Indirect Credit Supply: How Bank Lending to Private Credit Shapes Monetary Policy Transmission

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Indirect Credit Supply: How Bank Lending to Private Credit Shapes Monetary Policy Transmission Sharjil Haque, Young Soo Jang, Jessie Jiaxu Wang 2025-059 Please cite this paper as: Haque, Sharjil, Young Soo Jang, and Jessie Jiaxu Wang (2025). “Indirect Credit Supply: How Bank Lending to Private Credit Shapes Monetary Policy Transmission,” Finance and Economics Discussion Series 2025-059. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2025.059. 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.

Indirect Credit Supply: How Bank Lending to Private Credit Shapes Monetary Policy Transmission* Sharjil Haque† Young Soo Jang ‡ Jessie Jiaxu Wang § This Version: July 2025 Abstract Thispaperexamineshowbanks’financingofnonbanklendersaffectsmonetarypolicytransmission. Usingsupervisorybankloan-leveldataanddeal-levelprivatecredit data, we document an intermediation chain: Banks lend to Business Development Companies (BDCs)—large private credit providers—which then lend to firms. As monetarytighteningrestrictsbanklending,firmsturntoBDCsforcredit,prompting BDCstoborrowmorefrombanks. Thisintermediationchainraisesborrowingcosts, asbankschargeBDCshigherrates,whichBDCspassontofirms. Consistentwiththis pass-through,bank-reliantBDCsrespondmorestronglytomonetarytightening,and BDC-dependentfirmsgrowmorebutexhibitweakerinterestcoverageratios. Overall, whilebanklendingtononbanksmitigatescreditcontractionandsupportsinvestment during tightening, it amplifies monetary transmission by elevating borrowing costs andfinancialdistressrisk. Keywords: Banks and nonbanks; Monetary policy transmission; Business developmentcompanies(BDCs);Privatecredit;Creditchain *WethankAbhishekBhardwaj,MatthewDenes,BurtonHollifield,EricaJiang(discussant),YueranMa, Simon Mayer, Phillip Monin, Raghuram Rajan, Irina Stefanescu, seminar participants at the Federal ReserveBoard, VirginiaTech, andconferenceparticipantsattheNBERSummerInstituteCorporateFinance meeting,CMU-Pitt-PSUFinanceconferenceforfeedback.WealsothankChookaszianAccountingResearch CenteratChicagoBoothforaccesstotheBDCdata,andJoeYuke,KarlWirth,andJewonShinforresearch assistance. Theviewsexpressedinthispaperarethoseoftheauthorsanddonotnecessarilyrepresentthe viewsoftheFederalReserveBoardortheFederalReserveSystem. †FederalReserveBoardofGovernors. Email: sharjil.m.haque@frb.gov. ‡PennStateUniversity. Email: ykj5180@psu.edu. §FederalReserveBoardofGovernors. Email: jessie.wang@frb.gov.

1 Introduction Banklendingplaysakeyroleinhowmonetarypolicyshapestherealeconomy(Bernanke and Blinder, 1988; Kashyap and Stein, 2000). Typically, banks respond to tighter monetary policy by cutting lending and raising borrowing costs, leading to a contraction in credit supply. However, the financial landscape has evolved, with the rise of nonbank lenders—particularly in private credit—introducing new dynamics into this transmission mechanism. Private credit has been one of the fastest-growing segments of the U.S. financial system, with total assets reaching $1.1 trillion by 2023, a tenfold increase since 2009.1 Whilepriorresearchhasexplorednonbanks’roleinmonetarytransmission,lessis known about how their interactions with banks affect credit availability and borrowing costsduringtighteningcycles. This paper fills the gap by studying how banks’ financing of nonbank lenders shapes monetarypolicytransmission. WefocusonbanklendingtoBusinessDevelopmentCompanies (BDCs)—a rapidly growing segment of the private credit market that primarily lends to large and middle-market firms.2 BDCs provide an ideal setting to study the dynamics of monetary policy transmission through bank-nonbank interactions: (i) like banks,theyoriginatecredittofirms,butunlikebanks,theydonothaveaccesstodeposits, andinsteadpartiallyrelyonbankcreditlinestofinancelending,effectivelyextendingthe intermediationchain. (ii)BDCsmustdisclosedetailedportfolioinformationquarterly,allowingustomergeBDCinvestmentdatawithregulatorybankloandataandtracethefull credit chain. To our knowledge, thisis the first paper tostudy the credit flowfrom banks toBDCsandultimatelytofirms,anditsimplicationsformonetarypolicytransmission. We begin our analysis using the Federal Reserve’s supervisory Y-14 dataset, which provides detailed loan-level data on bank loans to both private and publicly listed U.S. firms.3 We document several novel facts about banks’ financing of BDCs. First, BDCs’ 1Weuse‘privatecredit’,‘privatedebt’,and‘directlending’interchangeablyforloansoriginatedandheld bynonbanklenders. Thegrowthofprivatecreditlikelystemsfromtighterbankregulation,anexpansion ofprivateequity, anddemandforflexibleloanproducts(Block, Jang, KaplanandSchulze,2024;Ereland Inozemtsev,2022). 2BDCsareclosed-endinvestmentfundswithover$310billionintotalassetsasof2023,makingthema significantcomponentofthenonbanklendingsector. 3The Y-14 dataset is an administrative, matched bank-firm-loan level dataset collected by the Federal 1

reliance on bank credit has grown significantly, more than doubling after the 2022 monetary tightening cycle compared to pre-2021 levels. Second, nearly 90% of bank lending to BDCs takes the form of credit lines, which are typically larger and offer greater creditor protection than loans to non-BDC borrowers. Finally, the market for bank lending to BDCsappearsconcentrated. Examining both banks and BDCs as credit sources, we document the evolution of aggregate credit volume and borrowing costs for nonfinancial businesses during the 2022 monetary tightening cycle.4 Two key patterns emerge: First, as bank credit to nonfinancial businesses slowed, lending to BDCs accelerated. BDCs maintained high utilization oftheirbankcreditlinesandincreasedlendingtononfinancialbusinesses,mitigatingthe aggregatecreditsupplycontraction. Second,bankschargedBDCshigherratepremiums, whichBDCspassedontoborrowers. Consequently,theweightedaverageinterestrateon combinedbankandBDCcreditsignificantlyexceededthatofbankcreditalone. Thisdual effect reveals that while BDCs help sustain credit volume during tightening, they simultaneouslyamplifymonetarytransmissionbyincreasingborrowingcostsfornonfinancial businesses. To quantify these patterns, we run regressions to estimate how bank loans to BDCs respond to monetary tightening relative to other loans. We control for credit risk using granular internal credit ratings from banks, comparing loans of the same rating, issued by the same bank, at the same time. During the 2022 monetary tightening cycle, banks increased lending to BDCs relative to non-BDC borrowers, with loan commitments to BDCsgrowing1.1percentagepointsmore,andtheircredit-lineutilizationrising18.6percentage points more than that of non-BDC borrowers. Banks also charged BDCs significantly higher rates, with the interest rate premium reaching 1.1 percentage points. These findings remain qualitatively robust across various monetary policy measures, includ- Reservesince2011aspartoftheDodd-FrankActStressTests. ItoffersthemostdetailedcoverageofU.S. firmswithbankloans,includingloancharacteristics,creditriskmetrics,andborrowerfinancials. 4Wefocusonthe2022cycleduetoitsunprecedentedspeedandmagnitudeofratehikesandtheresultingslowdowninbankcreditgrowth. Unlikethegradual2015tightening,whereratesrose225basispoints over three years amid continued loan expansion, the 2022 cycle saw a 525 basis point increase in just 18 months, triggering a sharp deceleration in bank credit supply (see the Federal Reserve’s H.8 data). This likely increased reliance on private credit, while liquidity pressures and deposit outflows made secured lendingtoBDCsmoreattractive. 2

ing changes in the effective Federal Funds rate and monetary policy shocks identified by Jarocin´ski and Karadi (2020) and Bauer and Swanson (2023). The simultaneous increase in both quantity and price of bank loans to BDC borrowers suggests heightened credit demandfromBDCs. A key channel for banks reallocating credit to BDCs during monetary tightening is the renegotiation of existing credit lines. We find that BDC borrowers actively renegotiate for higher commitments, with loan commitments rising 4.7 percentage points more than those of other borrowers on credit lines with limit expansions. BDC borrowers also increase credit line drawdowns sharply, particularly on expanded lines. These findings highlightcoordinationbetweenbanksandBDCborrowers: asBDCsdrawmorefromexisting credit lines, banks accommodate increased demand by raising credit limits on the most utilized loans, underscoring the role of renegotiation in credit reallocation during monetarycontractions. We find profitability—not risk-taking—drives banks’ reallocation of credit to BDCs through two key channels. First, during monetary tightening, BDC loans offer higher returns with lower risk due to greater collateralization, seniority, and lower loss given default. Second, banks may benefit from lower capital requirements on senior collateralized credit facilities to BDCs, enhancing the appeal of BDC lending over direct corporate lending,especiallyduringtightening.5 ToexamineBDCs’lendingstrategyduringmonetarytightening,wefocusonoverlapping borrowers—firms that hold both bank loans and BDC credit. Using a Khwaja and Mian(2008)-styleidentificationstrategy,whichcomparesloanswithinborrower,quarter, and loan type, we find BDC loans carry a rate premium of nearly 1.5 percentage points during tightening. This premium persists after controlling for borrower risk and loan characteristics including seniority, maturity, and performance. These results suggest that BDCs pass through higher bank fundingcosts to borrowers, amplifying monetary policy transmission via the price channel. Notably, firms increase BDC credit utilization dur- 5For example, banks that use internal estimates for risk-based capital requirements may benefit from lower loss given default on senior secured loans, provided that the underlying collateral meets certain eligibilitycriteria;seeBankforInternationalSettlements. Thisbenefitariseswhenthecapitalrequirements calculatedunderbanks’internalmodelsaremorebindingthanthoseunderthestandardizedapproach. 3

ing tightening, suggesting rising demand for private credit may drive BDCs’ increased relianceonbankfinancing. Despite higher borrowing costs, firms increase their demand for BDC credit during monetary tightening for two main reasons. First, BDCs absorb residual credit demand left unmet by bank rationing. We find that firms with higher bank loan utilization or shorterbankrelationships—proxiesforbankborrowingconstraint—becomesignificantly more reliant on BDC financing, particularly as monetary policy tightens. Second, BDC loans often include PIK provisions, allowing borrowers to defer interest payments. Such flexibility is especially valuable to borrowers during monetary tightening as the impact ofhigherratesmaterializesandcreditburdensbuildup. WenextexaminehowbanklendingtoBDCsinfluencesmonetarypolicytransmission along the intermediation chain. Specifically, we test whether BDCs’ reliance on bank financingaffectstheirresponsetomonetarytightening,andfindthat,indeed,morebankreliant BDCs exhibit stronger responses to tightening in both loan supply and pricing. During the 2022 tightening cycle, these BDCs expanded lending more aggressively and raisedinterestrateshigherthantheirlessbank-reliantcounterparts. Thispatternsupports apass-throughmechanism: bankspassonhigherfundingcoststoBDCs,whothenadjust loanpricingwhilemaintainingorevenexpandingcreditsupply. Finally,weexaminehowBDCs’creditsupplyduringmonetarytighteningaffectsfirmleveloutcomes,exploringthequantity-pricetradeoff. Wefindthatborrowersmorereliant on BDC credit exhibit greater capital expenditures and asset growth during monetary tightening, but also experience lower profitability and weaker interest coverage ratios. Additionally, firms that substitute BDC credit for bank debt in 2023 exhibit similar patterns: higher growth but increased leverage and default risk. These findings reveal the realconsequencesoftheprivatecreditintermediationchain: whileBDClendingsupports firm investments and growth during monetary tightening, it raises financial distress risk throughhigherborrowingcosts. Taken together, our findings reveal a nuanced impact of bank lending to nonbank lenders during monetary tightening. While this expansion mitigates aggregate contraction in credit supply, it amplifies monetary policy transmission by elevating borrowing 4

costs. This highlights a key tradeoff: private credit dampens the quantity channel by sustaining lending volume, yet intensifies transmission through the price channel. Such tradeoffhasimportantimplicationsonrealoutcomesofnonfinancialbusinesses. ContributiontotheLiterature. Ourpapercontributestoresearchonthebanklending channel of monetary policy, which has primarily focused on how banks’ direct lending to the corporate sector shapes policy transmission.6 We expand this work by showing thatbanksalsoadjustlendingtononbanklenders—suchasBDCs—whichinturnsupply credit to firms. Under this indirect credit supply mechanism, monetary policy affects aggregate credit not only through direct bank lending but also via shifts in credit allocation betweenbanksandnonbanks. Motivated by the post-GFC rise of nonbank lenders (Buchak, Matvos, Piskorski and Seru, 2018), recent work examines their role in monetary policy transmission (Elliott, Meisenzahl,PeydróandTurner,2019;Xiao,2020;Agarwal,Hu,RomanandZheng,2023; Elliott, Meisenzahl and Peydró, 2024; Cucic and Gorea, 2024). A key finding in this emergingliteratureisthatnonbanksattenuatetheimpactofmonetarytighteningbyproviding more credit when banks pull back. We refine this view by highlighting a pricequantitytradeoff: althoughnonbanksdampenthequantitychannelbymaintaininglending, they amplify the price channel by passing on higher borrowing costs. Our paper is thefirsttostudyhowcreditflowsfrombankstononbanks—andthentofinalborrowers— shapemonetarypolicytransmission. Weshowthatbanks’financingofnonbanksmatters, and that nonbanks dependent on bank funding exhibit stronger responses to monetary shocks.7 Finally,wecontributetothegrowingliteratureonprivatecreditanddirectlenders,focusingonBDCs,whicharelargeplayersinthismarket. Whilepriorworkexaminesdirect 6See, for example, Bernanke and Blinder (1988, 1992); Kashyap, Stein and Wilcox (1993); Jiménez, Ongena,PeydróandSaurina(2014);BernankeandGertler(1995);KashyapandStein(2000);Jiménez,Ongena, PeydróandSaurina(2012);BeckerandIvashina(2014);Drechsler,SavovandSchnabl(2017). 7Focusing on the mortgage market, Jiang (2023); Jiang, Matvos, Piskorski and Seru (2023); Agarwal et al.(2023)useshadowbank"callreports"andfindthatnonbanksoperatinginthissectorprimarilyrelyon short-termdebt. Severalrecentstudiesalsodocumenttheriseofbanklendingtononbankintermediaries like us (Acharya, Gopal, Jager and Steffen, 2024a; Gopal and Schnabl, 2022; Jiang, 2023; Javadekar and Bhardwaj, 2024; Acharya, Cetorelli and Tuckman, 2024b), but they do not examine the implications for monetarypolicytransmission. 5

lenders’ credit provision and its real effect, market discipline, lending terms, monitoring ability, and investment strategies (Davydiuk, Marchuk and Rosen, 2020a,b; Chernenko, ErelandPrilmeier,2022;Jang,2025;Blocketal.,2024;Chernenko,IalentiandScharfstein, 2024; Haque, Mayer and Stefanescu, 2024; Davydiuk, Erel, Jiang and Marchuk, 2024), less is known about how BDCs finance their lending, especially during tightening cycles. We address this gap by showing that bank credit lines are central to BDCs’ funding, and that BDCs actively renegotiate with banks to expand credit line limits in times of monetary tightening, reinforcing their role in monetary transmission. Related to our paper, Chernenko et al. (2024) argue that banks prefer lending to BDCs instead of direct middle-market lending because loans to BDCs are over-collateralized and thus require lower regulatory capital. We extend this view by showing that banks find lending to direct lenders particularly attractive during monetary tightening, as they can pass on interest rate increases more to BDCs than to non-BDC borrowers. This practice not only increasesprofitabilitybutalsobenefitsfromlowerloss-given-defaultrates. 2 Data and Empirical Facts We primarily utilize the Federal Reserve’s administrative matched bank-firm loan-level dataset for bank loan information and Refinitiv’s BDC Collateral dataset for BDC financials and investments. These datasets enable us to comprehensively track bank lending to various borrowers, including corporate entities and nonbank lenders like BDCs, as well as the subsequent credit allocation by BDCs to firms. All variables are defined in AppendixA.1. 2.1 Data Sources Matched Bank-Firm Loan Data from Federal Reserve’s Y-14. Our primary source is the Federal Reserve’s FR Y-14Q H.1 schedule on commercial loans (commonly referred to as the Y-14 data).8 This dataset covers detailed information on the universe of bilat- 8FordetailsonvariablescontainedinscheduleH.1andhowbanksarerequiredtoreportinformationto theFederalReserve,seethetablebeginningonpage170inthepubliclyavailablereportingform. 6

eral and syndicated loan facilities over $1 million in committed amounts held by Bank Holding Companies (BHCs) that are subject to the Federal Reserve’s Stress Tests.9 These reporting banks hold over 85% of total assets in the U.S. banking sector (Caglio, Darst and Kalemli-Özcan, 2021) and account for roughly 70–75% of all Commercial & Industrial(C&I)lending(Minoiu,ZarutskieandZlate,2021). The Y-14 data offers a granular view of loan contracting across a wide spectrum of firms on a quarterly basis. Besides committed and utilized loan amounts for each lending facility, the dataset captures key loan-level attributes, such as interest rates, spreads, maturity, priority in bankruptcy, collateral, and ex-ante estimates of loss given default (LGD), loan-type (e.g., credit line or term loan). The origination date allows us to separate new loans from existing ones each quarter. Banks also report financial, accounting, and balance sheet information for their borrowers over time annually. Additionally, we observe borrower-level risk measures, including internal credit ratings and time-varying defaultprobabilities,whichenableustocontrolforborrowerriskandruleoutrisk-based explanationsfordifferencesininterestratesbetweenloanstoBDCsandnon-BDCs.10 Our analysis primarily relies on quarterly loan-level data and annual borrower-level financials.11 Although reporting began in 2011Q3, we start our sample in 2012Q3 when coverage of banks improved significantly, and also to allow for a phase-in period for the structure of the collection and variables to stabilize. Appendix A.2 details our data cleaningandfiltrationprocedures. BDC Data from BDC Collateral. We use Refinitiv (LSEG)’s BDC Collateral dataset, whichcompilesmandatorySECfilingsintoaquarterlypanel,tocollectdataonallpublic and private BDCs and their portfolio investments from 2012Q3 to 2023Q4. The dataset covers190uniqueBDCs,ensuringbroadrepresentationofthesector. BecauseBDCsmust 9Aloanfacilityisalendingarrangementbetweenabankandaborrowerthatmayencompassmultiple loansofdifferenttypes, suchascreditlinesortermloans. Bankscategorizethefacilitytypebasedonthe loantypethatrepresentsthemajorityofthetotalcommitmentamount(Greenwald,KrainerandPaul,2024). 10Prior studies have shown that banks’ internal credit assessments are highly informative of borrower riskandfirmcharacteristics,andarestrongpredictorsofex-postdefault(WeitznerandHowes,2023;Lee, Li,MeisenzahlandSicilian,2019). 11Borrower-levelfinancialdataareavailableforapproximately60%offirmsinthedataset,withreporting beingmorefrequentamonglargerfirms. 7

reporttheirholdingstotheSEC,thedatasetisfreefromselectionbiasduetonon-random missing data. The dataset also assembles BDCs’ financial data from their SEC filings, includingforprivateBDCsnotcoveredbyCompustat. While BDCs invest in debt, equity, and structured products, their portfolios predominantlycomprisedebtinvestments. Forthesedebtinvestments,theBDCcollateraldataset provides detailed loan-level information, including borrower details (name and industry), contractual terms (par amount, interest rates, seniority, and loan type), and nonaccrual status. The dataset includes three interest rate measures: all-in yield, cash spread over the base rate, and Payment-in-Kind (PIK) spread—the latter accruing to the loan principalinsteadofbeingpaidincash.12 BDCreportsclassifythreetypesofloanseniority— first lien, second lien, and subordinated—and provide performance metrics such as fair value and non-accrual status (i.e., whether a loan is nonperforming). To complement this data, we hand-collect unique Taxpayer Identification Number (TIN) for our comprehensive list of BDCs from SEC filings. Our data cleaning and filtration procedures are detailedinAppendixA.3. Representativeness of BDC Collateral. To assess the representativeness of BDC borrowers in our dataset, we compare key variables across multiple private credit datasets. WereferenceJang(2025)’sproprietarydatabase,whichcoversasignificantshareofloans extended by both BDCs and private credit funds and is representative of private credit borrowersinPitchBook. AsshowninAppendixA.3,oursamplealignswithJang(2025)’s data in terms of the prevalence of first-lien loans and average loan interest rates, suggesting comparable representativeness.13 Furthermore, our sample’s distribution of loan amounts and spreads closely matches other studies, including Davydiuk et al. (2020a), who uses hand-collected data on BDCs, and Haque et al. (2024), who uses Pitchbook data. Wefindnoevidenceofsystematicdifferences—particularlyincreditrisk—between BDCCollateralborrowersandthosestudiedinotherprivatecreditresearch. Giventhese 12Unlike the Y-14 data, however, BDC Collateral does not contain financial information on BDCs’ borrowers(investees),asBDCsarenotrequiredbytheSECtodisclosesuchdetails. 13Loantypeandinterestratesaretheonlyvariablesdirectlycomparableacrossbothdatasets. Sinceboth datasetsrelyoninvestorholdingsdataratherthanissuancedata,theyprovideonlypartialcoverageofloan issuancedates,limitingourabilitytomeasureloanmaturityaccurately. 8

similarities, we believe our conclusions from examining BDC data likely extend to other typesofprivatecreditfunds. Matching BDCs to Y-14. We match individual BDCs that borrow directly from banks to the Y-14 data primarily using their TIN. This method identifies 133 BDCs, with an additional 9 matched using the “Fedmatch” algorithm from Cohen, Dice, Friedrichs, Gupta,Hayes,Kitschelt,Lee,Marsh,Mislang,Shatonetal.(2021),whichemploysstringmatching and probabilistic record linkage methods. This brings the total to 142 BDCs identified as borrowers from banks in the Y-14 data. Our matched sample includes 56 publicBDCs(80%ofallpublicBDCs)and86privateBDCs(74%ofallprivateBDCs),coveringapproximately75%ofallBDCsinoursampleand90%indollar-weightedterms.14 Matching BDC Borrowers to Y-14. To identify overlapping borrowers—firms that simultaneouslyholdbothbankloansandBDCcredit—wematchBDCborrowerstoborrowers intheY-14dataonaquarterlybasis. ThismatchingprocessutilizestheCohenetal.(2021) “Fedmatch” algorithm based on borrower name and industry. We then manually verify eachmatchforaccuracy. Monetary Policy Measures. Our primary focus is on the 2022 monetary tightening cycle, notable for its unprecedented speed and magnitude of rate hikes, and the accompanying slowdown in bank credit growth. To measure the stance of U.S. monetary policy, we primarily use two metrics: (1) a dummy variable for the 2022 tightening cycle (2022Q1–2023Q4)—including 2023Q4 despite rate hikes ending in July 2023, as rates remainedelevated—and(2)changesintheeffectiveFederalFundsrate. Recognizing that the Federal Funds rate is endogenous to broader economic conditions affecting both credit demand and supply, we also incorporate monetary policy shocks for robustness. Specifically, we use the shocks identified by Jarocin´ski and Karadi 14OurcarefulexaminationofnonbankfinancialintermediariesborrowingfromtheY-14banksconfirms that the unmatched BDCs had no outstanding commitments from Y-14 banks during our sample period. ManualverificationofSECcreditagreementsfurtherrevealsthatnearlythree-quartersoftheseBDCshave outstandingloansfromnon-Y-14banks(e.g., INGCapitalorNatixis); someothersappeartostrategically operatewithoutbankdebt,asindicatedbynamessuchas“[Redacted]UnleveredCorpBDC.” 9

(2020), which isolate unexpected monetary policy shifts using high-frequency changes in short-term interest rate derivatives around FOMC announcements. Additionally, we employ an updated version of the Bauer and Swanson (2023) measure, which similarly capturesunexpectedpolicychangesduringFOMCmeetings. 2.2 Empirical Facts Bank Lending to BDCs. Bank lending to BDCs saw steady growth throughout most of our sample period (Figure 1). A rapid increase began in 2021 and continued through the 2022 monetary tightening. Total bank loan commitments more than doubled since 2021, surpassing $60 billion. Table 1 reports loan-level summary statistics for key variables in our analysis, distinguishing between loans to BDCs (Panel A) and non-BDCs (Panel B). Below,wehighlightseveralkeycharacteristicsofbankloanstoBDCs. Bank loans to BDCs are significantly larger than those to non-BDC borrowers. The average committed loan to a BDC is approximately $90 million, while the median is $50 million—about 7 times and 14 times the respective sizes of loans to non-BDC borrowers. A similar pattern holds for the utilized loan amount. While interest rates are comparable across these two groups, a notable nuance emerges when examining the time series. Figure2plotstheweightedaverageinterestratesofloanstoBDCsandnon-BDCs,wherethe weights are the utilized loan amounts. Notably, BDC loans generally carry higher rates duringperiodsofmonetarytighteningrelativetoloanstonon-BDCs. More than 70% of bank loans to BDCs—and nearly 90% in dollar-weighted terms, as shown in Figure 1—are credit lines.15 These proportions are significantly higher than for loans to non-BDCs. Credit lines allow borrowers to draw funds up to a precommitted amount at a predetermined spread, enabling them to navigate adverse changes in aggregatelendingconditionsthroughunusedcreditlinecapacity. Notably,BDCsexhibithigher utilization rates of credit lines compared to non-BDCs reflecting their heavy reliance on creditlines. BankloanstoBDCstendtohaveshortermaturities. Bank loans to BDCs generally offer greater protection to creditors. Examining banks’ 15Acharyaetal.(2024a)alsodocumentthatbanksprovidecreditlinestononbanks,buttheirfocusdiffers fromours,concentratingonREITsandbanks’riskexposuretotheCREmarket. 10

ownex-anteestimatesoflossgivendefault(LGD),wefindthattheaverageLGDforBDC loansisabout10percentlowerthanfornon-BDCloans. Thisdifferencelikelystemsfrom the fact that BDC loans are more frequently collateralized, with banks holding first-lien, seniorsecuredpositionsinbankruptcy,asshowninTable1. Finally,themarketforbanklendingtoBDCsishighlyconcentrated,withonlyasubset of banks engaging in this specialized activity. On average, BDCs maintain borrowing relationships with 5.5 banks. At any given time, slightly over half of the banks in our sample participate in BDC lending. The distribution of this lending is markedly skewed: the top 5 banks account for 63% of the total committed loan amount to BDCs, while the top 10 banks represent about 84%. This level of concentration surpasses that observed in non-BDClending. BDCInvestmentsandFinancing. OursampleincludesallBDCs,whicharerequiredto fileSEC10-K/10-Qreportsdetailingtheirportfolioholdings. Asof2023Q4,BDCscollectivelyhold$318billionintotalassetsand$301billionintotalinvestments. BDCsprimarily lend to middle-market firms, which account for a third of private-sector GDP.16 Table 2presentssummarystatisticsforBDCloanportfolios. Theaverageloansizeis$11.31million with a maturity of about four years. BDC loans carry high interest rates, with an averageall-in-yieldof9.38%andaninterestspreadof7.19%,reflectingtheirfocusonriskier borrowerswhilealsoofferingflexibilityandrelationshiplendingbenefitssuchasthePIK provisions (Block et al., 2024; Jang, 2025). On average, about 9% of borrower-quarters exercise the PIK option to delay interest payments (12.5% weighted by loan amount), withprevalencesurgingduringperiodsofmarketstress,suchasCOVID-19andthe2022 monetarytightening(Figure3). While BDCs utilize both bonds and loans for debt financing, bank funding has been a critical source, and BDCs’ reliance on bank funding has grown significantly.17 Of the 190 BDCs in our sample, 142 borrowed from banks during our sample period and thus 16Middle-market firms, with annual sales between $10 million and $1 billion, comprise nearly 200,000 U.S.businesses;seehere. 17The average BDC leverage (Debt/Asset) is 0.4 in our sample. As shown in Figure A.1, leverage has steadily increased, partly driven by the Small Business Credit Availability Act (SBCAA) passed in March 2018(BallochandGonzalez-Uribe,2021). 11

appear in the Y-14 data. These bank-reliant BDCs dominate the credit market, providing fundingtoaround11,500firms—mostlyprivateentities—andaccountingforover90%of totalBDClending(AppendixFigureA.2). The increasing dependence on bank funding is evident in several metrics. First, the average ratio of Y-14 bank loan commitments to total debt for BDCs has risen over time, with notable growth during the 2022 monetary tightening cycle (upper panel, Figure 4). Second, interest expenses have become increasingly tied to bank loans, with spikes observed in 2022 during monetary tightening (lower panel, Figure 4). In addition, during the 2022 tightening, average net bank debt issuance was more than twice the magnitude ofnetequityissuanceandalmostsixtimesthatofnetbondissuance(untabulated).18 TwofactorscouldexplainBDCs’growingrelianceonbankfunding. First,bankfunding, particularly credit lines, offers readily available capital without delay when investment opportunities arise, making it the preferred source of financing for investments.19 Ouruntabulatedanalysissupportsthis,showingthatBDCs’investmentelasticity(marginal propensity to invest) with respect to bank debt is 0.995—indicating that BDCs use bank loans almost dollar for dollar to support their investments. This is higher than the elasticity for bonds (0.935) and equity (0.891). Second, the increasing dependence on bank financing across the BDC sector coincides with the rapid growth of private BDCs (Figure A.3). Private BDCs tend to rely more heavily on bank funding, consistent with their higher information asymmetry, necessitating more informationally-sensitive loans from banks(DiamondandDybvig,1983;DiamondandRajan,2001). Aggregate Credit Volume and Borrowing Costs for Nonfinancial Businesses. To understand the effect of monetary policy transmission, we examine facts on the aggregate creditvolumeandborrowingcostsfornonfinancialbusinessesduringthe2022monetary tighteningcycle,consideringbothbanksandBDCsascreditsources. First,whilebankcredittononfinancialbusinessesslowedandevencontractedduring 18Net debt issuance is current book debt minus lagged book debt, scaled by lagged assets. Net equity issuance is current book equity minus lagged book equity minus current net income, scaled by lagged assets. 19Inthemortgagemarket,shadowbanksalsodependheavilyonwarehousecreditlinestofinanceloan originations(Jiang,2023). 12

the 2022 tightening, bank lending to BDCs accelerated, suggesting a credit reallocation. BDCsmaintainedhighutilizationoftheirbankcreditlinesandincreasedlendingtononfinancialbusinesses,boostingaggregatecreditvolume(upperpanel,Figure5). BDCloan volume during this period nearly tripled compared to pre-2022 levels. This pattern indicates that BDCs’ increased borrowing from banks and subsequent lending mitigated the aggregatecreditsupplycontractionduringtightening. Second, while interest rates for BDC and non-BDC loans were comparable before the tightening, rates on BDC loans rose more sharply during the 2022 tightening cycle—to an average of 6.1% versus 5.4% for non-BDC borrowers—suggesting banks charged a premium to these nonbank lenders (Figure 2). BDCs passing on these higher costs to borrowers amplifies monetary policy transmission. The weighted average interest rate on combined bank and BDC credit carries a significant premium over bank loans alone (lower panel, Figure 5), reaching 1.0% during the 2022 tightening cycle, up from an average of 0.4% pre-2022. Thus, while BDCs mitigate credit supply contraction, they simultaneouslyamplifymonetarytransmissionbyincreasingborrowingcostsfornonfinancial businesses. 3 Bank Lending to BDCs during Monetary Tightening Thissectionpresentsourregressionresultsonthefirstpartoftheintermediationchain— bankslendingtononbankdirectlenders. Usinggranularsupervisorybankloandata,we estimate how bank lending to BDC borrowers differed from lending to other borrowers during the 2022 monetary tightening cycle. We examine both the quantity and pricing of bank loans to BDCs relative to other borrowers and uncover the underlying mechanisms drivingthesedifferences. 3.1 Regression Framework We aggregate the quarterly loan-level data into a bank-borrower-quarter panel to capture total credit provision for each borrower-bank pair, as borrowers can have multiple 13

outstanding loans from the same bank. Loan amounts are summed across committed or utilized amounts, while borrowing costs are interest rate averages weighted by utilized amounts. Ourbaselineregressionmodelis: Y = α+β (BDC × MP)+β BDC +X +FE +ϵ , (1) i,b,t 1 i t 2 i i,t−1 b,t i,b,t where Y represents borrower (i)-bank (b)-quarter (t) level outcomes, including: (i) i,b,t quarterly growth rate of loan commitments, (ii) loan utilization rate for credit lines, (iii) interest rate, weighted by utilized amounts, and (iv) credit risk measures (seniority in bankruptcy,collateralization,lossgivendefault,andprobabilityofdefault). Thestanceof monetarypolicy(MP)ismeasuredbyadummyforthe2022tighteningcycle(Tightening ) t t ∆ and changes in the effective Fed Funds rate ( FF). BDC is a dummy variable indicatt i ing whether the borrower is a BDC. We include lagged borrower-level control variables (X ) and fixed effects (FE ) to account for observed and unobserved heterogeneity, i,t−1 b,t respectively. Borrower-Level Controls. To account for observable time-varying differences between borrowers, we include lagged borrower-level characteristics (X ) to capture i,t−1 credit risk, bank loan usage, and debt structure. Derived from Y-14 data, these variables include bank-estimated probability of default, expected loss given default, total bank debt, share of term loans in total bank debt, and share of credit lines in total bank debt. FixedEffects. Weleveragebank-internalcreditratingsfromY-14data,whichwerefer to as credit rating, to compare loans with nearly identical credit risk levels. These ratings are granular, borrower-specific, and bank-dependent, derived from individual banks’ internal risk assessment models.20 As these ratings generally reflect borrower characteristics such as leverage or size and are updated over time—where poor loan performance typically results in a downgrade—recent studies have shown that they are highly informativeaboutborrowercreditriskandloanoutcomes(WeitznerandHowes,2023;Haque, 20Banks in our sample typically have 10 to 15 rating buckets, though some employ even more detailed creditratingclassifications. 14

MayerandWang,2023;Claessens,OngenaandWang,2024). Ourspecificationincludesbank×creditrating×year-quarterfixedeffects,ensuringcomparison of loans made by the same bank, within the same quarter, to BDC and non-BDC borrowers with identical internal credit ratings. These fixed effects account for timevaryinglenderheterogeneity,suchasdifferencesinbanks’internalriskassessmentmodels or capital ratios, which can influence lending decisions (Irani, Iyer, Meisenzahl and Peydro,2021),andbyconstruction,absorbthedirecteffectofmonetarypolicy(MP). We t double cluster standard errors at bank×borrower and year-quarter levels, with the sample periodspanning2012Q3–2023Q4. CoefficientofInterest. Ourprimarycoefficientofinterest, β ,capturesthedifferential 1 response of bank lending to BDC borrowers compared to other borrowers during monetarytightening,intermsofbothloanquantityandpricing. Weexpect β tobepositivefor 1 loan growth or utilization outcomes if bank lending to BDCs expands during tightening, potentiallymitigatingbroadercreditcontraction. Similarly,weexpectβ tobepositivefor 1 interest rate outcomes if bank lending to BDCs raises borrowing costs during tightening, amplifyingmonetarypolicytransmissionthroughtheinterestratechannel. 3.2 Baseline Results Ourresults,presentedinTable3,showthatbankssignificantlyincreasedlendingtoBDC borrowersandchargedthemhigherratesrelativetonon-BDCborrowersduringthe2022 monetary tightening cycle. Column (1) shows that loan commitments to BDCs grew by 1.1 percentage points more than those to other borrowers during the tightening cycle, withnosuchdifferencesinnon-tighteningperiods. Thiseffectiseconomicallysignificant, withitsmagnitudecomparabletothesamplemeanofloancommitmentgrowthforBDCs (Table1). Given that most bank loans to BDCs are credit lines, we examine utilization rates in Column (2). BDCsutilized credit lines 18.6 (=14.2+4.4) percentage pointsmore than non- BDCborrowersduringthetighteningcycle,asignificantincreasefromthe4.4percentage point difference in normal periods. This effect is economically large, considering the av- 15

eragecreditlineutilizationrateisaround50%. Column (3) shows that banks charged BDC borrowers higher interest rates, primarily driven by the 2022 tightening. The interest rate spread between BDC and non-BDC borrowerswidenedby1.1(=0.9+0.2)percentagepoints,aneconomicallymeaningfulincrease representing 25% of the unconditional mean of bank loan rates to BDCs (Table 1). This ratepremiumresultsinanadditionalannualloanexpenseof$0.3billionforBDCborrowers, accounting for 15% of their total bank loan expenses.21 Our strategy, incorporating lagged borrower credit risk measures (probability of default, expected LGD) as controls and granular bank-internal credit ratings as fixed effects, ensures comparison of loans withnearlyidenticalcreditrisklevels,rulingoutcreditriskdifferencesasanexplanation fortheratepremium. ∆ Columns (4)–(6) confirm these findings using changes in the Fed Funds rate ( FF) as t analternativepolicymeasure. Toaddresspotentialendogeneityconcernswithmonetary policy, we conduct robustness tests using monetary policy shocks from Jarocin´ski and Karadi(2020)andBauerandSwanson(2023)inSection6. Together, these findings indicate that during monetary tightening, banks reallocated credit toward BDCs, indirectly supporting credit supply while raising borrowing costs. The simultaneous increase in both quantity and price of bank credit for BDC borrowers suggestsheighteneddemandfromBDCs. 3.3 Renegotiation Through Credit Line Expansions Credit provision through credit lines involves active decision-making and renegotiation, with banks setting credit limits and borrowers deciding how much to draw. We next investigate whether banks shift credit to BDCs through renegotiation of existing loans or originationofnewloans. In Panel A of Table 4, we separate credit lines into three groups: pre-existing credit lines (Column 1), pre-existing credit lines with limit expansions (Column 2), and newly originated credit lines (Column 3). Our findings indicate that BDC borrowers predomi- 21By2023Q4,totalutilizedbankloansbyBDCsreached$27.24billion,withtotalbankloanexpensesof $2.05billion. 16

nantly obtain additional credit through renegotiation. Column (2) shows that loan commitments to BDCs grew by 3.6 (=4.7-1.1) percentage points more than those to other borrowers for pre-existing credit lines with limit expansions. These patterns hold when usingchangesintheFedFundsrateasameasureofmonetarypolicy.22 Thesepatternsalign withpriorresearchonloancontracting,showingthatbankloansarefrequentlyrenegotiatedas borrowersseekto adjustloantermsin responsetoupdated informationoncredit qualityandinvestmentopportunities(RobertsandSufi,2009;DenisandWang,2014). We then examine BDC borrowers’ credit utilization during monetary tightening, focusing on which loan types saw the greatest increase in drawdowns. Panel B of Table 4 examines the growth rate of utilized credit line amounts for all credit lines (Column 1), pre-existing credit lines (Column 2), and pre-existing credit lines with limit expansions (Column 3). BDC borrowers significantly increased credit line drawdowns during monetary tightening, with a substantial portion of the increase coming from credit lines that underwentlimitexpansions(Column3). Overall,Table4suggeststhatBDCborrowersprimarilyobtainadditionalcreditthrough renegotiation of existing credit lines, reflecting a coordinated interplay between banks and borrowers. BDC borrowers draw more from existing credit lines, while banks respond by increasing credit limits on the most utilized loans. Combined with the simultaneous rise in both quantity and price of bank credit for BDC borrowers, these results suggest that this trend is largely driven by heightened demand from BDCs seeking profitableinvestmentopportunities. 3.4 What Drives Banks’ Reallocation of Credit to BDCs? What drives banks’ reallocation of credit to BDCs? And does this credit expansion to BDCs, coupled with higher loan rates, reflect increased risk-taking by banks? Our analysis suggests that banks’ increased lending to BDCs is primarily driven by profitability ratherthanrisk-taking. First, loans to BDCs provide attractive returns relative to their credit rating without 22In fact, Column (6) shows that as the Fed Funds rate rises, new loan origination to BDCs declines relativetootherborrowers,reinforcingtheroleofcreditlineexpansionsincreditallocation. 17

exposing banks to additional credit risk. Although these loans carry higher interest rates (Table 3), banks appear to face lower credit risk. Table 5 reveals that, during monetary tightening,loanstoBDCsaremorelikelytobefirst-lienseniorsecuredandcollateralized, granting banks priority over borrower assets in the event of default. Moreover, these loans exhibit significantly lower loss given default (LGD) with no significant difference inex-antedefaultprobabilitiescomparedwithloanstootherborrowers. The rate premium with favorable risk profile is consistent with banks’ market power in the BDC bank lending market, as documented in Section 2.2. The highly concentrated market for bank lending to BDCs potentially confers upstream market power to banks. The limited number of banks engaged in BDC lending likely contributes to higher costs for BDCs accessing bank credit, similar to the mechanism Jiang (2023) finds in the mortgage market. This specialized market structure enhances banks’ ability to balance profitabilityandriskintheirBDClendingportfolios. Second,banksmaybenefitfromlowerfundingcostsduetofavorablecapitaltreatment for senior collateralized credit facilities extended to BDCs, particularly during monetary tightening. AsshowninTable5,loanstoBDCsaremorelikelytobebackedbycollateral, especially during the 2022 tightening cycle. Although banks are generally not capitalconstrained,issuingcollateralizedloansoffersaregulatoryadvantage,assuchloansmay carrylowercapitalrequirements,aligningwiththeargumentsinChernenkoetal.(2024). ThisfeaturefurtherenhancestheappealoflendingtoBDCsoverdirectcorporatelending. 4 BDC Lending during Monetary Tightening Having established that banks shift credit to BDCs and charge them higher interest rates during monetary tightening, we now examine the second part of the intermediation chain—BDC lending to firms. We aim to determine whether BDCs, in turn, charge their borrowers higher interest rates than banks, and if this effect intensifies during monetary tightening. IfBDCsfaceafundingcostpremiumfrombanksundercontractionarypolicy, weexpecttoobserveacorrespondingpremiuminBDClendingrates. 18

4.1 Khwaja-Mian Regressions on Overlapping Borrowers A key challenge in identifying this pass-through effect is that BDCs may select riskier borrowersunder-servedbybanks(Blocketal.,2024),makingitdifficulttoisolatewhether higher BDC lending rates stem from elevated funding costs or greater borrower risk. For example, Elliott et al. (2019) show that nonbanks expand credit supply during monetary contractionsbyincreasingrisk-taking,lendingtoborrowerswithhigherdefaultrisk. To address this concern, we identify "Overlapping Borrowers"—firms that simultaneously hold both bank loan commitments and BDC credit—by merging Y-14 data with BDC Collateral data. We employ a regression framework similar to Khwaja and Mian (2008) and Chodorow-Reich (2014). Our sample includes about 4,800 overlapping borrowers, with their numbers increasing in recent years, particularly during the 2022 monetary tightening cycle (Figure A.4). Appendix Table A.15 shows that overlapping borrowers tend to be larger, more leveraged, and have lower interest coverage ratios, cash holdings, and tangible assets compared to non-overlapping borrowers. These borrowers predominantly hold credit lines from banks and term loans from BDCs. Since borrowers mayincreasinglyprefermorepermanentformsoffinancing—suchastermloans—during periodsoftightening,itisimportanttocontrolforloantypeinouranalysis. To test whether BDC loans carry higher interest rates and amounts than bank loans— even within the same borrower, quarter, and loan type—we construct a loan-quarter panel dataset stacking bank- and BDC-originated loans to overlapping borrowers. We estimatethefollowingregression: Y = α+β (BDC × MP)+β BDC +X +FE +ϵ (2) l,t 1 l t 2 l l,t i,t,z l,t where Y represents loan (l)-quarter(t) level interest rate and loan amount for a given l,t loan l of type z at year-quarter t extended to borrower i. Loan type z includes credit line, term loan, or other forms of lending. BDC is a dummy variable equal 1 for BDC loans l and0forbankloans. MP capturesthestanceofmonetarypolicy. t We include borrower×year-quarter×loan type fixed effects (FE ) to control for timei,t,z varying borrower characteristics and systematic differences across loan types. This spec- 19

ification ensures comparison of BDC and bank loans within the same borrower, quarter, and loan type, isolating differences in pricing or amounts from borrower risk and demand variations. Loan-level controls (X ) include an indicator for non-accruing status, l,t loan maturity, and utilized amounts (when the dependent variable is interest rate, and viceversa). Standarderrorsaredouble-clusteredattheborrowerandyear-quarterlevel. PanelAofTable6presentstheresults. Column(1)confirmsthatBDC-providedloans carry significantly higher interest rates than bank loans, with a premium of 0.9 percentage points. The positive and statistically significant coefficient on BDC×Tightening indicates an even higher premium of 1.47 percentage points during monetary tightening. Column (2) further controls for loan seniority with fixed effects for first-lien senior secured, second-lien senior secured, and junior/unsecured debt, with the coefficient on BDC × Tightening remaining positive, stable, and significant. Given our strict fixed effects, this result cannot be attributed to borrower risk differences, loan type variations, or loan seniority. Overall, our results suggest that BDC lending rates increase more than bankratesduringmonetarytightening. Columns(5)and(6)examineloanamounts.23 Wefindthatfirmsincreasetheirutilization of BDC loans relative to bank loans during tightening. This suggests rising demand forprivatecredit,promptingBDCstoexpandtheirloansupplyandseekadditionalfundingfrombanks. Economic Significance The point estimates from Tables 3 and 6 allow us to assess the magnitudeofrateamplificationalongthebank-BDCandBDC-firmsegmentsoftheintermediationchain. AsnotedinSection3.2,bankschargeBDCsanadditional1.1percentage pointsininterestduringtightening. BDCs,inturn,passona1.47(=0.521+0.949)percentage point premium to firms (Table 6, Column 1). These magnitudes are sizable relative to prior studies. For instance, Erel, Liebersohn, Yannelis and Earnest (2023) find that online (fintech) banks charged borrowers 0.4 to 1.5 percentage points more than traditional banks during the 2022 tightening cycle, depending on loan type. Furthermore, as we show in Section 5.2, borrowers heavily reliant on BDC credit experience a 30% lower 23We use utilized amounts rather than commitments because BDC collateral reports only utilized amounts,eventhoughY-14providesboth. 20

interestcoverageratioduringtighteningrelativetothesamplemean. 4.2 Why Do Borrowers Prefer Private Credit over Bank Credit? TheincreasedutilizationofBDCloansathigherinterestratescomparedwithbankcredit suggests rising demand for private credit during monetary tightening. But why do borrowersdemandmoreBDCloansdespitetheirhighercosts? Weprovideevidencefortwo potentialexplanations. First, some borrowers turn to BDCs because they face constraints in securing additional bank credit, particularly when lending standards tighten.24 To test this, we reestimate Eq. (2) by dividing overlapping borrowers into those likely constrained in bank lending and those that are not, based on two measures. The first measure, borrowing capacity, considers a borrower-quarter bank loan constrained if the lagged utilization rate of bank loans exceeds the 75th percentile of the sample distribution, indicating nearly exhausted bank borrowing capacity. The second measure, relationship duration, deems a borrower-quarter bank loan constrained if the duration of its longest bank relationship is shorter than the sample mean across all bank-borrower pairs, suggesting less benefit from relationship banking (Petersen and Rajan, 1994). Bank loan-constrained borrowers face more difficulty in accessing further bank credit and may have to seek alternative financingathighercosts. Theresults,presentedinPanelsBandCofTable6,showthatour findings are more pronounced for bank loan-constrained borrowers, consistent with the banklendingconstraintsmechanism. Second, a significant benefit of BDC loans is the option to delay interest payments through PIK provisions. The use of PIK options tend to rise during monetary tightening, as the impact of higher rates materializes and credit burdens build up among borrowers (Figure 3). The flexibility offered by PIK provisions becomes increasingly valuable to borrowersduringchallengingfinancialconditions. Supportingthisnotion,Table7shows that borrowers with non-accruing loans are more likely to invoke PIK provisions during 24AccordingtotheSeniorLoanOfficerOpinionSurveyonBankLendingPractices(SLOOS),banklending standardsbegantighteningin2022Q3and,by2023,hadreachedlevelslastseenduringtheGlobalFinancial CrisisandtheCOVID-19pandemic. 21

monetary tightening episodes. This flexibility may explain part of the appeal of BDC loans, despite their typically higher interest rates compared to traditional bank loans.25 Additionaltotheabovereasons,BDCloansmayalsoofferotherbenefitsthataredifficult to quantify, such as faster loan approval, customized covenant structures tailored to borrowers’ needs, greater flexibility in renegotiation, and more stable relationships (Block et al.,2024;Jang,2025;DegerliandMonin,2024). 5 Pass-Through Along the Intermediation Chain This section examines how bank lending to BDCs influences monetary policy transmission along the intermediation chain. We test whether BDCs’ credit supply responses to monetary tightening depend on their reliance on bank funding, and how BDCs’ credit supplyshapesfirm-leveloutcomes. 5.1 BDCs’ Bank Reliance and Monetary Pass-Through If monetary policy transmits from banks to BDCs and subsequently to firms, we expect more bank-dependent BDCs to exhibit stronger responses to tighter policy. Specifically during the 2022 tightening cycle, we anticipate these BDCs to increase their loan supply whileraisingborrowingcosts,relativetolessbank-reliantBDCs. To test this, we leverage our unique dataset that merges granular Y-14 data on bank loans to BDCs with BDCs’ deal-level investment records and their financials. This enablesustotracethefullintermediationchain,providinganidealsettingtoexaminehow monetarypolicypropagatesthroughbanks’financingofBDCs. We measure a BDC’s reliance on bank financing using BankLoanExpenseShare , the i,t share of interest payments on bank loan over total interest expenses for BDC i in quarter t. This share has increased over time (Figure 4), with higher values reflecting greater 25SeethisFitchRatingsarticleontherisingtrendsofPIKfeaturesinprivatecredit. 22

relianceonbankcredit. Weestimatethefollowingmodel: Y = α+β (BankLoanExpenseShare ×MP)+β BankLoanExpenseShare +X +FE +ϵ , i,j,t 1 i,t t 2 i,t i,j,t b,t i,j,t (3) where Y represents BDC (i)-loan (j)-quarter (t) level outcomes, including loan amount i,j,t and interest rate. MP captures the stance of monetary policy. Controls X include total t i,j,t BDC assets (to isolate credit supply effects from portfolio expansion), loan maturity, and non-accrual status (to account for credit risk). To address unobservable heterogeneity, weincludeBDCfixedeffectstocontrolforBDC-specificpreferencesforcertainborrower types; year-quarter fixed effects to absorb time-varying macroeconomic conditions that couldimpactBDClending;andloan-typefixedeffectstoaccountforriskvariationsacross loan structures. Our coefficient of interest is β , which measures whether more bank- 1 reliantBDCsexhibitamplifiedlendingresponsestomonetarypolicychanges. Table 8 presents the results. Columns (1)–(2) examine loan amount, while columns (3)–(4)examineinterestrate. Thecoefficientestimatefor BankLoanExpenseShare × MP i,t t isconsistentlypositiveandstatisticallysignificantatthe1%levelacrossallcolumns,suggesting that higher reliance on bank loans is associated with greater loan amounts and higherinterestratesduringthe2022monetarytightening.26 Theestimatedeffectiseconomicallysignificant. ABDCwithaone-standard-deviation higher BankLoanExpenseShare (0.48) expands loan supply by 20.5% (0.48 × 0.428) and raises borrowing costs by 15 basis points (0.48×0.313) more than in non-tightening periods. This pattern aligns with a pass-through mechanism: as banks’ funding costs rise, they pass them to BDCs, who in turn reprice their loans while increasing lending where ratesremainattractive. One potential concern is that BDCs’ reliance on bank financing may be endogenous totheircharacteristics,suchasaccesstoalternativefundingsourcesorinvestmentstrategies. However, the persistence of BDCs’ bank reliance over time suggests that these re- 26Across all columns, the coefficient on BankLoanExpenseShare is negative and statistically significant, suggestingthat,undernormalconditions(i.e.,withnohikesintheFedfundrate),greaterrelianceonbank debtisassociatedwithsmallerloanamounts(Columns1–3)andlowerinterestrates(Columns4–6). This suggests bank-dependent BDCs may pursue more conservative lending strategies or favor stable, lowerriskpricing. 23

lationships are relatively stable and unlikely to be driven by short-term credit decisions. Additionally,relianceonbankfundingattheintensivemarginislikelyshapedbymarketwide conditions, particularly BDCs’ ability to substitute between bonds and loans—a factorwecontrolforusingyear-quarterfixedeffects. Tofurthervalidateourfindings,we conductrobustnesschecksusingalternativemeasuresofBDCs’bankreliance(Section6), as well as additional tests incorporating monetary policy shocks and alternative control variables. Ourresultsremainconsistentacrossallspecifications. Insum,ourevidencesuggeststhatBDCfundingstructureplaysacriticalroleinmonetarypolicytransmission. Duringthe2022tightening,morebank-reliantBDCsresponded moresharplytopolicychanges—expandinglendingmorewhilecharginghigherinterest rates. These findings underscore the importance of nonbank lenders and their funding structuresinshapingmonetarypolicytransmission. 5.2 BDC Credit and Real Effects on Firms Finally we examine how BDCs’ credit supply during monetary tightening shapes firmlevel outcomes, exploring the quantity-price tradeoff. Building on our previous findings that private credit dampens the quantity channel by maintaining lending and amplifies monetary tightening through the price channel, we investigate the net effect on firm performance and financial health. We expect firms more dependent on BDCs to invest and growmorerapidlywhilefacinghigherinterestexpensesrelativetocashflows,potentially amplifyingfinancialdistress. Firm-LevelRegressions Weconstructanunbalancedfirm-yearpanelusingannualborrowerlevel financials reported in Y-14 data.27 We measure a borrower’s BDC reliance using the ratiooftotalBDCcredittototalfirm-leveldebtandcreatea High BDC Relianceindicator forfirmsabovethesamplemean. Appendix Table A.15 compares overlapping borrowers with high and low BDC re- 27To maximize sample size, we relax our definition of overlapping borrowers by including those with BDCcreditinyeartandbankcreditineitheryeartort+1,asborrowers’financialsaretypicallyreported withalag. Resultsarerobusttoalternativeconstructions,andweprovideexternalvalidityusinganother datasetthatincludesprivatedebtborrowers(withorwithoutbankdebt)inSection6. 24

liance to non-overlapping borrowers (those in Y-14 that do not hold BDC credit). Highreliance firms are smaller in asset size and more likely to have negative ROA than their low-reliancecounterparts, consistentwithmiddle-market firmswithnegative cashflows depending more on nonbank loans (Chernenko et al., 2022; Haque et al., 2024). Overlappingborrowersgenerallyhavefewertangibleassetsthannon-overlappingborrowers, potentiallylimitingtheiraccesstobankcredit. Tounderstandtherealeffectsatthefirmlevel,weestimate: Y = α+β (HighBDCReliance ×MP)+β HighBDCReliance +X +FE +ϵ , i,t 1 i,t t 2 i,t i,t−1 j,t i,t (4) where Y represents firm(i)-year(t) outcomes: capital expenditure ratio, asset growth, i,t sales growth, ROA, and interest coverage ratio. MP captures monetary policy stance. t We include lagged firm-level controls X to account for observed firm heterogeneity i,t−1 and3-digitNAICS × Yearfixedeffectstoabsorbtime-varyingindustry-levelconditions. Table9reportsourresults. FirmsheavilyreliantonBDCcreditshowedgreatercapital expenditures during tightening, with the estimated 1.9% difference being economically significant given the 2.0% sample mean (Table A.15). Consistent with this pattern, these firms also exhibited substantially higher asset growth during tightening. We find no significant differences in sales growth, possibly due to lagged effect on sales. However, thesefirmsshowedlowerprofitability,suggestingincreasedcapitalinvestmentandasset growthdidnotimmediatelyboostearnings. Notably,firmsheavilyreliantonBDCcredit experienced a 30% lower interest coverage ratio during tightening relative to the sample mean,implyinghigherinterestexpensesandweakeningdebtservicecapacity.28 These results confirm that the quantity-price tradeoff from private credit provision creates economically meaningful real consequences at the firm level: BDC-reliant borrowers trade off greater investment and asset growth with higher financial distress from moreexpensiveBDCloans. 28UnreportedtestsconfirmhigherratiosofinterestexpensetolaggedsalesduringtighteningforBDCreliantborrowers,suggestingthedeclineinICRisindeeddrivenbyinterestexpenses. 25

WhichfirmsswitchfrombanktoBDCcredit? Whileourfirm-levelregressionsusethe High BDC RelianceindicatortoidentifyfirmsheavilyreliantonBDCcredit,thismeasure doesn’tnecessarilycaptureborrowersactivelysubstitutingBDCcreditforbankdebtduringtightening. Toexaminetherealeffectsofsuchsubstitution,weanalyzecross-sectional variationinborrowers’debtdynamicsfrom2022Q4to2023Q4,whentheeffectsofmonetarytighteningaremorepronounced. Wecategorizeborrowersintothreegroups: Switchers, who experienced negative growth in bank loan commitments but positive growth in BDC loans; Credit-Squeezed borrowers, who saw negative growth in both bank and BDC loans;andBank-Favoredborrowers,whoexperiencedbankcreditexpansion. Table 10 presents sample means for each group and reveals several key patterns.29 In terms of size and leverage, Switchers are smaller than Bank-Favored borrowers but larger than Credit-Squeezed borrowers. Switchers exhibit higher leverage (Debt/EBITDA of 5.52) compared to Credit-Squeezed borrowers (4.69), consistent with increased access to BDC credit. Regarding growth and investment, Switchers show substantially higher asset growth (45%) compared to both Credit-Squeezed (12%) and Bank-Favored (16%) firms, aligning with our earlier findings that BDC-reliant firms invest and grow more rapidly. Switchers also exhibit higher sales growth. However, despite this higher growth, Switchers show lower ROA (0.07) compared to Bank-Favored firms (0.09), suggesting stronger growthdidnotimmediatelyboostearnings. Notably, the growth of Switchers appears to come with weaker debt service capacity and financial distress risk. They have lower interest coverage ratios (1.65) compared to Credit-Squeezed(2.04)andBank-Favored(2.48)firms,consistentwithourregressionresults. Additionally,Switchershavehigherdefaultprobability(0.09)thanothergroups(0.08and 0.05)andlowercashholding,indicatingthatBDCsarefundingriskierborrowers. These patterns reinforce our earlier findings: BDC credit supports firm growth but imposes higher borrowing costs on borrowers during tightening. This growth has not translated into immediate profitability and appears to come with heightened risk of financialdistress. 29Given the small sample size due to our stringent definitions and focus on a single year, we report descriptivestatisticsratherthanregressionresults,acknowledgingthelimitationsofthisapproach. 26

6 Robustness This section presents a series of robustness tests to validate our findings and rule out alternativeexplanations. Robustness with Monetary Policy Shocks In our baseline analysis, we measure the stance of U.S. monetary policy using a dummy variable for the 2022 tightening cycle (2022Q1–2023Q4) and changes in the effective Federal Funds rate. However, both measures are endogenous to broader economic conditions that influence credit demand and supply. To address potential endogeneity concerns, we assess the robustness of our resultsusingmonetarypolicyshocksthatisolateunexpectedchangesinpolicy. Specifically, we incorporate the high-frequency shocks identified by Jarocin´ski and Karadi (2020), which use short-term interest rate derivative movements around FOMC announcementstoisolateunanticipatedmonetarypolicyshifts. Additionally,weemploy an updated version of the Bauer and Swanson (2023) shock measure, which similarly focuses on unexpected policy changes during FOMC meetings. Appendix Tables A.1– A.4confirmthatourmainresultsarelargelyrobustwiththesemonetarypolicyshocks. Other Monetary Policy Tightening Cycles Our economic narrative primarily focuses on the 2022 monetary tightening due to its unprecedented speed, the magnitude of rate hikes,andtheaccompanyingslowdowninbankcreditgrowth. Theonlyothertightening cycle within our sample period (2012Q3–2023Q4) is the 2015–2018 cycle, which spanned from 2015Q4 to 2018Q4. To test the generalizability of our findings, we conducted an exerciseusingadummyvariableforthe2015tighteningcycle. Appendix Table A.10 shows that the effects are largely insignificant. This suggests thatourproposedeconomicmechanismdependsonbothsharpratehikesandsignificant tighteninginbanklending. Incontrast,the2015–2018cyclefeaturedagradual225-basispoint increase over three years alongside continued loan expansion, with little evidence of tightening by banks. Without a contraction in bank credit, borrowers had no strong incentivetoturntoBDCs,preventingthemechanismdocumentedinourpaperfrommaterializing. This exercise underscores the necessary conditions for our proposed mech- 27

anism: a substantial monetary shock combined with a meaningful contraction in bank credit. Including Bank-Borrower Fixed Effects. Our baseline specification in Tables 3–5 includes bank×credit rating×year-quarter fixed effects, which effectively ensures that we compare loans made by the same bank, within the same quarter, to BDC and non-BDC borrowers with the same internal credit rating. These granular fixed effects account for borrower credit risk as well as any time-varying heterogeneity across lenders. However, concerns could still arise regarding the potential endogeneity of the bank-borrower match,whereunobservedtime-invariantrelationshipsbetweenbank-borrowerpairscould drive certain observed differences between BDC loans and non-BDC loans. To alleviate this concern, we check the robustness of our results by additionally incorporating bank times borrower fixed effects to strip out unobservable differences across bank-borrower pairs. AppendixTablesA.5–A.6confirmthatourmainresultsremainrobust. Alternative Definition of Non-BDCs. Our baseline analysis in Table 3 classifies Y-14 loan-leveldatabyborrowersintoBDCsandnon-BDCloanstoexaminehowbanklending to BDC borrowers differed from lending to other firms during the 2022 monetary tightening cycle. To ensure robustness, we test alternative definitions of non-BDC borrowers. Appendix Table A.7 confirms that our findings are little changed when we restrict the non-BDC sample to non-financial firms, excluding all BDCs and all Y-14 borrowers with a 3-digit NAICS code of 521 (Monetary Authorities-Central Bank) or 522 (Credit Intermediation and Related Activities). This restriction isolates our results from potential distortionsarisingfrombanklendingtoothernonbanks. RobustnessAcrossLoanTypes. Weconducttworobustnessteststoassesswhetherour findingsholdacrossdifferentloantypes. First, since credit line utilization can differ significantly from other loan types, our baseline estimates of utilization rates (Columns (2) and (5) in Table 3) focus on credit lines, which constitute the majority of bank loans to BDCs. We now extend our analysis 28

toincludeallloantypes. AppendixTableA.8confirmsthat,acrossvariousspecifications, BDC loan utilization remains higher than that of non-BDCs during monetary tightening, supportingourmainresults. Second, we extend our analysis in Table 5, which demonstrates lower credit risk for banks on BDC loans, to address potential concerns about loan type heterogeneity. To isolatetheeffectfromdifferencesbetweencreditlinesandtermloans,wenowrestrictthe analysis to credit lines only. As shown in Appendix Table A.9, our results remain robust underthisrestriction. Notably,someestimates,particularlyforlossgivendefault,become even more pronounced. This analysis strengthens our conclusion that BDCs effectively mitigate banks’ credit risk through enhanced collateralization and higher debt priority, enablingthemtosecurefundingevenduringperiodsofmonetarytightening. Sub-Sample Analysis for Public and Private BDCs. In Section 2.2, we showed that privatelyheldBDCshavedrivenmuchoftherecentgrowthindirectlending,coinciding with their increased reliance on bank loans. This raises the concern that differences in bankfinancingreliancebetweenprivateandpublicBDCsmaybeinfluencingourresults. Funding sources differ between private and public BDCs. Public BDCs raise capital primarilyfromretailinvestorsthroughbondandequityissuance,whileprivateBDCs—more akin to traditional private credit funds—depend on committed capital from high-networth individuals and institutional investors.30 With limited access to capital markets, privateBDCsrelymoreonbankcreditlinestoseizeinvestmentopportunities. Feestructures may also impact bank financing demand. Public BDCs typically charge higher performancefeesthanprivateBDCs(Turner,2019). Thus,despitehavingaccesstodrypowder, private BDCs may use bank credit lines strategically—not just to fund investments buttoenhanceperformance—dependingoncashflowtiming(AlbertusandDenes,2024). To test whether our baseline findings on bank lending to BDCs (Tables 3 and 5) hold 30A key advantage of public BDCs over private BDCs or traditional private credit funds is their ability todiversifyfundingsourcesbyincorporatingretailcapitalwhileenablingmanagerstochargehigherfees (Turner, 2019). More reputable fund managers are more likely to adopt the BDC structure (Jang, 2025). Indeed,manyprivateBDCstransitiontopublicstatusthroughanIPOasmanagersestablishatrackrecord ormergewithanexistingpublicBDC(O’Shea,BrownandWathen,2024). Forexample,MSCIncomeFund announced its IPO in January 2025, while Golub Capital BDC, Inc., a public BDC, merged with Golub CapitalBDC3,Inc. onJune2024,withtheformerasthesurvivingentity. 29

for both private and public BDCs, we re-estimate Eq. (1) on split samples. Appendix ∆ Tables A.11 (using Tightening as the monetary stance measure) and A.12 (using FF) t present the results. Across specifications, our key findings remain largely consistent for bothBDCtypes. WhilecoefficientestimatesinTableA.12suggestslightlystrongereffects for private BDCs, the overall patterns confirm the robustness of our results. Importantly, since private BDCs closely resemble traditional private credit funds, the robustness of ourfindingsforprivateBDCssuggestspotentialexternalvalidityforthebroaderprivate creditmarket. Alternative Controls and Measures for BDCs’ Reliance on Banks. In Section 5, we measure a BDC’s reliance on bank financing using BankLoanExpenseShare, the share of interest payments on bank loans relative to total interest expenses. To alleviate potential endogeneity concerns, we conduct robustness checks using alternative measures of bank reliance. Appendix Table A.13 reports results from re-estimating Eq. (3) with two alternative definitions of bank reliance: (1) High Bank Reliant (Bank Loan Ratio) is a dummy variable equal to 1 if a BDC’s utilized bank loan to total debt ratio is in the top quartile of the sample distribution. (2) High Bank Reliant (Utilization Rate) is a dummy variable equal 1 if a BDC’s bank loan utilization rate is in the top quartile of the sample distribution, indicating heavy credit line drawdowns. Across both definitions, our key findings on BDCloanamountsandinterestratesduringtighteningremainunchanged. Since BDC characteristics evolve over time, we further test an alternative model incorporating additional BDC-level controls in estimating Eq. (3). Specifically, we include: total BDC assets, BDC leverage, BDC net equity issuance as a share of total assets, bank loan commitment as a share of BDC’s total debt, and utilized bank loans as a share of BDC’stotaldebt. Theseadditionalcontrolshelpaccountfortime-varyingcharacteristics, such as funding structure and bank reliance. For example, including BDC leverage and net equity issuance helps control for fluctuations in equity financing. Appendix Table A.14confirmsthatourresultsremainrobustunderthesealternativespecifications. 30

Robustness with Khwaja-Mian Fixed Effect Specification To ensure that our findings in Section 4.1 based on the Khwaja-Mian triple fixed effect specification are not driven by highly specific variation, we investigate whether the results are robust to a more lenient fixed effects specification. In untabulated results, we confirm that our results hold under borrower × loan type and time fixed effects. Furthermore, our findings remain unchanged when controlling for fixed vs. floating rate loans, as well as a dummy variable indicating whether the base rate is tied to LIBOR, SOFR, PRIME, or another index. This robustness check helps address potential concerns about the generalizability of our resultsacrossdifferentloancharacteristicsandmarketconditions. External Validity of Real Effects: Beyond Overlapping Borrowers Our results in Section 5.2 are based on borrowers holding both bank loans and BDC credit, allowing us to observe their financials in Y-14. To provide external validity, we utilize a proprietary database from Jang (2025) that covers detailed borrower financials and includes loans from both BDCs and other private credit funds. This dataset allows us to assess the generalizabilityofourfindingsbeyondBDCsandborrowerswithbankloans,extendingour analysistoabroaderspectrumoftheprivatecreditmarket.31 We leverage a unique feature of the data that identifies the lead lender for each loan based on actual credit agreements.32 Private credit-originated loans (lead lender being privatecreditmanager)typicallyhaveminimalbankinvolvement(medianbankshareof 0%),comparedtosignificantbankparticipationinbank-originatedloans(93%). Weusea PrivateCredit indicator for borrowers of private credit-originated loans to identify firms exclusively borrowing from private credit managers. These borrowers are substantially smaller, more leveraged, and have fewer tangible assets than Y-14 borrowers (Panel of TableA.16). OurregressionresultsinTableA.16showsthatfirmsexclusivelyborrowingfromprivatecreditmanagersexperiencestrongergrowthinassetsandsalesduringthe2022tight- 31This data is sourced from an anonymous third-party valuation firm that provides loan appraisals for privatecreditmanagers. 32Abouttwo-thirdsoftheloanswereoriginatedbyprivatecreditmanagers,withtheremainderconsistingofbank-originatedloanssyndicatedtothem. 31

ening cycle, with no significant differences in ROA. This aligns with our earlier findings, suggesting that private credit reliance supported firm growth but did not immediately translateintohigherprofitability.33 7 Conclusion Thispaperoffersnewevidenceonhowbanks’financingofnonbanklendersshapesmonetary policy transmission. Our paper makes several contributions. First, by merging supervisory bank loan-level data with deal-level private credit data, we trace—for the first time—the flow of credit from banks to BDCs and ultimately to firms. We show that during monetary tightening, banks reallocate lending to BDCs by expanding credit line limits through renegotiations, indirectly supporting credit supply. However, because banks chargeBDCshigherinterestrates—rateswhicharethenpassedontoendborrowers—this intermediation chain raises borrowing costs and have important real effects on firms. In other words, while the extension of the credit chain mitigates the contraction in credit supply,italsoamplifiesthepricechannelofmonetarypolicy. Second, by directly observing individual bank loans to BDCs, we offer the first indepth look at the rapidly growing segment of bank loans to private credit funds. Our detailed data reveal why banks shift lending toward private debt lenders during monetary tightening. Specifically, these loans command higher interest rates yet exhibit lower loss-given-default—reflecting strong collateralization and seniority. This combination of increased profitability and lower risk underscores a key incentive behind the expanding bank–nonbanknexus. Overall, our findings underscore how connectivity between banks and nonbanks influencesmonetarypolicytransmissionandrealoutcomesofnonfinancialbusinesses. Althoughnonbanksattenuatethecontractionaryeffectsoftighteningbymaintainingcredit provision, higher borrowing costs mean that monetary policy still transmits effectively through the price channel. As nonbank lending continues to expand, these results pro- 33Wedonotfindsignificantresultsoncapitalinvestmentsorinterestcoverage,possiblyduetotherelativelylowercoverageofthesevariablesinthedata. 32

vide important insight on how future policy changes might propagate through increasinglycomplexintermediationchains. Looking ahead, our study suggests several avenues for further research. First, while wefocusonaperiodofpronouncedmonetarytighteningin2022,exploringwhetherthese transmission channels behave symmetrically during easing cycles would offer a more complete picture of the broader macroeconomic implications. Second, investigating how heterogeneity among nonbanks—such as varying funding structures, risk profiles, and regulatory frameworks—shapes their role in monetary policy transmission could yield important policy insights. As nonbank lending grows and evolves, understanding these dynamicswillbeessentialforbothresearchersandpolicymakers. 33

References Acharya, Viral V, Manasa Gopal, Maximilian Jager, and Sascha Steffen, “Shadow Always Touchesthe Feet: Implications of Bank CreditLines to Non-BankFinancial Intermediaries,”AvailableatSSRN4847858,2024. , Nicola Cetorelli, and Bruce Tuckman, “Where Do Banks End and NBFIs Begin?,” TechnicalReport,NationalBureauofEconomicResearch2024. Agarwal,Isha,MalinHu,RalucaA.Roman,andKelingZheng,“LendingbyServicing: MonetaryPolicyTransmissionthroughShadowBanks,”WorkingPaper2023. Albertus,JamesFandMatthewDenes,“Privateequityfunddebt: agencycostsandcash flowmanagement,”AvailableatSSRN3410076,2024. Balloch, Cynthia and Juanita Gonzalez-Uribe, “Leverage Limits in Good and Bad Times,”TechnicalReport,LSEWorkingPaper2021. Bauer, Michael D. and Eric T. Swanson, “A Reassessment of Monetary Policy Surprises andHigh-FrequencyIdentification,”NBERMacroeconomicsAnnual,2023,37,87–155. Becker, Bo and Victoria Ivashina, “Cyclicality of credit supply: Firm level evidence,” JournalofMonetaryEconomics,2014,62,76–93. Bernanke,BenSandAlanSBlinder,“Credit,money,andaggregatedemand,”1988. Bernanke, Ben S. and Mark Gertler, “Inside the Black Box: The Credit Channel of MonetaryPolicyTransmission,”JournalofEconomicPerspectives,1995,9(4),27–48. Bernanke,BenSandAlanSBlinder,“ofMonetaryTransmission,”TheAmericanEconomic Review,1992,82(4),901–921. Block, Joern, Young Soo Jang, Steven N Kaplan, and Anna Schulze, “A survey of privatedebtfunds,”ReviewofCorporateFinanceStudies,2024,13(2),335–383. Buchak, Greg, Gregor Matvos, Tomasz Piskorski, and Amit Seru, “Fintech, regulatory arbitrage, and the rise of shadow banks,” Journal of financial economics, 2018, 130 (3), 453–483. Caglio,CeciliaR,RMatthewDarst,andebnemKalemli-Özcan,“Risk-takingandmonetary policy transmission: Evidence from loans to smes and large firms,” Technical Report,NationalBureauofEconomicResearch2021. Chernenko,Sergey,IsilErel,andRobertPrilmeier,“Whydofirmsborrowdirectlyfrom nonbanks?,”TheReviewofFinancialStudies,2022,35(11),4902–4947. , Robert Ialenti, and David Scharfstein, “Bank Capital and the Growth of Private Credit,”TechnicalReport,HarvardUniversityWorkingPaper2024. Chodorow-Reich, Gabriel, “The employment effects of credit market disruptions: Firmlevelevidencefromthe2008–9financialcrisis,”TheQuarterlyJournalofEconomics,2014, 129(1),1–59. 34

, Olivier Darmouni, Stephan Luck, and Matthew Plosser, “Bank liquidity provision acrossthefirmsizedistribution,”JournalofFinancialEconomics,2022,144(3),908–932. € Claessens,Stijn,StevenOngena,andTengWang,“â œIfYouDon’tKnowMebyNow... € € â Banksâ ™ Private Information and Relationship Length,” Technical Report, CEPR DiscussionPapers2024. Cohen, Gregory J, Jacob Dice, Melanie Friedrichs, Kamran Gupta, William Hayes, IsabelKitschelt,SeungJungLee,WBlakeMarsh,NathanMislang,MayaShatonetal., “The USsyndicated loan market: Matching data,”Journal ofFinancial Research, 2021,44 (4),695–723. Cucic, Dominic and Denis Gorea, “Nonbank lending and the transmission of monetary policy,”TechnicalReport,DanmarksNationalbankWorkingPapers2024. Davydiuk, Tetiana, Isil Erel, Wei Jiang, and Tatyana Marchuk, “Common Investors Across the Capital Structure: Private Debt Funds as Dual Holders,” Working Paper Fisher College of Business WP No. 2024-21, Fisher College of Business; European CorporateGovernanceInstitute(ECGI)September302024. , Tatyana Marchuk, and Samuel Rosen, “Direct lenders in the US middle market,” AvailableatSSRN3568718,2020. , ,and ,“MarketDisciplineintheDirectLendingSpace,”AvailableatSSRN3729530, 2020. Degerli,AhmetandPhillipMonin,“PrivateCreditGrowthandMonetaryPolicyTransmission,”FEDSNotes,August2024. Denis, David J and Jing Wang, “Debt covenant renegotiations and creditor control rights,”JournalofFinancialEconomics,2014,113(3),348–367. Diamond, Douglas W and Philip H Dybvig, “Bank runs, deposit insurance, and liquidity,”Journalofpoliticaleconomy,1983,91(3),401–419. and Raghuram G Rajan, “Liquidity risk, liquidity creation, and financial fragility: A theoryofbanking,”JournalofpoliticalEconomy,2001,109(2),287–327. Drechsler,Itamar,AlexiSavov,andPhilippSchnabl,“Thedepositschannelofmonetary policy,”TheQuarterlyJournalofEconomics,2017,132(4),1819–1876. Elliott, David, Ralf Meisenzahl, José-Luis Peydró, and Bryce C Turner, “Nonbanks, banks, and monetary policy: US loan-level evidence since the 1990s,” in “Proceedings ofParisDecember2019FinanceMeetingEUROFIDAI-ESSEC”2019. , Ralf R Meisenzahl, and José-Luis Peydró, “Nonbank lenders as global shock absorbers: evidence from US monetary policy spillovers,” Journal of International Economics,2024,p.103908. Erel, Isil and Eduard Inozemtsev, “Evolution of Debt Financing Toward Less Regulated FinancialIntermediaries,”WorkingPaper,2022. 35

, Jack Liebersohn, Constantine Yannelis, and Samuel Earnest, “Monetary policy transmission through online banks,” Technical Report, National Bureau of Economic Research2023. Favara, Giovanni, Camelia Minoiu, and Ander Perez-Orive, “Zombie lending to us firms,”AvailableatSSRN4065886,2022. Gopal,ManasaandPhilippSchnabl,“Theriseoffinancecompaniesandfintechlenders insmallbusinesslending,”TheReviewofFinancialStudies,2022,35(11),4859–4901. Greenwald, Daniel L, John Krainer, and Pascal Paul, “The credit line channel,” Journal ofFinance,2024. Forthcoming. Haque, Sharjil, Simon Mayer, and Irina Stefanescu, “Private Debt versus Bank Debt in CorporateBorrowing,”AvailableatSSRN4821158,2024. , ,andTengWang,“HowPrivateEquityFuelsNon-BankLending,”AvailableatSSRN 4429521,2023. Irani, Rustom M, Rajkamal Iyer, Ralf R Meisenzahl, and Jose-Luis Peydro, “The rise of shadow banking: Evidence from capital regulation,” The Review of Financial Studies, 2021,34(5),2181–2235. Jang, Young Soo, “Are Direct Lenders More Like Banks or Arm’s-Length Investors?,” WorkingPaper,2025. Jarocin´ski, Marek and Peter Karadi, “Deconstructing Monetary Policy Surprises—The RoleofInformationShocks,”AmericanEconomicJournal: Macroeconomics,April2020,12 (2),1–43. Javadekar, Apoorva and Abhishek Bhardwaj, “How does bank lending to non-banks affectcreditallocationandsystemicrisk,”AvailableatSSRN,2024. Jiang, Erica, Gregor Matvos, Tomasz Piskorski, and Amit Seru, “Banking without Deposits: EvidencefromShadowBankCallReports,”WorkingPaper,NationalBureauof EconomicResearch2023. Jiang,EricaXuewei,“Financingcompetitors: Shadowbanks’fundingandmortgagemarketcompetition,”TheReviewofFinancialStudies,2023,36(10),3861–3905. Jiménez,Gabriel,StevenOngena,José-LuisPeydró,andJesúsSaurina,“Creditsupply and monetary policy: Identifying the bank balance-sheet channel with loan applications,”AmericanEconomicReview,2012,102(5),2301–2326. Jiménez, Gabriel, Steven Ongena, José-Luis Peydró, and Jesús Saurina, “Hazardous Times for Monetary Policy: What Do Twenty-Three Million Bank Loans Say about the EffectsofMonetaryPolicyonCreditRisk-Taking?,”Econometrica,2014,82(2),463–505. Kashyap, Anil K and Jeremy C Stein, “What do a million observations on banks say about the transmission of monetary policy?,” American Economic Review, 2000, 90 (3), 407–428. 36

Kashyap, Anil K., Jeremy C. Stein, and David W. Wilcox, “Monetary Policy and Credit Conditions: Evidence from the Composition of External Finance,” American Economic Review,1993,83(1),78–98. Khwaja, Asim Ijaz and Atif Mian, “Tracing the impact of bank liquidity shocks: Evidencefromanemergingmarket,”AmericanEconomicReview,2008,98(4),1413–1442. Lee,SeungJung,DanLi,RalfRMeisenzahl,andMartinJSicilian,“TheUSSyndicated TermLoanMarket: Whoholdswhatandwhen?,”2019. Minoiu, Camelia, Rebecca Zarutskie, and Andrei Zlate, “Motivating banks to lend? CreditspillovereffectsoftheMainStreetlendingprogram,”2021. O’Shea, Finian P., Michael C. Brown, and Jordan Wathen, “4Q24 BDC Scorecard: BeyondBDCs;PrivateCreditEyestheIntervalFund,”2024. Petersen, Mitchell and Raghuram Rajan, “The Benefits of Lending Relationships: EvidencefromSmallBusinessData,”JournalofFinance,1994,49(1),3–37. Roberts, Michael R and Amir Sufi, “Renegotiation of financial contracts: Evidence from privatecreditagreements,”JournalofFinancialEconomics,2009,93(2),159–184. Turner,David,“InvestorsaresoldonBDCfees,butit’sperformancethatcounts,”2019. Weitzner, Gregoryand CooperHowes, “Bankinformation productionover thebusiness cycle,”AvailableatSSRN3934049,2023. Xiao, Kairong, “Monetary Transmission through Shadow Banks,” The Review of Financial Studies,2020,33(6),2379–2420. 37

Figure1. The Rise of Bank Credit Line Lending to BDCs Notes: ThisfigureillustratesthevolumeandcompositionofbanklendingtoBDCs. Thedarkline withdiamondmarkersrepresentstheaggregatedollaramountofcommittedbankloans,whilethe graylinewithroundmarkersindicatestheaggregatedollaramountofcommittedbankcredit lines. Thebarsshowtheaggregatedollaramountofutilizedbankloans. Onaverage,the commitment-weightedshareofcreditlinesinbankloanstoBDCsis89%. Bankloansotherthan creditlinesprimarilyconsistoffrontingexposuresandtermloans. Thisfigureincludes142 uniquepublicandprivateBDCswithoutstandingbankcommitmentsand40banksoverthe sampleperiod2012Q3–2023Q4. 38

Figure2. Interest Rate on Bank Loans: BDCs versus non-BDCs Notes: ThisfigureplotstheaveragetimeseriesofinterestratesonbankloanstoBDCand non-BDCborrowers. Quarterlyaverageinterestratesareweightedbytheutilizedamount, aggregatedatthebank-borrowerpairlevel,andexpressedinpercentagepoints. Shadedarea representstherecentmonetarypolicytighteningcycle(2022Q1–2023Q4). Thesampleperiodis 2012Q3–2023Q4. 39

Figure3. Fraction of BDC loans with PIK spread Notes: ThisfigureplotsthefractionofBDCloanswithanon-zeropayment-in-kind(PIK)spread from2012Q3to2023Q4,weightedbyloanamount. PIKallowsborrowerstodeferinterest paymentsuntilmaturity,typicallyinexchangeforhigherinterestrates,providingborrowers withgreaterflexibilityincashflowmanagement. 40

Figure4. BDCs’ Financing: the increasing reliance on banks Notes: ThisfigureillustratesBDCs’increasingrelianceonbankloans. Theupperpanelplotsthe averageshareofbankloancommitmentsrelativetoBDC’stotaldebteachquarter. Thelower panelplotstheaverageBankLoanExpenseShare—theshareoftotalinterestexpenses attributabletointerestpaymentsonoutstandingbankloans—eachquarter. Thesampleperiodis 2012Q3–2023Q4,andtheanalysisincludes190BDCswithavailabledata. 41

Figure 5. Aggregate Credit Volume and Borrowing Costs for Nonfinancial Businesses Notes: Thisfigureillustratestheaggregatecreditprovisionandborrowingcostsfrombanksand BDCstononfinancialbusinesses. Theupperpanelplotsthetotalutilizedbankloans(black dashedline)andthecombinedtotalofutilizedbankandBDCloans(reddottedline),both expressedinbillionsofU.S.dollars. Thelowerpanelshowstheaverageinterestratesonbank loansandonthecombinedcreditfrombanksandBDCs,weightedbytheutilizedamountand expressedinpercentagepoints. Shadedarearepresentstherecentmonetarypolicytightening cycle(2022Q1–2023Q4). Nonfinancialbusiness4e2saredefinedasborrowersoutsidetheFinance andInsurancesector(i.e.,notin2-digitNAICScode52). Thesamplespans2012Q3–2023Q4 andincludesallY-14banksandBDCswithmatcheddeal-levelinformationfromBDCCollateral.

Table1. Summary Statistics: bank loans Count Mean SD Median PanelA:BankLoanstoBDCs CommittedLoanAmount(USDMn) 10,197 87.20 108.00 50.00 UtilizedLoanAmount(USDMn) 10,197 52.40 83.80 22.30 InterestRate(%) 10,197 4.32 2.13 3.73 CreditLineShare 10,197 0.72 0.41 0.99 TermLoanShare 10,197 0.02 0.12 0.00 UtilizationRate(onlycreditlines) 7,970 0.54 0.29 0.54 Maturity(Years) 10,197 5.71 2.53 5.00 ∆ Log(Loan) 7,232 0.01 0.31 0.00 1×(FirstLienSeniorSecured) 10,197 0.92 0.27 1.00 1×(Collateralized) 10,197 0.92 0.27 1.00 LossGivenDefault 7,809 0.29 0.16 0.30 ProbabilityofDefault 7,817 0.01 0.04 0.00 PanelB:BankLoanstoNon-BDCs CommittedLoanAmount(USDMn) 7,991,557 13.24 23.02 3.60 UtilizedLoanAmount(USDMn) 7,991,557 7.78 13.69 2.36 InterestRate(%) 7,991,557 3.89 1.89 3.57 CreditLineShare 7,991,557 0.33 0.45 0.00 TermLoanShare 7,991,557 0.31 0.44 0.00 UtilizationRate(onlycreditlines) 2,941,337 0.50 0.35 0.50 Maturity(Years) 7,056,000 7.57 5.41 5.50 ∆ Log(Loan) 5,290,219 -0.00 0.28 -0.00 1×(FirstLienSeniorSecured) 7,991,557 0.84 0.37 1.00 1×(Collateralized) 7,991,557 0.86 0.35 1.00 LossGivenDefault 5,762,070 0.33 0.19 0.33 ProbabilityofDefault 5,784,124 0.03 0.10 0.01 Notes: ThistablereportssummarystatisticsforbankloanstoBDCs(PanelA)andnon-BDCs (PanelB).Unlessotherwisestated,thedataisattheloan-year-quarterlevel,coveringthesample period2012Q3–2023Q4. CommittedLoanAmountisthereportedtotalloancommitmentina givencreditfacility. UtilizedLoanAmountisthereportedtotalloanutilizedamountinagiven creditfacility. InterestRateisthereportedinterestrateforaloan,expressedinpercentage points. CreditLineShareandTermLoanShareareborrower-timelevelaggregatesandreport thesharesofthatloantypeinagivenborrower’stotalbankdebt. UtilizationRateistheratioof utilizedtocommittedcreditandisdefinedonlyforcreditlines.Maturityisthedifferencebetween maturityandoriginationdateinyears. ∆ Log(Loan) isthelogchangeinloancommitmentat thebank-borrowerlevelinquarter t fromquarter t−1,expressedindecimal. FirstLienSenior Securedisadummyequalto1iftheborrowerpledgesafirstlienseniorsecuredclaimonagiven loan,and0iftheloanissecondlien,seniorunsecured,orcontractuallysubordinated. Collateralizedisadummyequalto1iftheborrowerpledgesanycollateralonagivenloan,and 0iftheloanisuncollateralized(unsecured). LossGivenDefaultistheexpectedloanlossrate upondefaultestimatedbythereportingbank. ProbabilityofDefaultisbanks’internal 43 estimatesofborrower’s1-yearaheadprobabilityofdefault.

Table2. Summary Statistics: BDCs and loan portfolio Count Mean SD Median PanelA:BDCLoan-level ParAmount(USDM) 460,192 11.31 27.65 3.95 Maturity(Years) 452,231 4.14 2.01 4.17 All-InYield(%) 438,545 9.38 2.89 9.36 CashSpread(%) 442,130 6.68 2.80 6.25 Cash+PIKSpread(%) 442,130 7.19 3.64 6.50 1×(NonaccrualLoan) 460,170 0.03 0.16 0.00 1× (BorrowerUsedPIKOption) 442,130 0.09 0.29 0.00 1×(FirstLienDebt) 460,192 0.83 0.37 1.00 1×(FixedRateLoan) 460,192 0.12 0.33 0.00 PanelB:BDC-level TotalAssets(USDM) 3,997 1479.04 3267.61 572.68 TotalDebt/TotalAssets 3,997 0.40 0.48 0.43 BankLoanCommitment/TotalAssets 3,997 0.14 0.33 0.06 BankLoanCommitment/TotalDebt 3,996 0.60 7.45 0.13 UtilizedBankLoan/TotalDebt 3,996 0.19 0.27 0.05 BankLoanExpenses 3,553 0.23 0.48 0.07 Notes: ThistablereportssummarystatisticsforBDCportfoliosaswellasBDCs’financials. Data isattheloan-quarterlevelforPanelAandBDC-quarterlevelforPanelB,coveringthesample period2012Q3–2023Q4. ParAmountisthereportedfacevalueoftheloan. Maturityisthe reportedmaturityoftheloanasoftheholdingdate(notoriginationdate)expressedinyears. All-in-Yieldisthereportedtotalinterestrateonagivenloan,expressedinpercentagepoints. CashSpreadisthestandardcreditspreadontheloanoverthebaserate. Cash+PIKSpread includestheadditionalspreadifagivenloanhasaPIKoption. I(FirstLienDebt)isanindicator variableequalto1iftheloanisafirstliendebtinvestment,and0otherwise. I(Non-Acrual)is anindicatorvariableequalto1iftheloanisnon-accruing,and0otherwise. I(BorrowerUsed PIKOption)isadummyvariableequaltooneifaborrowerexercisesthePIKoptionforagiven loaninagivenquarter. I(FixedRateLoan)isanindicatorvariableequalto1ifagivenloanis fixedinterestrate,and0otherwise. BankLoanExpenseistheshareoftotalinterestexpenses thatisattributabletointerestpaymentsonoutstandingbankloans. 44

Table3. Baseline Regressions: bank lending to BDCs vs. other borrowers ∆ ∆ LogLoan Utilization InterestRate LogLoan Utilization InterestRate (1) (2) (3) (4) (5) (6) BDC × Tightening 0.011 ∗∗ 0.142 ∗∗∗ 0.009 ∗∗∗ (0.004) (0.027) (0.002) BDC × ∆ FF 0.709 ∗ 9.372 ∗∗∗ 0.532 ∗∗∗ (0.400) (2.154) (0.112) ∗∗ ∗∗ ∗∗∗ ∗∗∗ BDC 0.001 0.044 0.002 0.003 0.071 0.004 (0.003) (0.019) (0.001) (0.003) (0.017) (0.001) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.286 0.480 0.028 0.286 0.480 N 3,653,826 1,712,362 3,468,670 3,653,826 1,712,362 3,468,670 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsregressionresultsfromEq. (1). Thesampleperiodis2012Q3–2023Q4,withdataatthe bank-borrower-quarterlevel. Tighteningisadummyequalto1forthe2022monetarytightening(2022Q1–2023Q4)and0otherwise. ∆ ∆ FF ischangesintheeffectivefederalfundsrate. BDC isadummyequalto1iftheborrowerisaBDC,and0otherwise. LogLoan isthelogchangeinloancommitmentbetweenbank b andborrower i intime t relativetotime t−1,andexpressedindecimal. Utilization istheratioofutilizedtocommittedbankcreditlines. InterestRate istheweightedaverageinterestrateacrossallutilized loansbetweenagiven b and i intime t,expressedindecimal. CreditRating isbank’sinternalcreditratings,whichisbank-specific, time-varying,andcapturesbank-estimatedex-antecreditrisk. Firmlevelcontrolsentertheregressionswithaone-periodlagand includebank-estimatedprobabilityofdefault,expectedlossgivendefault,shareoftermloansintotalbankdebt,shareofcreditlinesin totalbankdebt,andthenaturallogoftotalbankdebt. Standarderrorsaredoubleclusteredat bank×borrower andYearQtr level. 45

Table4. The Renegotiation Channel: panel A PanelA:IncreasesinCreditLineLimitsbyBanks ∆ Y : LogLoan (1) (2) (3) (4) (5) (6) i,b,t BDC × Tightening 0.012 ∗∗ 0.047 ∗∗∗ -0.622 (0.005) (0.013) (0.382) BDC × ∆ FF 0.810 ∗∗ 2.648 ∗∗∗ -28.915 ∗∗∗ (0.381) (0.925) (8.621) BDC 0.001 -0.011 0.483 0.004 0.002 0.159 (0.003) (0.008) (0.376) (0.003) (0.007) (0.198) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.046 0.256 0.204 0.046 0.256 0.204 CreditLineSample Existing LimitExpanded New Existing LimitExpanded New N 1,533,250 295,176 4,993 1,533,250 295,176 4,993 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThispanelreportsregressionresultsfromEq. (1). Thesampleperiodis2012Q3–2023Q4,withdataatthe bank-borrower-quarterlevel. Thesampleisrestrictedtocreditlinesonly. Columns(1)and(4)excludenewlyoriginatedcreditlines andexamineonlypre-existingcreditlines. Columns(2)and(5)restrictthesampletopre-existingcreditlinesconditionalonany positivechangeincreditlinecommitmentintime t relativeto t−1. Columns(3)and(6)focusonnewlyoriginatedcreditlinesonly. ∆ Tighteningisadummyequalto1forthe2022monetarytightening(2022Q1–2023Q4)and0otherwise. FF ischangesinthe ∆ effectivefederalfundsrate. BDC isandummyequalto1iftheborrowerisaBDC,and0otherwise. LogLoanisthelogchangein loancommitmentbetweenbank b andborrower i intime t relativetotime t−1,andexpressedindecimal. CreditRating isbank’s internalcreditratings,whichisbank-specific,time-varying,andcapturesbank-estimatedex-antecreditrisk. Firm-levelcontrolsenter theregressionswithaone-periodlagandincludebank-estimatedprobabilityofdefault,expectedlossgivendefault,shareoftermloans intotalbankdebt,shareofcreditlinesintotalbankdebt,andthenaturallogoftotalbankdebt. Standarderrorsaredoubleclusteredat bank×borrower andYearQtr level. 46

Table4. The Renegotiation Channel: panel B PanelB:CreditLineUtilizationbyBorrowers ∆ Y : LogCreditLineUtilization (1) (2) (3) (4) (5) (6) i,b,t BDC × Tightening 0.113 ∗ 0.105 ∗ 0.266 ∗∗∗ (0.061) (0.057) (0.080) BDC × ∆ FF 7.810 ∗∗ 5.853 15.883 ∗∗ (3.587) (3.823) (5.989) ∗ ∗ ∗∗ BDC 0.038 0.036 0.042 0.059 0.058 0.109 (0.035) (0.035) (0.052) (0.032) (0.031) (0.046) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.032 0.035 0.097 0.032 0.035 0.097 CreditLineSample All Existing LimitExpanded All Existing LimitExpanded N 1,667,709 1,533,250 295,187 1,667,709 1,533,250 295,187 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThispanelreportsregressionresultsfromEq. (1). Thesampleperiodisfrom2012Q3–2023Q4,withdataatthe bank-borrower-quarterlevel. Thesampleisrestrictedtocreditlinesonly. Columns(1)and(4)focusonallcreditlines. Columns(2) and(5)excludenewlyoriginatedcreditlinesandexamineonlypre-existingcreditlines. Columns(3)and(6)restrictthesampleto pre-existingcreditlinesconditionalonanypositivechangeincreditlinecommitmentintime t relativeto t−1. Tighteningisa ∆ dummyequalto1forthe2022monetarytightening(2022Q1–2023Q4)and0otherwise. FF ischangesintheeffectivefederalfunds ∆ rate. BDC isandummyequalto1iftheborrowerisaBDC,and0otherwise. Log(CLUtilization)isthelogchangeincreditline utilizationbetweenbank b andborrower i intime t,relativetotime t−1,andexpressedindecimal. CreditRating isbank’sinternal creditratings,whichisbank-specific,time-varying,andcapturesbank-estimatedex-antecreditrisk. Firm-levelcontrolsenterthe regressionswithaone-periodlagandincludebank-estimatedprobabilityofdefault,expectedlossgivendefault,shareoftermloansin totalbankdebt,shareofcreditlinesintotalbankdebt,andthenaturallogoftotalbankdebt. Allregressionsalsocontrolforthe contemporaneoustotalbankloancommitmentbetweenagivenbank-borrowerpair. Standarderrorsaredoubleclusteredat bank×borrower andYearQtr level. 47

Table5. Why Do Banks Prefer Lending to BDCs over Lending to Firms? 1stLienSeniorSecured Collateralized LossGivenDefault ProbabilityofDefault (1) (2) (3) (4) (5) (6) (7) (8) BDC × Tightening 0.103 ∗∗∗ 0.103 ∗∗∗ -0.027 ∗∗ 0.002 (0.020) (0.020) (0.011) (0.002) BDC × ∆ FF 5.910 ∗∗∗ 6.227 ∗∗∗ -2.397 ∗∗∗ 0.154 (1.534) (1.436) (0.593) (0.118) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.282 0.304 0.284 0.305 -0.083 -0.087 0.002 0.002 (0.017) (0.016) (0.017) (0.017) (0.013) (0.012) (0.001) (0.001) LaggedControls Y Y Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y Y Y R-squared 0.265 0.265 0.261 0.261 0.495 0.495 0.857 0.857 N 3,653,826 3,653,826 3,653,826 3,653,826 3,676,199 3,676,199 3,678,000 3,678,000 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsregressionresultsfromEq. (1). Thesampleperiodis2012Q3–2023Q4,withdataatthe bank-borrower-quarterlevel. 1stLienisadummyequalto1ifthelenderhasa1stlienseniorsecuredclaimontheborrower’sassetsin caseofdefault;Collateralizedisadummyequalto1ifthelenderhasacollateralizedclaimontheborrower’sassetsincaseofdefault; LossGivenDefaultisex-antebank-reportedestimateoflossgivendefault;ProbabilityofDefaultisex-antebank-reportedestimate ofdefaultprobabilityattheborrowerlevel;allfourvariablesaboveisobtainedbyaveragingacrossallcommitmentsbetween b and i in ∆ time t. Tighteningisadummyequalto1duringmonetarytightening(2022Q1–2023Q4)and0otherwise. FF ischangesinthe effectivefederalfundsrate. BDC isandummyequalto1iftheborrowerisaBDC,and0otherwise. CreditRating isbank’sinternal creditratings,whichisbank-specific,time-varying,andcapturesbank-estimatedex-antecreditrisk. Firmlevelcontrolsenterthe regressionswithaoneperiodlag. Incolumns(1)–(4),theseincludebank-estimatedprobabilityofdefault,expectedLGD,shareofterm loansintotalbankdebt,shareofcreditlinesintotalbankdebt,andnaturallogoftotalbankdebt. Incolumns(5)–(6),thecontrols remainthesame,exceptexpectedlossisomittedtoavoidusingthelaggeddependentvariable(asexpectedlossistheproductofloss givendefault,probabilityofdefault,andexpectedexposureatdefault). Incolumns(7)–(8),probabilityofdefaultisomittedforthesame reason,andexpectedlossisreplacedwithlossgivendefault. Standarderrorsaredoubleclusteredat bank × borrower andYearQtr level. 48

Table6. BDC Loans vs. Bank Loans to Overlapping Borrowers: panel A PanelA:FullSampleofOverlappingBorrowers InterestRate LoanAmount (1) (2) (3) (4) (5) (6) (7) (8) BDC × Tightening 0.521 ∗∗∗ 0.516 ∗∗∗ 0.252 ∗∗∗ 0.251 ∗∗∗ (0.116) (0.117) (0.093) (0.077) BDC×∆ FF 37.646 ∗∗∗ 37.156 ∗∗∗ 15.094 ∗∗∗ 14.957 ∗∗∗ (8.238) (8.342) (4.978) (4.945) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.949 0.978 1.107 1.135 -0.378 -0.372 -0.294 -0.288 (0.096) (0.097) (0.088) (0.087) (0.080) (0.064) (0.066) (0.067) LoanControls Y Y Y Y Y Y Y Y Firm × YrQtr × Loan-TypeFE Y Y Y Y Y Y Y Y DebtSeniorityFE N Y N Y N Y N Y R-squared 0.924 0.924 0.923 0.923 0.604 0.604 0.604 0.604 N 309,692 309,295 309,692 309,295 309,692 309,295 309,692 309,295 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsregressionresultsfromestimatingEq. (2). Thesampleperiodis2012Q3–2023Q4,withdataatthe loan-quarterlevel. InterestRateisthereportedinterestrate,expressedinpercentagepoints. LoanAmountisthenaturallogofthe utilizedamountoftheloan. BDCisadummyequalto1iftheloanisprovidedbyaBDC,0ifprovidedbyabank. Loan-typeFEis1 fortermloans,2forcreditlines,and3forotherloanstypes. DebtSeniorityFEareindicatorsforfirstlienseniorsecureddebt,second lienseniorsecureddebt,andjunior/unsecureddebt. Columns(1)–(4)controlforutilizedloanamount,maturity,andnon-accrual status,whileColumns(5)–(8)controlforloaninterestrate,maturity,andnon-accrualstatus. Standarderrorsaredouble-clusteredat theborrowerandYearQtrlevels. 49

Table6. BDC Loans vs. Bank Loans to Overlapping Borrowers: panel B PanelB:SplittingOverlappingBorrowersbyBankLendingConstraintsMeasuredbyBorrowingCapacity InterestRate LoanAmount (1) (2) (3) (4) (5) (6) (7) (8) BDC × Tightening 0.850 ∗∗∗ 0.380 ∗∗∗ 0.319 ∗∗∗ 0.204 ∗ (0.210) (0.128) (0.087) (0.115) BDC × ∆ FF 62.668 ∗∗∗ 23.576 ∗∗∗ 16.107 ∗∗∗ 12.960 ∗ (15.991) (7.847) (5.399) (6.455) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 1.074 0.896 1.279 1.041 -0.549 -0.325 -0.464 -0.248 (0.131) (0.113) (0.125) (0.094) (0.070) (0.100) (0.062) (0.080) LoanControls Y Y Y Y Y Y Y Y Firm × YrQtr × LoantypeFE Y Y Y Y Y Y Y Y DebtSeniorityFE Y Y Y Y Y Y Y Y R-squared 0.886 0.925 0.885 0.925 0.589 0.604 0.588 0.604 N 34,309 274,986 34,309 274,986 34,309 274,986 34,309 274,986 BankLoanConstrained Y N Y N Y N Y N ∗∗∗ ∗∗∗ ∗∗ H : Constrained-Unconstrained>0 0.470 39.092 0.115 3.147 0 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsregressionresultsfromestimatingEq. (2)bysplittingthesampleofoverlappingborrowersintothoselikely facingexantebanklendingconstraintsandthosethatdonot. Aborrower-quarterisclassifiedasexante"BankLoanConstrained"if theone-periodlaggedutilizationrateofbankloansexceedsthe75thpercentileofthesampledistribution,indicatingtheborrowerhas nearlyexhausteditsbankborrowingcapacity. Thesampleperiodis2012Q3–2023Q4,withdataattheloan-year-quarterlevel. Interest Rateisthereportedinterestrate,expressedinpercentagepoints. LoanAmountisthenaturallogoftheutilizedamountoftheloan. BDCisadummyequalto1iftheloanisprovidedbyaBDC,0ifprovidedbyabank. Loan-typeFEis1fortermloans,2forcredit lines,and3forotherloanstypes. DebtSeniorityFEareindicatorsforfirstlienseniorsecureddebt,secondlienseniorsecureddebt, andjunior/unsecureddebt. Columns(1)–(4)controlforutilizedloanamount,maturity,andnon-accrualstatus,whileColumns (5)–(8)controlforloaninterestrate,maturity,andnon-accrualstatus. H : Constrained-Unconstrained>0isbasedonaone-tailed 0 testtocomparethecoefficientdifferences. Standarderrorsaredouble-clusteredattheborrowerandYearQtrlevels. 50

Table6. BDC Loans vs. Bank Loans to Overlapping Borrowers: panel C PanelC:SplittingOverlappingBorrowersbyBankLendingConstraintsMeasuredbyRelationshipDuration InterestRate LoanAmount (1) (2) (3) (4) (5) (6) (7) (8) BDC × Tightening 0.661 ∗∗∗ 0.291 ∗ 0.263 ∗∗ 0.242 ∗∗ (0.154) (0.164) (0.117) (0.103) BDC × ∆ FF 51.413 ∗∗∗ 14.192 18.411 ∗∗∗ 9.078 (10.217) (10.267) (6.345) (5.986) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.983 0.992 1.171 1.098 -0.369 -0.463 -0.290 -0.369 (0.115) (0.153) (0.100) (0.131) (0.096) (0.087) (0.076) (0.083) LoanControls Y Y Y Y Y Y Y Y Firm × YrQtr × LoantypeFE Y Y Y Y Y Y Y Y DebtSeniorityFE Y Y Y Y Y Y Y Y R-squared 0.892 0.928 0.892 0.928 0.604 0.600 0.604 0.600 N 56,066 253,229 56,066 253,229 56,066 253,229 56,066 253,229 BankLoanConstrained Y N Y N Y N Y N ∗∗ ∗∗∗ ∗∗ ∗∗ H : Constrained-Unconstrained>0 0.370 37.221 0.021 9.333 0 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsregressionresultsfromestimatingEq. (2)bysplittingthesampleofoverlappingborrowersintothoselikely facingexantebanklendingconstraintsandthosethatdonot. Aborrower-quarterisclassifiedas"BankLoanConstrained"ifthe durationofitslongestbankrelationshipisshorterthanthesamplemeanacrossallbank-borrowerpairs. Bankrelationshipdurationis measuredattheborrower-bank-timelevelasthetotalnumberofquartersagivenbank-borrowerpairhasmaintainedanon-zeroloan commitment. Forborrowerswithmultiplelenders,thelongestrelationshipdurationamongallbanksisconsidered. InterestRateis thereportedinterestrate,expressedinpercentagepoints. LoanAmountisthenaturallogoftheutilizedamountoftheloan. BDCisa dummyequalto1iftheloanisprovidedbyaBDC,0ifprovidedbyabank. Loan-typeFEis1fortermloans,2forcreditlines,and3 forotherloanstypes. DebtSeniorityFEareindicatorsforfirstlienseniorsecureddebt,secondlienseniorsecureddebt,and junior/unsecureddebt. Columns(1)–(4)controlforutilizedloanamount,maturity,andnon-accrualstatus,whileColumns(5)–(8) controlforloaninterestrate,maturity,andnon-accrualstatus. H : Constrained − Unconstrained > 0isbasedonaone-tailedtestto 0 comparethecoefficientdifferences. Standarderrorsaredouble-clusteredattheborrowerandYearQtrlevels. 51

Table7. Flexibility Benefit of BDC Loans during Tightening: PIK options 1× (BorrowerUsedPIKOption) (1) (2) (3) (4) 1×(NonaccrualLoan) × Tightening 0.120 ∗∗∗ 0.096 ∗∗ (0.032) (0.044) 1×(NonaccrualLoan) × ∆ FF 6.694 ∗∗∗ 5.081 ∗ (2.046) (2.552) 1×(NonaccrualLoan) 0.127 ∗∗∗ 0.104 ∗∗∗ 0.153 ∗∗∗ 0.142 ∗∗∗ (0.019) (0.035) (0.019) (0.029) LoanControls Y Y Y Y FirmFE Y Y Firm × LoantypeFE Y Y YrQtrFE Y Y Y Y R-squared 0.574 0.614 0.574 0.614 N 431,981 289,172 431,981 289,172 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsevidenceontheflexibilitybenefitofBDCloansduringmonetary tighteningperiods,specificallyfocusingontheoptiontodelayinterestpaymentsthrough Payment-in-Kind(PIK)provisions. ThesamplecomprisesallBDCloansfromBDCCollateral, spanningfrom2012Q3to2023Q4,withdataattheloan-year-quarterlevel. Thedependent variable, 1× (BorrowerUsedPIKOption),isadummyvariableequaltooneifaborrower exercisesthePIKoptionforagivenloaninagivenquarter. 1×(NonaccrualLoan)inadummy variableequaltooneiftheloanisreportedasnon-accruinginagivenquarter. Tighteningisa ∆ dummyequalto1forthe2022monetarytightening(2022Q1–2023Q4)and0otherwise. FF is changesintheeffectivefederalfundsrate,expressedasadecimal. LoanControlsincludetheloan amount,remainingmaturity,interestrateandadummyforfixedrateloans. Standarderrorsare double-clusteredattheborrowerandYearQtrlevels. 52

Table8. BDCs’ Reliance on Bank Financing and Monetary Pass Through LoanAmount InterestRate (1) (2) (3) (4) BankLoanExpenseShare× Tightening 0.428 ∗∗∗ 0.313 ∗∗∗ (0.121) (0.0735) BankLoanExpenseShare× ∆ FF 21.55 ∗∗ 9.056 ∗∗ (10.47) (3.881) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ BankLoanExpenseShare -0.442 -0.247 -0.231 -0.0857 (0.126) (0.0879) (0.0633) (0.0390) Controls Y Y Y Y BDCFE Y Y Y Y YrQtrFE Y Y Y Y LoantypeFE Y Y Y Y R-squared 0.501 0.501 0.559 0.559 N 353,559 353,559 341,009 341,009 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsregressionresultsfromestimatingEq. (3). Thesampleperiodis 2012Q3–2023Q4,withdataattheBDC-loan-year-quarterlevel. Thedependentvariableis Loan Amount (thenaturallogoftheloan’sfacevalueindollars)incolumns(1)–(2)and Interest Rate (theinterestrateonagivenloan,expressedinpercentagepoints)incolumns (3)–(4). BankLoanExpenseShareistheshareoftotalinterestexpensesattributabletointerest paymentsonoutstandingbankloansforagivenBDCinagivenquarter. Tighteningisa ∆ dummyequalto1forthe2022monetarytightening(2022Q1–2023Q4)and0otherwise. FF is changesintheeffectivefederalfundsrate,expressedasadecimal. ControlvariablesincludeBDC totalassets,loanmaturity,andanindicatorvariablecapturingnon-accrualstatusoftheloan. Standarderrorsaredouble-clusteredattheborrowerandYearQtrlevels. 53

Table9. Real Effects of BDC Financing during Tightening Capex/ Asset Sales Interest TotalAssets Growth Growth ROA Coverage (1) (2) (3) (4) (5) High BDC Reliance × 0.019 ∗∗∗ 0.142 ∗∗∗ 0.009 -0.042 ∗∗∗ -1.840 ∗∗ Tightening (0.006) (0.043) (0.026) (0.009) (0.779) ∗∗∗ ∗∗ High BDC Reliance -0.006 -0.125 -0.032 0.006 0.454 (0.004) (0.021) (0.015) (0.005) (0.338) LaggedFirmControls Y Y Y Y Y 3-DigitInd × YearFE Y Y Y Y Y R-squared 0.163 0.182 0.208 0.305 0.263 N 4,810 4,758 4,774 4,860 4,830 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsevidenceontherealeffectsofBDCfinancingonborrowersduring monetarytighteningbyestimatingEq. (4). Thesamplecovers2012–2023,withdataatthe borrower-yearlevel. Thedependentvariableistheratioofcapitalexpendituresovertotalassetsin column(1),growthrateoftotalassetsoverthepreviousyearincolumn(2),growthrateofsales overthepreviousyearincolumn(3),ratioofEBITDAtobookvalueoftotalassetsincolumn(4), andtheinterestcoverageratio(EBITDA/InterestExpense)incolumn(5). HighBDCReliance isadummyequalto1iftheborrower’sBDCdebt/totaldebtisabovesamplemean. Tighteningis adummyequalto1forthe2022monetarytightening(2022–2023)and0otherwise. Laggedfirm controlsincludelogoftotalassets,totaldebt/totalassets,ROA,andcash/totalassetsincolumns (1)–(3). ROAisremovedasacontrolvariableincolumns(4)–(5)asdependentvariableincludes EBITDA.Industryfixedeffectsareatthe3-digitNAICSlevel. Standarderrorsareclusteredat theborrowerlevel. 54

Table10. Firm Characteristics in 2023: Switchers to BDC Credit vs. Others SwitchersfromBank Credit-Squeezedby Bank-Favored CredittoBDCCredit BanksandBDCs TotalAssets(USDMn) 3097 2480 3668 TotalDebt/EBITDA 5.52 4.69 6.16 Capex/TotalAssets 0.00 0.01 0.01 AssetGrowth 0.45 0.12 0.16 SalesGrowth 0.27 0.13 0.23 ROA 0.07 0.07 0.09 InterestCoverage 1.65 2.04 2.48 ProbabilityofDefault 0.09 0.08 0.05 Cash/Assets 0.04 0.06 0.05 Notes: Thistablepresentsthesamplemeanoffirm-levelcharacteristicsin2023foroverlapping borrowers,categorizedbasedonchangesintheircreditcompositionfrom2022Q4to2023Q4. Column(1)showsSwitchers,borrowerswhoexperiencednegativegrowthinbankloan commitmentsbutpositivegrowthinBDCloans;Column(2)showsCredit-Squeezed borrowers,thosewhosawnegativegrowthinbothbankandBDCloans;andColumn(3)shows Bank-Favoredborrowers,thosewhoexperiencedbankcreditexpansion. Thesampleislimitedto oneyearofoverlappingborrowerswithvalidfinancialdatareportedconcurrentwithloan informationin2023,yielding91firmsincolumn(1),110firmsincolumn(2),and243firmsin column(3). 55

Appendix A.1 Variable Definitions VariabledefinitionsarecategorizedintoY-14loan-level,Y-14firm-level,BDCloan-level, andBDC-levelvariables. Y-14Loan-LevelVariables. • CommittedLoanAmount: Reportedtotalloancommitmentinagivencreditfacility. • UtilizedLoanAmount: Reportedtotalloanutilizedamountinagivencreditfacility. • InterestRate: Reportedinterestrate,expressedasadecimal. Forregressionanalysis, itisaggregatedtothebank-borrower-quarterlevel,weightedbytheutilizedloan amount. • Maturity: Differencebetweenloanmaturitydateandloanoriginationdate, expressedinyears. • UtilizationRate: Ratioofutilizedloanamounttocommittedloanamount. • TermLoanShare: Shareoftermloancommitmentasafractionoftotalbankloan commitment. • CreditLineShare: Shareofcreditlinecommitmentasafractionoftotalbankloan commitment. • ∆ Log(Loan): Logquarterlychangeintotalloancommitmentatthebank-borrower level. • 1×(FirstLienSeniorSecured): Indicatorvariableequalto1iftheborrowerpledgeda firstlienseniorsecuredclaimonagivenloan. Forregressionanalysis,itis aggregatedtothebank-borrower-quarterlevel. • 1×(Collateralized): Indicatorvariableequalto1iftheborrowerpledgedany collateralonagivenloan. Forregressionanalysis,itisaggregatedtothe bank-borrower-quarterlevel. • LossGivenDefault: Expectedloanlossrateupondefaultestimatedbythereporting bank. Itisreportedasafractionbetween0and1. • ProbabilityofDefault: Banks’internalestimatesofborrower’s1-yearahead probabilityofdefault. • 1×(Term): Indicatorvariableequalto1iftheloanisatermloan. A1

Y-14Firm-LevelVariables. • TotalAssets: Bookvalueofcurrentyearassets(USDMn). • TotalDebt: Sumofallshort-termdebtandlong-termdebt(USDMn). • TotalDebt/TotalAssets: Ratiooftotaldebttototalassets. • BDCDebt/TotalDebt: RatioofBDCdebttoTotalDebt. • HighBDCReliance: Dummyvariableequalto1ifBDCDebt/TotalDebtisabove samplemean. • Cash/TotalAssets: RatioofCashandMarketableSecuritiestoTotalAssets,also referredinmaintextasliquidity. • InterestCoverage: RatioofEBITDAtointerestexpenseamount. • Tangibility: Ratiooftotaltangibleassetsovertotalassets. • Capex/TotalAssets: Ratioofcapitalexpendituresovertotalassets. • AssetGrowth: Growthrateoftotalassetsoverthepreviousyear. • Sales: Netsalesforthecurrentyear(USDMn). • SalesGrowth: Growthrateofsalesoverthepreviousyear. • ROA:RatioofEBITDAtobookvalueoftotalassets,alsoreferredinmaintextas earnings,orfirmprofitability. • 1×(ROA < 0): Dummyvariableequalto1ifEBITDA/ROAisnegativeforthe currentyear. • CreditRating: Bank’sinternalcreditratings,whichisattheborrower-level, bank-specific,time-varying,andcapturesbank-estimatedex-antecreditrisk. • BankLoanCommitment/TotalAssets: Ratioofbankloancommitmenttototalassets. BDCLoan-Level. • ParAmount: Reportedfacevalueoftheloan(USDMn). • All-InYield: Reportedtotalinterestrateonagivenloan,expressedinpercentage points. • 1×(Term): Indicatorvariableequalto1iftheloanisatermloan. • CashSpread: Standardcreditspreadontheloanontopofthebaserate,expressedin percentagepoints. • Cash+PIKSpread: SumofcashspreadandtheadditionalPIKspreadifagivenloan hasaPIKoption,expressedinpercentagepoints. A2

• Maturity: Reportedloanmaturityasoftheholdingdate(notorigination). • 1×(Non-accrual): Indicatorvariableequalto1ifthegivenloanisreportedas non-accruinginagivenquarter. • 1× (BorrowerUsedPIKOption): Indicatorvariableequalto1ifaborrower exercisesthePIKoptionforagivenloaninagivenquarter. BDC-LevelVariables. • BankLoanExpenseShare: ForagivenBDCinagivenquarter,theshareoftotal interestexpensesthatisattributabletointerestpaymentsonoutstandingbank loans. ConstructedbymergingBDCCollateralandY-14. • BankReliance: ForagivenBDCinagivenquarter,anindicatorvariableequalto1if theBDC’sbankloanutilizationrateisabovethe75thpercentileofthesample distributionacrossallBDCsandovertime. ConstructedbymergingBDC CollateralandY-14. • BDCLeverage: Theratiooftotaldebttototalassets. • BDCNetEquityIssuance: Thechangesinbookequity(NAV)minuscurrentquarter netincomescaledbylaggedtotalassets. MonetaryPolicySeries. • Tightening: Adummyvariableequal1forthe2022monetarytighteningcycle (2022Q1–2023Q4). ∆ • FF: ChangesintheFederalFundsrate,expressedasadecimal. • MPShock: ThesumofmonetaryPolicyshocksreportedbyJarocin´skiandKaradi (2020)andBauerandSwanson(2023)fromquarter t−1throughquarter t. A.2 Y-14 Data Cleaning • TheY-14H.1. datawasdownloadedinJanuary2024. FollowingGreenwaldetal. (2024)andChodorow-Reich,Darmouni,LuckandPlosser(2022),weidentify distinctfirmsusingTaxpayerIdentificationNumber(TIN),allowingustolinkthe samefirmacrossbanksandovertime. Thisaddressescaseswhenafirmborrows frommultiplebanks,whichmayusedifferentnamingconventionsforthesame borrower. • SomeborrowershavemissingTIN.Weapplyaname-standardizationalgorithmto obtainacleananduniformsetoffirmnames. IfaTINismissing,wefillinmissing observationsifthebankreportsaconsistentTINinanyportionoftheloandata; formulti-bankborrowersforwhichonebankdoesnotreportaTIN,weusea consistentTINreportedbyotherbanks. A3

• Unlessotherwisestated,allvariablesarewinsorizedatthe2.5and97.5percent levels,followingFavara,MinoiuandPerez-Orive(2022),tomitigatetheinfluence ofoutliersandpotentialreportingerrors. • Weexcludeobservationswithnegativeorzerovaluesforcommittedloanamount, negativevaluesforutilizedloanamount,orcaseswherecommittedloanamountis lessthanutilizedamount. • Wedropallfacilityrecordswithoriginationdatesbefore1990ormaturitiesgreater than30yearstominimizetheinfluenceofpotentialdataentryerrors. • Toensuredataaccuracyininterestratecalculations,weexcludeobservationswith interestratesbelow0.5percentorabove50percenttominimizetheinfluenceof potentialdataentryerrors. • Whenusingfirm’sreportedfinancialvariables,weexcludefinancialstatement informationifthefinancialstatementdateismissingorcomesafterthereporting date. Wealsoexcludelikelydataerrorsbyimposingthefollowingconditions: (i) EBITDAdoesnotexceednetsales,(ii)fixedassetsexceedtotalassets,(iii)cashand marketablesecuritiesdonotexceedtotalassets,(iv)long-termdebtdoesnot exceedtotalliabilities,(v)short-termdebtdoesnotexceedtotalliabilities,(vi) tangibleassetsdonotexceedtotalassets,(vii)currentassetsdonotexceedtotal assets,and(viii)currentliabilitiesdonotexceedtotalliabilities. • Becauseoftimingdifferencesbetweenreportedfinancialinformationandloan informationintheY-14,thefinancialdataarecollapsedatthefirm-yearlevelusing theyearofthereportedfinancialinformationandnotthecorrespondingloan information. A.3 BDC Collateral Data Cleaning • TheBDCCollateraldataisreportedattheBDCloan-quarterlevel,providing detailedinformationonborrower,lender,reportingperiod,paramountoftheloan, all-in-yield,maturity,seniority,loantype(termloan,revolver,unitranche,etc) investmenttype(equityvs. varioustypesofdebt: firstlien,secondlien, subordinated),non-accruingstatus,amongotherdetails. • Westartthesamplefrom2012Q3,excludeexposuresclassifiedas’Equity’,and retainonlydebtinvestments(i.e. loans). • Loanswithaparamountabovethe99thpercentilearedroppedtomitigatethe impactofpotentialoutliers. • Wedroploansmissinginterestratedata(i.e.,all-in-yield)andthosewherethelien isclassifiedas’Other’. Additionally,loanswithreportedinterestratesbelow0.5% orabove50%areexcluded. • Facilityrecordswithoriginationdatesbefore1990andmaturitiesexceeding30 yearsareremovedtominimizepotentialdataentryerrors. A4

• WeexcludeonelenderfromtheBDCsamplethat,basedondiscussionswith industryexperts,isnotatypicalmiddle-marketprivatecreditlenderbutinstead specializesinSBA-guaranteeddebtfinancingforverysmallbusinesses. Retaining thislenderdoesnotaffectourresults. • WealsoobtainBDCquarterlyfinancialdatafromBDCCollateral,whichprovides fullcoverageofmajorfinancialstatementitems—suchastotalassets,totaldebt, totalinvestments,cash,andnetincome. However,itdoesnotfullycoverinterest expense,asnotallBDCsreportthisitem. InBDCCollateral,interestexpenseis availablefor87.6%ofBDC-quarterobservations. • Tosupplementthis,wemergedatafromCompustatforpublicBDCs,increasing totalinterestexpensecoverageto92.5%. Wemanuallyverifyarandomlyselected sample,confirmingthattheremainingmissingobservationswereindeedreported aszeroinCompustatforpublicBDCsorintherespectiveSECfilingsforprivate BDCs. Notably,themedianleverageofBDCswithmissinginterestexpenseis 4.6%,significantlylowerthanthe44.2%medianleverageforthefullsample. This suggeststhattheseBDCsdonotreportinterestexpenseasamaterialexpense becausetheydonotrelyheavilyondebt. Accordingly,wesettheremaining missinginterestexpensedatatozero. Ourresultsremainunchangedwhen excludingthesemissingobservations. • ToevaluatewhetherBDCborrowersinRefinitiv’sBDCCollateraldataare representativeofthebroaderprivatecreditmarket,weperformabalancetest comparingkeycharacteristicsofBDCborrowersinoursamplewithprivatecredit borrowersfromJang(2025)overtheperiod2014Q3–2023Q4. Thetablebelow presentsthebalancetestresults,examiningborrowers’likelihoodofobtaininga first-lienloanandtheaverageinterestratespread(cash+PIK).Themean differencesbetweenthetwogroupsarenotstatisticallysignificant,evenatthe10% level(assumingunequalvariances). Balance Test: BDC borrowers vs. private credit borrowers BDC Privatecredit MeanDifference N Mean SD N Mean SD First-lien 11731 0.894 0.307 6605 0.895 0.306 -0.001 Interestratespread(%) 11731 7.362 2.776 6605 7.300 2.709 0.062 A5

A.4 Additional Figures FigureA.1. BDC Leverage Notes: ThisfigureplotsthetimeseriesoftheaverageBDCLeverage(Debt/Asset)weightedby BDCassets. Thesampleperiodis2012Q3–2023Q4. Thesampleincludes190BDCswith availabledata. A6

FigureA.2. Share of BDC Loans Held by Bank-Reliant BDCs Notes: ThisfigureplotstheshareoftotalBDCcreditprovidedbyY-14bank-reliantBDCs. Credit amountreferstothetotalparvalueofloansatagiventimeforeachBDC.Bank-reliantBDCsare thosewithanon-zeroloancommitmentfromabank. Thesampleperiodspans2012Q3to 2023Q4,with142uniquebank-reliantBDCs. A7

FigureA.3. Assets of Public and Private BDCs Notes: Thisfigureplotsthetotalassets(USDbillions)ofpublicandprivateBDCsovertime, basedonBDCCollateraldata. Thesampleperiodspans2012Q3–2023Q4. A8

FigureA.4. Overlapping Borrowers with both Bank Loans and BDC Credit Notes: Thisfigureshowsthenumberofuniqueoverlappingborrowersinoursampleovertime. OverlappingborrowersarefirmsthatsimultaneouslyholdbankloansandBDCcreditinagiven quarter,wherebankloansincludebothdrawnandundrawncommitments. Thesamplecontains 4,793uniqueoverlappingborrowers. A9

A.5 Additional Tables TableA.1. Robustness to Table 3 and Table 5: alternative monetary policy shocks ∆ LogLoan Utilization InterestRate 1stLien Collateralized LGD (1) (2) (3) (4) (5) (6) PanelA:MonetaryPolicyShockfromJarocin´skiandKaradi(2020) BDC × MPShocks 0.090 ∗∗ 0.573 ∗∗∗ 0.013 0.282 ∗∗ 0.325 ∗∗ -0.093 ∗∗ (0.038) (0.200) (0.013) (0.135) (0.133) (0.041) ∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.005 0.088 0.005 0.314 0.317 -0.096 (0.003) (0.018) (0) (0.017) (0.017) (0.014) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.286 0.480 0.265 0.261 0.508 N 3,653,826 1,712,362 3,468,670 3,653,826 3,653,826 3,628,607 PanelB:MonetaryPolicyShockfromBauerandSwanson(2023) BDC × MPShocks 0.047 ∗∗∗ 0.294 ∗∗ 0.007 0.146 ∗ 0.178 ∗∗ -0.043 ∗∗∗ (0.015) (0.131) (0.008) (0.086) (0.087) (0.009) ∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.004 0.085 0.005 0.313 0.315 -0.095 (0.002) (0.018) (0) (0.017) (0.017) (0.014) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.286 0.480 0.265 0.261 0.508 N 3,653,826 1,712,362 3,468,670 3,653,826 3,653,826 3,628,607 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforourbaselineresultsinTable3andTable5usingthemonetarypolicyshocksfrom Jarocin´skiandKaradi(2020)inPanelAandfromBauerandSwanson(2023)inPanelB. A10

TableA.2. Robustness to Table 6: alternative monetary policy shocks InterestRate LoanAmount (1) (2) PanelA:MonetaryPolicyShockfromJarocin´skiandKaradi(2020) BDC× MP Shock 1.909 ∗∗ 1.687 ∗∗∗ (0.752) (0.439) ∗∗∗ ∗∗∗ BDC 1.213 -0.265 (0.093) (0.063) R-squared 0.923 0.604 N 309,295 309,295 PanelB:MonetaryPolicyShockfromBauerandSwanson(2023) BDC × MP Shocks 1.059 ∗∗∗ 0.836 ∗∗ (0.389) (0.369) ∗∗∗ ∗∗∗ BDC 1.209 -0.267 (0.092) (0.064) R-squared 0.924 0.604 N 309,295 309,295 LoanControls Y Y Firm × Loantype × YrQtrFE Y Y DebtPriorityFE Y Y ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforresultsinTable6usingthemonetarypolicyshocks fromJarocin´skiandKaradi(2020)inPanelA,andthemonetarypolicyshocksfromBauerand Swanson(2023)inPanelB. A11

TableA.3. Robustness to Table 7: alternative monetary policy shocks 1× (BorrowerUsedPIKOption) (1) (2) (3) (4) 1×(NonaccrualLoan) × Jarocinski-KaradiShock 0.321 ∗∗ 0.193 (0.157) (0.191) 1×(NonaccrualLoan) × Bauer-SwansonShock 0.169 ∗∗ 0.136 ∗ (0.074) (0.072) 1×(NonaccrualLoan) 0.164 ∗∗∗ 0.151 ∗∗∗ 0.161 ∗∗∗ 0.150 ∗∗∗ (0.020) (0.031) (0.019) (0.030) LoanControls Y Y Y Y FirmFE Y Y Firm × LoantypeFE Y Y YrQtrFE Y Y Y Y R-squared 0.574 0.614 0.574 0.614 N 431,981 289,172 431,981 289,172 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforresultsinTable7usingthemonetarypolicyshocks fromJarocin´skiandKaradi(2020)andthemonetarypolicyshocksfromBauerandSwanson (2023). A12

TableA.4. Robustness to Table 8: alternative monetary policy shocks LoanAmount InterestRate (1) (2) (3) (4) BankLoanExpense × Jarocinski-KaradiShock 1.985 ∗∗∗ 0.798 ∗∗ (0.668) (0.353) BankLoanExpense × Bauer-SwansonShock 0.612 0.448 ∗ (0.527) (0.249) ∗∗∗ ∗∗∗ ∗∗ ∗∗ BankLoanExpense -0.305 -0.234 -0.105 -0.0929 (0.0754) (0.0814) (0.0431) (0.0431) LoanControls Y Y Y Y BDCFE Y Y Y Y YrQtrFE Y Y Y Y LoantypeFE Y Y Y Y R-squared 0.501 0.501 0.559 0.559 N 353,559 353,559 341,009 341,009 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforresultsinTable8usingthemonetarypolicyshocks fromJarocin´skiandKaradi(2020)andthemonetarypolicyshocksfromBauerandSwanson (2023). A13

TableA.5. Robustness to Table 3: including bank-borrower fixed effects ∆ ∆ LogLoan Utilization InterestRate LogLoan Utilization InterestRate (1) (2) (3) (4) (5) (6) BDC × Tightening 0.018 ∗∗ 0.089 ∗∗∗ 0.007 ∗∗∗ (0.008) (0.026) (0.002) BDC × ∆ FF 0.552 4.600 ∗∗∗ 0.325 ∗∗∗ (0.383) (1.347) (0.081) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y Bank × BorrowerFE Y Y Y Y Y Y R-squared 0.121 0.744 0.855 0.121 0.744 0.855 N 3,630,073 1,698,343 3,444,201 3,630,073 1,698,343 3,444,201 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforourbaselineresultsinTable3byincludingbank-borrowerfixedeffects. A14

TableA.6. Robustness to Table 5: including bank-borrower fixed effects 1stLienSeniorSecured Collateralized LossGivenDefault ProbabilityofDefault (1) (2) (3) (4) (5) (6) (7) (8) BDC × Tightening 0.039 ∗∗∗ 0.039 ∗∗∗ -0.012 ∗∗ -0.001 (0.014) (0.013) (0.005) (0.001) BDC × ∆ FF 1.826 ∗∗ 2.150 ∗∗ -0.672 ∗∗∗ -0.026 (0.815) (0.915) (0.238) (0.081) LaggedControls Y Y Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y Y Y Bank × BorrowerFE Y Y Y Y Y Y Y Y R-squared 0.882 0.882 0.903 0.903 0.910 0.910 0.915 0.915 N 3,630,073 3,630,073 3,630,073 3,630,073 3,630,073 3,630,073 3,630,073 3,630,073 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforourbaselineresultsinTable5byincludingbank-borrowerfixedeffects. A15

TableA.7. Robustness to Table 3: alternative definitions of other borrowers ∆ ∆ LogLoan Utilization InterestRate LogLoan Utilization InterestRate (1) (2) (3) (4) (5) (6) BDC × Tightening 0.011 ∗∗ 0.141 ∗∗∗ 0.009 ∗∗∗ (0.005) (0.027) (0.002) BDC × ∆ FF 0.726 ∗ 9.450 ∗∗∗ 0.542 ∗∗∗ (0.409) (2.111) (0.114) ∗∗∗ ∗∗ ∗∗∗ ∗∗∗ BDC 0.001 0.052 0.002 0.003 0.079 0.004 (0.003) (0.019) (0.001) (0.003) (0.017) (0.001) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.029 0.290 0.480 0.029 0.290 0.480 N 3,590,663 1,676,659 3,415,275 3,590,663 1,676,659 3,415,275 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsrobustnesschecksforTable3byrestrictingthegroupofnon-BDCborrowerstoprimarilynon-financial firms,includingallY-14borrowersthatareneitherBDCsnorhaving3-digitNAICScode521(MonetaryAuthorities-CentralBank) or522(CreditIntermediationandRelatedActivities). A16

TableA.8. Robustness to Table 3: utilization rate for all loan types UtilizationRate (1) (2) (3) (4) BDC × Tightening 0.108 ∗∗∗ 0.089 ∗∗∗ (0.024) (0.025) BDC × ∆ FF 5.803 ∗∗ 5.500 ∗∗ (2.687) (2.171) ∗ ∗∗ BDC 0.012 0.022 0.035 0.040 (0.020) (0.021) (0.020) (0.020) LaggedControls Y Y Y Y Bank × CreditRating × YrQtrFE N Y N Y Bank × CreditRatingFE Y N Y N YrQtrFE Y N Y N R-squared 0.475 0.503 0.475 0.503 N 3,653,826 3,653,826 3,653,826 3,653,826 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsrobustnesschecksforColumns(2)and(5)ofTable3byexpandingthe sampletoallloantypesandincludingalternativefixedeffects. A17

TableA.9. Robustness to Table 5: restricting sample to existing credit lines 1stLienSeniorSecured Collateralized LossGivenDefault ProbabilityofDefault (1) (2) (3) (4) (5) (6) (7) (8) BDC × Tightening 0.107 ∗∗∗ 0.105 ∗∗∗ -0.086 ∗∗∗ 0.005 (0.021) (0.021) (0.016) (0.003) BDC × ∆ FF 7.235 ∗∗∗ 6.937 ∗∗∗ -5.625 ∗∗∗ 0.315 (1.721) (1.645) (1.215) (0.202) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.298 0.318 0.297 0.316 -0.036 -0.053 0.001 0.001 (0.018) (0.019) (0.018) (0.018) (0.007) (0.009) (0.001) (0.001) LaggedControls Y Y Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y Y Y R-squared 0.298 0.298 0.299 0.299 0.496 0.496 0.872 0.872 N 1,533,250 1,533,250 1,533,250 1,533,250 1,533,250 1,533,250 1,533,250 1,533,250 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsrobustnesschecksforTable5byrestrictingtheanalysistocreditlines,thepredominantlendingformfor BDCs. A18

TableA.10. The 2015 Tightening Cycle for Table 3 and Table 5 DeltaLogLoan Utilization InterestRate LGD 1stLien Collateralized (1) (2) (3) (4) (5) (6) BDC × 2015Tightening 0.000352 -0.0579 ∗ -0.000112 -0.0252 ∗ -0.0244 -0.0339 ∗ (0.00693) (0.0304) (0.000804) (0.0125) (0.0191) (0.0194) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.00358 0.0671 0.00466 -0.0837 0.293 0.297 (0.00396) (0.0240) (0.000548) (0.0142) (0.0192) (0.0199) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.029 0.248 0.352 0.488 0.255 0.254 N 2,773,907 1,212,497 2,715,007 2,757,866 2,773,907 2,773,907 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsanalternativeanalysisofourbaselineresultsfromTables3and5. Weintroduceanewdummyvariable, 2015Tightening,tocapturethetighteningcyclethatoccurredfrom2015Q4to2018Q4. Toisolatetheeffectsofthisspecificcycleand avoidinterferencefromthe2022tightening,werestrictoursampleperiodto2012Q3–2021Q4,endingimmediatelybeforethestartof the2022tighteningcycle. A19

TableA.11. Robustness to Table 3 and Table 5: public and private subsample analysis DeltaLogLoan Utilization InterestRate LGD 1stLien Collateralized (1) (2) (3) (4) (5) (6) PanelA:PubliclylistedBDCs BDC × Tightening 0.010 ∗ 0.143 ∗∗∗ 0.009 ∗∗∗ -0.011 0.090 ∗∗∗ 0.086 ∗∗∗ (0.006) (0.028) (0.002) (0.013) (0.021) (0.021) ∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.001 0.020 0.002 -0.096 0.288 0.291 (0.004) (0.020) (0.001) (0.018) (0.019) (0.020) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.286 0.480 0.508 0.265 0.262 N 3,652,310 1,710,969 3,467,202 3,627,104 3,652,310 3,652,310 PanelB:PrivateBDCs BDC × Tightening 0.009 0.092 ∗∗ 0.009 ∗∗∗ -0.039 0.138 ∗∗∗ 0.144 ∗∗∗ (0.010) (0.043) (0.001) (0.023) (0.039) (0.040) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.004 0.127 0.003 -0.069 0.260 0.261 (0.008) (0.031) (0.001) (0.017) (0.034) (0.034) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.287 0.480 0.509 0.266 0.262 N 3,649,988 1,708,842 3,464,951 3,624,798 3,649,988 3,649,988 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforourbaselineresultsinTable3andTable5usingsubsampleanalysisforpublicand privateBDCandmonetarystancemeasuredusingthetighteningdummy. A20

TableA.12. Robustness to Table 3 and Table 5: public and private subsample analysis DeltaLogLoan Utilization InterestRate LGD 1stLien Collateralized (1) (2) (3) (4) (5) (6) PanelA:PubliclylistedBDCs BDC×∆ FF -0.219 8.373 ∗∗∗ 0.527 ∗∗∗ -1.201 ∗∗∗ 4.344 ∗∗∗ 4.350 ∗∗∗ (0.267) (2.164) (0.114) (0.230) (1.463) (1.331) ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.004 0.043 0.003 -0.097 0.305 0.306 (0.003) (0.019) (0.001) (0.018) (0.018) (0.019) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.286 0.480 0.508 0.265 0.262 N 3,652,310 1,710,969 3,467,202 3,627,104 3,652,310 3,652,310 PanelB:PrivateBDCs BDC×∆ FF 2.283 ∗∗ 9.210 ∗∗∗ 0.497 ∗∗∗ -3.625 ∗∗ 8.872 ∗∗∗ 9.710 ∗∗∗ (0.942) (2.544) (0.117) (1.415) (2.673) (2.661) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ BDC 0.003 0.147 0.006 -0.078 0.302 0.304 (0.006) (0.024) (0.001) (0.014) (0.029) (0.029) LaggedControls Y Y Y Y Y Y Bank × CreditRating × YrQtrFE Y Y Y Y Y Y R-squared 0.028 0.287 0.480 0.509 0.266 0.262 N 3,649,988 1,708,842 3,464,951 3,624,798 3,649,988 3,649,988 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforourbaselineresultsinTable3andTable5usingsubsampleanalysisforpublicandprivate BDCandmonetarystancemeasuredbythechangeinfedfundsrates. A21

TableA.13. Robustness to Table 8: alternative bank reliance measures LoanAmount InterestRate (1) (2) (3) (4) High Bank Reliant × Tightening 0.197 ∗∗ 0.234 ∗∗∗ 0.326 ∗∗∗ 0.285 ∗∗∗ (0.0793) (0.0781) (0.0846) (0.0713) ∗∗∗ ∗∗ ∗∗∗ High Bank Reliant -0.234 -0.162 -0.0274 -0.200 (0.0653) (0.0769) (0.0636) (0.0599) Controls Y Y Y Y BDCFE Y Y Y Y YrQtrFE Y Y Y Y LoantypeFE Y Y Y Y R-squared 0.506 0.506 0.563 0.563 N 363,931 363,931 350,861 350,861 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablereportsrobustnesstestsforTable8usingalternativemeasuresforBDCs’ relianceonbanks. ForColumns(1)and(3),HighBankReliantisadummyvariableequalto1 ifaBDC’sutilizedbankloantototaldebtratioisinthetopquartile(abovethe75thpercentile)of thesampledistribution. ForColumns(2)and(4),HighBankReliantisadummyvariableequal to1ifaBDC’sbankloanutilizationrateisinthetopquartile(abovethe75thpercentile)ofthe sampledistribution. TherestofthespecificationisidenticaltoTable8. A22

TableA.14. Robustness to Table 8: alternative controls LoanAmount InterestRate (1) (2) (3) (4) BankLoanExpense× Tightening 0.313 ∗∗∗ 0.427 ∗∗∗ (0.111) (0.114) BankLoanExpense×∆ FF 21.880 ∗∗ 7.724 ∗ (8.531) (3.940) ∗∗∗ ∗∗ ∗∗∗ ∗∗ BankLoanExpense -0.487 -0.266 -0.311 -0.134 (0.146) (0.100) (0.097) (0.051) Controls Y Y Y Y BDCFE Y Y Y Y YrQtrFE Y Y Y Y LoantypeFE Y Y Y Y R-squared 0.497 0.497 0.537 0.537 N 353,329 353,329 339,846 339,846 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsrobustnesschecksforTable8usinganalternativesetofBDC-level controls,includingtotalBDCassets,BDCleverage,BDCnetequityissuance(changesinbook equity(NAV)minuscurrentquarternetincome)asashareoftotalassets,bankloancommitment asashareofBDC’stotaldebt,andutilizedbankloansasashareofBDC’stotaldebt. A23

TableA.15. Summary Statistics for Borrowers Financials Variable Count Mean SD Median PanelA:OverlappingBorrowerswithLowBDCReliance TotalAssets(USDMn) 6,450 2,875 5,537 1,088 TotalDebt/TotalAssets 6,450 0.58 0.22 0.55 BDCDebt/TotalDebt 6,450 0.05 0.06 0.02 Cash/TotalAssets 6,450 0.05 0.06 0.03 InterestCoverage 6,424 3.25 5.23 2.32 Tangibility 6,450 0.51 0.28 0.47 Capex/TotalAssets 6,386 0.02 0.08 0.01 AssetGrowth 3,887 0.14 0.50 0.01 SalesGrowth 3,874 0.15 0.42 0.06 ROA 6,450 0.10 0.08 0.09 1×(ROA < 0) 6,450 0.07 0.25 0.00 PanelB:OverlappingBorrowerswithHighBDCReliance TotalAssets(USDMn) 2,813 311 912 124 TotalDebt/TotalAssets 2,809 0.46 0.28 0.44 BDCDebt/TotalDebt 2,813 0.60 0.29 0.53 Cash/TotalAssets 2,813 0.08 0.11 0.03 InterestCoverage 2,727 5.45 12.87 1.91 Tangibility 2,813 0.54 0.29 0.53 Capex/TotalAssets 2,759 0.02 0.07 0.01 AssetGrowth 1,570 0.11 0.44 0.00 SalesGrowth 1,583 0.12 0.33 0.04 ROA 2,813 0.09 0.11 0.08 1×(ROA < 0) 2,813 0.19 0.39 0.00 PanelC:NonOverlappingBorrowers TotalAssets(USDMn) 722,006 638 3265 16 TotalDebt/TotalAssets 716,931 0.35 0.27 0.30 Cash/TotalAssets 716,931 0.12 0.15 0.06 InterestCoverage 639,113 21.72 40.90 7.82 Tangibility 716,931 0.91 0.17 1.00 Capex/TotalAssets 666,364 0.02 0.07 0.00 AssetGrowth 530,937 0.48 3.72 0.05 SalesGrowth 529,649 0.59 4.89 0.05 ROA 716,931 0.17 0.20 0.12 1×(ROA < 0) 722,006 0.09 0.29 0.00 Notes: Thistablepresentsfirm-levelsummarystatisticsforthreegroupsofborrowersduring 2012–2023,withdataattheborrower-yearlevel: OverlappingborrowerswithlowBDCreliance (BDCDebt/TotalDebtbelowsamplemean)inPanelA;OverlappingborrowerswithhighBDC reliance(BDCDebt/TotalDebtabovesamplemean)inPanelB;Non-overlappingborrowersin PanelC.Overlappingborrowerssimultaneouslyholdbothbankloansandprivatecredit. Non-overlappingborrowersareallothernon-financialborrowers. Thesampleincludes3,693 uniqueoverlappingborrowerswithavailablefiArm24financialdata.

TableA.16. Real Effects of BDC Credit during Tightening: external validity PanelA:SummaryStatistics Variable Count Mean SD Median BorrowersofPrivateCredit-OriginatedLoans TotalAssets(USDMn) 4636 357 458 214 TotalDebt/TotalAssets 4636 0.63 0.31 0.56 Capex/TotalAssets 3751 0.02 0.03 0.01 AssetGrowth 4636 0.17 0.51 0.00 SalesGrowth 4636 0.14 0.29 0.08 ROA 4597 0.10 0.10 0.09 InterestCoverage 3470 2.07 2.62 1.67 BorrowersofBank-OriginatedLoans TotalAssets(USDMn) 2016 857 978 473 TotalDebt/TotalAssets 2016 0.65 0.30 0.59 CapEx/Assets 1594 0.02 0.03 0.01 AssetGrowth 2016 0.14 0.46 0.00 SalesGrowth 2016 0.10 0.25 0.06 ROA 1990 0.10 0.10 0.09 InterestCoverage 1415 2.00 2.58 1.72 PanelB:RegressionResults Capex/ Asset Sales Interest TotalAssets Growth Growth ROA Coverage (1) (2) (3) (4) (5) PrivateCredit× -0.001 0.059 ∗∗ 0.026 ∗∗∗ 0.008 0.230 Tightening (0.001) (0.024) (0.006) (0.004) (0.131) ∗∗ ∗∗ PrivateCredit 0.000 -0.028 0.011 -0.006 -0.228 (0.001) (0.012) (0.005) (0.006) (0.139) LaggedFirmControls Yes Yes Yes Yes Yes Ind × YearFE Yes Yes Yes Yes Yes R-squared 0.42 0.12 0.08 0.10 0.05 N 5345 6652 6652 6667 4939 ∗ p <.10,∗∗ p <.05,∗∗∗ p <.01 Notes: ThistablepresentsexternalvalidityforTable9usingaproprietarydatabasefromJang (2025). Thesamplecovers2014–2023attheborrower-yearlevel. PanelAprovidessummary statisticsforborrowersofprivatecredit-andbank-originatedloans. PanelBpresentsregression results. DependentvariablesfollowTable9. PrivateCredit isadummyequalto1forprivate credit-originatedloansand0forbank-originatedloans. Tighteningisadummyequalto1forthe 2022monetarytightening(2022–2023)and0otherwise. Laggedfirmcontrolsincludelogoftotal assets,totaldebt/totalassets,ROA,cash/totalassets,andNetPP&E/Assets. ROAisexcludedas acontrolincolumns(4)-(5)wherethedependentvariableincludesEBITDA.IndustryFEare basedonthedatabase’sclassificationsystem: BusinessServices,Consumer,Energy,Finance, Healthcare,Industrials,TMT(Technology,Media,andTelecommunication),andOthers. A25 Standarderrorsaretwo-wayclusteredattheborrowerandyearlevel.