Less Bank Regulation, More Non-Bank Lending

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Less Bank Regulation, More Non-Bank Lending Mary Chen, Seung Jung Lee, Daniel Neuhann, Farzad Saidi 2023-026 Please cite this paper as: Chen, Mary, Seung Jung Lee, Daniel Neuhann, and Farzad Saidi (2023). “Less Bank Regulation, More Non-Bank Lending,” Finance and Economics Discussion Series 2023-026. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.026. 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.

Less Bank Regulation, More Non-Bank Lending* Mary Chen† Seung Jung Lee‡ Daniel Neuhann§ Farzad Saidi¶ April 28, 2023 Abstract BankderegulationintheformoftherepealoftheGlass-SteagallActfacilitatedthe entry of non-bank lenders into the market for syndicated loans during the pre-2008 credit boom. Institutional investors disproportionately purchase tranches of loans originated by universal banks able to cross-sell loans and underwriting services to firms (as permitted by the repeal). A shock to cross-selling intensity increases loan liquidityatoriginationandovertime. Themechanismisthatnon-loanexposuresensuremonitoringevenwhenbanksretainsmallloanshares. Ourfindingscomplement theconventionalviewthatregulatoryarbitragecausedtheriseofnon-banklenders. JELclassification: G20,G21,G23,G28 Keywords: Non-banklending,bankderegulation,creditsupply,loanliquidity,industrial organizationoffinancialmarkets *This paper subsumes parts of a previous working paper by Neuhann and Saidi titled “Bank Deregulation and the Rise of Institutional Lending.” We thank Bo Becker, Max Bruche, Itay Goldstein, Vasso Ioannidou, Victoria Ivashina, Erica Jiang, Vincent Maurin, Ralf Meisenzahl, Anthony Saunders, Victoria VanascoandVikrantVig,aswellasseminarparticipantsatUniversityofPennsylvania,LondonBusiness School(FinancialIntermediationTheoryWorkshop,theLondonBusinessSchoolSummerFinanceSymposium, and the 2nd EuroFIT Research Workshop on Syndicated Loans), Cambridge Judge Business School, ESMT Berlin, University of Zurich, the Jackson Hole Finance Conference, the 11th Annual Cambridge- PrincetonConference,andtheNotreDameConferenceonCurrentTopicsinFinancialRegulationformany helpful suggestions. We also thank Kelsey Shipman and Martin Sicilian for excellent research assistance. Saidi acknowledges funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)throughCRCTR224(ProjectC03). Theviewsexpressedhereinarethoseoftheauthorsandnotnecessarily those of the Federal Reserve System. The Shared National Credit data used here are confidential andwereprocessedsolelywithintheFederalReserveBoard. MaryChenworkedwiththeSharedNational CreditdataonlywhileasaresearchassistantattheBoardofGovernorsoftheFederalReserveSystem. †FederalReserveBankofBoston. mary.chen@bos.frb.org ‡FederalReserveBoardofGovernors. seung.j.lee@frb.gov §UniversityofTexasatAustin. daniel.neuhann@mccombs.utexas.edu ¶UniversityofBonn&CEPR.saidi@uni-bonn.de

1 Introduction The credit boom that culminated in the 2008 financial crisis was driven in part by the growing participation of non-bank institutional investors in capital markets (sometimes referredtoasshadowbanks),includingtheriseofsecuritization(IvashinaandSun,2011; Gorton and Metrick, 2013). In the market for corporate credit, these developments were mostdramaticinthecontextofsyndicatedlending,whereinstitutionalinvestorsincreasinglyboughtandsecuritizedloantranchesoriginatedbybanks.1 From the perspective of traditional financial intermediation theory, the entry of non-banks as passive lenders in corporate lending markets presents a puzzle. In the traditional view, monitoring by an informed intermediary is essential to lending. Hence, bank loans are hard to sell to other market participants with lower holding costs because banks have no incentive to monitor loans once they are sold. Why were institutional investors increasingly willing to buy bank loans despite these concerns? A common explanationisregulatoryarbitrage: sincebanksaresubjecttocostlyregulations,therearegains fromtradewithinstitutionsthatdonotfacetheseconstraints(Irani,Iyer,Meisenzahl,and Peydro´,2021). Thus,regulatorybank-capitalconstraintstippedthebalancetowardsmore participationbypassivelendersdespitethedrawbacksofworsemonitoring. Inthispaper,wearguethattheregulatory-arbitrageviewmaynotexplainthefull extentofnon-bankentry. WeusedatafromtheSharedNationalCredit(SNC)programto showthatbankde-regulationintheformoftherepealoftheGlass-SteagallActfacilitated the entry of non-bank intermediaries into the market for corporate credit during the pre- 2008 credit boom. More specifically, we argue that the repeal permitted the formation of universal banks that can efficiently monitor even when they retain relatively low loan sharesbecausetheyrealizeeconomiesofscopeacrosslendingandunderwritingservices. Institutional investors may therefore trust universal banks to monitor borrowers at loan shares at which they would not trust a stand-alone commercial bank to do so. Thus, our main insight is that specific forms of deregulation may increase, rather than decrease, thedegreeofcomplementarityincreditprovisionbetweenbanksandnon-bankfinancial 1IvashinaandSun(2011)estimatethatnearly70%oftheincreaseinsyndicated-loanissuanceduringthe 2001-2007creditboomisaccountedforbyinstitutionalfunding. NadauldandWeisbach(2012)showthat term-loanBfacilities,whichrepresentamajorityofthesyndicated-loantranchesobtainedbyinstitutional investors,weremuchmorefrequentlysecuritizedthanotherfacilities. Therefore,institutionalinvestment insyndicatedloansisanimportantstepinthesecuritizationprocess. 1

institutions. This leads to a more nuanced view of the effects of regulation on gains from tradeacrossfinancialinstitutions. On the theoretical side, we develop this argument using a simple repeated version of the Holmstro¨m and Tirole (1997) model of informed lending, where the scope for repeated interactions improves monitoring efficiency by reducing the cost of collecting information over time. An increase in monitoring efficiency then allows universal banks toselllargerloansharestouninformedinvestors,suchasinstitutionalinvestors,without compromising monitoring and loan quality. Testing these implications for loan liquidity requires exogenous variation in economies of scope. According to our proposed mechanism, this can be achieved using a shock to cross-selling opportunities, which serve to broadenintermediationrelationships(DruckerandPuri,2005). Suchashockisprovided byaparticularderegulatoryeventin1996,namelytheremovalofinformationalfirewalls between investment and commercial banking divisions among existing universal banks (NeuhannandSaidi,2018),leadingtoasharpincreaseincross-sellingintensity. Accordingly, we use this deregulatory episode to test whether after the shock to cross-selling, there was an increase in loan liquidity at origination and over time for universal-bankoriginatedloans,butnotforloansarrangedbyothertypesoffinancialinstitutions,suchaspure-playcommercialbanks. Documentingsuchaneffectconvincingly requires data on the entire life cycle of loans. This is because many non-bank investors purchasestakesinsyndicatedloansinthesecondarymarket,shortlyaftertheinitialsyndicationprocess (Lee,Li, Meisenzahl, andSicilian, 2019). Hence,we useSharedNational Credit(SNC)data,allowingustotrackloansharesovertime(asin,forinstance,Iraniand Meisenzahl, 2017; Bruche, Malherbe, and Meisenzahl, 2020). This dataset also offers the best available coverage of loan shares at issuance, which overcomes a crucial weakness of more common datasets on syndicated loans, typically DealScan, that has hindered research on loan-contracting mechanisms that need to be tested using lead and participant sharesasoutcomevariables. Overall,weestimatebank-scopederegulationtohavegeneratedadditionalliquidityinthesyndicated-loanmarketamountingupto$388bn,andatleast$119bnevenwhen holding credit demand fixed, over the period 1996-2008. This is sizable compared to the 2008 total amount of approximately $1.9tn (all in 2023 dollars) of corporate bonds outstanding on the books of some of the largest institutional investors, namely life and P&C 2

insurers(Becker,Opp,andSaidi,2022).2 We establish this finding in two steps. First, we focus on syndicate formation at origination. To estimate the treatment effect of expanding lead arrangers’ banking scope, we estimate a difference-in-differences specification at the loan level. In doing so, we compare the distribution of loan shares in syndicates arranged by at least one universalbankleadarrangerascomparedtoloansnotarrangedbyanyuniversalbanksbeforeand after1996. Ourkeyfindingisthatfollowingthe1996deregulation,universalbanks(UBs) retained around five-percentage-point smaller lead shares than did commercial banks (CBs),therebyfreeingupspaceforinstitutionalinvestorstoenterasparticipants. To strengthen the link to our theoretical mechanism, we turn to the cross-section of loans at origination. Our theory builds on differences in the cost of monitoring. This suggests that the effect should be particularly pronounced for firms with high ex-ante monitoring costs (where potential efficiency gains are large). We find this to be true: the drop in universal banks’ lead shares after 1996 amounts to ten percentage points (i.e., twice as large as our baseline effect) for borrower firms with higher sales-growth volatility (as a measure of ex-ante risk), and is non-existent in the subsample of safe firms. We also distinguish between term loans and credit lines. While the estimated effect on average lead shares is similar across credit lines and term loans, we find a stronger effect on total lead shares, i.e., the sum of all lead arrangers’ shares, for term loans. This partially reflects the fact that term loans tend to have more lead arrangers on average, but also points to an increase in the overall liquidity of such syndicated loans by increasing the total participant share by 6.5 percentage points—larger than our baseline estimate. Term loans could thus be viewed as the natural point of entry for institutional investors asparticipantsinsyndicatedloans,andparticularlysoafterthe1996deregulation. The second step is to evaluate the composition of syndicates over time. To do this, we exploit the fact that our data cover the trading of syndicated loans on the secondarymarket,andmoveouranalysistothemoregranularloan-yearbylendercategory level. This also enables us to control for time-varying unobserved heterogeneity at the loan level, including borrowers’ demand and loan quality (Irani and Meisenzahl, 2017). Consistent with the view that universal bank-originated loans are more liquid, we find that universal banks sell more of their lead share over time. Interestingly, the institution 2Notethatextrapolatingfromourresults, itseemsplausiblethattheadventofuniversalbankingmay haveincreasedalsotheextensivemarginofcreditsupply,butwecannotestimatethisdirectly. 3

playing the role of lead arranger is more likely to switch over time as well: institutional investors are more likely to become lead arrangers over the life of the loan if the original loanwasarrangedbyauniversalbank. More specifically, we estimate that this way institutional investors gain, on average, 0.9 percentage points in universal bank-originated syndicated loans, as opposed to syndicated loans arranged by other types of financial institutions, per year since the issuanceofaloan. Thisestimateincreasesto1.1percentagepointsperyearforcreditlines. This can be rationalized when firms’ observed behavior during the run-time of the loan produces information that can be used to monitor it more easily over time. In the case of creditlines,suchvaluableinformationisgeneratedbyfirms’draw-downbehavior. In contrast to lead shares, institutional investors’ participant shares of universal bank-originated loans do not increase as much over time (0.6 percentage points per year on average). In line with our findings for the cross-section of loans at the time of origination, however, we find a much stronger effect for term loans, such that institutional investors’participantshareincreasesby1.5percentagepointsperyearforuniversalbankoriginated term loans. This confirms our conjecture that term loans are the natural point ofentryforinstitutionalinvestorsasparticipantsinsyndicatedloans. Overall,wefindclearsupportforthehypothesisthatloansoriginatedbyuniversalbank lead arrangers are substantially more liquid. The advent of universal banking thus presentsacrucialentrypointforinstitutionalinvestorsinthemarketforcorporatecredit. Our findings suggest a visceral role for bank scope in determining the overall industrial organizationofcorporatecreditmarkets. Relatedliterature Our paper is most closely related to the literature linking changes in bank regulation to the entry of non-bank financial intermediaries (such as so-called shadow banks). Buchak, Matvos, Piskorski, and Seru (2018) study the market for residential mortgages and find that regulation accounts for roughly 60% of shadow banking growth, while Irani, Iyer, Meisenzahl,andPeydro´ (2021)arguethatbankcapitalregulationinducesless-capitalized banks to reduce loan retention in the market for corporate credit. These papers have in common the notion that tighter regulation leads to more migration of financial intermediationtothenon-banksector. Wecomplementthisviewbyarguingthatcertainformsof 4

de-regulation may induce such behavior as well. More broadly, Begenau and Landvoigt (2022), Buchak, Matvos, Piskorski, and Seru (2022), and Jiang (2023) consider the evolutionofbankboundariesrelativetoshadowbanksundervariousformsofregulation. Our specific focus is on the market for syndicated corporate loans. In this context, our paper is closely related to Blickle, Fleckenstein, Hillenbrand, and Saunders (2021) whoalsouseSNCdatatoshowthatleadarrangerssometimesselltheirentireleadshares, withoutanyadverseconsequencesforloanperformance. Ourevidenceisconsistentwith theirs, and our proposed mechanism provides a theory of why lead arrangers can sell their loan shares, namely that they have repeated interactions with the borrower firm. In thiscontext,ourresultscanbeinterpretedasshowingwhyuniversalbankswerethekey entrypointintolendingbynon-bankintermediaries. Our paper is also related to multiple strands of the literature on credit markets and heterogeneous financial intermediaries. First, we contribute to the literature documenting circumstances that favor non-bank entry, such as split control rights that mitigate bargaining frictions (Berlin, Nini, and Yu, 2020), and exit into lending, such as loan renegotiation (Beyhaghi, Nguyen, and Wald, 2019). Fleckenstein, Gopal, Gutie´rrez, and Hillenbrand (2021) point to the cyclicality of non-bank lending, while Aldasoro, Doerr, andZhou(2022)showthatnon-banksfinanceriskierfirmsglobally. Incontrasttothesepapers,wefocusonvariationthatallowstoascertainachannel through which banking deregulation facilitates non-bank lending. The key enabler is the effectofthederegulationofbankscope,andthesubsequentriseofuniversalbanking(see, among others, the seminal work by Puri, 1996; Gande, Puri, Saunders, and Walter, 1997; Drucker and Puri, 2005), on the distribution of shares retained by syndicate lenders— a relevant object of study already in early papers on syndicated loans (e.g., Sufi, 2007; Ivashina, 2009). The most important advance that we attempt to make is to account for heterogeneity in bank scope among syndicate lenders, differentiating at the very least betweenuniversalandpurecommercial/investmentbanks. Our main theoretical conjecture relates to universal banks’ monitoring efficiency in their role as lead arrangers. As such, our paper relates to previous work studying informed lenders in syndicated loans (e.g., Gustafson, Ivanov, and Meisenzahl, 2021, using similar data as we do) and the effect of their presence on loan liquidity (Santos and Shao,2018). Bylinkingthismonitoringadvantagebyuniversalbankstotheentryofnon- 5

bank lenders into the market for syndicated loans, our proposed mechanism is related to supply-sidedrivenexplanationsthatcanrationalizelowerleadshares,andsubsequently largerresidualparticipantshares,atorigination(asinourmodel)orloansalesafterorigination (for which we provide empirical evidence that is consistent with related theoretical work, such as Gryglewicz, Mayer, and Morellec, 2022). Also related is Hu and Varas (2022), who consider a dynamic version of the Holmstro¨m and Tirole (1997) model with limited commitment to loan retention. They show that banks may sell loan stakes and monitorlessovertime. Incontrasttoourpaper,theydonotconsiderbankheterogeneity inmonitoringexpertisedrivenbyeconomiesofscope. Ourfocusonthesupplysideofloanscomplementsstudiesthatfocusondemandside factors determining what firms borrow from non-banks (most notably Chernenko, Erel, and Prilmeier, 2022). Importantly, in our analysis of loan dynamics over the life cycle, we explicitly control for time-varying unobserved heterogeneity at the loan level, capturing borrower-level loan demand, and estimate differential effects for institutional (non-bank)investorsvs. banks. 2 Theoretical Framework We now introduce a simple model of informed lending based on Holmstro¨m and Tirole (1997) and use it to derive empirical predictions. There is a firm with capital A which f requires I − A in external funding to finance a project of size I. The project yields a f return of R if it is successful and 0 if it fails. The firm is run by an entrepreneur who can deliberately reduce the probability of success to enjoy a private benefit B. Shirking reduces the probability of success to p from p , with ∆ p = p − p > 0. The firm L H H L canobtainfinancingfromtwosources: outsideinvestors,whoareuninformedinthesense thattheydonotpossessanymonitoringexpertise,andintermediaries,whocanreducethe private benefit of shirking from B to b by exerting privately costly effort. The outside optionforoutsideinvestorsandintermediariesisaninvestmentwithnetrateofreturnγ. To allow for relationships and economies of scope, we repeat the model twice and let intermediaries become more efficient monitors over time. The cost of monitoring is c if the intermediary has not monitored the firm in the past, and c < c if it has. An H L H intermediarythathasmonitoredthefirminthepastissaidtobeexperienced. 6

Firms, intermediaries, and investors receive an endowment of A , A , and A , f m u respectively, at the beginning of each period and consume at the end of each period. Contracts are short-term, and the firm applies for funding anew at the beginning of each period. A is large enough such that outside investors can supply all required funds that u are not supplied by intermediaries, but outside investors are not willing to invest unless thefirmismonitoredbyanintermediary,theleadarranger. We start our analysis in the second period. Let the project’s payoffs be divided up sothat R +R +R = R,where R , R ,and R denotethereturnsaccruingtothefirm, f m u f m u theintermediary,andoutsideinvestors,respectively. Assumingthatthefirmismonitored byanintermediary,thefirm’sincentiveconstraintisR ≥ b ,whileanintermediarywith f ∆p cost c ∈ {c ,c } preferstomonitorif R ≥ c . L H m ∆p Let I denote the capital lent to the firm by the intermediary. Since monitoring is m costly, firms prefer uninformed to informed intermediary capital if possible and borrow just enough from intermediaries in order to ensure monitoring incentives. In the context of syndicated lending, I can be interpreted as the lead arranger’s loan share, or lead m share. If there are no experienced intermediaries, perfect competition among intermediaries implies that the participation constraint binds. Hence, the lead share is IH = p L c H m γ∆p andthepromisedpaymentis RH = c H. m ∆p Plentiful intermediary capital is not sufficient to dissipate all rents when there is an experienced intermediary since she can use her cost advantage to undercut all competitors and still earn excess profits. The worst case for the firm is that the experienced intermediary acts as a monopolist and offers exactly the same terms as an inexperienced intermediary, retaining all rents for herself. In this case, the incentive constraint is slack and the rent is equal to the difference in monitoring costs, ∆ c ≡ c −c . The best case is H L that she behaves competitively and invests IL = p L c L in exchange for payment RL = c L . m γ∆ m ∆p Theexactdivisionofthesurplusisimmaterialtoouranalysis. Hence,weassumethatthe experienced intermediary offers a weighted average of the “monopolist” and “perfectcompetition” contracts, with her bargainingpower 0 < µ < 1 determining the weight on themonopolistcontract: p (c +µ ∆ c) c +µ ∆ c I ∗ = H L and R (µ) = L . m ∆ m ∆ γ p p ∆ This contract delivers rents µ c to the experienced intermediary, and it lowers the lead 7

share because the experienced intermediary is a more efficient monitor. Lending experiencethusmakesbankloanscheaperandmoreliquid. We now turn to the first period where all intermediaries face monitoring cost c . H The firm’s problem is the same as above. However, intermediaries take into account that being experienced tomorrow has the promise of additional rents which we summarize by v(µ). We parameterize the intermediary’s probability of being the firm’s monitor tomorrow conditional on monitoring the firm today by α ∈ [0,1]. We use α to reflect the probability of repeated interactions, and interpret universal banking as a positive shock to α. Sincethis raises thevalue ofmonitoring, the scopefor relationshipbanking leads to a lower effective cost of monitoring, cˆ(α) = c −αµ ∆ c. The intermediation contract and H theincentiveconstraintarethengivenby: p cˆ cˆ I ∗∗ = L and R ∗∗ ≥ . m ∆ m ∆ γ p p Hence, the promise of future rents relaxes financial constraints for the firm today, and makesbankloansmoreliquidbyreducingtheleadshare. 2.1 Empirical Predictions We now describe the model’s empirical content. We consider banks to be informed intermediaries and institutional investors (non-bank intermediaries) to be uninformed investors. Hence,theleadshareis I ,andtheparticipantshareis I− A − I . Weinterpret m f m the advent of universal banking and subsequent deregulation of bank scope as a shock toα,theprobabilityofrepeatedinteractionsforaninformedintermediary. Thetreatment effectofuniversalbankinginagivenperiodis ∗∗ ∆ ∂I p µ c m = − L . ∆ ∂α γ p We refer to loans where a universal bank is the lead arranger as UB-led, and those where a commercial (or any non-universal) bank is the lead arranger as CB-led. We then have thefollowingempiricalpredictionsatorigination. EmpiricalPrediction1(Loansharesatorigination) RelativetoCB-ledloans, (i) UB-ledloanshavelowerleadsharesandhigher(total)participantshares. 8

(ii) The UB treatment effect is larger for risky firms with high default risk, p , and for opaque L ∆ firmswhereeconomiesscopeinmonitoring, c,areparticularlylarge. Repeated interactions are more likely to occur over longer time horizons, and economies ∆ ofscopeinmonitoringaccumulateaslendingrelationshipsdeepen. Thissuggeststhat c increases disproportionately over time for UB-led loans, allowing uninformed investors suchasinstitutionalinvestorstoincreasetheirloanshareovertime. EmpiricalPrediction2(Lifecycle) The participation (in any capacity) by institutional investorsinloansinitiallyarrangedbyUBsincreasesovertime. ∆ Unless the total size of the loan varies during its run-time, the fact that an increase in c duetouniversalbankingleadstoasmallerleadshare I canbeinterpretedasatransferof m leadsharesfromuniversalbankstoa(new)groupofinstitutionalinvestors. Alternatively, itcanbeinterpretedasreflectingimprovedliquidityofparticipantshares,evencompared to primary market trading. This would, in turn, imply a transfer of participant shares from equally uninformed lenders—even if they are universal banks because they were notleadarrangers—toinstitutionalinvestors. An important source of heterogeneity between loans is whether the loan is a term loan or a credit line. A term loan is more likely to require monitoring of specific firm actions, while a credit line is more likely to require monitoring of the firm as a whole. A credit line that has not yet been drawn down can be considered otherwise equivalent to a term loan. As firms draw down on their credit lines over time, this discloses additional information. Thus, it should become easier to monitor the firm over time as its fundamental type is revealed (Botsch and Vanasco, 2019). This suggests that universal banks can disproportionately reduce their lead share in credit lines over time, and previously uninformed investors that have observed the firm over time can take over their lead shares, whereas these institutional investors are more likely to enter term loans as participants. EmpiricalPrediction3(Life-cycleheterogeneity) LeadsharesinitiallyheldbyUBsdropover time,andmoresoforcreditlinesthanfortermloans. Institutionalinvestorsaremorelikelytobuy lead shares of UB-arranged loans in credit lines, whereas they increase their participant share in termloansovertime. 9

3 Empirical Strategy and Data Wenextdiscussouridentificationstrategybasedonthebank-scopederegulationfollowing the stepwise repeal of the Glass-Steagall Act. Then, we will describe our administrativedataonsyndicatedloansandsampleselection. 3.1 Identification Strategy An important prerequisite for estimating the impact of bank scope on syndicate structures is a setting that provides variation in bank scope. The stepwise repeal of the Glass- Steagall Act constitutes such a setting (Neuhann and Saidi, 2018). The Glass-Steagall Act of 1933 imposed a separation of commercial banking (deposit taking and lending) and investmentbanking(especiallyunderwritingofcorporatesecurities). StartingApril30,1987,commercialbankswereallowedtobecomeuniversalbanks, and generate up to 5% of their gross revenues from underwriting and dealing in securities other than corporate debt and equity. The first major step of the repeal took place in January and September 1989, which is when commercial banks could generate a higher fraction (10% in 1989, which increased to 25% in 1996) of their revenues through underwritingactivities,includingunderwritingofcorporatedebtandequity. Commercialbanks became universal banks typically by opening so-called Section 20 subsidiaries for these purposes. Anotherpossibilitywastoacquireaninvestmentbank. While this first step towards universal banking led to an increase in bank size by allowingbankstoengageinbothcommercialandinvestmentbanking,firewallsseparatingthetwoactivitiesremainedinplace. Someoftheinformationalandfinancialfirewalls within bank-holding companies were, however, abolished by the Federal Reserve Board inasecondsteponAugust1,1996. Theeliminationofthesefirewallsbetweencommercial bankingandsecuritiesdivisionsenableduniversalbankstocross-sellloansandnon-loan products, which used to be severely restricted, not to say forbidden, under the Federal Reserve Act (Sections 23A and B). Furthermore, the removal of informational firewalls allowedforthepossibilityofsharingnon-publiccustomerinformationbetweencommercialbankingandsecuritiesdivisions. We wish to test whether banks of wider (deregulated) scope retain smaller shares when arranging syndicated loans. In our model, the underlying mechanism is that uni- 10

Figure1: FractionofSyndicatedLoansCross-soldbyUniversalBanks Notes: Thisfigureplotsforeachyearfrom1987to2002thefractionofsyndicatedloansarrangedbyatleast one universal bank (from DealScan) that is observed to have also served as the lead underwriter of any equityordebtoffering(asrecordedinSDC)bythesameborrowerfirmsanytimefromthesameyearthe loanwasissuedupuntiltheendofthefourthyearthereafter. versalbankshavedeeperbank-firmrelationships,forexamplethroughcross-sellingloans and non-loan products. Thus, we hypothesize that universal, rather than commercial or investment, banks retain smaller shares of loans when their ability to enter deeper bankfirmrelationshipsisstrengthened. We use the 1996 deregulation as a shock to universal banks’ ability to cross-sell loansandunderwritingservicesandreapinformationaleconomiesofscopethisway. As argued in Neuhann and Saidi (2018), the proportion of cross-sold loans increased significantlyforuniversalbanks,ratherthaninvestmentbanks,afterthe1996deregulation. We useRefinitivDealScanandSecuritiesDataCompany(SDC)Platinumdatatovalidatethis assumption.3 Figure 1 shows that the proportion of syndicated loans with a universal- 3Data are from Refinitiv, Dealscan and LoanConnector, Wharton Research Data Services, https:/wrdsweb.wharton.upenn.edu/wrds/;andRefinitiv,ThomsonONEInvestmentBankingandDealsmoduleand SDCPlatinum,http://www.thomsonone.com/. 11

bank lead arranger granted to firms whose debt or equity was underwritten by the same universal bank in the subsequent five years (from t until year-end t+4) increased substantially around 1990, shortly after the revenue limit was elevated for the first time, and thenagainaroundthemid-1990s.4 Against this background, we employ a difference-in-differences strategy akin to Neuhann and Saidi (2018) around 1996 for treated universal banks vs. other banks that were unaffected in their scope of banking activities. In a first step, we analyze the syndicatestructureofloansarrangedbythesedifferentgroupsofbanks. Eachsyndicatedloan isapackagethatconsistsofoneormultiplefacilitieswhich,inturn,consistofloanshares provided by one or multiple syndicate lenders. To estimate the effect on total or average lead shares (across all arrangers a) at the package level l (representing a syndicated loan l granted at date t to firm f in industry i(f)), we estimate the following regression specification: Leadshare = β Arrangedbyuniversalbank × After(1996) l 1 l t +β Arrangedbyuniversalbank +µ +δ +ϵ , (1) 2 l a j(f)t l where the dependent variable is either the total or the average share (in %) of the loan retained by all lead arrangers, Arranged by universal bank is an indicator variable for l whetheranyoneoftheleadarrangersisauniversalbankatthetimeofissuance, After(1996) t is an indicator for whether the loan was issued after 1996, µ denotes bank fixed effects a for all lead arrangers of loan l, and δ denotes borrower firm f’s (two-digit) industry j(f)t byyearfixedeffects. Standarderrorsareclusteredatthelevelofallleadarrangers. AsArrangedbyuniversalbank isaloan-levelcharacteristicthatreflectswhetherany l one of the lead arrangers is a universal bank, we can separately include fixed effects for eachleadarrangerassociatedwithloanl. Eveninthe(common)caseofasyndicatedloan havingonlyoneleadarranger,wecanestimateacoefficientonArrangedbyuniversalbank l inthepresenceofarrangerfixedeffectsbecausewetrackcommercialbanksthatmayhave opted to become universal banks after their first loan transaction in the data. As a result, Arrangedbyuniversalbank can vary within certain types of arrangers, namely commerl cialbanksthateventuallybecomeuniversalbanks. Thedifference-in-differencesestimate 4We would not expect to find a clear effect in 1996 or 1997, however, as one can only noisily infer the actual timing of cross-selling loans and underwriting services from the issue dates of the two types of financialassets. 12

β is then identified using commercial banks that became universal banks prior to the 1 deregulation and, therefore, experienced an expansion in the scope of their activities in 1996. That is, to estimate β and β , a given lead arranger a needs to be observed in at 1 2 leastthreeinstances: whenitwasstillacommercialbank(capturedbythearrangerfixed effects), after it opted to become a universal bank but before the 1996 deregulation (β ) 2 and, finally, as a universal bank after the 1996 deregulation (β ). The omitted category 1 consists of other types of lenders, including commercial and investment banks but also institutionallenders,whosescopedidnotincreasefollowingthe1996deregulation. As our data cover secondary-market trading of syndicated loans, we can also analyze the development of loan shares held over time, and differentiate by three types of lenders: universal banks, other banks, and institutional investors (or non-banks). In particular, we are interested in the development of shares held by institutional investors of loans initially arranged by universal banks. To this end, we estimate the following regressionspecificationattheloan-yearbylendercategorylevel lit: Share = β Institutionalinvestor ×Arrangedbyuniversalbank ×Yearssinceissue lit 1 i l lt +β Institutionalinvestor ×Arrangedbyuniversalbank 2 i l +β Institutionalinvestor ×Yearssinceissue +θ +ϕ +ϵ , (2) 3 i lt lt it lit wherethedependentvariableisthetotalshare(in%)ofloanl retainedbyparticipantsor lead arrangers in lender category i in year t, Institutional investor is an indicator variable i forloansharesheldbyinstitutionalinvestors,asopposedtotheremainingtwocategories oflenders(universalandnon-universalbanks),Arrangedbyuniversalbank isanindicator l variable for whether any one of the lead arrangers is a universal bank at the time of issuance, Years since issue is the difference in years between t and the year in which loan lt l isissued, θ and ϕ denote,respectively,loanbyyearandlendercategorybyyearfixed lt it effects. Standarderrorsareclusteredatthelevelofallleadarrangers. 3.2 Data Description Our main object of analysis relates to the distribution of loan shares within syndicates, at origination as well as over time. As pointed out by, among others, Bruche, Malherbe, and Meisenzahl (2020), loan shares are poorly filled in the standard database on syn- 13

dicated loans, DealScan. What is more, the DealScan database covers only the primary syndicated-loanmarketand,assuch,doesnottrackloansharesthatareeventuallytraded in the secondary market. This is of particular relevance for institutional investors, many of which enter syndicated loans through acquiring loan shares of term loans in the secondarymarket(IvashinaandSun,2011;NadauldandWeisbach,2012). To address these challenges, we use data from the Shared National Credit (SNC) program,whichwasestablishedin1977bytheBoardofGovernorsoftheFederalReserve System, the Federal Deposit Insurance Corporation, and the Office of the Comptroller of theCurrency tofacilitatereviewsof largesyndicatedloans. Upuntil 2017,SNCincluded loans larger than $20 million that are shared by three or more supervised institutions.5 Information about a loan is provided by a designated bank—usually an agent bank. One or more agent banks are generally responsible for coordinating participating lenders, negotiating the contractual details, and preparing adequate loan documentation. Once the loanismade,agentbanksarealsoresponsibleforloanservicing. We use annual data with report dates from 1992 to 2012. Starting in 1992, the data are available in a consistent format. After 2012 the behavior of agent banks and institutional lenders in the syndicated-loan market was heavily influenced by the Leveraged Lending Guidance (Calem, Correa, and Lee, 2020; Aramonte, Lee, and Stebunovs, 2022). Asoftheendof2012,theSNCdatabasecoveredapproximately9,300syndicatedloansto 5,800 borrowers, for a total of $3 trillion in drawn credit and commitments (commitment isthemaximumamountlendersagreetoprovide). The SNC database includes the full syndicate membership for each recorded loan as well as each lead arranger’s and each participant’s outstanding and committed shares of the loan. Moreover, because regulators collect the same loan across time, syndicate membership changes can be observed in these data.6 Since some loans are reported once but reviewed multiple times, we only keep the observations with the most recent review date. We group SNC loan types into “Term,” “Revolver” and “Other,” and we drop loans classified as “Other.” We drop any observations that have a different loan type than the loan type at origination. We drop loan-years that do not have any identified lead arrangers. We also drop loan-lender-year observations with negative loan amounts. 5Startingin2018,thethresholdincreasedto$100millionandthefrequencyofdatasubmissionsmoved toquarterly. 6Loanamountsareinflationadjustedto1992dollars. 14

Finally,wedroploanswithoriginationdatesthatoccurafterthereportdate. We distinguish between commercial/investment and universal banks by matching,usingtheRpackagefedmatch(Cohen,Dice,Friedrichs,Gupta,Hayes,Kitschelt,Lee, Marsh,Mislang,Shaton,Sicilian,andWebster,2021),namesofSNClendersandnamesof top holders of SNC lenders to a list of names of all commercial banks and the dates they became universal (from Neuhann and Saidi, 2018). This enables us to classify lenders in thesyndicated-loanmarketascommercialbanksthatwerenotyetuniversalbanksatthe time of loan issuance, universal banks, commercial banks that never became universal banks, investment banks, and non-banks. We use the DealScan lender IDs of universal banks,soweaugmentouruniversal-bankindicatorusingaDealScan-SNClendermatch. If any of these methods matches to the list and the report date is after the date the bank became universal, then the lender is marked as a universal bank in the SNC data. SNC also has general entity types that categorize lenders on a syndicate as a U.S. bank, a foreign bank, or a non-bank. We use this broad definition to assume that non-banks are institutionalinvestors.7 We use a random forest model to predict lead arrangers in SNC. The verified data (training data) of 15,515 loan-level observations is from a SNC-DealScan match and uses the lead-arranger information in DealScan. If a lender is not a domestic bank, foreign bank,financecompany,orbroker-dealer,weassumethatitisnotaleadarranger. Theindependent variables of the model are SNC entity type, commitment share, commitment total,adummyforrecessionduringorigination,originationyear,reportyear,timedifferencebetweenreportdateandorigination,andadummyforwhetherthelenderwasever a lead arranger for the borrower on another loan. We make some changes to the SNC entity type based on our own entity categorizations. If we classify a lender as a domestic bank, foreign bank, finance company, or broker-dealer, but a more granular form of SNC entity type categorizes it as a “Domestic Entity Other” (DEO) or “Foreign Entity Other” (FEO), then we change the SNC entity type to “National Bank” (NAT), “Foreign Bank” (FBK),“FinanceCompany”(FNC),or“SecuritiesBroker/Dealer”(SBD),respectively. We classify lead arrangers using a decision threshold of a false positive rate of 0.1. Therefore,weareconservativeinlimitingfalsepositives,oridentifyingalenderasalead arranger when it is not. Agent banks in the SNC data are automatically classified as lead 7Similar distinctions are made in Calem, Correa, and Lee (2020) and Aramonte, Lee, and Stebunovs (2022). 15

arrangers outside of the model (given a model prediction of 1). We take only the top ten lead arrangers by model prediction. In the verified set, the largest number of lead arrangers for one loan was 29, but there were only 58 out of 15,515 with more than ten leadarrangers. Finally, we estimate separate models for term loans and revolver loans (credit lines). For the verified data, the model correctly identifies most of the actual lead arrangers for term loans. This is more difficult to achieve for revolvers because they attract predominantly banks, rather than other types of investors, and banks are more likely to be lead arrangers. However, increasing the false positive rate does not lead to material changesinourresultsatthegranularloan-yearbylendercategorylevel(see(2))because theinstitutionalsharesoftheloansdonotchangemuchdependingonthefalse-positiveratethreshold. Summarystatistics Table 1 presents summary statistics for both levels of analysis. In the top panel, we include summary statistics for variables employed in our loan-level (or package-level l) analysis. The average lead share per institution is somewhat smaller than the average share retained by all lead arrangers of a given loan, reflecting the fact that the average number of lead arrangers per syndicated loan is greater than one (1.5). Out of all loans duringoursampleperiodfrom1992to2012,78.7%haveatleastoneleadarrangerthatis auniversalbankatthetimeofissuance. In the bottom panel, we consider the more granular (and dynamic) loan-year by lendercategorylevellit. Aswesummarizetotalleadorparticipantsharesforeachlender category separately, a given observation can indicate a zero share if no such lender participates in any capacity in a given loan. As such, the average values for total lead and participant shares are naturally smaller (and do not add up to 100% either) than they are inourloan-leveldataset. Incontrast,thesummarystatisticsforArrangedbyuniversalbank l remainroughlysimilaracrossthetwolevelsofobservation. 16

Figure2: TotalFlowandStockofInstitutionalHoldingsofSyndicatedLoans (a)Flow (b)Stock Notes: The top panel plots the dollar volume of SNC loans held by institutional investors for those loans originatedduringeachyearfrom1992to2012bywhetherthesyndicatedloanswerearrangedbyatleast oneuniversalbank(UB-led)ornot(allothersyndicatedloans). Thebottompanelplotsthedollarvolume ofSNCloansheldbyinstitutionalinvestorsatyear-endfrom1992to2012bywhetherthesyndicatedloans werearrangedbyatleastoneuniversalbank(UB-led)ornot(allothersyndicatedloans). 4 Results We start with graphical evidence. The top panel of Figure 2 shows that the flow participation (in any capacity) by institutional investors in syndicated loans has been steadily increasing since the late 1990s up until the Great Financial Crisis, after which it showcases some cyclicality (Fleckenstein, Gopal, Gutie´rrez, and Hillenbrand, 2021). In terms of the stock of such institutional holdings, the maximum was reached around the Great 17

Figure3: FlowandStockofUB-ledvs. OtherSyndicated-loanHoldingsoutofInstitutionalSyndicated-loanPortfolio (a)Flow (b)Stock Notes: Thetoppanelplotstheshareofuniversalbank(UB)ledsyndicatedloansandtheshareofothersyndicatedloansininstitutionalinvestors’syndicated-loanportfolioforSNCloansoriginatedduringeachyear from1992to2012. Thesharesadduptooneeachyear. Thebottompanelplotstheshareofuniversalbank (UB)ledsyndicatedloansandtheshareofothersyndicatedloansininstitutionalinvestors’syndicated-loan portfolioforSNCloansattheendofeachyearfrom1992to2012. Thesharesadduptooneeachyear. FinancialCrisisinoursample(seebottompanelofFigure2). Inbothinstances,thetrends pertain to universal bank-arranged syndicated loans rather than loans arranged by other types of financial institutions. Our empirical evidence speaks to these developments insofar as the 1996 deregulation enabled universal banks to arrange syndicated loans at smallerleadshares,therebyfreeingupspaceforinstitutionalinvestorstoparticipate. TheshareofUB-ledsyndicatesininstitutionalinvestors’syndicated-loanportfolio is roughly stable since 1996 (see Figure 3). In conjunction with the fact that institutional- 18

investorparticipationinanysyndicatedloansincreasedduringthesameperiod,thissuggests that institutional investors’ demand for UB-led loans was constant, while the supply thereofincreasedaroundthe1996deregulation. We next turn to formal tests. To shed light on the supply-driven explanation for increased institutional participation, weestimate specification (1)at theloan level onour sample of syndicated loans, and use the time window from 1992 to 2002 around the 1996 deregulation. Westartin1992soastoexcludethetwoprecedingderegulatoryeventsthat relaxed universal banks’ revenue limits on underwriting and, thus, cross-selling loans andnon-loanproducts. Table2presentstheresults. InlinewiththefirstpartofEmpiricalPrediction1,we findthatloansarrangedbyanyuniversalbank(s)carryafivepercentage-pointsmallertotalandaverageleadshareafterthe1996deregulation(columns1and2). Theseestimates holduptoincludingindustry-yearfixedeffectsincolumns3and4. Comparedtosample averages,totalandaverageleadsharesinUB-ledloansarelowerbyapproximately13.7% and 17.5%, respectively. This represents a substantial increase in loan liquidity in the aftermath of the 1996 deregulation, amounting to $388bn (in 2023 dollars) over the period 1996-2008,8 ifoneassumesthatthisfreed-upspaceisusedentirelybynon-banks. Note that the positive coefficient on Arranged by universal bank is estimated using l commercial banks that chose to become a universal bank before the 1996 deregulation. As such, their taking larger shares reflects their increased lending capacity associated withtheirexpansion inbanksizeand operations. Importantly, before1996informational firewallsarestillinplace,whichgreatlyinhibituniversalbanks’abilitytobenefitfrominformational economies of scope by cross-selling loans and non-loan products. The effect of the latter is reflected solely by the coefficient on the interaction term, β in (1), control- 1 ling for (and, thus, conditional on) all other properties of universal banks that come into existencewhencommercialbanksdecidedtoswitchtouniversalbankingbefore1996. The estimated effects for total and average lead shares are relatively similar because most syndicated loans in the U.S. have only one lead arranger. This is particularly true for credit lines, which we consider in columns 1 and 2 of Table 3. For term loans, we instead find a particularly large effect on total lead shares (see columns 3 and 4 of the 8To arrive at this estimate, we multiply for each loan issued during the relevant time period the total volume with an indicator for whether any one of the lead arrangers was a universal bank and then with 5.0%,correspondingtoourestimateinTable2. 19

Figure4: LoanSharesHeldbyInstitutionalInvestorsbyYearforAllSyndicatedLoans (a)LeadArrangers (b)Participants Notes: The top panel plots the fraction of SNC loans held by institutional investors as lead arrangers of thoseloansoriginatedduringeachyearfrom1992to2012bywhethertheloanswereinitiallyarrangedby atleastoneuniversalbank(UB-led)ornot(allothersyndicatedloans). Thebottompanelplotsthefraction ofSNCloansheldbyinstitutionalinvestorsasparticipantsofthoseloansoriginatedduringeachyearfrom 1992to2012bywhethertheloanswereinitiallyarrangedbyatleastoneuniversalbank(UB-led)ornot(all othersyndicatedloans). same table). Since the estimated effect for average lead shares is similar for credit lines and term loans, the difference in the effect on total lead shares is accounted for by the factthattermloanstendtohaveahighernumberofleadarrangersonaverage. Sincethe overallliquidityofsyndicatedloansisdeterminedbythetotalparticipantshare,ourfindingsthusindicatethattermloansarethenaturalpointofentryforinstitutionalinvestors, 20

andparticularlysoafterthe1996deregulation. The second part of Empirical Prediction 1 implies that treatment effects should be particularlylargeforriskyandopaquefirmsforwhichmonitoringcostsarehighexante. Consistentwiththisprediction,wefindtheleadshare-reducingeffectofuniversalbanks toexistonlyforriskyborrowers,asmeasuredbythelatter’ssix-yearsales-growthvolatility(seecolumns1and2vs. columns3and4inTable4). Forriskyborrowers,UB-ledsyndicatesseeadeclineintheirtotalandaverageleadsharesbyninetotenpercentagepoints followingthe1996deregulation. Thisindicatesasubstantialincreaseinloanliquidityfor thetypesoffirmsthatwerepreviouslyparticularlydifficulttosyndicate. Next, we turn to the evolution of lead and participant shares over time, which are tracked at an annual frequency in our data. Empirical Prediction 2 suggests that lead arrangersinUB-ledloansmayreducetheirloansharesovertime,therebyfurtherincreasing loan liquidity. This can occur in two ways: either by sales of lead shares to new lenders who remain lead arrangers (thus leaving lead shares unchanged at the loan level), or by salestonewlenderswhoactasparticipants(therebyreducingtheoverallleadshare). Tables 5 and 6 examine the evolution of total and average lead shares over a fiveyearhorizonusingthesameregressionspecificationasabove. Weestimatevirtuallyconstantdifference-in-differencesestimatesacrossallfiveyears. Thisindicatesthatthetreatment effect on lead and participant shares does not vary over time. In particular, lead sharesarenotconvertedtoadditionalparticipantsharesovertime. This leaves the possibility that lead shares themselves become more liquid when theloanisoriginatedbyauniversal-bankleadarranger. Figure4showsthatinstitutional investors enter UB-led syndicates both as lead arrangers and participants, though the shareasleadarrangersisquitesmall. To investigate this further, we estimate (2) at the granular loan-year by lender category level. In these regressions, we can control for time-varying unobserved heterogeneity at the loan level, subsuming any developments that may affect all lenders in a syndicated loan equally, including borrower-level shocks and loan quality (Irani and Meisenzahl, 2017). We also control for lender category by year fixed effects. This rules out that our results are driven by unobserved shocks affecting all investments by a given lender. In Table 7, we test whether institutional investors purchase lead shares in initially 21

Figure5: LeadSharesHeldbyInstitutionalInvestorsbyYear (a)SyndicatedRevolvers (b)SyndicatedTermLoans Notes: The top panel plots the fraction of SNC revolver loans held by institutional investors as lead arrangersofthoseloansoriginatedduringeachyearfrom1992to2012bywhethertheloanswerearranged byatleastoneuniversalbank(UB-led)ornot(allothersyndicatedloans). Thebottompanelplotsthefraction of SNC term loans held by institutional investors as lead arrangers of those loans originated during eachyearfrom1992to2012bywhethertheloanswerearrangedbyatleastoneuniversalbank(UB-led)or not(allothersyndicatedloans). UB-led loans over time. We find that institutional investors join as lead arrangers later duringtheloan’srun-time,withanon-negligibleintensityatalmostonepercentagepoint foreachyearafterloanissuance,asreflectedbythecoefficientonthetripleinteractionin column1. Thatis,thesharearrangedbyuniversalbanksdropsovertime,allowinginstitutional investors to step in as lead arrangers. This is in line with the phenomenon that lead arrangers at times sell their shares in the secondary market (Blickle, Fleckenstein, 22

Figure6: ParticipantSharesHeldbyInstitutionalInvestorsbyYear (a)SyndicatedRevolvers (b)SyndicatedTermLoans Notes: ThetoppanelplotsthefractionofSNCrevolverloansheldbyinstitutionalinvestorsasparticipants ofthoseloansoriginatedduringeachyearfrom1992to2012bywhethertheloanswerearrangedbyatleast oneuniversalbank(UB-led)ornot(allothersyndicatedloans). ThebottompanelplotsthefractionofSNC term loans held by institutional investors as participants of those loans originated during each year from 1992to2012bywhethertheloanswerearrangedbyatleastoneuniversalbank(UB-led)ornot(allother syndicatedloans). Hillenbrand,andSaunders,2021),whichwecomplementbypointingoutthatthederegulationofbankscopeenablesinstitutionalinvestors,andnotjustotherbanks,topurchase themaswell. EmpiricalPrediction3furthersuggeststhattheloantype(termloanorcreditline) is an important determinant of whether institutional investors enter a syndicate as lead arrangers or participants. Figures 5 and 6 show that institutional investors enter syn- 23

dicates initially arranged by universal banks as lead arrangers slightly more for credit lines(thoughthesharesareverysmall),andasparticipantspredominantlymoreforterm loans. Inlinewiththegraphicalevidence,thecoefficientonthetripleinteractioninTable 7islargerforcreditlinesthanfortermloans(columns2and3),andforriskierborrowers (columns 4 and 5). The latter reflects our conjecture in Section 2.1 that observing firms that borrow from universal-bank lead arrangers reduces asymmetric information over time especially for credit lines with firms’ observable draw-down behavior. This is es- ∗∗ ∆ pecially valuable if the borrower is risky (as I is decreasing in c at a greater rate for m higherlevelsof p inourmodel). L In Table 8, we then consider participant instead of lead shares. The coefficient on the triple interaction in column 1 indicates that the total participant share, i.e., the sum of all participant shares, in UB-led loans held by institutional investors does not increase much over time. At a rate of 0.56 percentage points per year, one would require a runtime of nine years to match the reduction in the lead share (of 5 percentage points) in Table2wherewe,however,donotcontrolforborrowers’demandaswedoinTable8by includingloanbyyearfixedeffects. Thisestimatelendssupporttotheideathatifinstitutionalinvestorsstepinasparticipants, they tend to do so shortly after the initial syndication process (Lee, Li, Meisenzahl, and Sicilian, 2019). In line with our Empirical Prediction 3, this is different for term loans (column 3), where institutional investors tend to hold larger participant shares in UB-ledloansovertimethanisthecaseforcreditlines(column2). Wealsofindastronger, albeitborderlineinsignificant,effectforriskyborrowers(columns4and5). Using these more conservative estimates, and accounting for the possibility that institutional investors may not only enter syndicated loans as participants but can also take over as lead arrangers, we assess bank-scope deregulation to have generated additional liquidity amounting to at least $119bn (in 2023 dollars) over the period 1996-2008. This figure incorporates the average time of entry by institutional investors, assuming thatitisuniformlydistributedovertherun-timeoftheloan.9 9Inparticular,wemultiplyforeachloanissuedduringtherelevanttimeperiodthetotalvolumewithan indicatorforwhetheranyoneoftheleadarrangerswasauniversalbank,with1.48%,whichcorrespondsto thesumofthetwocoefficientsinthefirstcolumnofTables7and8,andfinallywiththemaximumrun-time ofagivenloaninthedatadividedbytwo. 24

5 Conclusion Banking deregulation in the form of the repeal of the Glass-Steagall Act was instrumental in driving the growth of non-bank lenders in the market for corporate credit. The formation of universal banks created economies of scope in bank lending that allowed passiveinvestorstobuylargerloanshareswithoutcompromisingloanquality. Ourfindings complement the conventional view that the rise of non-bank lenders was driven by regulatory arbitrage due to tight banking regulations. More broadly, our results indicate that regulation shapes the industrial organization of lending markets in complex ways, so that the tightness of regulation is not a sufficient statistic for predicting the migration of specific activities to the non-bank sector. Thus, our findings offer important insights forthedesignforsystemicpolicies. 25

References ALDASORO, I., S. DOERR, AND H. ZHOU (2022): “Non-bank Lending During Financial Crises,”BISWorkingPaper. ARAMONTE, S., S. J. LEE, AND V. STEBUNOVS (2022): “Risk Taking and Low Longer- Term Interest Rates: Evidence from the U.S. Syndicated Term Loan Market,” Journal of Banking&Finance,138,105511. BECKER, B., M. M. OPP, AND F. SAIDI (2022): “Regulatory Forbearance in the U.S. Insurance Industry: The Effects of Removing Capital Requirements for an Asset Class,” ReviewofFinancialStudies,35(12),5438–5482. BEGENAU, J., AND T. LANDVOIGT (2022): “Financial Regulation in a Quantitative Model oftheModernBankingSystem,”ReviewofEconomicStudies,89(4),1748–1784. BERLIN, M., G. NINI, AND E. G. YU (2020): “Concentration of Control Rights in LeveragedLoanSyndicates,”JournalofFinancialEconomics,137(1),249–271. BEYHAGHI, M., C. NGUYEN, AND J. K. WALD (2019): “Institutional Investors and Loan Dynamics: Evidence from Loan Renegotiations,” Journal of Corporate Finance, 56, 482– 505. BLICKLE, K., Q. FLECKENSTEIN, S. HILLENBRAND, AND A. SAUNDERS (2021): “The MythoftheLeadArranger’sShare,”NYUSternWorkingPaper. BOTSCH, M., AND V. VANASCO (2019): “Learning by Lending,” Journal of Financial Intermediation,37,1–14. BRUCHE, M., F. MALHERBE, AND R. R. MEISENZAHL(2020): “PipelineRiskinLeveraged LoanSyndication,”ReviewofFinancialStudies,33(12),5660–5705. BUCHAK, G., G. MATVOS, T. PISKORSKI, AND A. SERU (2018): “Fintech, Regulatory Arbitrage,andtheRiseofShadowBanks,”JournalofFinancialEconomics,130(3),453–483. (2022): “Beyond the Balance Sheet Model of Banking: Implications for Bank RegulationandMonetaryPolicy,”JournalofPoliticalEconomy,forthcoming. CALEM, P., R. CORREA, AND S. J. LEE (2020): “Prudential Policies and their Impact on CreditintheUnitedStates,”JournalofFinancialIntermediation,42,100826. CHERNENKO, S., I. EREL, AND R. PRILMEIER (2022): “Why Do Firms Borrow Directly fromNonbanks?,”ReviewofFinancialStudies,forthcoming. COHEN, G. J., J. DICE, M. FRIEDRICHS, K. GUPTA, W. HAYES, I. KITSCHELT, S. J. LEE, W. B. MARSH, N. MISLANG, M. SHATON, M. SICILIAN, AND C. WEBSTER(2021): “The U.S. Syndicated Loan Market: Matching Data,” Journal of Financial Research, 44(4), 695– 723. 26

DRUCKER, S., AND M. PURI (2005): “On the Benefits of Concurrent Lending and Underwriting,”JournalofFinance,60(6),2763–2799. FLECKENSTEIN, Q., M. GOPAL, G. GUTIE´RREZ, AND S. HILLENBRAND(2021): “Nonbank LendingandCreditCyclicality,”NYUSternWorkingPaper. GANDE, A., M.PURI, A.SAUNDERS, ANDI.WALTER(1997): “BankUnderwritingofDebt Securities: ModernEvidence,”ReviewofFinancialStudies,10(4),1175–1202. GORTON, G., AND A. METRICK (2013): “Securitization,” in Handbook of the Economics of Finance, ed. by G. Constantinides, M. Harris, and R. Stulz, vol. 2A of Handbooks in Economics,pp.1–70.Elsevier. GRYGLEWICZ, S., S. MAYER, AND E. MORELLEC (2022): “Screening and Monitoring of CorporateLoans,”SwissFinanceInstituteResearchPaperNo.21-82. GUSTAFSON, M. T., I. T. IVANOV, AND R. R. MEISENZAHL (2021): “Bank Monitoring: EvidencefromSyndicatedLoans,”JournalofFinancialEconomics,139(2),452–477. HOLMSTRO¨M, B., AND J. TIROLE (1997): “Financial Intermediation, Loanable Funds, and theRealSector,”QuarterlyJournalofEconomics,112(3),663–691. HU, Y., AND F. VARAS (2022): “Intermediary Financing without Commitment,” UNC WorkingPaper. IRANI, R. M., R. IYER, R. R. MEISENZAHL, AND J.-L. PEYDRO´ (2021): “The Rise of ShadowBanking: EvidencefromCapitalRegulation,”ReviewofFinancialStudies,34(5), 2181–2235. IRANI, R. M., AND R. R. MEISENZAHL (2017): “Loan Sales and Bank Liquidity Management: Evidence from a U.S. Credit Register,” Review of Financial Studies, 30(10), 3455– 3501. IVASHINA, V. (2009): “Asymmetric Information Effects on Loan Spreads,” Journal of FinancialEconomics,92(2),300–319. IVASHINA, V., AND Z. SUN (2011): “Institutional Demand Pressure and the Cost of CorporateLoans,”JournalofFinancialEconomics,99(3),500–522. JIANG, E. X. (2023): “Financing Competitors: Shadow Banks’ Funding and Mortgage MarketCompetition,”ReviewofFinancialStudies,forthcoming. LEE, S. J., D. LI, R. R. MEISENZAHL, AND M. J. SICILIAN (2019): “The U.S. Syndicated TermLoanMarket: WhoHoldsWhatandWhen?,”FEDSNotes. NADAULD, T. D., AND M. S. WEISBACH (2012): “Did Securitization Affect the Cost of CorporateDebt?,”JournalofFinancialEconomics,105(2),332–352. NEUHANN, D., AND F. SAIDI (2018): “DoUniversalBanksFinanceRiskierButMoreProductiveFirms?,”JournalofFinancialEconomics,128(1),66–85. 27

PURI, M. (1996): “Commercial Banks in Investment Banking: Conflict of Interest or CertificationRole?,”JournalofFinancialEconomics,40(3),373–401. SANTOS, J. A. C., AND P. SHAO (2018): “LoanOwnershipandLiquidityintheSecondary LoanMarket,”FederalReserveBankofNewYorkWorkingPaper. SUFI, A. (2007): “Information Asymmetry and Financing Arrangements: Evidence from SyndicatedLoans,”JournalofFinance,62(2),629–668. 28

Tables Table1: SummaryStatistics Packagelevel Mean Std. dev. 5th pctl 95th pctl N Totalleadshare(in%) 37.102 27.122 4.256 92.5 28,830 Averageleadshare(in%) 26.617 18.454 4.245 62.5 28,830 Numberleads 1.506 1.096 1 4 28,830 Arrangedbyuniversalbank 0.787 0.410 0 1 28,830 Loan-yearbylendercategorylevel Mean Std. dev. Min Max N Participantshare(in%) 21.017 22.874 0 66.665 417,096 Leadshare(in%) 13.812 21.367 0 60 417,096 Arrangedbyuniversalbank 0.702 0.457 0 1 417,096 Thetoppanelpresentssummarystatisticsatthepackagelevel; thevariablescorrespondtothose employedinTables2to6. Thebottompanelpresentssummarystatisticsattheloan-yearbylender categorylevel;thevariablescorrespondtothoseemployedinTables7and8. 29

Table2: EffectofUniversal-bankDeregulationonLeadArrangers—PackageLevel Totalleadshare Avg. leadshare Totalleadshare Avg. leadshare Arrangedbyuniversalbank -5.087*** -4.663*** -5.011*** -4.502*** ×After(1996) (1.376) (1.351) (1.317) (1.297) Arrangedbyuniversalbank 10.476*** 7.123*** 10.041*** 6.684*** (1.335) (1.429) (1.296) (1.371) Lead-arrangerFE Y Y Y Y Industry-yearFE N N Y Y N 28,830 28,830 28,830 28,830 AdjustedR2 0.48 0.20 0.49 0.22 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2002. Observationsareatthepackagelevel,correspondingtoloan l issuedatdate t. Thedependentvariable incolumns1and3isthetotalshare(in%)oftheloanretainedbyallleadarrangers,andisdefined between0and100. Thedependentvariableincolumns2and4istheaverageshare(in%)ofthe loanretained byall leadarrangers, andis definedbetween 0and100. Arrangedby universalbank l is an indicator variable for whether any one of the lead arrangers is a universal bank at the time of issuance. After(1996) is an indicator for whether the loan in question was issued after 1996. t Lead-arrangerfixedeffectsindicatetheinclusionofbankfixedeffectsforallleadarrangersatthe packagelevel. Industry-yearfixedeffectsarebasedontwo-digitNAICScodesofborrowerfirms. Robuststandarderrors(clusteredatthelevelofallleadarrangers)areinparentheses. 30

Table 3: Effect of Universal-bank Deregulation on Lead Arrangers: Credit Lines vs. TermLoans—PackageLevel Totalleadshare Avg. leadshare Totalleadshare Avg. leadshare Sample Creditlines Creditlines Termloans Termloans Arrangedbyuniversalbank -4.310*** -4.417*** -6.472*** -4.630*** ×After(1996) (1.253) (1.278) (1.919) (1.780) Arrangedbyuniversalbank 10.200*** 6.565*** 9.555*** 7.313*** (1.333) (1.417) (1.832) (1.728) Lead-arrangerFE Y Y Y Y Industry-yearFE Y Y Y Y N 18,233 18,233 10,597 10,597 AdjustedR2 0.52 0.21 0.48 0.26 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2002. Observationsareatthepackagelevel,correspondingtoloanl issuedatdatet. Inthefirsttwocolumns, thesampleislimitedtosyndicatedloansthatareclassifiedascreditlinesthroughouttheirentire run-time. In the last two columns, the sample is limited to syndicated loans that are classified as term loans throughout their entire run-time. The dependent variable in columns 1 and 3 is the totalshare(in%)oftheloanretainedbyallleadarrangers,andisdefinedbetween0and100. The dependentvariableincolumns2and4istheaverageshare(in%)oftheloanretainedbyalllead arrangers, and is defined between 0 and 100. Arranged by universal bank is an indicator variable l forwhetheranyoneoftheleadarrangersisauniversalbankatthetimeofissuance. After(1996) t isanindicatorforwhethertheloaninquestionwasissuedafter1996. Lead-arrangerfixedeffects indicate the inclusion of bank fixed effects for all lead arrangers at the package level. Industryyearfixedeffectsarebasedontwo-digitNAICScodesofborrowerfirms. Robuststandarderrors (clusteredatthelevelofallleadarrangers)areinparentheses. 31

Table 4: Effect of Universal-bank Deregulation on Lead Arrangers: Risky vs. Safe Borrowers—PackageLevel Totalleadshare Avg. leadshare Totalleadshare Avg. leadshare Sample Highvolatility Highvolatility Lowvolatility Lowvolatility Arrangedbyuniversalbank -9.815*** -9.119*** 1.068 1.418 ×After(1996) (3.165) (2.734) (2.574) (1.972) Arrangedbyuniversalbank 10.421*** 9.073*** 7.311* 5.018* (3.202) (2.712) (3.735) (2.857) Lead-arrangerFE Y Y Y Y Industry-yearFE Y Y Y Y N 1,833 1,833 1,540 1,540 AdjustedR2 0.61 0.28 0.60 0.20 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2002. Observationsareatthepackagelevel,correspondingtoloanl issuedatdatet. Inthefirsttwocolumns, the sample is limited to the top quarter of firms in terms of their six-year sales-growth volatility from t−6 to t−1. In the last two columns, the sample is limited to the bottom quarter of firms in terms of their six-year sales-growth volatility from t−6 to t−1. The dependent variable in columns1and3isthetotalshare(in%)oftheloanretainedbyallleadarrangers, andisdefined between0and100. Thedependentvariableincolumns2and4istheaverageshare(in%)ofthe loanretained byall leadarrangers, andis definedbetween 0and100. Arrangedby universalbank l is an indicator variable for whether any one of the lead arrangers is a universal bank at the time of issuance. After(1996) is an indicator for whether the loan in question was issued after 1996. t Lead-arrangerfixedeffectsindicatetheinclusionofbankfixedeffectsforallleadarrangersatthe packagelevel. Industry-yearfixedeffectsarebasedontwo-digitNAICScodesofborrowerfirms. Robuststandarderrors(clusteredatthelevelofallleadarrangers)areinparentheses. 32

Table 5: Effect of Universal-bank Deregulation on Total Lead Shares: Dynamic Effects—PackageLevel 1year 2years 3years 4years 5years Arrangedbyuniversalbank×After(1996) -4.996*** -4.945*** -4.892*** -4.867*** -4.863*** (1.298) (1.279) (1.279) (1.275) (1.277) Arrangedbyuniversalbank 9.758*** 9.563*** 9.441*** 9.421*** 9.412*** (1.309) (1.301) (1.298) (1.293) (1.291) Lead-arrangerFE Y Y Y Y Y Industry-yearFE Y Y Y Y Y N 28,830 28,830 28,830 28,830 28,830 AdjustedR2 0.49 0.48 0.48 0.48 0.48 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2002. Observationsareatthepackagelevel,correspondingtoloan l issuedatdate t. Thedependentvariable isthetotalshare(in%)oftheloanretainedbyallleadarrangers,definedbetween0and100,and is measured one to five years after loan issuance (across columns). Arranged by universal bank is l an indicator variable for whether any one of the lead arrangers is a universal bank at the time of issuance. After(1996) is an indicator for whether the loan in question was issued after 1996. t Lead-arrangerfixedeffectsindicatetheinclusionofbankfixedeffectsforallleadarrangersatthe packagelevel. Industry-yearfixedeffectsarebasedontwo-digitNAICScodesofborrowerfirms. Robuststandarderrors(clusteredatthelevelofallleadarrangers)areinparentheses. 33

Table 6: Effect of Universal-bank Deregulation on Average Lead Shares: Dynamic Effects—PackageLevel 1year 2years 3years 4years 5years Arrangedbyuniversalbank×After(1996) -4.520*** -4.484*** -4.481*** -4.490*** -4.500*** (1.284) (1.263) (1.267) (1.263) (1.265) Arrangedbyuniversalbank 6.656*** 6.661*** 6.641*** 6.652*** 6.653*** (1.377) (1.362) (1.358) (1.353) (1.352) Lead-arrangerFE Y Y Y Y Y Industry-yearFE Y Y Y Y Y N 28,830 28,830 28,830 28,830 28,830 AdjustedR2 0.23 0.23 0.23 0.23 0.23 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2002. Observationsareatthepackagelevel,correspondingtoloan l issuedatdate t. Thedependentvariable is the average share (in %) of the loan retained by all lead arrangers, defined between 0 and 100, andismeasuredonetofiveyearsafterloanissuance(acrosscolumns). Arrangedbyuniversalbank l is an indicator variable for whether any one of the lead arrangers is a universal bank at the time of issuance. After(1996) is an indicator for whether the loan in question was issued after 1996. t Lead-arrangerfixedeffectsindicatetheinclusionofbankfixedeffectsforallleadarrangersatthe packagelevel. Industry-yearfixedeffectsarebasedontwo-digitNAICScodesofborrowerfirms. Robuststandarderrors(clusteredatthelevelofallleadarrangers)areinparentheses. 34

Table7: UniversalBanksandInstitutional-investorLeadArrangers—Within-loanVariationoverTime Leadshare Sample All Creditlines Termloans Highvolatility Lowvolatility Institutionalinvestor×Arrangedbyuniversalbank×Yearssinceissue 0.923*** 1.133*** 0.606** 0.487* 0.416 (0.216) (0.266) (0.242) (0.267) (0.351) Institutionalinvestor×Arrangedbyuniversalbank -1.003 -1.017 1.516 -1.402 0.116 (2.077) (2.302) (1.563) (2.150) (2.895) Institutionalinvestor×Yearssinceissue -0.586*** -0.410* -0.596*** -0.237 0.084 (0.216) (0.244) (0.214) (0.285) (0.308) Loan-yearFE Y Y Y Y Y Lender-category-yearFE Y Y Y Y Y N 417,096 252,732 164,364 23,271 23,271 AdjustedR2 0.08 0.15 0.02 0.19 0.23 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2012. Observations are at the loan-year by lender category level lit, corresponding to the shares of loan l held by three types of lenders i, namely universal banks, other banks, and institutional investors (or non-banks), in year t. In columns 2 and 3, the sample is limited to loans classified as credit lines and term loans, respectively, in year t. In columns 4 and 5, the sample is limited to the top and bottom quarters, respectively, of firms in terms of their six-year sales-growth volatility from t−6 to t−1. The dependent variable is the total share (in %) of loan l retained by lead arrangers in lender category i in year t, and is defined between 0 and 100. Institutional investor is an indicator variable for loan shares i heldbyinstitutionalinvestors,asopposedtotheremainingtwocategoriesoflenders(universalandnon-universalbanks). Arrangedby universal bank is an indicatorvariable for whether anyone of the lead arrangersis a universal bankat the time of issuance. Years since l issue isthedifferenceinyearsbetweentandtheyearinwhichloanl isissued. Robuststandarderrors(clusteredatthelevelofalllead lt arrangers)areinparentheses. 35

Table8: UniversalBanksandInstitutional-investorParticipants—Within-loanVariationoverTime Participantshare Sample All Creditlines Termloans Highvolatility Lowvolatility Institutionalinvestor×Arrangedbyuniversalbank×Yearssinceissue 0.555* 0.368 1.538*** 1.255 -0.028 (0.333) (0.310) (0.563) (0.861) (0.509) Institutionalinvestor×Arrangedbyuniversalbank -3.760** -1.733 1.682 -2.571 -5.342 (1.859) (1.973) (3.166) (3.324) (4.035) Institutionalinvestor×Yearssinceissue -2.219*** -1.503*** -2.410*** -2.303*** -1.670*** (0.190) (0.156) (0.286) (0.350) (0.439) Loan-yearFE Y Y Y Y Y Lender-category-yearFE Y Y Y Y Y N 417,096 252,732 164,364 23,271 23,271 AdjustedR2 0.03 0.29 -0.12 0.04 0.15 The sample consists of syndicated loans granted to U.S. firms anytime from 1992 to 2012. Observations are at the loan-year by lender category level lit, corresponding to the shares of loan l held by three types of lenders i, namely universal banks, other banks, and institutional investors (or non-banks), in year t. In columns 2 and 3, the sample is limited to loans classified as credit lines and term loans,respectively,inyeart. Incolumns4and5,thesampleislimitedtothetopandbottomquarters,respectively,offirmsintermsof theirsix-yearsales-growthvolatilityfromt−6tot−1. Thedependentvariableisthetotalshare(in%)ofloanlretainedbyparticipants in lender category i in year t, and is defined between 0 and 100. Institutional investor is an indicator variable for loan shares held by i institutionalinvestors,asopposedtotheremainingtwocategoriesoflenders(universalandnon-universalbanks). Arrangedbyuniversal bank isanindicatorvariableforwhetheranyoneoftheleadarrangersisauniversalbankatthetimeofissuance. Yearssinceissue isthe l lt differenceinyearsbetweentandtheyearinwhichloanl isissued. Robuststandarderrors(clusteredatthelevelofallleadarrangers) areinparentheses. 36