sponsorshipstatus. Thesample is approximately evenly split between sponsored and non-sponsored loans, although the numberofnon-sponsoredloan-timeobservationsissomewhathigher. WeobservethatPE-sponsored loansarelargerthannon-PEloans. ThemeanconcordanceratingishigherforPE-sponsoredloans relativetonon-PE,indicatingthattheseloansareriskieronaverage. Giventheimportanceofconcordance ratings in our formal analysis, we provide a further breakdown of our sample, split by rating and borrower type in the Online Appendix Table A2. Importantly, we observe differences betweenPEandnon-PEloans: only16percentofPE-backedloansreceivean“Investment-Grade Pass”ratingwhile22.5percentofnon-PEloanshavethisrating,consistentwithpriorstudiesdocumentingthatsponsoredloansaremostlyleveragedloans. WealsoobserveagreatershareofPE loans classified as “Special Mention,” “Doubtful,” “Substandard,” or “Loss” (19.1 vs. 16.5 percent). These patterns confirm that PE-backed loans are in general more risky, and are consistent withpriorstudiesthatdocumentsponsorsdonotrandomlytargetcompanies. Next, wealsoobserveloanspreadsforasubsetofourloans. Consistentwiththedistribution of loan ratings mentioned above, we see mean credit spreads are higher in PE relative to non-PE and comparable to those reported in Axelson et al. (2013). The final rows of Panel A and B in Table1showaround13.9percentofourloan-timeobservationsareflaggedashavingacovenant thatwaseitherwaivedorresetbythelender. Thismeansthecovenantwouldhavebeenviolated hadthewaiverorresetnotbeengranted. For the PE sample, we also compute a variable we call Total PE Sponsor-Bank Exposure. This measureaggregatesalloutstandingloans(utilizedportionsofcommitments)fromportfoliocompaniesthatarefundedbythesamePEsponsor(e.g.,KKR)andLeadBankpairatdatet,capturing agivenlender’stotaldebtexposuretoagivensponsor’sportfoliocompanies. Weusethismeasure insomeofourtestsdiscussedsubsequently. 1.3 KeyFacts: CovenantdesignandviolationinPE-backedloans Weproceedbyestablishingfourkeyfactsinthedata. Fact 1. Over 50 % of PE-sponsored loans include a maximum leverage ratio and/or other 10

traditionalperformance-basedcovenants. Considering that covenants in syndicated loans are often tied to a firm’s current earnings, we choose to classify covenants into performance-based and non-performance-based following Christensen and Nikolaev (2012). Performance-based covenants include debt-to-EBITDA ratios (leverage ratio or senior leverage ratio covenant), interest coverage ratios, debt service coverage ratios,fixedchargecoverage,andothervariableswiththegeneralcharacteristicthatthecovenant must capture some measure of earnings before interest and taxes. Non-performance covenants in our sample include negative, affirmative, current ratio, or other balance-sheet-related capital covenants. Overall, we find more than 50 percent of the PE sample includes performance-based covenantsthataredirectlylinkedtotheborrower’searnings.9 The SNC database includes a description of each covenant type, which we report in Table 2. Themostfrequentloancovenantinthesampleisthemaximumleverageratiocovenant,whichis presentinatleast29percentofthesample,consistentwithIvashinaandKovner(2011)whoalso find the same covenant is present in 29 percent of their data using the DealScan database. The second most frequent maintenance covenant is the interest coverage ratio. As Panel A in the tableshows,around55.8percentofthePEsampleincludesatleastoneofthefollowingcovenants: leverage/seniorleverageratio,interestcoverage,debtservicecoverageratio,orfixedchargecoverage ratio. We also find that negative covenants are quite common. They are present in at least 20percentofallPE-sponsoredloans. When we examine non-PE loans in Panel B, we find that 62 percent of the loans have the fourperformance-basedcovenantsmentionedabovebutfeaturerelativelyfewernegativeoraffirmative covenants. Since performance-based covenants subject borrowers to stricter monitoring, weviewthehighernumberofperformance-basedcovenantsandfewernegativeandaffirmative covenantsasevidenceconsistentwithpriorstudiesthatdocumentPEsponsorsgetmoregenerous covenantsrelativetonon-PE(DemirogluandJames,2010;BeckerandIvashina,2016). Fact2. Loanswithperformance-basedcovenantsareheldbyconcentratedsyndicates. Recentstudiesshowleadbanksoftenselltheirloanstonon-bankinstitutionalinvestors(Blickle, Fleckenstein, Hillenbrand, and Saunders, 2020) and that there has been a sharp increase in deals 9ThemediannumberofcovenantsperSNCloanis1,whilethe75thpercentileshowsthreecovenantsperloanin bothPE-backedandnon-PE-backedsamples. 11

with split control rights (Berlin et al., 2020) in the syndicated loan market. As shown by Berlin et al. (2020), split control credit agreements delegate the exclusive right to monitor and renegotiate financial covenants to banks. We examine if most performance-based covenants (e.g., leverage ratio or interest coverage ratio) in the private equity sample are held within concentrated syndicates—i.e., syndicates with few non-bank institutional lenders like CLOs and loan funds— suchthatbanksretainsubstantialrenegotiationandbargainingpoweruponcovenantviolation. Foragivenloan,weexaminetherelationbetweenthenumberofnon-bankinstitutionallenders in a syndicate and the likelihood of the loan having performance-based covenants. In the SNC’s Loan Participant data, we can observe the names of each non-bank investor (e.g. XYZ CLO Ltd. or ABC Credit Fund) holding a given loan at a given time. Therefore, since we can observe the number of institutional lenders holding a given loan at a given time, we create buckets of the number of non-bank institutional investors and compute the share of loans with performancebased covenants in each bucket (see Figure 1). The chart is restricted to PE-backed loans only as prior studies argue that loans with PE sponsors are more likely to be covenant-lite (Badoer et al., 2021; Becker and Ivashina, 2016). We see that when the number of institutional lenders is ten or less, 60 percent of the loans have performance-based covenants. However, when loans have 20 or more non-bank investors, this share drops sharply to 38 percent. Further, the share drops to 33 percent when the number of non-bank investors rises to 50 or more. Consistent with split control rights, these patterns indicate banks generally retain the loan tranches with traditional performance-based covenants to renegotiate and bargain effectively when borrowers are in distress. Table1showsPE-sponsoredloanshaveahighernumberoflendersintheirsyndicatestructure. For example, the median number of Institutional/Non-Bank Lenders is 12 in the PE-sponsored loans sample and 10 in the non-PE, while the difference in means is much higher. However, we also see a significant share of the PE sample has relatively few non-bank investors. For example, thedistributionsofthenumberoflendersaswellasthenumberofnon-bankinvestorsinTable1 show that close to half of all PE-sponsored loans have ten or fewer non-bank lenders. Further, one-quarterofPE-sponsoredloanshaveonlysixorfewerlendersatagiventime. Finally,wesplitoursamplebyloantype,whichweclassifyasrevolvingcreditfacilities,term loans,andotherloans. Wefindmorethan90percentoffacilitiesarerevolversandtermloansfor 12

bothtypesoffirms,consistentwithpriorstudies. PE-sponsoredborrowershavemoretermloans consistentwithAxelsonetal.(2013). Inaddition,asFigure2shows,PE-sponsoredfirmsexhibita moreevensplitofthetwomajortypesofloanfacilitiesrelativetonon-PE.Thisisagainconsistent with split control rights in which covenant-lite term loans are typically paired with a revolving creditfacilitytopreservemonitoringandrenegotiationpowerwiththesubsetoflendersthathold therevolver. Fact 3. PE-backed borrowers have a higher covenant violation rate relative to non-PEbackedborrowersbutreceivemorewaivers. Chodorow-ReichandFalato(2022)foundnearlyone-thirdofloansintheSNCsampleviolated covenants during the global financial crisis (GFC). Our analysis focuses on the post-GFC period andsplitsthesamplebyPEandnon-PEfirms. Figure3plotstheshareofcovenantviolationsfor firms backed by PE sponsors. We plot the trend for all covenants as well as performance-based covenants. Both types of loans exhibit similar trends, but performance-based covenants are violated more often since they are more sensitive to macroeconomic conditions. For example, both trends exhibited sharp spikes during the calendar years 2015 and 2016, potentially due to the oil price shock of 2014 or the Federal Reserve ending its quantitative easing program. This effect is stronger for performance-based covenants. Since then, we have seen a declining trend until the COVID-19pandemicwhenthecovenantviolationraterosetonearly25percentforperformancebased covenants. This estimate is comparable with the survey evidence from Gompers et al. (2022), whofound that22.7 percentof PE-backedfirms violated covenants during thepandemic. ComputingasimpleaverageovertimeshowsthatPE-sponsoredloansexhibitanaverage(annual) violationrateof18.0percentforallcovenantsand21.9percentforperformance-basedcovenants. Figure 4 plots the same variables for non-PE-owned firms. The spike in violations in 2015– 2016 was much less pronounced for non-PE. The decline in violations in 2020 is simply a lag effect, as a larger share of non-PE loans were examined in February 2020, before the onset of the pandemic. Itthusdisplaysalargejumpinviolationsin2021duetoreviewsconductedinAugust 2020. Non-PE loans violate covenants 16.0 percent of the time for all covenants and 20.4 percent forperformance-basedcovenants. Toexaminethesepatternsmoreformally,weestimateasimplelinearprobabilitymodelwhere the dependent variable takes the value of 1 if a covenant is violated at a given point in time and 13

0 otherwise. We include several loan-level controls, including loan amount, utilization rate, maturity, indicators for loan type, loan purpose, and risk rating. In the next section, we outline our benchmark analysis and describe our loan-level controls and fixed effects in further detail (Section2.1). Eq. (1)showstheequationweestimateinageneralform. Thedependantvariableis(i) 1×(Violated), an indicator taking the value of 1 if any covenant in a loan j between bank-firm pair [b,i] attime t isviolatedand0otherwiseand(ii) 1×(Waiver),anindicatortakingthevalue of1ifacovenantiswaivedorreset,whichmeanstheborrowerwouldhavebeeninviolationofa covenanthadthelendernotgrantedawaiver.10 OurkeyvariableofinterestisanindicatorofPE ownership. 1(Y ) = α+β PE +FEs+Controls+ϵ (1) j,b,i,t 1 i,t j,b,i,t We report these results in Table 3. In columns (1)-(3), comparing PE and non-PE loans that are of the same type (i.e., revolvers or term loans) and risk profile, originated by the same bank to borrowers in the same industry-time, we find PE-backed loans have a higher probability of covenant violation. Our estimates suggest PE-backed firms have approximately 4 to 5 percent highercovenantviolationrates. ThiscouldbeattributabletothefactthattheleverageratioinPEbackedfirmsisrelativelyhigher. Next,columns(4)-(6)examineifPE-backedloansareassociated withgreatercovenantwaiversorresetsgrantedbythelender. Asthecoefficientestimatessuggest, PE-backedloansareassociatedwitharounda4percenthigherprobabilityofreceivingacovenant violation waiver or covenant reset from the lender, controlling for lender, borrower, and loan contract differences. These effects are meaningful, considering the unconditional probability of receivingawaiverinourfullloan-timesampleis13.9percent.11 Fact4. BothPEandnon-PEborrowersfrequentlyrenegotiateloancommitmentsoutsideof distress. Next,weconstructavariableAmendmentOutsideDistress,whichtakesthevalueof1ifagiven loan’s dollar commitment is changed in period t relative to period t−1 conditional on the loan remainingcovenant-compliant(i.e., outsideofdistress). Table1showsloansarefrequentlyrene- 10Asalreadymentioned,inourformalanalysis,weusebothcases(i.e.,flaggedviolationsandwaiversreportedin theSNCdatabase)ascovenantviolationstoseparatetheeventofbreachingacovenantfromthesubsequentresolution. 11Awaivercancomewithconditions,andaborrowermaystillfaceadverseconsequences. 14

gotiated. Conditionalonborrowersremainingoutsideofcovenantbreach,forbothPEandnon-PE borrowers, we observe loan commitments alone are renegotiated around 33 percent of the time. Whenwecollapsealluniqueloanstocounttheshareofloansthatgothroughatleastoneamendment to loan commitment outside of covenant violation, we find PE loans are renegotiated 52 percent of the time, and non-PE loans are renegotiated 46 percent of the time. We want to emphasizethatthisisstillalowerboundintermsoftotalamendmentsbecausesomeloansexperience changestootherloanterms,suchasspreads. Overall,thispatternisconsistentwithpriorstudies thatdocumentloansareoftenrenegotiatedoutsideofdistress(RobertsandSufi,2009b;Denisand Wang,2014;Roberts,2015). 2 Empirical Strategy 2.1 BenchmarkAnalysis We discuss our benchmark analysis in this section and establish the following key results: (i) PE-backedborrowersexperienceasmallerreductionincreditcommitmentuponcovenantviolation relative to comparable non-PE borrowers, (ii) this limited-punishment effect is present in both covenant-heavyandcovenant-liteloans,(iii)limited-punishmentisstrongeramongbanksthathave morecapital,and(iv)evenoutsideofdistress,PE-backedfirmsareabletorenegotiatemorefavorableloanoutcomesrelativetocomparablenon-PE-backedfirms. Our goal is to examine if ex-post enforcement behavior following covenant violations varies systematicallyduetoPE-ownershipstatus. ThekeyempiricalchallengeisthatPEownershipand covenantviolationsarenon-randomandlikelydeterminedinresponsetoborrower-specificcredit risk. Moreover, macroeconomic and bank-specific factors may simultaneously drive covenant violationsandloanoutcomes. Ourbaselineanalysiscomparestheeffectofviolationsonoutcomes between observably similar loans with similar credit risk issued by the same bank, such that the loansdifferonlybyPE-sponsorshipstatus. Unlessotherwisestated,allregressionsareestimated attheloan-timelevel,wheretimeisattheyear-quarterlevel. Webeginwiththefollowingbaseline specification: Y = β PE +β Violate +β PE ×Violate +Z +X +η +θ +ϵ (2) j,b,i,t 1 i,t 2 j,t 3 i,t j,t j,b,i,t j,b,i,t b,t z,t j,b,i,t 15

Thedependantvariableisalternatively(i) Log(Commitments),thenaturallogarithmofcredit commitment in loan facility j issued by bank b to firm i in time t, and (ii) an indicator variable 1 (Credit Reduced) that takes the value of 1 if total committed credits between a given bank-firm pair are reduced in a given time-period t relative to t−1. Our preferred measure is (i) because ourhypothesisisthatthePE-sponsorshipeffectpotentiallymattersmoreattheintensivemargin uponacontractualbreach. Inadditionaltestsdiscussedsubsequently,wealsoexaminetheeffect ofPE-backingonloaninterestratespreadsandloanmaturityuponcovenantviolation. Violate takesthe valueof 1if anycovenant isbreached ina givenloanin thecurrent orany j,t ofthepreviousfourquartersrelativetothedatethatagivencreditcommitmentisobserved. This definition is consistent with prior studies, which show the effects of covenant violation on debt issuance can persist long after the actual violation (Roberts and Sufi, 2009a). Our key variable of interest is PE×Violate , which captures the marginal effect of PE-ownership on loan outcomes j,t conditionalonacovenantviolation. WeestimateEquation(2)overthesampleperiod2012–2021. FollowingGustafson,Ivanov,andMeisenzahl(2021),alsobasedonSNCdata,standarderrorsare clusteredatthebank-timelevel.12 We consider a carefully selected array of fixed effects to absorb confounding borrower and lender risk factors. In particular we include bank-time (η ) and sector-time (θ ) fixed effects. b,t z,t Unlessotherwisestated,alltimefixedeffectsareattheyear-quarterleveloftheSNCreportdate. ThevectorZ includesindicatorsforloanpurpose,loantype(creditline,termloans,andother), j,b,i,t loanoriginationyear-quarter,covenanttype(i.e.,performance-basedvs. non-performance-based covenants),and,perhapsmostimportantly,loanconcordanceratingwhichcapturestime-varying borrower risk.13 Concordance ratings capture credit risk based on a careful appraisal of hard information (e.g., leverage ratio, EBITDA, etc.) and soft information related to a borrower’s repaymentcapacity. X includesaloan’stime-to-maturityandutilizationrate. Ourmainregresj,b,i,t sionsdonotincludeinterestratespreadasanexplanatoryvariablebecausedoingsoreducesour sample by more than 50%. Instead, we report robustness tests using interest rate spread in section 4. As mentioned above, (η ) allows us to examine observably similar loans issued by the b,t 12Wealsoverifiedourresultsareunchangedwithstandarderrorsdouble-clusteredatthebankandindustry×Year- Quarterlevel,followingIvanovandWang(2022). 13For loans with multiple covenants, we classify it as performance-based if it has at least one performance-based covenant. 16

samebanktoborrowerswithinthesameindustry-timethatdifferonlybyPEstatus. Thus,wecan rule out confounding effects from bank-specific channels. Since PE status varies over time, we also use firm fixed effects in some of our specifications. Finally, in one specification, we also add bank×borrowerfixedeffectstofurthercontrolforunobservedtime-invariantfactorsthatarespecific to a bank-firm relationship, such as banks’ private or soft information on borrowers’ creditworthinessandbanks’portfoliospecializationinparticulartypesofborrowers(Chodorow-Reich, 2014). Finally, in our robustness tests reported in subsection 4.3, we show that our main result is robusttomatchingPEloanstonon-PEloansbasedonfirm-levelvariableswithinthesame2-digit industry(leverage,size,probabilityofdefault,andEBITDA). Table4reportsourbenchmarkresultswherethedependentvariableis(i) Log (Commitments) at the loan-level in Panel A, and (ii) 1 (Credit Reduced) in Panel B. Examining the estimates in column (1) in Panel A, we see that β is negative, indicating that violation of a covenant reduces 2 credit commitment, consistent with prior studies. In terms of economic significance, covenant violationsreducecommitmentsby11.6percent.14 Importantly, β ispositiveandsignificant. The 3 estimateindicatesthereductionincreditcommitmentuponviolationisonly4.53percentifafirm has a PE sponsor. Taken together, we can infer that the mitigating effect of PE ownership on lenders’enforcementactionsisquitestrong. Wefindsimilarresultswhenwelookatcolumns(2) to (6) with variations in fixed effects. We want to emphasize that the inclusion of Concordance Ratings means our identifying variation comes from within Investment-Grade loans and within Non-Investment-Gradeloans(Non-Investment-GradeloansareprimarilyLeveragedloans). Panel B reports the results where the outcome is 1 (Credit Reduced), an extensive margin measure. As we estimate the probability of credit reduction, our hypothesis is that β > 0 and 2 β < 0. Consistent with our hypothesis, we find that covenant violations raise the probability of 3 creditreductions. Thequantitativeeffectisquitelarge—rangingfrom6.7percentto8.7percent, depending on the set of controls. But the significant and positive sign on β again indicates a 3 limited-punishmenteffect. Inourrobustnesstests,weshowthattheseresultsholdwhenthecontrolgroupisconstructed using a matching methodology primarily following Bernstein et al. (2019) or Boucly et al. (2011) and controlling for time-varying firm-level variables for the set of loans for which we can merge 14(e(−0.124)−1)×100=−11.66. 17

oursamplewiththeFederalReserve’sFRY-14Qdata. Wediscusstheseresultsinsubsection4.3. 2.2 InstrumentalVariable Despiteourrichsetofcontrols,wecannotcompletelyruleoutnon-randommatchingofborrower characteristics and covenant violations. We address this concern by employing an instrumental variable research design, largely following Ivanov and Wang (2022) and Chodorow-Reich and Falato (2022). The excluded instrument is the strictness of the lender’s supervisor at the time of loanorigination.15 Banksupervisorsfrequentlymeetwithbankmanagementtodiscussbothspecificissuesrelatedtobankactivitiesandmoregeneralperspectivessuchasindustryoutlookand analyze internal reports with the goal of reducing failure risk relative to what banks themselves mightchoose(Hirtleetal.,2020). Ourrelevanceconditionisthatloansmadeunderstrictersupervisorshavetightercovenantsand,therefore,haveagreaterpropensityforcovenantviolation. Ourexclusionrestrictionisbasedontwosourcesofquasi-exogenousvariationinsupervisory strictness at loan origination, which we argue only affects credit commitments through covenant tightness. First, federal supervisors have been shown to be stricter than state supervisors, and there exists a pre-determined periodic rotation between them (Agarwal et al., 2014; Chodorow- Reich and Falato, 2022). Second, within each regulatory-district × supervisor-type combination, supervisorswithvaryinglevelsofleniencyarequasi-exogenouslyassignedtobanks(Ivanovand Wang,2022). Moreover, we explicitly control for other loan characteristics that could be affected by strict supervisors (e.g., loan risk and utilization rates). By controlling for supervisor strictness during the life of the loan, the instrument is valid because of the variation at loan origination. Because we can compare observably identical PE and non-PE loans within each federal district similar to Ivanov and Wang (2022), we circumvent the issue of banks sorting into different regulatory settings. Taken together, the variation in supervisory strictness at origination stemming from a pre-determinedrotationpolicyandsupervisors’personalitytraitsisunlikelytobecorrelatedwith unobservedborrowercharacteristics. Using the SNC data, we identify a strict supervisor at loan origination if the examiner-in- 15Banksupervisionhasexpandedsubstantiallyfollowingtheglobalfinancialcrisisof2007–2008.Forexample,postcrisisreformshaveledtoadditionalsupervisoryprogramsthroughbankstresstesting,morestringentregulatorymonitoringofriskylending,andothermacro-prudentialreforms(IvanovandWang,2022). 18

charge during the loan origination year-quarter is classified as Strict. We use examiners’ history ofassigning“Fail”or“Pass”ratingstodifferentloanfacilitiestodefineagivenexaminerasStrict. Specifically, an examiner is classified as Strict if their total number of assigned Fail ratings to different loans is greater than the sample median. Figure 6 plots the distribution of an examiner’s propensitytofailaloanatagivenpointintime. Wenotemostexaminerstendtofailaround10-15 percent of the loan facilities they are assigned to. We then re-estimate our benchmark regression usingexaminerstrictnessatloanoriginationasaninstrumentforacovenantviolation. We find that the first-stage relationship of strictness at loan origination on covenant violation isquitestrong. Havingastrictsupervisoratoriginationincreasesthelikelihoodofaviolationby 6.9percentagepoints.16 Table 5 reports the main results from our IV estimation. We alternate between district and district × year fixed effects to ensure that our identifying variation does not come from a small subsetofobservationsthatmaynotholdintheaggregate.17 Webeginwith 1(Credit Reduced)as the main outcome of interest. Similar to our benchmark regressions, violations lead to a higher probabilityofreductionofloancommitment. However,theeffectisentirelyoffsetiftheborrower is PE-owned. In fact, summing up the coefficients on Violate and PE×Violate in columns (1) and(2)showslendersraisecommitmentstoPEborrowers. Thiscouldbearesultofrenegotiation betweenPEinvestorsandlendersandupdatedinformationrelatedtoexpectedperformance. The PE×Violateestimatesarealsosignificantwhenweexaminethevolumeofloancommitments. Tofurtherexaminethelimitedpunishmenteffect, weincludeanadditionaloutcomevariable incolumns(5)and(6): thenaturallogarithmofloanmaturityexpressedinthenumberofquarters. We find that creditors substantially lower loan maturity, consistent with findings related to the acceleration of loan repayment upon covenant violation. However, the positive interaction on PE×Violate suggests this effect is substantially mitigated by the presence of PE sponsors. We alsore-estimateaversionofthisanalysiswithoutdistrictordistrict×yearfixedeffectstoexploit alargersampleandfindsimilarresults. ThesefindingsarereportedinTableA3. 16Recallthatintheunconditionalcovenantexamsample,theprobabilityofviolationisaround20percentatagiven pointintime.Thus,ourfirst-stagerelationshipiseconomicallymeaningful. 17WefollowIvanovandWang(2022)andchoosetouseDistrict×YearFEinsteadofDistrict×ReportDateFE. 19

2.3 CovenantHeterogeneityandEnforcementBehavior How does ex-ante covenant design affect ex-post enforcement upon violation? One potential explanationofourbaselineresultisthatitisdrivenbycreditorcoordinationcostsorhighbargaining frictionsduetotheriseofcovenant-liteloans. BeckerandIvashina(2016)showthatasignificant amount of recent corporate loan issuance has been “cov-lite.” They argue that weaker covenant enforcement makes these loans riskier because they lack traditional maintenance covenants such astheleverageratiocovenant. Ontheotherhand,Berlinetal.(2020)arguethatcovenant-liteloans arecloselyassociatedwith‘splitcontrolrights’,whichstillallowlenderstodisciplineborrowersin amannersimilartotraditionalmaintenancecovenants. Thereasonisthatdealswithsplitcontrol rights usually market the covenant-lite tranche to institutional investors, while a relatively small setofbanks-includingtheleadbank-holdsontothecovenant-heavytranche. Thus, lendersin a specific tranche with traditional maintenance covenants retain high bargaining power in loan renegotiationfollowingaviolation. To test if our results are driven only by high bargaining frictions generated by covenantlite loans, we split our sample into “Covenant-Heavy" and “Covenant-Lite" loans. We define a covenant-heavyloanasonethathasatleastoneofthefollowingtraditionalfinancialmaintenance covenants: (i) leverage ratio or senior leverage ratio, (ii) interest coverage ratio, (iii) debt service coverageratio, and(iv)fixedchargecoverageratio. Weclassifyallotherloansas“covenant-lite” (i.e.,loansthatdonothaveanyofthefourmaintenancecovenantsmentioned). Thisclassification effectivelycapturesperformanceandnon-performancecovenantsoutlinedearlierinsection1. We thenseparatelyestimateEq. (2)forcovenant-heavyandcovenant-liteloans. WereporttheseresultsinTable6. Ourspecificationissimilartothebenchmarkregressionand controlsforthesupervisoryriskrating,whichallowsustocompareoutcomeswithininvestmentgrade-rated loans and non-investment-grade-rated loans. Columns (1) and (2) report results on the covenant-heavy sample. We again document that PE sponsorship almost entirely offsets the effects of covenant violations. Importantly, our results are robust to a host of controls and fixed effects, including firm fixed effects. Recall from Figure 1 that syndicates holding loans with performance covenants are relatively more concentrated. This is consistent with split control rights, whichgrantmonitoringandnegotiationtoasmallsubsetoflenders. Wefindsimilarresultswhen 20

we examine the estimates using the covenant-lite sample. While the results from the covenantlite sample could be related to creditor coordination costs, we cannot draw the same conclusion in the covenant-heavy sample due to split control rights. Overall, our interpretation is that the limited-punishmentchannelcannotbefullyexplainedbytheriseofcovenant-liteloans. Asanalternateexercise,werunthefollowingtripleinteractionspecification: Y = β PE +β Violate +β PE ×Violate ×Covlite +Z +X +η +θ +ϵ j,b,i,t 1 i,t 2 j,t 3 i,t j,t j j,b,i,t j,b,i,t b,t z,t j,b,i,t (3) InEq. (3),wetestifourresultsarestrongerforcovenant-liteloans. Weincludealllowerorder terms, such as PE×Violate, but omit them from display for brevity. Table A4 reports the results of this test and shows that the triple interaction is insignificantly different from 0 in all of the specifications.18 2.4 Banks’EquityCapitalandEnforcementBehavior Next, we examine if differences in a bank’s capital constraints shape the patterns documented by our baseline analysis as major lenders in the corporate loan market face increased regulatory scrutiny since the 2008 financial crisis (Irani et al., 2021). For example, Chernenko, Erel, and Prilmeier(2022)showthattwo-thirdsofnonbanklendingfollowingGFCcanbeattributedtobank regulations that constrain banks’ ability to lend to unprofitable and highly leveraged borrowers. ThissectionasksiflimitedpunishmenttowardshighlyleveragedPE-sponsoredfirmsispositively relatedtoagivenbank’scapitalratiobecauseahigherex-antecapitalratioallowsgreaterspaceto takeonadditionalrisk. Analternative—andnotmutuallyexclusive—reasonisthatbankswith higher capital ratios may be more skilled at adjusting their credit exposure to a given borrower uponaccrualofnewinformation. WemergeoursamplewithdatafromFRY-9C,whichcontainsinformationonbank-levelvariables. We measure bank capital as the ratio of bank equity to assets (Tier 1 capital) in the year preceding the year a loan covenant is reviewed by SNC examiners, leading to 28,550 unique loantime observations. Figure 5 shows that the equity-to-assets ratio is concentrated around 9 to 13 18Wealsoverifythatourbenchmarkresultgoesthroughwhenwesplitthesamplebyrevolvingcreditfacilitiesand termloans: i.e.,wedocumentlimited-punishmenteffectwithinrevolvingcreditfacilitiesandwithintermloans. This resultisavailableuponrequest. 21

percent. Wedocumentthatbanksinoursamplehaveamedianequity-to-assetsratioof11percent. We assign a bank to have High Equity Capital if it has an equity-to-assets ratio above the sample medianof11percentanddefineaLowEquityCapitalbanksymmetrically. Wewanttoemphasize that our sample does not cover the 2008 financial crisis and that higher regulatory restrictions since then have led nearly all banks in our sample to be well above their minimum capital requirements. Furthermore, a bank may have to carry more regulatory capital if it is identified as a global systemically important bank or GSIB (see, Favara, Ivanov, and Rezende, 2021). Thus, a “LowEquityCapital”lenderisastrictlyrelativedefinitionforthisparticularexerciseandshould notnecessarilybeinterpretedasasignalofbankhealth. WesplitoursampleandestimateourbenchmarkregressionfromEq. (2)foreachtypeofbank. Columns(1)to(3)ofTable7focusonhigh-capitalbanksandshowqualitativelysimilarresultsto our baseline. The estimate in column (1) implies that high-capital lenders reduce commitments for all borrowers but less so for PE-backed borrowers following a covenant violation. Column (4) presents the first specification for lenders with relatively lower capital and shows a sharply diverging pattern. In particular, while we find these lenders do reduce commitments to all borrowersfollowingviolations,theinsignificantcoefficientontheinteractionterm PE×Violatesuggeststhat,unlikehigh-capitallenders,low-capitallenderswhoseregulatoryconstraintsaremore likelytobind, donotshowanyleniencytowardsPE-backedborrowers. Theinteractiontermbecomes weakly significant at the 10 percent level when we include origination year-quarter fixed effects(finalcolumn). Inourrobustnesstestsinsection4,wealsouseanalternativethresholdfor classifyinglendershavinghighorlowcapitalanddocumentsimilarpatterns. 2.5 RenegotiationandContractualFlexibilityOutsideofDistress Onequestionrelatedtoourlimited-punishmentbenchmarkresultsishowloanrenegotiationoutside of distress affects debt enforcement during distress. Credit agreements are frequently renegotiatedoutsideofdistressordefault(RobertsandSufi,2009b;Roberts,2015). Forexample,loans that are frequently renegotiated favorably outside of distress might also experience limited punishmentafteracovenantviolation. In subsection 1.3, we confirm that all firms, regardless of PE-sponsorship status, frequently 22

renegotiate loan commitments outside of distress. Now, we investigate the role of renegotiation outsideofdistressmoreformallyandestimatethefollowingspecification: Y =β PE +β Violate +β PE ×Violate + j,b,i,t 1 i,t 2 j,b,i,t 3 j,t j,b,i,t β PE ×AmendmentOutside Distress +β AmendmentOutside Distress + 4 i,t j,t 5 j,t Z +X +η +θ +ϵ (4) j,t j,b,i b,t z,t j,b,i,t InEq. (4),thevariable AmendmentOutside Distress andtheinteractiontermwith PE capture the effect of renegotiation and the marginal effect of PE sponsorship on the loan amount. We reporttheseresultsinTableA5. Theindividualterm AmendmentOutside Distressdisplaysmixed results: itispositiveandsignificantinonlytwoofourspecificationsbutisinsignificantincolumns (2)to(5). Interestingly,weobservethecoefficientontheinteractionterm,i.e.,β ,isconsistentlypositive, 4 large, and significant, implying renegotiations involving a PE sponsor lead to more favorable outcomes than renegotiations without a sponsor. Since we are comparing loans of similar risk, one cannot simply interpret the positive coefficient to be driven by the better performance of PE-sponsored firms. One interpretation is that PE sponsors increase the borrower’s bargaining power during loan renegotiations, consistent with evidence from Liu (2021). Consequently, PEbackedborrowershavemorefinancialflexibilityoutsideofdistress. Weprovidefurtherevidence consistent with higher bargaining power in Section 3.2. The estimates on PE×Violate are still positive while that on Violate is negative, suggesting our benchmark results are not necessarily drivenbyrenegotiationsoutsidedistress. 3 Mechanism 3.1 Mechanism1: SponsorReputationandRelationshipRent Malenko and Malenko (2015) and Buccola (2023) argue a sponsor evaluating how it wants its portfolio companies to deal with financial distress is not facing a one-shot game, but is a repeat player. Therefore,alendermightdisplaylimitedpunishmentifahigh-reputationsponsorbacksa firmsincethesponsorhasincentivestopreservefavorableloantermsinfuturebuyoutdealsand 23

is experienced in resolving distress (Bernstein et al., 2019; Johnston-Ross et al., 2021) or provide operational support (Gompers et al., 2022; Block, Jang, Kaplan, and Schulze, 2023). For a lender, expectedgainsfromrepeatedLBOsalsoincludevariousfee-basedincome,suchasunderwriting feesorcommitmentfeeswhichwediscussinsection4. Thus,totheextentthattheexpectedgains from preserving relationship rent with reputed sponsors surpass the cost of enforcing written contracts,lendersarelikelytobemorelenientinenforcingcontractsfollowingcovenantbreaches. We test our hypothesis by constructing two measures of sponsor reputation. First, we constructreputationasafunctionofthemarketshareofthedealvolumeheldbyaPEsponsorinthe U.S.syndicatedloanmarket,consistentwithDemirogluandJames(2010). Werankoursponsors intermsofthetotalnumberofdealsexecutedintheSNCsample. Wethenclassifythetop50sponsors (out of over 600 PE sponsors) as High Reputation sponsors. Cumulatively, these 50 sponsors hold around 63 percent of the market share in terms of deal volume in our sample. As a simple validation exercise, we confirm that more than 70 percent of the top 50 sponsors that appear in our sample have also appeared in the top 50 PE sponsor list in the Private Equity International (PEI) global 300 Private Equity Firm Ranking in 2019 and 2020. Therefore, our measure captures bothafund’sactivityinthesyndicatedloanmarketaswellastheamountofequitycapitalsponsors raised as an indicator of future activity. For confidentiality reasons, we are prevented from disclosingthenamesofuniquesponsorsbackingSNCloans. Second, for robustness, we construct a continuous measure of reputation as the natural logarithmofoneplusthetotalnumberofdealsexecutedbyaPEsponsor,i.e. Log (1 + no.of deals), similar to Badoer et al. (2021). To the extent that reputational capital is increasing in the number ofdeals,weexpecttoseegreaterlenderleniencyforsponsorswithagreaternumberofdeals. Were-estimateourbenchmarkspecificationwherewereplacePE×ViolatewithReputation× Violate. Allothercontrols,including PEandViolate,areasdiscussedbefore. Weestimatethisregressiononbothofourbenchmarkoutcomesofinterest,Log(Commitments),and 1× (CreditReduced). Table 8 reports our results where the interaction is between High Reputation indicator and Violate. We expect the interaction effect to be positive and negate the negative effect of covenant violations on credit commitments. Columns (1) and (2) show a strong positive effect on our interactioneffectofinterest. Whileviolationsleadtosignificantreductionsincommittedcredit,we observe lenders are much more lenient when a borrower is backed by a high-reputation private 24

equity sponsor. We observe qualitatively similar patterns when we look at 1(Credit Reduced) in columns(3)and(4). Notethatour PE indicatorabsorbsstandardPEeffects,allowingustodisi,t entangle the effect of reputation on lender enforcement. We observe similar patterns using the secondreputationmeasure,discussedinsubsection4.6. Finally, for completeness and robustness purposes, we also estimate a triple interaction specification with PE×High Reputation×Violate as the key variable of interest, outlined in Eq. (5). Theseregressionsincludealllower-orderinteractionsthatarenotabsorbedbyfixedeffects,which wedonotdisplayforbrevity. Y =β PE +β Violate +β PE ×Reputation ×Violate + j,b,i,t 1 i,t 2 j,b,i,t 3 i,t j,t j,b,i,t Z +Other interactions+X +η +θ +ϵ (5) j,t j,b,i b,t z,t j,b,i,t We report these results in Table A6 of the Online Appendix. We find qualitatively similar results. Thepointestimatesthemselvessuggestwhenwelookathigh-reputationsponsors,lenders do not display any punishment at all in the sense that the entire negative effect of violation is negated(theunreportedinteractionbetween PEandViolateisinsignificantinthisspecification). 3.2 Mechanism2: LoanRenegotiationandBargainingPower In this section, we propose a related mechanism behind our results: we examine if limited punishmentcouldalsobeexplainedbythehigherbargainingpowerofPEsponsorsduringloanrenegotiations. For example, Liu (2021) finds evidence of superior bargaining power of PE-owned hospitalsvis-a-visinsurers,whileBernsteinetal.(2019)providesurveyevidencethatsponsorsdirectlyhelpwithloanrenegotiationwithbankersandlawyerswhenportfoliofirmsareindistress. We show evidence of superior bargaining power in two ways. First, we exploit the fact that a covenant violation immediately gives the lender more bargaining power by design. Moreover, thisbargainingpowerislikelytobepreservedwhenagivensyndicateishighlyconcentrated(i.e., it has relatively few lenders, especially non-bank lenders such as CLOs and hedge funds). Giannetti and Meisenzahl (2022) define concentrated syndicates in a similar way. If our benchmark results go through when the syndicate is concentrated, it implies PE sponsors dampened creditor enforcement even when lenders have high bargaining power. Since we can control for loan 25

characteristics,ourinterpretationisthatsponsorsraisetheportfoliocompany’sbargainingpower (relativetodistressednon-PEborrowers)whenrenegotiatingwithconcentratedsyndicates. For the purposes of this particular analysis, recall from Table 1 that our sample contains an adequate number of loans with a relatively small number of institutional lenders (likely due to splitcontrolrights). 19 Wedefineanewtime-varyingvariableConcentrated,whichtakesthevalue of1ifthetotalnumberofinstitutionallendersinagivensyndicateatagivenpointintimeisless than or equal to the median number of institutional lenders in the full sample (i.e., 11). We also verify that our results are nearly identical when we use the 25th percentile of the total number of institutional lenders (which is 6 in our sample). We estimate the following triple-differences specification. Y =β PE +β Violate +β Concentrated ×Violate + j,b,i,t 1 i,t 2 j,b,i,t 3 j,t j,b,i,t β PE ×Concentrated ×Violate +β Concentrated+ 4 i,t j,t j,b,i,t 5 Z +Other interactions+X +η +θ +ϵ (6) j,t j,b,i b,t z,t j,b,i,t We report these results in Panel A of Table 9. First, focusing on columns (1) and (2), we seethatViolate×Concentrated isnegative,implyingthat,uponcovenantviolation,concentrated syndicates reduce commitments more than dispersed syndicates. However, the positive sign on PE×Concentrated×Violate in columns (1) and (2) implies that PE-sponsorship status dampens the credit commitment reduction. We interpret these findings as successful renegotiation of loan contracts by sponsors (relative to non-PE borrowers), indicating higher bargaining power of PEsponsoredfirmsrelativetonon-PEborrowers. Wenote,however,thattheextensivemargineffect, ascapturedby 1(Credit Reduced),isinsignificant,implyingthebargainingpowereffectofPEon creditreductionismorepronouncedattheintensivemargin. Second,agivenlendercouldbeheavilyreliantonagivensponsorfordealflow,whichwould raisethesponsor’sbargainingpowerduringloanrenegotiation. Weproxyasponsor’sbargaining power vis-a-vis lenders by aggregating the dollar value of all outstanding LBO loans between a given lender (lead bank) and a given sponsor’s portfolio companies at time t. Our expectation is thatthehigheragivenbank’sexposuretoagivensponsor’sLBOactivity(capturingthelender’s 19Berlinetal.(2020)showthenumberoflendersinloanswithsplitcontrolrightsismuchlessthanthosewithoutit. 26

historical reliance on a given sponsor for deal flow), the higher the sponsor’s bargaining power vis-a-visthebank. Specifically,weuseavariableintroducedearlier: TotalPEsponsor-bankexposure. Thismeasurecapturesabank’stotalutilizedloancommitmentsbyallportfoliocompaniesbacked byagivenPEsponsor. Basedonthismeasure,wethendefineanindicatorvariableHighExposure thattakesthevalueofoneifthesponsor-leadbankexposureattimetisequaltoorgreaterthanthe samplemedianand0otherwise. AsreportedinTable1,themediantotalsponsor-bankexposure amountisUSD2.25B.Thus,HighExposuretakesavalueof1ifthelender’stotalLBOexposurefor all companies backed by a specific PE sponsor (e.g., KKR) is greater than or equal to USD 2.25B. Since the main variable of interest is only available for PE-backed loans by definition, our test is restricted to the PE sample only. We estimate a variant of our benchmark regression for only the PE-sample, where the key variables of interest are Violate and the interaction High Exposure× Violate. Panel B of Table 9 reports these results. For both of our outcome variables, we see covenant violationsleadtoreductionsincreditcommitmentasbefore. However, thiseffectissignificantly dampened if a given lender was heavily reliant on a given sponsor for continued deal flow. This resultisagainconsistentwithasponsor’sbargaingpowerdampeningcreditorenforcement. 3.3 LoanPerformancePost-Violation: DowngradesandDefaults Animportantquestionrelatedtoboththesponsor’sreputationandhighbargainingpowermechanisms is whether PE-backed loans fail more often relative to non-PE loans that also violate covenants. In this section, we examine loan performance conditional on covenant violation. We estimateEq. (2)withloanperformanceasthedependentvariable. Wemeasureloanperformance using (i) loan downgrades and (ii) realized defaults. We want to emphasize that we do not arguePE-sponsoredloansoutperformnon-PEloansunconditionally,butratherthatrealizeddowngrades and defaults are not meaningfully higher in PE relative to non-PE, conditional on covenant violations. Loan Downgrades: We measure loan performance at both the extensive and the intensive margin. First, we construct an indicator that takes the value of 1 if a loan is classified as Special Mention, Substandard, Doubtful, or Loss, and 0 otherwise. Second, we compute the natural log- 27

arithmof1plusthetotaldollaramountofthecredit’scommittedexposurewherethefinalexam ratingisSpecialMention,Substandard,Doubtful,orLoss. WereportourfirstsetoftestsinTable10,focusingoncolumns(1)to(3). Notsurprisingly,the coefficientonViolateispositivesincecovenantviolationsandtheprobabilityofloandowngrades are positively correlated. We also see that PE is positively related to the probability of being downgraded. This supports the view that PE-sponsored loans are riskier, as documented earlier through higher covenant violation rates. Interestingly, when we examine the interaction effect, we find it is not statistically significant across any of our specifications. One explanation is that PE sponsors enhance operational support (Gompers et al., 2022) and distress resolution-related activities (Hotchkiss, Smith, and Strömberg, 2021) when a firm is in distress in order to preserve reputation and bargaining power with lenders. We find a qualitatively similar result when we lookattheintensivemarginusingthedollarvolumeofloansthathavebeenclassifiedasdoubtful, specialmention,substandard,orloss. Loan Defaults: Finally, we also examine default rates. Following Giannetti and Meisenzahl (2022),weusetheinformationonthenumberofdaysthatanypayment(interestorprinciple)for agivenloanispastdue. Specifically,wedefineanindicatorvariable Defaultthattakesthevalue of 1 if a loan is past due for 60 days or more and 0 otherwise. We then estimate our benchmark specificationtoexamineifthedefaultrateisdifferentfollowingcovenantbreachforPE-sponsored loans. WereportresultsonthedefaultrateinTable11. Similartothepreviousresult,columns(1)and (2)showthatcovenantviolationsarepositivelyrelatedtotherealizeddefaultrate. However,PE× Violateisnegativelycorrelatedwiththedefaultrate,indicatingthatPE-specificmechanismsmay reduce the likelihood of defaults upon covenant violation. Again, these findings are consistent withPEsponsorseitherresolvingdistressmoreefficientlyasarguedinBernsteinetal.(2019)and Hotchkiss et al. (2021), or through operational expertise in distress (Gompers et al., 2022). We want to emphasize that we do not claim that PE-sponsored loans are unconditionally associated with lower default probability. Rather, conditional on covenant violations, which can be broadly interpreted as early phases of distress, PE-sponsored loans are less likely to eventually default comparedtocomparablenon-PEloansthatalsoviolatecovenants. 28

4 Additional Results and Robustness 4.1 EquityInfusion Amechanismrelatedtobothsponsorreputationandhigherbargainingpoweristheabilitytoinjectequityindistress. Forexample,higherreputedsponsorslikelyhavehigherabilitytoinjectequitygiventhattheyobtainmorecapitalfromlimitedpartners. Priorresearchhasshownsponsors arelikelytoinjectequitytohelptheirportfoliocompaniesovercomeliquidityproblems(Bernstein etal.,2019;Hotchkissetal.,2021). Toidentifyevidenceofequityinjection,weagainreadthrough the SNC loan covenant schedule. Bank examiners provide detailed descriptions of a borrower’s actionstoensurecovenantcomplianceandcrucially,whatcorrectiveactionsweretakentocurea covenantviolation. Inthesedescriptions,examinersexplicitlymentioniftheborrowerreceivedan equityinjectiontocuretheviolation(orundertookothercorrectiveactions,suchascost-cutting). An example of a typical description is outlined below, with identifying information removed for confidentialitypurposes. A covenant default occurred on [Date]. The default occurred because the leverage ratio of X exceeded the covenant limit. [Lender] issued a default letter [Date]. Company X injected [Dollarvalue]inequitytocurethedefaultandtook[OtherActions]. Thecombinationofthese actionsproducedanadjustedEBITDAof[Dollarvalue],effectivelycuringthedefault. Wethenuseasimpletext-searchalgorithmtoidentifyinstancesofequityinjectionusingwords such as “Injected” or “Infused” and their variants. After manually verifying the accuracy of our algorithm, we create an indicator variable 1(Capital Injection) that takes the value of 1 if a loan is identified to have received an equity injection at a given point in time and 0 otherwise. In our full baseline sample, we identify around 1,700 loan-time observations with an equity injection at a given time, which is around 4 percent of the sample. Recall from Figure 3 and Figure 4 that the average rate of violation in the full covenant sample is around 20 percent. The relatively lowerfrequencyofequityinfusionsimpliesitisunlikelytobetheonlymechanisminexplaining enforcementbehavior. WeinvestigatethesedescriptivefindingsthroughaformaltesttoassesswhetherPE-sponsored firms are associated with more instances of equity injections upon violation. We re-estimate Eq. 29

(2) with 1(Capital Injection) as an outcome variable to do so. Our key variables of interest are again Violate and PE×Violate. These results are reported in Panel A of Table 12. First, across most of our specifications, we see that Violate is positively related to equity injections. However, thequantitativeeffectisrelativelylow. Specification(2)alsoshowsthatPE-backedfirmsaremore likelytoinjectequity. However, thisestimateisalsoquantitativelysmallandisnotrobustacross otherspecifications. Ourinterpretationisthatwhileequityinjectionisindeedamechanismfirm, especially a PE-backed one, undertakes to cure covenant violations, it is not the dominant mechanism at play. Consistent with this view, Gompers et al. (2022) showed PE managers provided more operational support (e.g., providing strategic guidance, reducing costs, or connecting companies with potential customers, suppliers, or strategic partners) compared to equity injection, whenfirmswereindistressduringtheCOVID-19pandemic. Weacknowledge,however,thatone limitation of our measure of equity injection, 1(Capital Injection), is that it cannot capture the intensivemargineffect,whichcouldalsobesystematicallydifferentforPE. 4.2 OtherOutcomes: LoanSpreadandMaturity DoPE-backedfirmspayhigherspreadsorfaceshorterloanmaturitiesasatrade-offforretaining higheraccesstocredit? Wenowestimateourbenchmarkequationonloanspreadsandmaturity. Table 13 reports these results. Data on loan spreads is available for a smaller set of loans in the SNC, leading to a much smaller estimation sample in columns (1) to (4). Spreads are defined in basis points over LIBOR. We generally find that covenant violations lead to higher loan spreads: theeffectisquitelargerangingfrom34toasmuchas50basispointsdependingonourcontrols. Forexample,incolumn(2),whenweincludesector-time,origination-time,andvariousloancontrols, the estimate on Violate is significant at the 1 percent level and stands at 38.32. However, consistent with limited punishment towards PE, we see PE×Violate is -28.94, implying that the spreadincreaseismuchlessforPE-sponsoredloans. Whilecolumns(1)and(2)showevidenceof limited punishment, we do not find the same results when we include further fixed effects such as bank-time or firm fixed effects. Overall, we do not see any specification where PE×Violate is positiveandsignificant. Recall from subsection 2.2, that we detected limited-punishment in loan maturity reduction 30

using our instrumental variable analysis. We now also run regressions similar to our benchmark regression (i.e., without an instrument)on loan maturity. Wereport these incolumns (5) to(8) of Table13. Acrossallspecifications,Violateisnegativelyrelatedtoloanmaturity,suggestinglenders generally reduce loan maturity upon covenant violation. In columns (5) and (7) we see that the effect is mitigated by PE-backing since the interaction terms are positive. However, we find that thisresultisnotrobusttotheinclusionoffirmfixedeffectsororiginationyear-quarterfixedeffects. While we cannot conclusively draw the conclusion that PE leads to limited punishment in terms of loan maturity from this result, we again fail to detect evidence that lenders are substantially shortening maturities for PE-backed loans upon covenant violation (which would have required PE×Violatetobesignificantandnegative). Our overall interpretation is that there is no significant evidence that lenders are extracting surplus through harsher contractual terms in exchange for a lower reduction in credit commitment, and there is suggestive evidence that limited punishment towards PE also exists when we lookatspreadsormaturity. 4.3 Matched-SampleAnalysisusingfirm-levelvariables One concern is that our loan-level sample does not explicitly control for time-varying firm-level risk factors (e.g., debt ratio) on which lenders could condition their decisions, even though we includesupervisoryriskratings. Tomitigatethisconcern, wenowmatchourSNCsampletothe FederalReserve’sFRY14-Qdataoncommercialloanswhichcontainsdetailedfirm-levelbalance sheet information and has been used extensively in prior studies (e.g. Brown, Gustafson, and Ivanov(2021);Chodorow-Reich,Darmouni,Luck,andPlosser(2022)). TheFRY-14Qdataconsists of information on all loan facilities with over USD 1 million in the committed amount held by Bank Holding Companies (BHCs) in the U.S. and began in 2012 to support the Dodd-Frank Act Stress Tests (DFAST). The key advantage of the FR Y-14Q is the extensive coverage of private firmsthatborrowfromU.S.banks,alongwithinformationontheirbalancesheetsandaccounting statements. For example, Caglio, Darst, and Kalemli-Özcan (2021) find more than 90 percent of firmsintheY-14dataareprivate. Giventhedifferencesinfirmnamingconventionsandbanksthatarerequiredtoreportinfor- 31

mationtotheY-14relativetoSNC,weareabletomergearound50percentofourbaselinesample of PE and non-PE firms. Using this merged sample, we now construct the control group (consisting of non-PE-backed firms) to match PE-backed firms on observable characteristics. Specifically, for all PE-backed firms in our data, we select at most five non-PE-backed firms in the FR Y-14Q sample in the pre-buyout year that (i) belong to the same two-digit NAICS code and have (ii) EBITDA, (iii) book assets, (iv) leverage ratio (debt/assets), and (v) 1-year ahead probability of default within a 20 percent bracket around corresponding value for the PE-backed firm. The matching variables and general methodology broadly follow Bernstein et al. (2019), Boucly et al. (2011) and Haque et al. (2022). The only difference is that we also match the 1-year ahead probability of default as estimated by the reporting bank, which we believe is particularly relevant to ourresearchquestion. Crucially,defaultprobabilityestimatescapturemarketinformationthatis otherwiseunavailableforprivatefirms. Wealsoincludefirm-levelcontrolvariables. TableA7reportssummarystatisticsforthefullmergedSNC-Y14QsampleinPanelA.Asthe tableshows,PE-backedfirmshavehigherdebtontheirbooksandahigherprobabilityofdefault. Importantly,thedebtratioof52percentinPEisconsistentwithpriorstudiessuchasBrown(2021) orGornalletal.(2021). BothfirmtypesaresimilarintermsofsizeandEBITDA.PanelBrestricts the summary stats (means) to the sample of observations where PE and non-PE are matched as describedabove. Table14PanelAre-estimatesourbenchmarkregressionsattheloanlevelwherenowthenon- PE loans are matched according to the methodology described above. Further, instead of supervisory risk-rating, all regressions now include the following firm-level controls: Debt/Asset ratio, EBITDA/Assets ratio, Log (Total Assets) and an indicator variable, 1∗(Public), which controls for whether a firm is publicly-traded in a given year. We observe that our key result on limitedpunishment is robust to the matched control group as can be seen in columns (1) to (3). One caveatisthatoursamplesizedecreasessignificantlyduetothemergewithFR-Y14andthematchingexercise. Toalleviateconcernsrelatedtoasmallersamplesize,were-estimateourbenchmark regressions with the same firm-level controls as above, but without any matching in Panel B. We obtainamuchlargersamplewithoutmatching. Weagainseeourbenchmarkresultisunchanged incolumns(1)and(2)wheretheoutcomeisLog(Commitments). For 1×(CreditReduced)theresult issignificantinourmoststringentspecificationincolumn(4)butisinsignificantincolumn(3)in 32

PanelB.Overall,theseresultshighlighttherobustnessofourfindingstoamatchingmethodology thatisbasedonpriorstudies. 4.4 AlternativeDefinitionofaCovenantBreach Our benchmark definition of covenant violations includes both violations reported by the lender andwaivers. Inthissection,wedepartfromthisdefinitionandexcludecovenantwaiversorresets as a type of covenant violation, and re-estimate our benchmark results in Table 4. This leads to a much lower number of violations. However, we find that our results remain unchanged using bothofourmainoutcomevariables(i.e.,Log(Commitments)and 1×(Credit Reduced)). Wereport theseresultsinTableA8. 4.5 HighEquityCapitalBanks: AlternativeThreshold InTableA9oftheOnlineAppendix,werepeatourexerciseinsection4.3withalternatethresholds classifying a bank as having high or low capital. Specifically, High Equity Capital bank is now definedasbeinginthetopquartileofthesampledistributioninsteadofthemedian. Wefindour resultsarenearlyunchanged. 20 4.6 AlternativeMeasureofReputation Werepeatourtestsonreputationusingacontinuousmeasureofreputationasoutlinedintheprevious section. Table A10 reports the results where the interaction is between Violate and the naturallogarithmofoneplusthenumberofdeals. Sincethismeasureincludesfundswithrelatively fewerdeals,weexpectthelender-leniencyeffecttobesmallerbutsignificantnonetheless. Consistent with our hypothesis, the interaction term is positive when the outcome is Log(Commitments) in columns (1) and (2) and negative for 1×(Credit Reduced) in columns (3) and (4). While the estimateonViolateisquitesimilarinmagnitudecomparedtothoseinTable8,weseethattheinteractioneffect,whilehighlysignificantandpositive,issmallerinmagnitudeinTable8. Theeffect is particularly pronounced when we look at 1×(Credit Reduced). Compared to Table 8, where 20Wealsousethetotalrisk-basedcapitalratioasanalternatemeasureoflendercapitaltochecktherobustnessofour resultsinTable7.ThisisdefinedasthesumofTier1andTier2capital(plusTier3capitalwhereapplicable)overtotal risk-weightedassets. Wefindsimilarresultsusingthesamplemediantoclassifylenderswithhigh-orlow-risk-based capital.Theseresultsareavailableuponrequest. 33

weobservedthattheentirecovenantviolationeffectismitigatedbyhigh-reputationsponsors,we see the limited-punishment effect is much smaller since this measure captures sponsors that are belowthetop50or100rankedfundsinoursample. 4.7 Fee-basedIncome Therepeated-dealsargumentstemsfromtheideathathigh-reputationsponsorsaremoreskilled inresolvingdistress(Hotchkissetal.,2021;Block,Jang,Kaplan,andSchulze,2022)andgenerating value (Gompers et al., 2022). Additionally, lenders might expect repeated deals to generate a highervolumeoffee-basedincomewiththesamesponsorinfutureLBOsasitiswell-knownthat banks earn various types of fees during the loan syndication process (Blickle et al., 2020; Bruche, Malherbe,andMeisenzahl,2020). Followingthisargument,ifsponsoredloansareassociatedwith higher fee-based income at the intensive margin, lenders are even more likely to display limited punishmentinexpectationofhigherfutureincome. We obtain data from DealScan and provide suggestive evidence that this is indeed the case: PE-sponsored loans are associated with higher upfront fees (which include underwriting fees) and commitment fees. Figure A1 in the Online Appendix plots commitment fees and upfront feesforPEandnon-PEloansbetweentheperiodof2012and2021, measuredinbasispoints. We observe both fee types are higher for sponsored loans, and the effect is particularly pronounced for commitment fees. We also estimate a simple regression that confirms a positive correlation between PE-sponsorship and fee-based income using DealScan data only, as reported in the Online Appendix Table A11. This positive correlation is potentially related to PE deals being more complex relative to non-PE deals, thus requiring higher fee rates. Higher fees in turn are consistent with limited-punishment or greater continuation lending when sponsored-borrowers are in distress. 5 Conclusion ThispaperexamineshowPEsponsorsshapetheenforcementofdebtcontractsinthesyndicated loan market, using covenant violations as an empirical setting. By combining supervisory data from the Shared National Credit Program with LBO information from Preqin, we build a novel 34

loan-level dataset of PE-sponsored borrowers, their covenants, covenant compliance, and postviolationoutcomes. WefindthatPE-backedborrowersviolatecovenantsmoreoftenthannon-PEbacked borrowers. Yet, lenders do not reduce the stock of available credit to PE-backed borrowers as much as they do when non-PE firms violate covenants. This limited-punishment effect is presentinbothcovenant-heavyandcovenant-liteloansbutisstrongerforwell-capitalizedbanks. Wealsofindsimilarpatternswhenwelookatotherloantermssuchasmaturityandinterestrate spread,althoughourresultsarestrongestforloancommitments. Weshowthatourresultisdriven by two related mechanisms: (i) a repeated-deals mechanism as lenders and sponsors frequently interact in credit markets, which incentivizes lenders to preserve relationship rent, and (ii) the high bargaining power of PE sponsors in renegotiating loan contracts, particularly when a given lenderisheavilyreliantonasponsorfordealflow. Ourresultsareconsistentwithrecentdiscussions by Buccola (2023) who argue that equity sponsors rather than senior lenders have practical control over the way that distressed companies respond to their financial problems. Further, we provide novel descriptive facts about loan contracts in buyouts, such as amendments outside of distressanddifferencesinnon-bankparticipationinPE-sponsoredvsnon-PEloans. Ourdetailedloan-leveldatabaseallowsustoovercomestandardendogeneityconcernsrelated to covenant violations. In particular, our baseline research design compares credit outcomes followingcovenantviolationsforreasonablycomparableloanswithsimilarcreditriskissuedbythe same bank to borrowers in the same sector who differ only by PE-sponsorship status. We also exploit bank examiner personality traits in an instrumental variable setting, where the excluded instrument is the strictness of the bank supervisor at the time of loan origination. Finally, to further mitigate endogeneity concerns, we deploy a matching methodology following prior studies andshowthatourresultsholdwhenwematchPEtonon-PEloansbasedondebt,assets,EBITDA, anddefaultprobability. Overall, we uncover a novel mechanism that affects lenders’ enforcement behavior following acontractualbreach. Ourfindingssuggestsponsorsdampentheeffectofstate-contingentallocation of creditor control rights, allowing PE-backed firms more financial flexibility in distress and thereforedirectlyaffectingthecorporatedebtpolicyofalargeshareofborrowersinthesyndicated loanmarket. 35

References Achleitner,A.-K.,R.Braun,B.Hinterramskogler,andF.Tappeiner(2012). Structureanddeterminantsoffinancialcovenantsinleveragedbuyouts. ReviewofFinance16(3),647–684. Agarwal,S.,D.Lucca,A.Seru,andF.Trebbi(2014). Inconsistentregulators: Evidencefrombanking. TheQuarterlyJournalofEconomics129(2),889–938. Axelson,U.,T.Jenkinson,P.Strömberg,andM.S.Weisbach(2013). Borrowcheap,buyhigh? The determinantsofleverageandpricinginbuyouts. TheJournalofFinance68(6),2223–2267. Badoer, D. C., M. Emin, and C. M. James (2021). Contracting costs and reputational contracts. AvailableatSSRN3536458. Barber, B.M.andA.Yasuda(2017). Interimfundperformanceandfundraisinginprivateequity. JournalofFinancialEconomics124(1),172–194. Becher, D. A., T. P. Griffin, and G. Nini (2022). Creditor control of corporate acquisitions. The ReviewofFinancialStudies35(4),1897–1932. Becker, B. and V. Ivashina (2016). Covenant-light contracts and creditor coordination. Riksbank ResearchPaperSeries(149),17–1. Berlin, M., G. Nini, and G. Y. Edison (2020). Concentration of control rights in leveraged loan syndicates. JournalofFinancialEconomics137(1),249–271. Bernstein, S., J.Lerner, andF.Mezzanotti(2019). Privateequityandfinancialfragilityduringthe crisis. TheReviewofFinancialStudies32(4),1309–1373. Blickle, K., Q. Fleckenstein, S. Hillenbrand, and A. Saunders (2020). The myth of the lead arranger’sshare. FRBofNewYorkStaffReport(922). Block,J.,Y.S.Jang,S.N.Kaplan,andA.Schulze(2023). Asurveyofprivatedebtfunds. Technical report,NationalBureauofEconomicResearch. Block, J. H., Y. S. Jang, S. N. Kaplan, and A. Schulze (2022). A survey of private debt funds. AvailableatSSRN. Boucly, Q., D. Sraer, and D. Thesmar (2011). Growth LBOs. Journal of Financial Economics 102(2), 432–453. Bräuning,F.,V.Ivashina,andA.Ozdagli(2022). High-yielddebtcovenantsandtheirrealeffects. NationalBureauofEconomicResearch. 36

Brown, G. (2021). Debt and leverage in private equity: A survey of existing results and new findings. InstituteforPrivateCapital,WorkingPaper,RetrievedfromtheUniversityofNorthCarolina atChapelHill,InstituteforPrivateCapital. Brown,J.R.,M.T.Gustafson,andI.T.Ivanov(2021). Weatheringcashflowshocks. TheJournalof Finance76(4),1731–1772. Bruche,M.,F.Malherbe,andR.R.Meisenzahl(2020). Pipelineriskinleveragedloansyndication. TheReviewofFinancialStudies33(12),5660–5705. Buccola, V. S. (2023). Sponsor control: A new paradigm for corporate reorganization. U. Chi. L. Rev.90,1. Caglio, C.R., R.M.Darst, andKalemli-Özcan(2021). Risk-takingandmonetarypolicytransmission: EvidencefromloanstoSMEsandlargefirms. NationalBureauofEconomicResearch. Carey, M. and M. B. Gordy (2021). The bank as grim reaper: Debt composition and bankruptcy thresholds. JournalofFinancialEconomics142(3),1092–1108. Chava, S. and M. R. Roberts (2008). How does financing impact investment? The role of debt covenants. TheJournalofFinance63(5),2085–2121. Chernenko,S.,I.Erel,andR.Prilmeier(2022). Whydofirmsborrowdirectlyfromnonbanks? The ReviewofFinancialStudies35(11),4902–4947. Chodorow-Reich, G. (2014). The employment effects of credit market disruptions: Firm-level evidencefromthe2008–9financialcrisis. TheQuarterlyJournalofEconomics129(1),1–59. Chodorow-Reich,G.,O.Darmouni,S.Luck,andM.Plosser(2022).Bankliquidityprovisionacross thefirmsizedistribution. JournalofFinancialEconomics144(3),908–932. Chodorow-Reich,G.andA.Falato(2022). Theloancovenantchannel: Howbankhealthtransmits totherealeconomy. TheJournalofFinance77(1),85–128. Christensen, H. B. and V. V. Nikolaev (2012). Capital versus performance covenants in debt contracts. JournalofAccountingResearch50(1),75–116. Cohen,G.J.,J.Dice,M.Friedrichs,K.Gupta,W.Hayes,I.Kitschelt,S.J.Lee,W.B.Marsh,N.Mislang, M.Shaton, M.Sicilian, andC.Webster(2021). TheUSsyndicatedloanmarket: Matching data. JournalofFinancialResearch44(4),695–723. Davis,S.J.,J.Haltiwanger,K.Handley,R.Jarmin,J.Lerner,andJ.Miranda(2014). Privateequity, jobs,andproductivity. AmericanEconomicReview104(12),3956–3990. Davis,S.J.,J.Haltiwanger,K.Handley,B.Lipsius,J.Lerner,andJ.Miranda(2021). Theeconomic effectsofprivateequitybuyouts. AvailableatSSRN3465723. 37

Demiroglu,C.andC.M.James(2010). TheroleofprivateequitygroupreputationinLBOfinancing. JournalofFinancialEconomics96(2),306–330. Denis, D. J. and J. Wang (2014). Debt covenant renegotiations and creditor control rights. Journal ofFinancialEconomics113(3),348–367. Diamond, D. W. (1991). Monitoring and reputation: The choice between bank loans and directly placeddebt. JournalofPoliticalEconomy99(4),689–721. Falato,A.andN.Liang(2016). Docreditorrightsincreaseemploymentrisk? Evidencefromloan covenants. TheJournalofFinance71(6),2545–2590. Favara, G., I. Ivanov, and M. Rezende (2021). Gsib surcharges and bank lending: Evidence from uscorporateloandata. JournalofFinancialEconomics142(3),1426–1443. Favara, G., E. Morellec, E. Schroth, and P. Valta (2017). Debt enforcement, investment, and risk takingacrosscountries. JournalofFinancialEconomics123(1),22–41. Fracassi, C., A. Previtero, and A. Sheen (2022). Barbarians at the store? Private equity, products, andconsumers. TheJournalofFinance77(3),1439–1488. Giannetti,M.andR.Meisenzahl(2022). Ownershipconcentrationandperformanceofdeterioratingsyndicatedloans. SwedishHouseofFinanceResearchPaper(21-16). Gompers,P.A.,S.N.Kaplan,andV.Mukharlyamov(2022). PrivateequityandCovid-19. Journal ofFinancialIntermediation51,100968. Gornall, W., O. Gredil, S. T. Howell, X. Liu, and J. Sockin (2021). Do employees cheer for private equity? the heterogeneous effects of buyouts on job quality. Working Paper (December 24, 2021). Gryglewicz,S.andS.Mayer(2022). Dynamiccontractingwithintermediation: Operational,governance,andfinancialengineering. JournalofFinance,Forthcoming. Gustafson, M. T., I. T. Ivanov, and R. R. Meisenzahl (2021). Bank monitoring: Evidence from syndicatedloans. JournalofFinancialEconomics139(2),452–477. Haque, S., Y. S. Jang, and S. Mayer (2022). Private equity and corporate borrowing constraints: Evidencefromloanleveldata. AvailableatSSRN4294228. Hirtle,B.,A.Kovner,andM.Plosser(2020). Theimpactofsupervisiononbankperformance. The JournalofFinance75(5),2765–2808. Hotchkiss,E.S.,D.C.Smith,andP.Strömberg(2021).Privateequityandtheresolutionoffinancial distress. TheReviewofCorporateFinanceStudies10(4),694–747. 38

Irani, R. M., R. Iyer, R. R. Meisenzahl, and J.-L. Peydro (2021). The rise of shadow banking: Evidencefromcapitalregulation. TheReviewofFinancialStudies34(5),2181–2235. Ivanov, I. and J. Wang (2022). Bank supervision, corporate credit supply, and bank monitoring. ManagementScience,Forthcoming. Ivashina, V. and A. Kovner (2011). The private equity advantage: Leveraged buyout firms and relationshipbanking. TheReviewofFinancialStudies24(7),2462–2498. Johnston-Ross,E.,S.Ma,andM.Puri(2021). Privateequityandfinancialstability: evidencefrom failedbankresolutioninthecrisis. Technicalreport,NationalBureauofEconomicResearch. Kaplan,S.N.andP.Stromberg(2009). Leveragedbuyoutsandprivateequity. JournalofEconomic Perspectives23(1),121–46. Kiyotaki,N.andJ.Moore(1997). Creditcycles. JournalofPoliticalEconomy105(2),211–248. Lian, C. and Y. Ma (2021). Anatomy of corporate borrowing constraints. The Quarterly Journal of Economics136(1),229–291. Liu,T.(2021). Bargainingwithprivateequity: Implicationsforhospitalpricesandpatientwelfare. AvailableatSSRN3896410. Malenko, A. and N. Malenko (2015). A theory of LBO activity based on repeated debt-equity conflicts. JournalofFinancialEconomics117(3),607–627. Nini,G.,D.C.Smith,andA.Sufi(2009).Creditorcontrolrightsandfirminvestmentpolicy.Journal ofFinancialEconomics92(3),400–420. Nini, G., D. C. Smith, and A. Sufi (2012). Creditor control rights, corporate governance, and firm value. TheReviewofFinancialStudies25(6),1713–1761. Roberts,M.R.(2015). Theroleofdynamicrenegotiationandasymmetricinformationinfinancial contracting. JournalofFinancialEconomics116(1),61–81. Roberts,M.R.andA.Sufi(2009a).Controlrightsandcapitalstructure: Anempiricalinvestigation. TheJournalofFinance64(4),1657–1695. Roberts, M. R. and A. Sufi (2009b). Renegotiation of financial contracts: Evidence from private creditagreements. JournalofFinancialEconomics93(2),159–184. Shive, S. and M. Forster (2022). Sponsor reputation and capital structure dynamics in leveraged buyouts. AvailableatSSRN3781879. Smith Jr, C. W. and J. B. Warner (1979). On financial contracting: An analysis of bond covenants. JournalofFinancialEconomics7(2),117–161. 39

A Variable definitions 40

Variable Definition Source DependentVariables CapitalInjection Theindicatorvariablethattakesthevalueof1ifaloanreceivesan SNC equityinfusionand0otherwise. CreditReduced Theindicatorvariablethattakesthevalueofoneiftotalcommitted SNC creditsbetweenagivenbank-firmpairarereducedinagiventime periodrelativetothepriorperiod. CommitmentFee Afeepaidtolendersonundrawnamountsunderarevolvingcredit DealScan oratermloanpriortodraw-down. Thisfeeisusuallyreferredtoas a“ticking”feeontermloans. (DaysPastDue>=60) Anindicatorvariablewhichtakesthevalueof1ifaloanpaymentis SNC pastduefor60daysormore. LoanSpread SpreadoverLIBORexpressedinbasispoints.Obtainedusingtextual SNCandauthors’calculaanalysis from SNC Credit View from variables related to payment tions scheduleandrepaymenttermsdescription. Log(Commitments) Thenaturallogarithmofthecommitmentamountofagivencredit SNC facility. Log(LoanMaturity) Thenaturallogarithmoftheloanmaturitywhichismeasuredasthe SNC differencebetweentheoriginationdateandmaturitydate. Log(1+Non-PassAmount) Thenaturallogarithmofthetotaldollaramountofacredit’scom- SNC mitted exposure where the final exam rating is a Special Mention, Substandard,Doubtful,orLoss. Substandard/Doubtful Indicatorvariablethattakesthevalueof1ifaloanisclassifiedas SNC substandardorisunderspecialmentionand0otherwise. UpfrontFee Afeepaidbytheissueratthetimeofdealclosure. Itisoftentiered, DealScan withtheleadarrangerreceivingalargeramountinconsiderationfor structuringand/orunderwritingtheloan. ControlVariables AmendmentOutsideDistress Isanindicatorvariablethattakesthevalueof1ifaloancommitment SNCandauthors’calculaamountischangedinperiodtrelativetoperiodt−1outsideofour tions definitionofcovenantviolation. ConcordanceRating Anumericalriskratingthatfederalsupervisorsassigntoeachcredit SNC facility at a given point in time. Lower ratings denote lower risk in a particular credit facility. We use the variable Adjusted ConcordanceRating,whichisbasedontheReportedConcordanceRatingbut isupdatedtoreflectmissingorinvalidratinginformationreported bytheAgentBank. Aratingof1isInvestmentGradePass,2isNon- Investment-Gradepass,3isLowestRatedPass,4isSpecialMention,5is Substandard,6isDoubtfuland7isLoss. Weinterchangeablyusethe terms“SupervisoryRiskRatings”and“ConcordanceRating”.Foran exampleofhowloanqualityismappedfromtheagentbank’sinternalratingtosupervisoryrating,seethisreportingformbytheSNC office. Covenant-LiteLoan A loan that has none of the following traditional financial mainte- SNCandauthors’calculanance covenants: (i) leverage ratio or senior leverage ratio, (ii) in- tions terestcoverageratio,(iii)debtservicecoverageratio,and(iv)fixed chargecoverageratio. Covenant-heavyloansaredefinedsymmetrically. Default Anindicatorvariablethattakesthevalueof1ifanypaymentrelated SNCandauthors’calculatoagivenloanis60daysorlongerpastdueand0otherwise. tions HighReputation Anindicatorthattakesthevalueof1ifaPEsponsorisrankedwithin SNCandauthors’calculathetop50ofallsponsorsintermsofmarketshareofdealvolumein tions thefullSNCsample. LiquidityCovenants Anindicatorvariablethattakesthevalueof1ifacovenantexplicitly SNC mentionsitcontainsliquiditycovenantssuchasthecurrentratioin thedescription. LoanPurpose An indicator variable that takes the value of one for Acquisition SNC and/or Merger Financing, General Corporate Purpose, Refinancing/Consolidation,etc. LoanTime-to-Maturity Thedifferencebetweentheloanmaturitydateandthereviewdate SNC (inyears)ofagivencreditfacility. LoanType Anindicatorvariablethattakesthevalueofonefordifferentloan SNC facilitiessuchasrevolvingcreditlines,termloans,orotherloans. NegativeCovenants Anindicatorvariablethattakesthevalueof1ifacovenantexplicitly SNC mentionsnegativecovenantsinthedescription. Non-performance-basedCovenant Capturesprimarilynegativecovenants(e.g.,equitypaymentlimita- SNC tions),affirmativecovenants(e.g.,financialreportingtothelender), minimumcurrentratiorequirement,andmaximumcapitalexpenditurelimits. 41

PE Indicatorvariablethattakesthevalueof1ifaloanissponsoredbya SNC PEfirmand0otherwise. Performance-basedCovenant An indicator variable that takes the value of 1 for any one of the SNC followingcovenants: debt-to-EBITDAratio,seniordebt-to-EBITDA ratio, interestcoverageratio, fixedchargecoverageratio, debtservicecoverageratios,levelofEBITDA,minimumprofitabilityrequirements, debt-to-equity ratio, loan-to-value ratio, and net worth requirements. ProbabilityofDefault Bankestimatedprobabilityofdefaultforagivenborrower.Reported FRY-14Q default probabilities are typically forward-looking one-year ahead projections. Public Anindicatorvariablewhichtakesthevalueof1ifafirmispublicly FRY-14Q tradedonastockexchangeinagivenyear. Total Number of Institutional Thenumberofinstitutionallenders(e.g.,CLOs,hedgefunds,ordi- SNCandauthors’calcula- Lenders rectlenders)thatinvestinagivenloansyndicateatagivenpointin tions time.Thisvariableiscomputedonlyforloanswithatleastoneinstitutionalinvestoratanytime. TotalNumberofLenders Thenumberoflendersinagivenloansyndicateatagivenpointin SNC time. TotalPESponsor-BankExposure Thesumofalloutstandingutilizedcommitmentbyallportfoliocom- SNCandauthors’calculapaniesthatarefundedbyagivenPEfund-bankpairatobservation tions datet. TotalRisk-BasedCapitalRatio Aratioofthetotalrisk-basedcapitaloverRisk-WeightedAssets,con- FR Y-9C and authors’ calstructedattheBankHoldingCompany×Timelevel. culations Violate Anindicatorvariablethattakesthevalueof1ifaloanbreachesa SNC covenantorrequiresawaiveroramendmentinordertostaycompliantand0otherwise. Inrobustnesstests,weexcludewaiversand resets. UtilizedExposure Theoutstandingdrawnamountunderagivenlineofcreditinmil- SNC lionsofUSdollars. UtilizationRate The outstanding drawn amount divided by the total commitment SNC amount. 42

Figure1: SyndicateConcentrationandShareofPerformance-basedCovenantsinPE (a) Notes: This chart shows the relationship between the number of lenders in a syndicate and the likelihood of the loan having performance-based covenants. The y-axis plots buckets of the number of non-bank investors holding a given loan at a given point in time, and the x-axis plots the share of loans in each bucket that have performancebasedcovenants.Performance-basedcovenantsaredefinedinAppendixAandcontainmostlytraditionalmaintenance covenantssuchasmaximumleverageratio,interestcoverageratio,fixedchargecoverageratio,ordebtservicecoverage ratio. ThesampleisrestrictedtoPE-sponsoredloans. 43

Figure2: ShareofCommitmentsbyLoanandFirm-Type (a)Notes:ThischartplotstheshareofdifferenttypesofloanswithinthePEandnon-PEsampleintheSNCdatabase. Loan types are grouped into term loans, credit lines, and other types of facilities. All variables are defined in AppendixA. 44

Figure3: ProbabilityofViolatingaCovenant: PEFirms (a)Notes:ThischartplotstheshareofloansthatareviolatedinagivenyearforfirmsbackedbyPEsponsors.Thegrey lineplotsthetrendforalltypesofcovenants,whilethedashedblacklinerestrictsthesametoonlyperformance-based covenants. Performance-basedcovenantsaredefinedinAppendixA. 45

Figure4: ProbabilityofViolatingaCovenant: Non-PEFirms (a) Notes: This chart plots the share of loans that are violated in a given year for firms not backed by PE. The grey lineplotsthetrendforalltypesofcovenants,whilethedashedblacklinerestrictsthesametoonlyperformance-based covenants. Performance-basedcovenantsaredefinedinAppendixA. 46

Figure5: BankCapital: EquitytoAssetsRatio (a) Notes: This chart plots the distribution of the lender’s capital, proxied by the equity to assets ratio, using a histogramof20equal-widthbins. Thesampleisrestrictedtothemergedsamplethatincludesthelender’sfinancial informationfromFR-Y9C,leadingto28,550uniqueloan-timeobservations. 47

Figure6: ExaminerStrictness (a)Notes: Thischartplotsthedistributionofexaminerstrictness. Examinerstrictnessismeasuredastheshareoffail ratingsassignedbyagivenexaminer-in-charge. Thus,foragivenexaminer-in-charge,itismeasuredasthenumber offailratingsoverhertotalnumberofexams. Thebenchmarksamplehas540uniqueexaminers. 48

Table1: SummaryStatistics PanelA:PE-backed N Mean Stdev p50 p25 p75 Commitments(USDMn) 19,189 492 743 250 95 600 Maturity(Years) 19,189 6.1 7.7 5 5 7 UtilizationRate 19,189 0.62 0.42 0.85 0.13 1 ConcordanceRating 19,189 2.5 1.2 2 2 3 TotalPESponsor-BankExposure(USDBn) 19,188 10.7 19.7 2.25 0.32 12.5 AmendmentOutsideDistress 19,188 0.33 0.47 0 0 1 TotalNumberofLendersinSyndicate 19,189 86.2 195 10 6 27 TotalNumberofInstitutional/Non-BankLenders 16,947 96.9 206 12 7 46 LoanSpread(bps) 3,832 322 169 300 200 425 CovenantViolationsWaivedorReset(%) 3,420 13.9 - - - - PanelB:Non-PE-backed Commitments(USDMn) 24,481 403 664 198 75 465 Maturity(Years) 24,481 6.1 3.36 5 5 7 UtilizationRate 24,481 0.61 0.41 0.73 0.16 1 ConcordanceRating 24,481 2.3 1.2 2 2 3 TotalNumberofLendersinSyndicate 24,481 44.3 115 8 5 17 TotalNumberofInstitutional/Non-BankLenders 20,781 51.4 123 10 6 20 AmendmentOutsideDistress 24,481 0.32 0.46 0 0 1 LoanSpread(bps) 4,532 307 154 300 200 400 CovenantViolationsWaivedorReset(%) 2,686 13.9 - - - - (a)Notes: Thistablereportssummarystatisticsofloan-timeobservationsincludedinthebenchmarksamplefromthe SharedNationalCredit. Thesummarystatisticspresentedherepertaintoloansthathavebeensampledandthathave availableinformationforallloanandborrowercharacteristics. Timeisdefinedattheyear-quarterlevel. Allvariables aredefinedinAppendixA. 49

Table2: CovenantTypeandDollarVolume Commitment(MnUSD) PanelA:PE-backedloans Freq(%) Mean Median Leverage/SeniorLeverageRatio 29.3 405 200 NegativeCovenants 20.0 635 365 InterestCoverageRatio 13.3 428 234 AffirmativeCovenants 10.6 650 350 FixedChargeCoverage 9.9 237 117 CurrentRatio 4.6 617 393 SpringingCovenant 4.5 450 200 DebtServiceCoverageRatio 3.3 254 147 NetWorthCovenant 2.1 339 210 MaximumCapitalExpenditure 2.0 170 85 LoantoValue 0.3 489 380 PanelB:Non-PE-backedloans Leverage/SeniorLeverageRatio 29.5 433 200 NegativeCovenants 9.4 486 250 InterestCoverageRatio 15.8 404 200 AffirmativeCovenants 4.3 430 231 FixedChargeCoverage 12.8 223 110 CurrentRatio 4.7 464 185 SpringingCovenant 2.7 383 250 DebtServiceCoverageRatio 4.0 195 102 NetWorthCovenant 4.0 314 160 MaximumCapitalExpenditure 3.8 209 100 LoantoValue 0.4 218 125 (a)Notes:Thistablereportsthefrequencyofdifferenttypesofloancovenants,splitbetweenthePE-backedloansample (PanelA)andthenon-PE-backedloansample(PanelB)intheSNCdatabase. Wealsoreportthedistributionofloan amountssecuredbyeachcovenantandborrowertype. AllvariablesandcovenantsaredefinedinAppendixA. 50

Table3: ProbabilityofViolatingaCovenantandSubsequentResolution Y 1(Violated) 1(ViolationWaived) j,i,b,t (1) (2) (3) (4) (5) (6) PE 0.0391∗∗∗ 0.0384∗∗∗ 0.0445∗∗∗ 0.0398∗∗∗ 0.0397∗∗∗ 0.0454∗∗∗ (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) R-squared 0.102 0.117 0.129 0.0934 0.108 0.120 BankxTimeFE Y Y Y Y Y Y SectorxTime N Y Y N Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFE N N Y N N Y N 43,491 43,481 43,478 43,491 43,481 43,478 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: Thistablereportsestimatesofalinearprobabilitymodel. Incolumns(1)-(3),thedependantvariableisanindicatortakingthevalueof1ifacovenantis violatedatagivenpointintimeand0otherwise. Incolumns(4)-(6),thedependantvariableisanindicatortakingthevalueof1ifaloancovenantwouldhavebeen noncompliantbutforacovenantwaiverorresetgrantedbythelender. PE isanindicatorvariabletakingthevalueof1ifaloaninvolvesaPE-ownedborrower and 0 otherwise. Sector-time fixed effects are defined at the 2-digit NAICS level. Time FEs are at the year-quarter level of the SNC review date. Loan controls includeutilizationrate,totalloancommitmentinlogs,time-to-maturity,andindicatorsforsupervisoryriskrating,loantype(creditlines,termloans,etc.),and loanpurpose. AllvariablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 51

Table4: BenchmarkResults: CovenantBreachandCreditorEnforcement Panel A : Log(Commitments) (1) (2) (3) (4) (5) (6) Violate -0.124∗∗∗ -0.116∗∗∗ -0.113∗∗∗ -0.111∗∗∗ -0.0958∗∗∗ -0.280∗∗∗ (0.027) (0.027) (0.027) (0.027) (0.026) (0.038) PE×Violate 0.0776∗∗ 0.0686∗∗ 0.0682∗∗ 0.0678∗∗ 0.0680∗∗ 0.124∗∗∗ (0.034) (0.034) (0.034) (0.034) (0.034) (0.045) R-squared 0.752 0.754 0.756 0.756 0.767 0.398 BankxTimeFE Y Y Y Y Y Y SectorxTimeFE N Y Y Y Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFE N N Y Y Y Y Covenant-typeFE N N N Y Y Y FirmFE N Y Y Y N N Bank-FirmFE N N N N Y N N 42,874 42,864 42,861 42,861 42,801 43,478 Panel B : 1(Credit Reduced) (1) (2) (3) (4) (5) (6) Violate 0.0842∗∗∗ 0.0874∗∗∗ 0.0818∗∗∗ 0.0812∗∗∗ 0.0766∗∗∗ 0.0671∗∗∗ (0.018) (0.019) (0.018) (0.018) (0.018) (0.014) PE×Violate -0.0854∗∗∗ -0.0844∗∗∗ -0.0792∗∗∗ -0.0793∗∗∗ -0.0808∗∗∗ -0.0538∗∗∗ (0.026) (0.027) (0.026) (0.026) (0.026) (0.020) R-squared 0.165 0.176 0.181 0.181 0.187 0.0642 BankxTimeFE Y Y Y Y Y Y SectorxTimeFE N Y Y Y Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFE N N Y Y Y Y Covenant-typeFE N N N Y Y Y FirmFE N Y Y Y N N Bank-FirmFE N N N N Y N N 36,560 36,548 36,545 36,545 36,496 37,274 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a) Notes: This table reports the benchmark results where the dependent variable is (i) the natural logarithm of loan commitment at time t in Panel A, and (ii) 1(CreditReduced)inPanelB.PEisanindicatorvariabletakingthevalueof1ifaloaninvolvesaPE-ownedborrowerand0otherwise. Sector-timefixedeffects aredefinedatthe2-digitNAICSlevel. TimeFEsareattheyear-quarterleveloftheSNCreportdate. Loancontrolsincludeutilizationrate,totalloancommitment in logs, time-to-maturity, and indicators for supervisory risk rating, loan type (credit lines, term loans, etc.), and loan purpose. Covenant types are split into performance-basedandnon-performance-based. AllvariablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 52

Table5: InstrumentalVariable: ExaminerStrictnessatLoanOrigination 1(Credit Reduced) Log(Commitments) Log(Maturity) (1) (2) (3) (4) (5) (6) Violate 0.389∗ 0.457∗∗ -2.624∗∗∗ -1.823∗∗∗ -2.821∗∗∗ -3.327∗∗∗ (0.204) (0.198) (0.492) (0.457) (0.287) (0.396) PE×Violate -0.674∗∗ -0.609∗∗ 1.241∗ 0.381∗ 0.859∗∗ 1.460∗∗∗ (0.320) (0.306) (0.733) (0.218) (0.397) (0.446) DistrictFE N Y N Y N Y DistrictxYearFE Y N Y N Y N BankFE Y N Y N Y N BankxTimeFE N Y N Y N Y LoanControls Y Y Y Y Y Y SectorFE Y Y Y Y Y Y First-StageF-Stat 19.3 21.5 23.7 20.9 24.0 33.1 N 28,254 28,217 33,124 33,087 33,124 33,093 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a) Notes: This table reports instrumental variable regression estimates where the outcomes are 1 (Credit Reduced), Log(Commitments), and the natural logarithm of loan maturity expressed in number of quarters (Log(Maturity)). The excluded instrument is the strictness of the lender’s supervisor at the time of loanorigination. Sectorfixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEsareattheyear-quarterleveloftheSNCreportdate. Loancontrolsinclude utilizationrateandtime-to-maturityin Columns(1)-(4)andutilizationrateand Log(commitments)incolumns(5)and(6). Theyalsoincludeindicatorsfor supervisoryriskrating,loantype,andloanpurposeinallspecifications. AllvariablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Time level. 53

Table6: EnforcementandCovenantHeterogeneity Y : Log(Commitments) Covenant-HeavySample Covenant-LiteSample (1) (2) (3) (4) Violate -0.0871∗∗∗ -0.0788∗∗ -0.187∗∗∗ -0.180∗∗∗ (0.033) (0.033) (0.057) (0.058) PE×Violate 0.0902∗∗ 0.0800∗ 0.198∗∗∗ 0.183∗∗ (0.045) (0.045) (0.077) (0.077) R-squared 0.764 0.770 0.780 0.784 FirmFE Y Y Y Y BankxTimeFE Y Y Y Y SectorxTimeFE N Y N Y LoanControls Y Y Y Y N 21,948 21,934 19,575 19,557 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: ThistablereportsOLSestimatesofthebaselineequation,estimatedseparatelyonloansthatareclassified as“Covenant-Heavy”andthosethatareclassifiedas“Covenant-Lite.” Covenant-Heavyloansarethosethathaveat least one of the following financial covenants: maximum leverage/senior leverage ratio, interest coverage ratio, debt service coverage ratio, and fixed charge coverage ratio. Covenant-lite loans are those that have none of the financial covenants listed above. Loan controls include utilization rate, time-to-maturity, and indicators for supervisory risk rating, loan type (credit lines, term loans, etc.), loan purpose, and loan origination year-quarter. All variables are definedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 54

Table7: CovenantViolations,PEpresence,andBankCapital Y : Log(Commitments) HighEquityCapital LowEquityCapital (1) (2) (3) (4) (5) (6) Violate -0.392∗∗∗ -0.382∗∗∗ -0.374∗∗∗ -0.262∗∗∗ -0.267∗∗∗ -0.226∗∗∗ (0.050) (0.050) (0.050) (0.067) (0.063) (0.061) PE×Violate 0.202∗∗∗ 0.184∗∗∗ 0.182∗∗ 0.0848 0.0928 0.129∗ (0.069) (0.069) (0.073) (0.081) (0.080) (0.072) R-squared 0.377 0.401 0.409 0.338 0.376 0.408 BankxTimeFE Y Y Y Y Y Y SectorxTimeFE N Y Y N Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFE N N Y N N Y N 14,320 14,311 14,308 14,153 14,129 14,126 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes:ThistablereportsOLSestimateswherethedependentvariableisthenaturallogarithmofloancommitment between a given firm-bank pair at time t. High Equity Capital lenders are defined as those with an equity-to-assets ratio above the sample median in the year preceding a covenant violation. Low Equity Capital banks are defined symmetrically. PE is an indicator variable taking the value of 1 if a loan involves a PE-owned borrower and 0 otherwise. Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEsareattheyear-quarterlevelof the SNC report date. Loan controls include utilization rate, total loan commitment in logs, time-to-maturity, and indicators for supervisory risk rating, loan type (credit lines, term loans, etc.), and loan purpose. All variables are definedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 55

Table8: SponsorReputationandCreditorEnforcement Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) Violate -0.275∗∗∗ -0.267∗∗∗ 0.0559∗∗∗ -0.267∗∗∗ (0.034) (0.034) (0.012) (0.034) High Reputation 0.252∗∗∗ 0.256∗∗∗ 0.00363 0.00472 (0.026) (0.025) (0.008) (0.008) Violate×High Reputation 0.152∗∗∗ 0.152∗∗∗ -0.0545∗∗ -0.0593∗∗∗ (0.049) (0.049) (0.022) (0.022) R-squared 0.389 0.394 0.0515 0.0610 BankxTimeFE Y Y Y Y SectorxTimeFE Y Y Y Y OriginationYr-QtrFE N Y N Y LoanControls Y Y Y Y N 43,490 43,480 37,285 37,275 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a) Notes: This table reports OLS estimates where the dependant variables are the same as in the baseline. High Reputationisaproxyforasponsor’sreputationandtakesthevalueof1ifthesponsorisrankedwithinthetop50 fundsinthebaselinesampleintermsofmarketshareofdealvolumeintheUSsyndicatedloanmarket. Inaddition to the controls listed above, all regressions also include an indicator for PE-backed firms. Loan controls include utilization rate, total loan commitment in logs, time-to-maturity, and indicators for supervisory risk rating, loan type(creditlines, termloans, etc.), andloanpurpose. AllvariablesaredefinedinAppendixA. Standarderrorsare clusteredattheBank×Timelevel. 56

Table9: LoanRenegotiationandPEBargainingPower Panel A : Concentrated Syndicates Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) PE×Violate × Concentrated 0.240∗∗∗ 0.252∗∗∗ -0.00844 -0.0138 (0.079) (0.079) (0.044) (0.044) Violate × Concentrated -0.374∗∗∗ -0.374∗∗∗ 0.0551∗ 0.0544∗ (0.056) (0.056) (0.030) (0.030) Concentrated -1.023∗∗∗ -1.017∗∗∗ 0.0496∗∗∗ 0.0502∗∗∗ (0.020) (0.020) (0.009) (0.009) R-squared 0.567 0.572 0.0715 0.0752 BankxTimeFE Y Y Y Y SectorxTimeFE Y Y Y Y OriginationYr-QtrFE N Y N Y LoanControls Y Y Y Y N 37,555 37,551 32,277 32,275 Panel B : Reliance on Deal Flow Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) Violate -0.193∗∗∗ -0.184∗∗∗ 0.0456∗∗ 0.0453∗∗ (0.044) (0.045) (0.023) (0.023) Violate×High Exposure 0.151∗∗ 0.148∗∗ -0.0719∗∗ -0.0723∗∗ (0.063) (0.063) (0.031) (0.031) High Exposure 0.355∗∗∗ 0.352∗∗∗ 0.0172∗ 0.0134 (0.027) (0.027) (0.010) (0.010) R-squared 0.461 0.471 0.0787 0.0787 BankxTimeFE Y Y Y Y SectorxTimeFE Y Y Y Y OriginationYr-QtrFE N Y N Y LoanControls Y Y Y Y N 19,097 19,100 16,551 16,557 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: ThistablereportstwotestsrelatedtothehighbargainingpowerofPEsponsorsvis-a-vislenders. PanelA reportstriple-differenceestimateswherethedependentvariableisthesameasthebaseline. Concentratedisaproxy forasyndicate’sownershipconcentrationandtakesthevalueof1ifthetotalnumberofinstitutionallendersinagiven loantimeislessthanthesamplemedian,0otherwise.Allregressionsincludelower-orderinteractionsandcontrolsfor loantime-to-maturity, utilizationrate, andtheactualnumberofinstitutionallenders. InPanelB, High Exposure capturesalender’stotalloanexposuretoaspecificPEsponsorthrougheveryoutstandingLBOdeal. Loancontrols also include indicators for supervisory risk rating, loan type (credit lines, term loans, etc.), and loan purpose. All variablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 57

Table10: LoanPerformance: Downgrades 1(Substandard/Doubtful) log (1 + Non Pass Amount) (1) (2) (3) (4) (5) (6) Violate 0.204∗∗∗ 0.189∗∗∗ 0.192∗∗∗ 4.722∗∗∗ 4.370∗∗∗ 4.463∗∗∗ (0.016) (0.016) (0.016) (0.273) (0.279) (0.279) PE 0.0158∗∗∗ 0.0128∗∗ 0.0268∗∗∗ 0.457∗∗∗ 0.392∗∗∗ 0.695∗∗∗ (0.006) (0.005) (0.005) (0.113) (0.107) (0.107) PE×Violate 0.0184 0.0140 0.00276 -0.0764 -0.154 -0.358 (0.023) (0.022) (0.023) (0.389) (0.398) (0.400) R-squared 0.160 0.196 0.214 0.162 0.208 0.227 FirmFE Y Y Y Y Y Y BankxTimeFE Y Y Y Y Y Y SectorxTimeFE N Y Y N Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFE N N Y N N Y N 43,491 43,481 43,478 43,491 43,481 43,478 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: ThistablereportsOLSestimateswherethedependantvariablecapturesloanperformance. Incolumns(1)to(3),wemeasureperformancethroughan indicatorthattakesthevalueof1ifaloanisclassifiedasSpecialMention,Substandard,Doubtful,orLoss. Incolumns(4)to(6),weusethenaturallogarithmof 1plusthedollaramountofaloanfacility’scommittedexposurewherethefinalexamratingisSpecialMention,Substandard,Doubtful,orLoss. Loancontrolsare thesameasinthebaseline. AllvariablesaredefinedinSectionAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 58

Table11: LoanDefaults Y : 1×(Days Past Due >= 60) (1) (2) (3) (4) Violate 0.0143∗∗∗ 0.00926∗∗ 0.0143∗∗∗ 0.00964∗∗ (0.004) (0.004) (0.004) (0.004) PE×Violate -0.0107∗ -0.0135∗∗ -0.0105∗ -0.0140∗∗∗ (0.006) (0.005) (0.006) (0.005) R-squared 0.168 0.477 0.159 0.479 FirmFE N Y N Y BankxTimeFE Y Y Y Y SectorxTime Y Y Y Y LoanControls Y Y Y Y OriginationYr-QtrFEs N N Y Y N 43,478 42,861 43,478 42,861 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: Thistablereportsregressionestimateswherethedependentvariableisanindicatorthattakesthevalueof1 ifaloanpaymentispastduefor60daysormore.Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel.Time FEsareattheyear-quarterleveloftheSNCreportdate. Loancontrolsarethesameasinthebaseline. Allvariables aredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 59

Table12: CapitalInjection 1×(Capital Injection) (1) (2) (3) (4) Violate 0.0178∗∗ 0.0165∗∗ 0.0165∗∗ 0.00597 (0.007) (0.007) (0.007) (0.005) PE×Violate 0.0148 0.0174∗ 0.0153 0.00707 (0.011) (0.010) (0.010) (0.007) R-squared 0.0298 0.0450 0.102 0.865 FirmFE N N N Y BankxTimeFE N N Y N SectorxTime Y Y Y Y OriginationYr-QtrFE N Y Y N LoanControls Y Y Y Y N 43,660 43,657 43,478 43,046 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a) Notes: This table reports regression estimates using the baseline equation for equity injection. The dependent variable is an indicator that takes the value of 1 if a loan received equity infusion and 0 otherwise. Equity Infusion is identified from the SNC data as described in section 4. Sector-time fixed effects are defined at the 2-digit NAICS level. TimeFEsareattheyear-quarterleveloftheSNCreportdate. Loancontrolsarethesameasinthebaseline. All variablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 60

Table13: OtherOutcomes: LoanSpreadsandMaturity Loan Spreads Loan Maturity (1) (2) (3) (4) (5) (6) (7) (8) Violate 50.36∗∗∗ 38.32∗∗∗ 34.43∗∗∗ -2.345 -1.952∗∗∗ -1.953∗∗∗ -1.675∗∗∗ -0.774∗∗∗ (8.122) (7.674) (6.801) (8.071) (0.269) (0.199) (0.266) (0.275) PE×Violate -35.66∗∗∗ -28.94∗∗ -17.65 0.549 1.105∗∗ -0.331 1.032∗∗ -1.845 (12.693) (12.014) (11.077) (14.026) (0.447) (0.349) (0.431) (1.603) R-squared 0.189 0.247 0.397 0.756 0.0366 0.200 0.0936 0.303 FirmFE N N N Y N N N Y BankxTimeFE N N Y N N N Y N SectorxTime Y Y Y Y Y Y Y Y OriginationYr-QtrFE N Y Y N N Y N N LoanControls Y Y Y Y Y Y Y Y N 8,334 8,324 8,262 6,962 43,660 43,657 43,481 42,864 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: Thistablereportsregressionestimatesusingthebaselineequationwithadditionaldependantvariables: loanspreads(overLIBOR,expressedinbasis points)andloanmaturity(expressedinanumberofquarters). Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEsareattheyear-quarter leveloftheSNCreportdate. Incolumns(1)to(4),loancontrolsincludeloanutilizationrateandtime-to-maturity. Incolumns(5)to(8),loancontrolsinclude loanutilizationrateandLog(Commitments). Loancontrolsalsoincludeindicatorsforloantype. AllvariablesaredefinedinAppendixA. Standarderrorsare clusteredattheBank×Timelevel. 61

Table14: MatchedSampleAnalysisusingFirm-levelFactors PanelA:MatchedControlSampleandfirmcontrols Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) Violate -0.224∗∗∗ -0.211∗∗∗ 0.116∗∗∗ 0.109∗∗∗ (0.060) (0.062) (0.038) (0.039) PE×Violate 0.210∗∗ 0.199∗∗ -0.0968∗ -0.0896 (0.082) (0.084) (0.055) (0.056) R-squared 0.785 0.789 0.245 0.278 N 9,335 9,307 8,093 8,066 FirmFE Y Y Y Y BankxTimeFE Y Y Y Y SectorxTime N Y N Y LoanControls Y Y Y Y Firm-levelControls Y Y Y Y PanelB:Onlyfirm-controlswithoutmatching Violate -0.189∗∗∗ -0.181∗∗∗ 0.0898∗∗∗ 0.0913∗∗∗ (0.047) (0.048) (0.031) (0.032) PE×Violate 0.180∗∗∗ 0.183∗∗∗ -0.0739 -0.0800∗ (0.064) (0.065) (0.046) (0.047) R-squared 0.780 0.783 0.229 0.251 N 16,207 16,190 13,934 13,913 FirmFE Y Y Y Y BankxTimeFE Y Y Y Y SectorxTime N Y N Y LoanControls Y Y Y Y Firm-levelControls Y Y Y Y ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: ThistablereportsOLSestimatesofthebenchmarkregression,augmentedwithamatchingprocedureand firm-level controls from the FR Y-14Q in Panel A. PE loans are matched to non-PE loans based on firm size (Log (TotalAssets),Debt/Assets,EBITDA/Assets,and1-yearaheadprobabilityofdefaultinthepre-buyoutyearwithin thesame2-digitNAICSindustry. InPanelB,thecontrolgroupisnotmatchedbutincludesfirm-levelcontrols. Firm controlsinbothpanelsincludeDebt/Assets,EBITDA/Assets,Log(TotalAssets)and1∗(Public). Sector-timefixed effects are defined at the 2-digit NAICS level. Time FEs are at the year-quarter level of the SNC report date. Loan controlsarethesameasinthebaseline,exceptfortheomissionofthesupervisoryriskrating.Allexplanatoryvariables aredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. 62

Online Appendix A1

FigureA1: Fee-BasedIncome (a) Notes: This chart plots the distribution of the Loan Commitment Fee (upper chart) and Upfront Fee for sponsoredandnon-sponsoredsyndicatedloans,usingdatafromDealScanbetween2012-2021. Thisdataonlyrepresents DealScan and is not merged with SNC. Total deals in the top chart is 149,459 and in the bottom chart is 125,955. Dataistruncatedatthe99percentlevel. AllexplanatoryvariablesaredefinedinAppendixA. A2

TableA1: LoansbyIndustry(%) NAICSCode Desc. PE Non-PE 2 Mining,UtilitiesandConstruction 13.1 16 3 Manufacturing 21.8 21.1 4 Trade,TransportationandWarehousing 14.9 16.5 5 IT,Finance,ProfessionalandManagementServices 37.7 33.8 6 EducationandHealthCare 5.6 4.7 7 Arts,EntertainmentandAccommodation 5.3 5.8 Others 1.6 2.1 (a)Notes: Thistablereportsloan-timeobservationsby1-digitNAICScode,splitbyPEandNon-PEloans. TableA2: ShareofLoansbyConcordanceRatingsandBorrowerTypes ConcordanceRating Description Pass/Fail PE Non-PE 1 InvestmentGradePass Pass 16.2% 22.5% 2 Non-InvestmentGradePass Pass 47.8% 43.5% 3 LowestRatedPass Pass 16.9% 17.5% 4 SpecialMention Fail 8.5% 7.5% 5 Substandard Fail 9.3% 7.7% 6 Doubtful Fail 0.9% 0.7% 7 Loss Fail 0.4% 0.6% (a)Notes: ThistablereportstheshareofobservationsbySupervisoryRiskRating, alsocalledConcordanceratings, splitbyborrowertype. Thisratingisusedtocontrolforborrowerriskintheempiricalanalysisinthispaper. Wealso addthecolumnPass/Failtoclarifyratingsthatcorrespondtoapassrating.Concordanceratingisa7-scalenumerical ratingfullydefinedinAppendixA. A3

TableA3: ExaminerStrictnessatLoanOrigination: RobustnessTest 1(Credit Reduced) Log(Commitments) Log(Maturity) (1) (2) (3) (4) (5) (6) Violate 0.439∗∗ 0.403∗ -1.413∗∗∗ -0.276 -3.410∗∗∗ -3.285∗∗∗ (0.204) (0.210) (0.468) (0.501) (0.389) (0.388) PE×Violate -0.817∗∗∗ -0.681∗∗ 0.808∗∗ -0.541 1.778∗∗∗ 1.534∗∗∗ (0.310) (0.309) (0.355) (0.747) (0.456) (0.450) BankFE Y Y N Y Y Y SectorFE Y N Y N Y N SectorxTimeFE N Y N Y N Y LoanControls Y Y Y Y Y Y N 30,516 30,507 35,700 35,679 35,687 35,679 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a) Notes:Thistablereportsinstrumentalvariableregressionestimateswheretheoutcomesareindicatorsforcreditreduction,Log(Commitments),andthenatural logarithmofloanmaturityexpressedinanumberofquarters. Theexcludedinstrumentisthestrictnessofthelender’ssupervisoratthetimeofloanorigination. ThekeydifferencefromTable5isthatweexcluderegulatorydistrictfixedeffects. Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEsareat theyear-quarterleveloftheSNCreportdate.Loancontrolsincludeutilizationrate,totalloancommitmentinlogs,time-to-maturity,andindicatorsforsupervisory riskrating,loantype(creditlines,termloans,etc.),andloanpurpose. AllexplanatoryvariablesaredefinedinAppendixA. Standarderrorsareclusteredatthe Bank×Timelevel. A4

TableA4: Aretheresultsstrongerforcovenant-liteloans? (1) (2) (3) Violate -0.117∗∗∗ -0.115∗∗∗ -0.0991∗∗∗ (0.031) (0.031) (0.031) PE×Violate 0.0911∗∗ 0.0867∗∗ 0.0768∗ (0.040) (0.040) (0.040) PE×Violate×1(Covlite) 0.0138 0.0179 0.0278 (0.077) (0.077) (0.077) R-squared 0.749 0.750 0.754 FirmFE Y Y Y BankxTimeFE Y Y Y SectorxTime N N Y LoanControls Y Y Y OriginationYr-QtrFE N Y Y N 42,874 42,871 42,864 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes:Thistablereportstripleinteractionspecificationstoexamineifthelimitedpunishmenteffectisstrongerfor Covenant-liteloansinoursample. Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEsareat theyear-quarterleveloftheSNCreportdate. Alllower-ordertermsareincludedbutomittedfromdisplayforbrevity. Loancontrolsincludeutilizationrate,totalloancommitmentinlogs,time-to-maturity,andindicatorsforsupervisory riskratingandloantype(creditlines,termloans,etc.).AllexplanatoryvariablesaredefinedinAppendixA.Standard errorsareclusteredattheBank×Timelevel. A5

TableA5: SensitivityofBenchmarkResulttoLoanRenegotiationOutsideofDistress Y : Log(Commitments) (1) (2) (3) (4) (5) (6) j,i,b,t Violate -0.269∗∗∗ -0.109∗∗∗ -0.106∗∗∗ -0.0988∗∗∗ -0.0899∗∗∗ -0.238∗∗∗ (0.040) (0.027) (0.027) (0.027) (0.027) (0.039) PE×Violate 0.168∗∗∗ 0.134∗∗∗ 0.134∗∗∗ 0.134∗∗∗ 0.135∗∗∗ 0.201∗∗∗ (0.048) (0.035) (0.035) (0.035) (0.036) (0.047) AmendmentOutside Distress 0.0571∗∗ 0.00898 0.0105 0.0117 -0.00463 0.0500∗∗ (0.025) (0.018) (0.018) (0.018) (0.017) (0.025) PE×AmendmentOutside Distress 0.188∗∗∗ 0.187∗∗∗ 0.187∗∗∗ 0.185∗∗∗ 0.188∗∗∗ 0.189∗∗∗ (0.042) (0.035) (0.035) (0.035) (0.035) (0.041) R-squared 0.373 0.755 0.757 0.757 0.768 0.404 BankxTimeFE Y Y Y Y Y Y SectorxTimeFE N Y Y Y Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFEs N N Y Y Y Y Covenant-typeFE N N N Y Y Y FirmFE N Y Y Y N N Bank-FirmFE N N N N Y N N 43,491 42,864 42,861 42,861 42,801 43,478 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: ThistablereportstherobustnesstestofTable4,bytestingifthepropensityofamendmentsoutsideofdistressaffectsrenegotiationanddebtenforcement uponcovenantviolation. Amendment_outside_distresstakesthevalueof1ifaloancommitmentamountischangedinperiodtrelativetoperiodt−1outside ofourdefinitionofcovenantviolation. Allothercontrolsarethesameasthebaseline. Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEs are at the year-quarter level of the SNC report date. Loan controls include utilization rate, total loan commitment in logs, time-to-maturity, and indicators for supervisoryriskrating,loantype(creditlines,termloans,etc.),andloanpurpose. Covenanttypesaresplitintoperformance-basedandnon-performance-based. AllexplanatoryvariablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. A6

TableA6: SponsorReputationandCreditorEnforcement: TripleInteractionSpecification Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) Violate×PE×High Reputation 0.293∗∗∗ 0.296∗∗∗ -0.0559∗∗∗ -0.0548∗∗∗ (0.067) (0.065) (0.021) (0.021) Violate -0.293∗∗∗ -0.289∗∗∗ 0.0543∗∗∗ 0.0566∗∗∗ (0.039) (0.038) (0.012) (0.012) R-squared 0.386 0.391 0.0515 0.0610 BankxTimeFE Y Y Y Y SectorFE Y N Y N SectorxTimeFE N Y N Y LoanControls Y Y Y Y N 43,490 43,480 37,285 37,275 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes:ThistablereportsOLSestimateswherethedependantvariableisthenaturallogarithmofloancommitment between a given firm-bank pair at time t in columns (1) and (2), and an indicator 1(Credit Reduced) in columns (3)and(4). HighReputationisaproxyforasponsor’sreputationandtakesthevalueof1ifthesponsorisranked within the top 50 funds in the baseline sample in terms of market share of deal volume in the US syndicated loan market. Inadditiontothecontrolslistedabove,allregressionsalsoincludeanindicatorforPE-backedfirmsaswellas lower-orderinteractionsbutareomittedfromdisplayforbrevity. AllvariablesaredefinedinAppendixA. Standard errorsareclusteredattheBank×Timelevel. A7

TableA7: Firm-levelComparison N PE N Non-PE PanelA:SNC-FRY14QMergedSample Log(Size) 7,137 20.9 9665 20.8 Debt/Assets 7,137 0.52 9665 0.47 EBITDA/Assets 7,137 0.12 9665 0.12 ProbabilityofDefault 7,137 0.05 9665 0.037 PanelB:SNC-FRY14QMergedSamplewithMatching Log(Size) 4,012 20.7 5643 20.7 Debt/Assets 4,012 0.51 5643 0.46 EBITDA/Assets 4,012 0.12 5643 0.12 ProbabilityofDefault 4,012 0.05 5643 0.04 (a) Notes: This table reports firm-year level summary statistics (means) of standard financial variables for PE and non-PE firms. The sample is constructed by merging the SNC database with the FR Y-14Q schedule H1 using the string matching algorithm outlined in Cohen et al. (2021) based on borrower name and industry. Panel A reports the full merged sample of SNC and FR Y-14Q and Panel B restricts the merged sample to loans that were matched followingthemethodologydescribedinsection4. A8

TableA8: BenchmarkTestwithAlternateViolationDefinition: RobustnessTest Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) Violate -0.369∗∗∗ -0.354∗∗∗ 0.113∗∗∗ 0.111∗∗∗ (0.082) (0.081) (0.029) (0.029) PE -0.0256 -0.0227 -0.00310 -0.00341 (0.017) (0.017) (0.006) (0.006) PE×Violate 0.226∗∗ 0.238∗∗ -0.126∗∗∗ -0.126∗∗∗ (0.105) (0.106) (0.044) (0.044) R-squared 0.397 0.401 0.0639 0.0642 BankxTimeFE Y Y Y Y SectorxTimeFE Y Y Y Y LoanControls Y Y Y Y Covenant-typeFE N Y N Y N 43,478 43,478 37,274 37,274 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes:ThistablereportsOLSestimateswherethedependentvariableisthenaturallogarithmofloancommitmentbetweenagivenfirm-bankpairattimet.The onlydifferencefromthebenchmarkregressionsisthatweexcludecovenantwaiversorresetsinourdefinitionofcovenantviolations. PEisanindicatorvariable takingthevalueof1ifaloaninvolvesaPE-ownedborrowerand0otherwise. Sector-timefixedeffectsaredefinedatthe2-digitNAICSlevel. TimeFEsareatthe year-quarterleveloftheSNCreportdate. Loancontrolsincludeutilizationrate,totalloancommitmentinlogs,time-to-maturity,indicatorsforsupervisoryrisk rating,loantype(creditlines,termloans,etc.),andloanpurpose. Covenanttypesaresplitintoperformance-basedandnon-performance-based. Allexplanatory variablesaredefinedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. A9

TableA9: CovenantViolations,PEpresence,andBankCapital: AlternateThresholds HighEquityCapital LowEquityCapital (1) (2) (3) (4) (5) (6) Violate -0.470∗∗∗ -0.432∗∗∗ -0.468∗∗∗ -0.220∗∗ -0.246∗∗ -0.184∗∗ (0.072) (0.073) (0.073) (0.098) (0.095) (0.092) PE×Violate 0.307∗∗ 0.241∗∗ 0.272∗∗ 0.134 0.173 0.175 (0.119) (0.122) (0.129) (0.103) (0.111) (0.114) R-squared 0.380 0.437 0.467 0.257 0.314 0.354 BankxTimeFE Y Y Y Y Y Y SectorxTimeFE N Y Y N Y Y LoanControls Y Y Y Y Y Y OriginationYr-QtrFE N N Y N N Y N 5,704 5,661 5,657 6,698 6,651 6,648 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a) Notes:ThistablereportsOLSestimateswherethedependentvariableisthenaturallogarithmofloancommitment betweenagivenfirm-bankpairattimet,usingalternatedefinitionsoflenders’capitalposition. HighEquityCapital lenders are defined as those with equity-to-assets ratio in the top quartile of the sample (12.7 percent), while Low Equity Capital lenders are defined as those with equity-to-assets ratio in the bottom quartile (9.8 percent) of the sample. PE is an indicator variable taking the value of 1 if a loan involves a PE-owned borrower and 0 otherwise. Sector-time fixed effects are defined at the 2-digit NAICS level. Time FEs are at the year-quarter level of the SNC report date. Loan controls include utilization rate, total loan commitment in logs, time-to-maturity, and indicators forsupervisoryriskrating,loantype(creditlines,termloans,etc.),andloanpurpose. Allexplanatoryvariablesare definedinAppendixA. StandarderrorsareclusteredattheBank×Timelevel. A10

TableA10: Sponsors’DealVolume,CovenantViolations,andLoanCommitments Log(Commitments) 1(Credit Reduced) (1) (2) (3) (4) Violate -0.322∗∗∗ -0.315∗∗∗ 0.0675∗∗∗ 0.0709∗∗∗ (0.042) (0.042) (0.015) (0.015) Log(1 + No.of Deals) 0.101∗∗∗ 0.102∗∗∗ 0.00184 0.00234 (0.009) (0.009) (0.003) (0.003) Violate×Log(1 + No.of Deals) 0.0502∗∗∗ 0.0505∗∗∗ -0.0146∗∗ -0.0163∗∗∗ (0.014) (0.014) (0.006) (0.006) R-squared 0.389 0.394 0.0515 0.0610 BankxTimeFE Y Y Y Y SectorxTimeFE Y Y Y Y OriginationYr-QtrFE N Y N Y LoanControls Y Y Y Y N 43,490 43,480 37,285 37,275 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes:ThistablereportsOLSestimateswherethedependantvariableisthenaturallogarithmofloancommitment betweenagivenfirm-bankpairattimetincolumns(1)and(2),andanindicator1(CreditReduced)incolumns(3) and(4). ThekeyvariableofinterestistheinteractionbetweenViolateandln(1 + no.of deals). Thelatteristhe naturallogarithmof1plusthetotalnumberofdealsexecutedbyaPEsponsor.Inadditiontothecontrolslistedabove, allregressionsalsoincludeanindicatorforPE-backedfirms. AllvariablesaredefinedinAppendixA.Standarderrors areclusteredattheBank×Timelevel. A11

TableA11: SupplementaryAnalysis: SponsoredLoansandFee-BasedIncomeusingDealScan Upfront Fees Commitment Fees (1) (2) (3) (4) PE 23.68∗∗∗ 23.51∗∗∗ 15.78∗∗∗ 15.91∗∗∗ (2.788) (2.850) (0.928) (0.951) R-squared 0.0780 0.109 0.214 0.257 BankFE Y N Y N YearFE Y N Y N BankxYearFE N Y N Y N 125,510 124,783 149,047 148,166 ∗ p<0.10,∗∗ p<0.05,∗∗∗ p<0.01 (a)Notes: ThistablereportsOLSestimateswherethedependentvariableiseitherUpfrontfeesorCommitmentfees of syndicated loans at origination only, using DealScan. It documents that PE-sponsored loans are associated with higherfeeratesrelativetonon-PE.BothfeetypesaremeasuredinbasispointsandaredefinedfullyinAppendixA. Standarderrorsareclusteredatthebankandyearlevel. A12