Institutional investors and loan dynamics: Evidence from loan renegotiations

Journal of Corporate Finance 56 (2019) 482–505

Contents lists available at ScienceDirect

Journal of Corporate Finance journal homepage: www.elsevier.com/locate/jcorpfin

Institutional investors and loan dynamics: Evidence from loan renegotiations☆

T

Mehdi Beyhaghia, Ca Nguyenb, John K. Waldb,

⁎

a b

Federal Reserve Bank of Richmond, USA The University of Texas at San Antonio, USA

ARTICLE INFO

ABSTRACT

Keywords: Nonbank institutional investors Loan renegotiation Loan path Syndicated loans Funding risk

We examine the probability of exit for different types of investors in the syndicated loan market, as well as how the entry and exit of different types of investors is associated with changes in loan characteristics. Nonbanks, particularly CLOs, closed-end funds, and mutual funds, are more likely than bank lenders to exit the syndicate rather than to participate in the renegotiated loan. For mutual funds, greater net fund outflows imply a greater likelihood of exit, and this finding is consistent with nonbank lending creating greater systemic risk (Stein 2013). For most nonbanks, the likelihood of an exit increases if the financial condition of the borrower improves and the potential for higher spreads wanes. Controlling for borrower risk, the addition of most nonbank institutions, in contrast to banks, is accompanied by an increase in loan spreads, but no significant increase in the number or tightness of covenants.

JEL classifications: G21 G23 G32

1. Introduction An existing literature examines the relation between nonbank institutional lenders and loan characteristics at the time of loan initiation (Nandy and Shao, 2010; Ivashina and Sun, 2011; Lim et al., 2014). A separate literature considers how syndicated loans are renegotiated (Roberts and Sufi, 2009; Roberts, 2015; Nikolaev, 2018; Berlin et al., 2017). We extend these literatures by examining the decisions of different types of investors to exit the lending syndicate prior to renegotiation. In particular, we show that nonbanks, such as mutual funds, which are subject to funding shocks, are more likely to exit the lending syndicate prior to renegotiation. This finding is consistent with the theoretical literature about funding shocks faced by bank versus nonbank lenders, and this finding supports Stein's (2013) discussion of the potential for nonbank syndicated loan investments to create systemic macroeconomic risk. Moreover, while the existing literature has examined how nonbanks lend to different types of borrowers and participate in different loans than banks, we show how exit and entrance decisions by different types of nonbank lenders are related to changes in loan

☆ We thank the editor (Douglas Cummings) and two anonymous referees, as well as Aziz Alimov, Lamont Black, Bob DeYoung, David Dicks, Yunjeen Kim, Fred Malherbe, Greg Nini, Gordon Roberts, and participants at the Federal Reserve Board of Governors, the American Finance Association Meetings in Philadelphia, the World Bank Conference on Long-Term Lending: Determinants and Effects in Washington DC, the International Finance and Banking Society meeting at Oxford University, the Northern Finance Association Meetings in Halifax, the Midwest Finance Association Meetings in Chicago, the Lone Star Finance Conference in Waco, and the Financial Management Association Meetings in Boston for helpful comments. This project was funded in part by the University of Texas at San Antonio, Office of the Vice President for Research. The views expressed in this article are solely those of the authors. They do not necessarily reflect the views of the Federal Reserve Bank of Richmond or the Federal Reserve System. ⁎ Corresponding author. E-mail addresses: [email protected] (M. Beyhaghi), [email protected] (C. Nguyen), [email protected] (J.K. Wald).

https://doi.org/10.1016/j.jcorpfin.2019.03.005 Received 13 July 2018; Received in revised form 25 March 2019; Accepted 26 March 2019 Available online 27 March 2019 0929-1199/ © 2019 Elsevier B.V. All rights reserved.

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

characteristics. By focusing on changes in loans over time, this analysis further mitigates potential selection issues across borrowers and loans. At certain points during the life of a syndicated loan contract (the loan path), we can observe changes in the loan ownership structure. At these points, we see changes in the composition of the lenders, the risk of the borrower, as well as revisions in the terms of the loan contract. By tracking syndicated loan contracts over time, we study how nonbank institutions are different from their commercial bank counterparts in their decisions to exit a lending syndicate. That is, we consider whether different types of nonbank institutions exit lending syndicates at different frequencies and under different conditions. We examine whether these exit decisions are related to the lenders' characteristics, such as differences in the nonbanks' funding liquidity, and whether different types of lenders react differently to changes in the borrower's financial condition.1 We also investigate the relation between the participation of a particular type of lending institution and revisions in the amount, maturity, and spread of the loan, as well as in the tightness of covenants.2 In our analysis, we use detailed information that we manually cross-check based on a representative sample of over 4369 loans that go through 7408 rounds of loan renegotiations between 1987 and 2013. Our data indicates that close to one fifth of the syndicated loans in our sample have at least one type of nonbank institutional investor in their original lending syndicate.3 We categorize the various types of nonbank lenders as finance companies, investment banks, hedge funds/private equity funds, open-end mutual funds, closed-end funds, insurance companies, collateralized loan obligations (CLO), and other.4,5,6 We hypothesize that the probability of exit will vary with funding liquidity. Thus we expect that commercial banks, which use government guaranteed deposits for funding, are less likely to exit than mutual funds or CLOs. Table 1 summarizes the types of lenders and their sources of funding. Consistent with this hypothesis, we find that of the original lending syndicate members, nonbanks are significantly more likely to exit the syndicate than commercial banks. We also find that the lender identity matters in exit behavior. CLOs, closed-end funds, and open-end mutual funds have respectively 12.4%, 9.7%, and 8.4% higher chances of exiting a syndicate than commercial banks after controlling for other factors. The likelihood of exit is 8.1% greater for hedge funds/private equities and 2.2% greater for investment banks. The rate of exit for finance companies, insurance companies, and other lenders is not significantly different from that for commercial banks.7 We also investigate whether there is heterogeneity in how nonbanks react to a change in borrower risk. We use different measures of changes in borrower risk including revisions in Standard and Poor's credit ratings and whether there were any covenant violations (technical defaults) in the prior quarter. Interestingly, we find that nonbanks are generally more likely to exit when borrowers become less risky. This finding is significant for all types of nonbanks with the exceptions of investment banks and finance companies. For example, a one-step improvement in credit rating increases the likelihood of exit for a hedge fund/private equity by an average of 10.7 percentage points. We also find that the likelihood of exit for hedge funds, private equity firms, and CLOs decreases when the borrowing firm has violated a financial covenant in the prior quarter. These findings are consistent with the notion that nonbank institutions have a higher appetite for risk and generally seek riskier, and therefore higher-yielding, investments (Nandy and Shao, 2010; Lim et al., 2014; Taylor and Sansone, 2006). Our results also show that exiting nonbank institutions are most frequently replaced by commercial banks. Moreover, we show that for those nonbanks where changes in funding are directly observable, decreases in funding imply a significant increase in the probability of exit. Specifically, using quarterly data on fund flows from the CRSP Mutual Fund Database, we find that higher net fund outflows in the prior six or nine months imply significantly higher likelihoods of mutual fund lenders exiting the lending syndicate. This finding is consistent with the theory and with Stein (2013), who suggests that open-end investment vehicles such as mutual funds are subject to demandable equity, and that therefore they are more likely to exit their investments (loan ownerships) than banks.8 These exits are more likely to occur because investors in these vehicles can seek to withdraw their funds with very short notice. We further provide an analysis of how loan characteristics change around the addition and deletion of different types of nonbank

1 In the equity market where the role of institutional investors in corporate governance is more intensely discussed, several studies have discussed the effect of an institution's funding liquidity risk on an institution's decision to exit an investment. See for example, Back et al. (2015), and McCahery et al. (2016). 2 A lender exits a lending syndicate by assigning or transferring its share of the loan to another syndicate member, by selling its share to another loan owner on the secondary market, or by requesting debt repayment from the borrower. Our results hold whether or not we consider those loans that are available on the secondary market. 3 In the rest of the paper, we use the term “nonbank” as shorthand for “nonbank institutional investor.” 4 Bank in our study means commercial bank, an institution that is primarily financed by deposits and is FDIC insured. 5 Typical closed-end funds in our study are loan funds such as Van Kampen American Capital Prime Rate Income Trust, Prime Income Trust, and Morgan Stanley Dean Witter Prime Income Trust. 6 Examples of other investors include Answett Worldwide Aviation Services, the Bill and Melinda Gates Foundation, Nortel Networks Inc., Textron, and the Whitehall Corporation. 7 In a related finding, Irani and Meisenzahl (2017) consider which banks are more likely to exit a syndicate. They find that banks which rely on wholesale funding and short-term funding are more likely to exit. 8 Nonbank institutions in general face fewer regulations than banks. At the same time, they do not have access to the government's protective facilities such as the Federal Reserve System as the lender of last resort and to the FDIC as the insurer of liabilities. Banks are mainly financed by deposits whereas nonbank institutions are funded by a variety of non-deposit instruments ranging from redeemable shares and securities to insurance policies and limited partnerships.

483

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 1 Institutional investors as lenders: sources of funding and amount of regulation. Type of lender

Main source of funding

Relative funding liquidity risk

Amount of regulation

Commercial Banks (depository institutions) Investment Banks Finance Companies Insurance Companies Open-end mutual funds Closed-end funds Hedge funds/mutual funds

Deposits (FDIC insured)

Moderate

High

Securities Parent Company Policies/securities Redeemable shares Securities Capital provided by deep-pocket investors with long-term lockup periods in forms of limited partnership Securities

Moderate/High Low Low High Moderate/High Low

Moderate Low/Moderate High High Moderate/High Low

Moderate/High

Low

CLOs

lenders. We find that spreads increase more with the addition of a nonbank lender, and this relation holds regardless of whether we examine a change from an all-bank to a partly nonbank syndicate, a change in the number of nonbank syndicate members, or a change in the share of the syndicate provided by nonbanks. We also find that the addition of both bank and nonbank lenders is on average associated with an increase in the amount of credit available to the borrower and with an extension of loan maturity. Additionally, we find that new nonbanks (other than insurance companies) do not have a significant association with covenant tightness. The rest of this study is structured as follows: Section 2 provides a review of the theoretical literature and discusses our hypotheses. Section 3 briefly compares our findings with the existing empirical literature. Section 4 details the construction of our data set. Section 5 presents our empirical findings, and Section 6 concludes. 2. Theory and hypotheses Our study is partly motivated by Stein's (2013) observation that when long-term assets, such as corporate loans, are held by institutions with short-term demandable claims, then shocks can create fire sales or other systemic spillovers (see, for instance, Shleifer and Vishny, 1992).9 Additionally, as Brunnermeier and Pedersen (2008) show, funding liquidity is a driver of market liquidity and risk premiums, and thus we expect to see relations between lenders' funding liquidity and both the frequency of exits from loan syndicates and the magnitude of loan spreads. That is, nonbank syndicate lenders will be more likely to exit rather than renegotiate if they face liquidity shocks. Our analysis is also linked to the literature on bank renegotiation and the special role that bank financing plays. For instance, Hart and Moore (1988) and Aghion and Bolton (1992) argue that contracts are inherently incomplete, and thus renegotiations are inevitable. Chemmanur and Fulghieri (1994) show that banks can establish a reputation for renegotiating rather than liquidating risky debt, and this gives borrowers an incentive to use bank loans rather than bonds even if loan spreads are higher than public bond spreads. Another literature stresses the special nature of banks, either because banks provide lower cost monitoring on behalf of investors (Diamond, 1984), because banks can observe borrowers' information (Fama, 1985), or because banks can charge fees for additional services outside of the loan contract (Gorton and Kahn, 2000). Banks can also develop close relationships with borrowers over time, which reduces information asymmetry and facilitates monitoring and screening (Boot, 2000; Boot and Thakor, 2000). In terms of implications for our study, these theories suggest that nonbank syndicate lenders will have fewer incentives to develop a reputation for renegotiating rather than liquidating loans, and thus they will be more likely to exit the syndicate rather than renegotiating the loan. Thus, our first hypothesis, which follows from both the liquidity funding literature and the special role that banks play, is: Hypothesis I. Nonbank institutional syndicate lenders are more likely to exit the syndicate rather than to renegotiate the loan. Moreover, we expect nonbank lenders subject to greater liquidity shocks to exit more frequently. For mutual funds, we can also observe heterogeneous funding shocks, and thus we can test the relation between shocks and exit decisions directly. We therefore test the hypothesis: Hypothesis II. The probability of exit is higher for nonbank lenders subject to greater liquidity shocks. In particular, mutual fund syndicate lenders are more likely to exit the syndicate after a decrease in funding. In order to provide a more complete picture regarding how these exit decisions are related to firm characteristics, we examine 9 Stein (2013) writes, “…what we'd really like to know, for any given asset class…is this: What fraction of it is ultimately financed by short-term demandable claims held by investors who are likely to pull back quickly when things start to go bad? It is this short-term financing share that creates the potential for systemic spillovers in the form of deleveraging and marketwide fire sales of illiquid assets…. If relatively illiquid junk bonds or leveraged loans are held by open-end investment vehicles such as mutual funds or by exchange-traded funds (ETFs), and if investors in these vehicles seek to withdraw at the first sign of trouble, then this demandable equity will have the same fire-sale-generating properties as short-term debt.”

484

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

how changes in borrower risk and loan covenant violations affect the decisions of nonbanks to exit the syndicate. In addition to examining loan exit decisions, our study considers how changes in syndicate composition, and particularly the addition or deletion of nonbank syndicate members, are associated with changes in loan amount, maturity, spread, and covenant tightness. While Ivashina and Sun (2011) find that greater institutional funding in 2001 to 2007 led to lower rates on nonbank loans, Lim et al. (2014) find that loans with nonbank creditors pay a higher interest rate relative to loans with only bank creditors in the same package. Higher spreads for nonbank loans are consistent with nonbanks' larger appetite for risk and attraction to high-yield investments as documented by Taylor and Sansone (2006) and Lim et al. (2014). Higher spreads are also consistent with nonbanks entering the syndicate when the firm is liquidity constrained and bank funding is not available as described by Nandy and Shao (2010) and Lim et al. (2014). Alternatively, nonbanks may hold less diversified loan portfolios than banks, and therefore they may be pricing part of the idiosyncratic risk of the loan. Another explanation is that banks provide discounts to borrowers for the potential to earn other fees from ongoing relationships (Standard and Poor's, 2013). Further, the theoretical relation described by Brunnermeier and Pedersen (2008) between funding liquidity and risk premiums implies that spreads will be higher for loans where investors are more subject to funding liquidity. Thus, our third hypothesis states: Hypothesis III. The addition of nonbank lenders is associated with higher spreads, whereas the deletion of nonbank lenders is associated with lower spreads. Garleanu and Zwiebel (2009) show that an asymmetric information problem between borrowers and lenders can cause covenants to be written tightly, and renegotiations to be relatively frequent. Garleanu and Zwiebel also show that tighter covenants are written into the contract when there is greater asymmetric information, when information acquisition is more costly, and when renegotiation costs are greater. While most of these arguments focus on banks and rely on bank's expertise as information collectors and delegated monitors, as discussed before, Sufi (2007) and Drucker and Puri (2009) argue that nonbank institutions may not be as adept at collecting private information as banks. They also suggest that nonbank institutional investors are less engaged in corporate governance than commercial banks. Given this discussion, as nonbank lenders are expected to have higher information acquisition and renegotiation costs, and less interest in maintaining a long-term relationship with the borrower, we expect loan contracts with nonbanks to be written in ways which are more likely to avoid renegotiations. That is, nonbanks are less likely to write a contract with a shorter maturity or tighter covenants as these would require the nonbanks to participate in additional information acquisition and renegotiation, which we expect to be more expensive for nonbanks than banks. This leads to the following hypotheses about nonbanks with respect to loan maturity and covenants: Hypothesis IV. The addition of nonbank lenders is associated with longer maturity loans. Hypothesis V. The addition of nonbank lenders is associated with less tight covenants.

3. Relation to existing empirical research Prior research shows that loan ownership is a key factor affecting the cost of debt for a firm (Ivashina, 2009; Nandy and Shao, 2010; Ivashina and Sun, 2011; Nadauld and Weisbach, 2012; Lim et al., 2014), and that ownership also affects the cost of financial distress and bankruptcy outcomes (Ivashina et al., 2016). Lenders can also influence future capital expenditures, cash holdings, payout policy, and financing decisions through collateral requirements, performance pricing, and the use of financial covenants on the loans they grant.10 The implications for how nonbank loan ownership affects the cost of debt for borrowers have been mixed in the literature. For instance, Carey et al. (1998) find that finance companies make riskier loans than banks, while Nandy and Shao (2010) find that nonbanks in general are more likely to lend to riskier borrowers, and that nonbanks receive higher spreads even after correcting for the additional risk. Examining firms with both bank and nonbank creditors, Lim et al. (2014) find that loans with nonbank creditors pay a higher interest rate than loans with only bank creditors. In contrast, Ivashina and Sun (2011) find that higher institutional funding in 2001 to 2007, due to an increase in the supply of credit, caused interest rates on nonbank loans to be lower than similar loans funded by banks. Our results show that spreads increase more with the addition of a nonbank lender, and this relation holds regardless of whether we examine a change from an all-bank to a partly nonbank syndicate, a change in the number of nonbank syndicate members, or a change in the share of the syndicate provided by nonbanks. However, this result could be driven by either supply or demand factors. That is, nonbanks may demand a higher rate than commercial banks when they join a syndicate causing an increase in rates, or higher yields may cause more nonbanks to join the syndicate. Note that some of the explanations we proposed above allow for multiple interpretations of this relation. For example, the funding liquidity, lower diversification, and lack of other fees may be more consistent with nonbanks demanding higher yields, while the greater risk appetite of nonbanks is consistent with nonbanks being attracted to higher yields. Our results on the exit of mutual funds following a decrease in net fund inflows support the funding liquidity interpretation, while the results on the exit of other nonbank lenders in response to borrowers' improving conditions support the latter interpretation of a greater appetite for risk. Comparing syndicated loans with nonbank participants in their original syndicate against syndicated loans with only-bank 10 See studies by Shleifer and Vishny (1992), Chava and Roberts (2008), Drucker and Puri (2009), Nini et al. (2009), Bradley and Roberts (2015), and Roberts (2015).

485

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

participants in their original syndicate creates a potential selection problem. Different types of institutions may prefer loans from different borrowers with different loan characteristics, thus it is unclear whether there is any causal effect between lender type and contract characteristics. In order to avoid the selection issue with different borrowers, Lim et al. (2014) consider loan packages for one borrower that include a loan facility with at least one nonbank investor as well as a loan facility with only bank investors. Lim et al. are able to show that nonbank loans are priced with premiums relative to bank-only loans in the same loan package. Our approach differs from Lim et al. (2014) because not only do we control for the same borrower, but we also consider the same loan over time. That is, we analyze how the terms of a given loan change in response to changes in the lending syndicate composition. Hence, not only are we able to exclude the self-selection effects which occur from nonbanks lending more to certain types of borrowers, but we are also able to exclude the self-selection effects which occur because nonbanks prefer to participate in certain types of loans. Working with a loan path is advantageous because it allows us to isolate the relation between a change in loan ownership and how the loan contract is renegotiated. Our result of higher spreads for renegotiated loans that include nonbanks are consistent with the findings in Nandy and Shao (2010) and Lim et al. (2014), but not with the findings in Ivashina and Sun (2011), although our findings are in the context of same loan renegotiations rather than across-loan comparisons. Our results on nonbanks exiting loans more frequently than banks, and our analysis with exogenous shocks to mutual fund liquidity, have no close comparison in the existing literature. Our results can also be used to reject some of the alternative explanations on why nonbanks behave differently from banks in the loan market. Our covenant renegotiation results show that nonbanks are not more actively involved in loan management and monitoring than banks. We find that the number of covenants decreases when nonbanks continue to participate in the lending syndicate or when new nonbanks are introduced to an all-bank syndicate. This result supports the arguments by Sufi (2007) and Drucker and Puri (2009) that nonbank institutions may not be as adept at collecting private information as banks and suggests that nonbank institutional investors are less engaged in corporate governance than commercial banks. This finding is also consistent with the notion that renegotiations and monitoring are more costly for these institutions than for banks.11 Further, our analysis adds to the literature on loan renegotiations. Some of these studies on renegotiation focus on the changes in the borrowing firm's characteristics or market conditions as the trigger for loan renegotiations (Roberts and Sufi, 2009; Roberts, 2015). Other studies argue that because contracts are inherently incomplete (Hart and Moore, 1988; Aghion and Bolton, 1992) renegotiations are an important element of updating debt contracts, and that renegotiations are therefore inevitable regardless of the firm's and market's performance. These studies show that ex ante firm and contract characteristics are a good predictor of the frequency and intensity of future renegotiations (Garleanu and Zwiebel, 2009; Paligorova and Santos, 2016; Dou, 2016; Nikolaev, 2018). While these two groups of studies provide different views on what typically triggers renegotiations and to what extent renegotiations are related to the borrowing firm's characteristics, there is a consensus in the literature on two issues: (i) loan renegotiations happen frequently (Denis and Wang, 2014; Roberts, 2015) and (ii) an increase in the size of the lending syndicate and the presence of nonbank institutions in the original lending syndicate makes renegotiations more difficult or more costly (Bolton and Scharfstein, 1996; Nikolaev, 2018). The existence of nonbank lenders increases the coordination problem as different lenders have more varied objectives, and thus renegotiations are more difficult.12 Our study adds a dimension to this line of research by showing that a significant portion of the nonbank institutions in the original lending syndicate leave when a renegotiation occurs. Moreover, the decision to leave is related to both the performance of the borrower and the individual characteristics of the lenders. Thus the presence of some nonbanks in the original lending syndicate does not necessarily have a direct effect on the renegotiated terms, as the original lenders may choose to exit before renegotiations are complete. The degree to which nonbank lenders choose to exit a lending syndicate rather than engage in renegotiation has not, to our knowledge, been previously examined. 4. Data and sample selection Our analysis consists of two types of tests. In the first type, our objective is to investigate whether nonbank institutions (or different types of nonbank institutions) are more likely to exit a lending syndicate or to engage in loan renegotiations relative to commercial banks. In the second type of tests, we aim to discover how loan terms and covenants are modified when different types of nonbank institutions enter or exit the lending syndicate, or alter their share in the loan. The dependent variable in our first group of tests is a lender's likelihood of exiting the lending syndicate. In our second group of tests, the dependent variables are the changes in the loan contract features (amount, maturity, spread, and covenants) after the renegotiation.13 We decompose the syndicate lenders into nine categories: commercial banks and eight nonbank categories. We then analyze differences in each category's approach in choosing renegotiation over exit and the marginal effect of the entrance or exit of each 11 Consistent with higher costs of renegotiations for nonbank institutions, Berlin et al. (2017) find that the rise of nonbank institutional investors has made loan renegotiations more costly in general. Moreover, Saavedra (2018), and Berlin et al. discuss the recent development in loan contracting to make renegotiation easier. An example is the development of covenant-lite loan deals (2% of total loan issuance) and split control rights (about 0% in 2009 but rising sharply thereafter). 12 The effect of nonbank institutions is more extreme around a potential bankruptcy as shown by Gilson et al. (1990) and Demiroglu and James (2015). 13 We consider renegotiations outside financial distress or default.

486

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

nonbank category on loan terms. Further, we control for a host of variables including changes in borrowing firms' characteristics, market conditions, and initial loan contract terms in addition to industry, time, loan type, and loan purpose fixed effects. We also conduct several tests to ascertain the robustness of our results. We use data on net fund inflows for the mutual funds in our sample to examine how funding shocks affect mutual fund loan portfolios.14 4.1. Sample selection We start with the sample of all corporate loans in Loan Pricing Corporation's DealScan database that are initiated between 1984 and 2013 (276,862 loans). DealScan contains loan packages between a borrower and either a syndicate of lenders or a single lender. Loan packages are typically composed of several individual loan facilities that can differ based on type (term loan versus line of credit), size, security, maturity, spread, syndicate structure, and other loan characteristics. We focus on loans that belong to U.S. borrowers and are denominated in U.S. dollars (130,722 loans). We restrict the sample to loans belonging to non-financial, nonutility, public U.S. borrowers with available financial and market value data at the time of loan initiation and loan renegotiation. We also limit the sample to all borrowers with book assets greater than $10 million. We use the DealScan–Compustat link provided by Michael Roberts which extends through 2013 (Chava and Roberts, 2008) to acquire financial and stock price information from Standard & Poor's Compustat data set (35,240 loans after this step). Next, we require all loan facilities to have non-missing, nonnegative, non-zero loan amounts (principal), maturity, and interest spread (28,526 loans after this step). Lastly, we exclude loans less than $1 million and those missing lender information. This leaves us with a sample of 28,302 loans. Details on all variables used in this study are provided in Table 2. 4.2. Loan path construction As mentioned above, our paper differs from existing studies of institutional loan investors in that we consider the dynamic role of nonbank institutions for loans that are renegotiated. We obtain information on the terms of the renegotiation from one of two different methods. For our primary method, we search DealScan for any information that corresponds to renegotiated contracts. DealScan reports information on loan amendments in the separate Facility Amendment Table. In addition to reporting quantitatively the magnitude of a change with respect to the loan amount, maturity, and interest spread, the Facility Amendment Table also provides a description of all the other modifications based on the information that DealScan collects from, among other sources, the SEC filings, and voluntary disclosures by lenders and borrowers. We carefully read and use all the descriptions provided in the “comment” column for all the loans in our final sample to construct our data. To complement the first method, we also re-examine all the loans that are identified in DealScan as new loans to check whether they are in fact renegotiated versions of another loan in the data. Using a sample of 1000 loans within 1996–2005, Roberts and Sufi (2009) find that many of the loan renegotiations (47%) generate independent observations in DealScan.15 Therefore, our second method involves re-examining loans to identify observations that correspond to renegotiated contracts. We identify the loan path built through the first method as Amended loans, and the loan path identified using the second method as Refinanced loans. Loan path construction through each method is explained in detail in Internet Appendix I. Fig. 1 demonstrates an example of a loan with three renegotiation rounds. The final sample consists of 7408 renegotiation rounds constituting 4369 loan paths. The first method identifies 3745 amendments, while the second method captures 3663 refinanced loans that are classified as new loans by DealScan. In comparison, Michael Robert's hand-collected data has 501 loan paths and 1773 renegotiation rounds. 4.3. Lender identification There are a total number of 3046 unique lenders in our sample. We classify lenders initially into nine groups: (1) Commercial Bank; (2) Investment Bank; (3) Finance Company; (4) Insurance Company; (5) Open-end Mutual Fund; (6) Closed-end Fund; (7) Hedge Fund/Private Equity; (8) Collateralized Loan Obligations; and (9) Other, which includes lenders that do not belong to any of the above categories. Our identification strategy in general is similar to Lim et al. (2014) with the following three main exceptions. First, instead of putting all banks in one group, we distinguish between commercial banks and investment banks. The main reason is that in this paper we specifically care about the source of funding. Commercial banks rely on deposits as the primary source of financing, and they are eligible for FDIC insurance. Second, we distinguish between closed-end funds and open-end funds due to the higher vulnerability of open-end funds to financing shocks as described in Internet Appendix II. Third, we identify CLOs as another main group because of their increasing importance in the corporate loan market. To identify lender types, we first use the information provided by DealScan under “Institution Type.” Then we manually check 14 Among nonbank institutions, mutual funds are specifically regulated and are required to frequently report detailed information on funding flows and portfolio composition to the public. 15 Most of the loans whose lending syndicate structure changes during renegotiations are reported as new loans in DealScan. On the other hand, DealScan reports the change in syndicate structure for some loans in the “comment” column of the Facility Amendment table, but this data is not available for all loans. Hence, if the information regarding the change in lending syndicate of a loan is missing in the “comment” column, we assume the lending syndicate of the loan remains same. An analysis of only the subsample reported as new loans provides results consistent with the reported tables, but potentially introduces a selection bias.

487

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 2 Variable descriptions. Variables

Description

Main data source

Commercial Bank Nonbank Institution Investment Bank Finance Company Insurance Company Open-end Mutual Fund Closed-end Fund Hedge Fund/Private Equity CLO Other

An indicator variable that equals 1 if the lender is a commercial bank. An indicator variable that equals 1 if the lender is not a commercial bank. An indicator variable that equals 1 if the lender is an investment bank. An indicator variable that equals 1 if the lender is a finance company. An indicator variable that equals 1 if the lender is an insurance company. An indicator variable that equals 1 if the lender is an open-end mutual fund. An indicator variable that equals 1 if the lender is a closed-end fund. An indicator variable that equals 1 if the lender is a hedge fund or a private equity firm. An indicator variable that equals 1 if the lender is a CLO. An indicator variable that equals 1 if the lender is not categorized as a commercial bank, investment bank, finance company, insurance company, open-end mutual fund, closed-end fund, hedge fund, private equity firm, or CLO. Example: Bill and Melinda Gates Foundation. An indicator that equals 1 if the lender is one of the original loan arrangers. The number of lenders in the syndicate Net growth in fund assets beyond reinvested dividends as in Sirri and Tufano (1998):

DealScan, DealScan, DealScan, DealScan, DealScan, DealScan, DealScan, DealScan, DealScan, DealScan,

Loan Arranger Number of Lenders Net Fund Inflows

Net fund inflowi, t =

All-bank to nonbank Nonbank to all-bank Nonbank to nonbank Changes in No of commercial banks Changes in No of nonbank lenders Changes in No of investment banks Changes in No of finance companies Changes in No of insurance companies Changes in No of open-end mutual funds Changes in No of closed-end funds Changes in No of hedge funds/ private equities Changes in No of CLOs Changes in No of other No of Months Since Last Renegotiation Bank Loan Holdings Nonbank Loan Holdings Loan Sale Loan terms Amount/Assets Borrowing Base

mtnai, t

mtnai, t 1 (1 + mreti, t ) mtnai, t 1

Other Other Other Other Other Other Other Other Other Other

sources sources sources sources sources sources sources sources sources sources

DealScan, DealScan, CRSP

100

where mtnai,t is the monthly total net assets of fund i at time t and mreti,t is the total returns of fund i at the end of month t. An indicator variable that equals 1 if a lending syndicate with only commercial bank lenders transitions into a lending syndicate with at least one non-commercial bank lender. An indicator variable that equals 1 if a lending syndicate with at least one non-commercial bank lender transitions into a lending syndicate with only commercial bank lenders. An indicator variable that equals 1 if a lending syndicate with at least one non-commercial bank lender keeps its lender structure or transitions into a lending syndicate that still has a noncommercial bank lender. The difference between the number of commercial banks present in the lending syndicate when loans are renegotiated and the number of commercial banks in the lending syndicate before the renegotiation takes place. The difference between the number of non-commercial bank lenders present in the lending syndicate when loans are renegotiated and the number of non-commercial bank lenders in the lending syndicate before the renegotiation takes place. The difference between the number of investment banks present in the lending syndicate when loans are renegotiated and the number of investment banks in the lending syndicate before the renegotiation takes place. The difference between the number of finance companies present in the lending syndicate when loans are renegotiated and the number of finance companies in the lending syndicate before the renegotiation takes place. The difference between the number of insurance companies present in the lending syndicate when loans are renegotiated and the number of insurance companies in the lending syndicate before the renegotiation takes place. The difference between the number of open-end mutual funds present in the lending syndicate when loans are renegotiated and the number of open-end mutual funds in the lending syndicate before the renegotiation takes place. The difference between the number of closed-end funds present in the lending syndicate when loans are renegotiated and the number of closed-end funds in the lending syndicate before the renegotiation takes place. The difference between the number of hedge funds and private equity firms present in the lending syndicate when loans are renegotiated and the number of hedge funds and private equity firms in the lending syndicate before the renegotiation takes place. The difference between the number of CLOs present in the lending syndicate when loans are renegotiated and the number of CLOs in the lending syndicate before the renegotiation takes place. The difference between the number of “other” lenders present in the lending syndicate when loans are renegotiated and the number of “other” lenders in the lending syndicate before the renegotiation takes place. The definition of “other” is provided above. Number of months since the most recent renegotiation.

DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources DealScan, Other sources

The percentage of loan amount committed by bank lenders present in the lending syndicate. The percentage of loan amount committed by nonbank lenders present in the lending syndicate. An indicator variable that equals 1 if the facility is available for sale in the secondary loan market at some time.

DealScan DealScan Loan Syndication and Trading Association

Loan materiality computed as loan facility size divided by borrowing firm's assets. An indicator variable that equals 1 if the facility contains a borrowing base.

DealScan DealScan

(continued on next page)

488

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 2 (continued) Variables

Description

Main data source

Covenant

An indicator variable that equals 1 if the facility has at least one net worth or financial covenant. Most common covenants in the loan credit agreements include: Maximum Net Worth, Maximum Tangible Net Worth, Maximum Capital Expenditures, Minimum Interest Coverage Ratio, Minimum Fixed Charge Coverage Ratio, and Maximum Debt-to-EBITDA. Total number of covenants. See Covenants. Maturity of the facility in months. Number of lenders participating in the facility. An indicator variable that equals 1 if the facility contains a performance pricing grid. An indicator variable that equals 1 if loan type is 364-day Facility, Revolver/Line < 1 Yr., Revolver/Line≥ 1 Yr., Revolver/Term Loan, Bridge Loan, Guidance Line (Uncommitted), Limited Line, Multi-Option Facility or Standby Letter of Credit. An indicator variable that equals 1 if the facility is secured. Loan's all-in-drawn spread. An indicator variable that equals 1 if term loan type is Term Loan A. An indicator variable that equals 1 if term loan type is Term Loan, Term Loan B, Term Loan C, Term Loan D, or term loans with higher letter designation.

DealScan

An indicator variable that equals 1 if loan purpose is Corporate purposes, Capital expenditure, or Equip. Purchase. An indicator variable that equals 1 if loan purpose is Work. cap., CP backup, Debtor-in-poss., Dividend Recap, Recap., Exit financing. An indicator variable that equals 1 if loan purpose is Debt Repay. An indicator variable that equals 1 if loan purpose is Takeover, Acquis. Line, LBO or Merger.

DealScan

The standard deviation of changes in borrower's quarterly EBITDA/Assets over the past eight quarters. EBITDA to the total debt of the borrowing firm. Total debt to assets of the borrowing firm. Natural log of the borrowing firm's assets. Operating earnings to assets of the borrowing firm. S&P long term issuer credit rating (AAA, AA+, etc. and unrated). Market value (calculated by the sum of the book total debt and market value of equity) to assets of the borrowing firm. An indicator variable that equal 1 if borrower violates financial covenant in the quarter prior to renegotiation

Compustat

Moody's Yield on seasoned BAA-rated bonds minus yield on AAA-rated bonds.

Federal Reserve Bank of St. Louis FDIC US. Bureau of Economic Analysis CRSP

Number of covenants Maturity No of Lenders Performance Pricing Revolver Secured Spread Term Loan A Term Loan B Loan Purpose Corporate Working Capital Debt Repayment Takeover Borrowing firm characteristics Earnings Volatility EBITDA/Debt Leverage Log(Assets) Profitability Rating Tobin's Q Violation in Prior Quarter Market Conditions Aggregate Credit Spread Banking Sector Leverage GDP

Total liabilities to total book assets for commercial banks in the United States. GDP growth rate (2009 dollar).

Market Return Dependent Variables ΔAmount (%)

Quarterly return on the CRSP value weighted index.

ΔMaturity (%) ΔSpread (%) Change in number of covenants Covenant tightening index

Exit

DealScan DealScan DealScan DealScan DealScan DealScan DealScan DealScan DealScan

DealScan DealScan DealScan

Compustat Compustat Compustat Compustat Compustat Compustat Professor Sufi's website

The difference between loan amounts after and before renegotiation divided by loan amount before renegotiation. The difference between loan maturities after and before renegotiation divided by loan maturity before renegotiation. The difference between loan spreads after and before renegotiation divided by loan spread before renegotiation. The change in the number of covenants during a renegotiation round. The index receives a value of 1, 2 or 3 based on the overall change in loan covenants in a loan renegotiation from borrower perspective. If relatively more covenants become tighter, then the index is assigned a value of 3. If relatively more covenants are loosened, then the index receives a value of 1. If the number of loosened covenants equals the number of tightened covenants or the change is ambiguous, then the index receives a value of 2. A new covenant is considered a tightening and removal of a covenant is considered a loosening. An indicator that equals one if a lender that was a lending syndicate member at the time of loan origination has exited the syndicate before the first round of renegotiations.

lenders' primary four-digit Standard Industrial Classification (SIC) codes to reclassify the few institutions that are misclassified by DealScan. We further manually check all the lender names and look for keywords that indicate lender type (e.g., “CLO” in the lender name). Lastly, we use Capital IQ, Moody's Investors Service (for CLOs), Bloomberg, SEC filings, and other news content (using Google search) to manually recheck those nonbank institutions. Internet Appendix III provides more details on our lender identification method.

489

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Renegotiation Round 1

Round 2

Round 3

First renegotiation (time 1)

Second renegotiation (time2) Origination (time 0) Third renegotiation (time 3)

6/8/2007

9/12/2007

6/1/2008

10/30/2008

Fig. 1. A Demonstration of how Loan Paths Are Created. This figure demonstrates a 5-year $100 million revolving credit facility as it goes through three rounds of renegotiations. The loan facility was granted to NaviSite, a provider of hosting, application management and managed cloud services for enterprises on June 8, 2007.

5. Methods and empirical results 5.1. Institutional types and the likelihood of exit: descriptive statistics We perform our first group of tests, examining lenders' exit decisions, by focusing on the first renegotiation round. That is, we consider each lender's exit from the syndicate between when the loan is originated and when the result of the first loan renegotiation is reported. If a lender is in the lending syndicate at the time of loan origination and is not present in the lending syndicate at the time the renegotiation outcomes are reported, then we assume that the lender has exited the syndicate. Table 3 presents the summary statistics for all the variables used in our exit analyses. The unit of observation is loan-lender. Panel A shows that on average 24% of lenders exit prior to the first renegotiation. Panel B exhibits the institutional profile of lenders participating in a loan syndicate at origination. As expected, commercial banks account for the largest group of lenders in the US syndicated loan market (83%). The most common types of nonbank institutions are finance companies, investment banks, and CLOs which account for, respectively, six, five and three percent of all lenders. Insurance companies, open-end funds, closed-end funds, hedge fund and private equity funds appear less frequently, with each group constituting approximately 1% or less of all lenders in the sample. Table 3 also shows that approximately 12% of the lenders are a part of the loan arranging team. A median lending syndicate includes 14 members, and the mean number of lenders is 17. Panel B of Table 3 also reports the net fund inflow for the sample of mutual funds lenders in our sample (in percentages). Net fund inflows are calculated using quarterly information from the CRSP Mutual Fund Database (details provided in the next subsection). There are 149 unique open-end mutual funds in our sample and we manually match 88 of these funds to CRSP data. Panel B reports that the mean fund flows for mutual funds range from −1.31% to 0.78%. Panel B also shows that on average 24.5% of the loans have a record in the secondary loan data, i.e., at least one loan dealer posted a bid price or an ask price. Panel C describes the original terms of each loan. Syndicated loans have an average of half a billion dollars in principal, maturity is 54.3 months on average, and average spread is 170.02 bps over LIBOR. More than half of the loans are secured. Around 77% of loans have at least one covenant. Only 7% of loans report what collateral is used to back the loan, and 66% of loan contracts include performance pricing provisions that automatically adjust the interest rate during the life of the loan based on the performance and financial health of the borrowing firms. Panel D reports summary statistics on borrowing firm characteristics as reported in the quarter prior to loan origination. The book size of assets for an average firm in our sample is slightly more than six billion dollars. The average borrower's cash flow to debt, measured as the ratio of earnings before interest, tax, depreciation and amortization (EBITDA) to total debt, is slightly less than half. Total debt accounts for approximately 36% of firm assets, while average Tobin's Q is approximately 1.47. An average firm has a profitability ratio (net income to total assets) of 4% at the time of loan origination and an earnings volatility, calculated as the standard deviation of first differences in EBITDA/Assets, of 2%. Moreover, the average probability that a firm violates a financial covenant (technical default) prior to the first loan renegotiation is about 9.2%. Panel E shows the changes in borrowing firm characteristics between origination and the quarter before the first renegotiation round. In general, borrowing firms grow larger and employ more debt in their capital structure. However, other performance measures such as EBITDA/Debt, profitability, credit rating, and Tobin's Q decrease on average at the renegotiation time. The last panel (Panel F) demonstrates the changes in market conditions between loan origination and the first renegotiation. All macroeconomic factors, including GDP, stock market returns, banking sector leverage, and aggregate credit spread deteriorate slightly or remain unchanged on average during the sample period. Detailed descriptions of each variable are provided in Table 2. 5.2. Probability of exit: regression analysis We estimate the probability of exit as a function of the institutional type of a lender, controlling for the changes in the borrowing firm's financial health, the changes in macroeconomic environment, the initial loan features, and other controls. Model (1) 490

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 3 Summary statistics for probability of exit.

Panel A: Probability of Lender's Exit Exit Panel B: Lender Characteristics Commercial Bank Nonbank Institution Investment Bank Finance Company Insurance Company Open–end Mutual Fund Closed-end Fund Hedge Fund/Private Equity CLO Other Loan Arranger Number of Lenders Net Fund Inflows in the Prior 3 Months Net Fund Inflows in the Prior 6 Months Net Fund Inflows in the Prior 9 Months Net Fund Inflows in the Prior 12 Months Loan Sale Panel C: Prior Level of Loan Characteristics Amount (mil) Maturity (month) Spread (bps) Secured Borrowing Base Performance Pricing Covenant Panel D: Prior Level of Firm Characteristics Assets (bil) EBITDA/Debt Leverage Tobin's Q Profitability Earnings Volatility Violation in Prior Quarter Panel E: Change in Firm Characteristics Log(Assets) EBITDA/Debt Leverage Tobin's Q Profitability Earnings Volatility S&P Credit Rating Panel F: Change in Macroeconomic factors GDP Market Return Banking Sector Leverage Aggregate Credit Spread

N

Mean

SD

Min

Median

Max

38,537

0.243

0.429

38,537 38,537 38,537 38,537 38,537 38,537 38,537 38,537 38,537 38,537 38,537 38,537 215 210 210 204 36,584

0.831 0.169 0.054 0.056 0.006 0.012 0.006 0.006 0.025 0.003 0.115 16.767 0.304 −1.306 0.783 0.188 0.245

0.374 0.374 0.227 0.229 0.079 0.110 0.080 0.075 0.156 0.057 0.319 11.164 2.372 12.244 3.462 2.303 0.430

1 −4.355 −86.423 −5.403 −6.242

14 −0.002 −0.049 0.255 −0.003

45 10.931 16.659 26.388 8.243

38,537 38,537 38,537 38,537 38,537 38,537 38,537

566.442 54.3 170.019 0.551 0.072 0.663 0.765

926.789 22.092 114.221 0.497 0.259 0.473 0.424

1 9 15

300 60 150

24,000 156 700

38,537 38,537 38,537 38,537 38,537 38,537 22,770

6.045 0.483 0.358 1.469 0.036 0.017 0.092

26.005 2.966 0.225 0.942 0.024 0.023 0.290

0.012 −0.999 0.001 0.229 −0.103 0.001

1.712 0.102 0.329 1.189 0.035 0.01

781.818 38.893 1.376 6.193 0.133 0.242

38,537 38,537 38,537 38,537 38,537 38,537 38,537

0.202 −0.210 0.029 −0.155 −0.004 −0.001 −0.186

0.398 2.435 0.141 0.667 0.025 0.011 0.928

−0.898 −29.676 −0.432 −3.396 −0.154 −0.076 −5

0.1 −0.005 0.006 −0.042 −0.002 0 0

1.566 10.842 0.678 2.161 0.129 0.074 2

38,537 38,537 38,537 38,537

−0.001 −0.012 −0.002 0.103

0.008 0.12 0.005 0.463

−0.028 −0.322 −0.019 −2.02

−0.001 −0.006 −0.002 0.02

0.024 0.335 0.011 2.47

This table presents the summary statistics for the sample of 38,537 lenders before the first renegotiation is completed. The data is from 1987 to 2013. All continuous variables are winsorized at the 0.5% and 99.5% levels. Variable definitions are provided in Table 2.

demonstrates this specification.

Pr (Exit )L, B, C =

( + +

7

1i,

Lender Typei +

2 Loan

ArrangerL, B, C +

Market ConditionsB, C + Other Controls +

L, B , C )

3 Loan

CharacteristicsB, C +

4

Firm CharacteristicsB, C (1)

here, Φ denotes the cumulative normal distribution and subscript L represents the Lth lender in the syndicate. B represents the borrower id and C represents loan contract id among the borrower's portfolio of loans. i represents one of the nine possible lender types. The dependent variable is the probability that a lender withdraws from a loan syndicate. The variable of interest is the institutional type of the lender, with Commercial Banks, the reference group, omitted from the analysis. Loan Arranger L, B, C indicates whether lender L is among the loan arrangers for loan C. Loan CharacteristicsB,C is the vector of Loan C initial terms. ∆Firm CharacteristicsB,C and ∆Market ConditionsB,C are changes in firm characteristics and market conditions from loan C initiation to the time of the first loan renegotiation. As control variables, we consider loan characteristics before renegotiation, including the loan amount, maturity, spread, secured 491

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 4 Probability of exit. All loans (1) Nonbank Institution Investment Bank

0.161*** [0.041] (0.043)

Finance Company Insurance Company Open-end Mutual Fund Closed-end Fund Hedge Fund/Private Equity CLO Other Controls for Loan Arrangers Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Pseudo R2 Observations

Yes Yes Yes Yes Yes Yes Yes 0.180 38,537

Unquoted loans (2)

0.087** [0.022] (0.035) 0.069 [0.018] (0.044) 0.177 [0.045] (0.128) 0.327*** [0.084] (0.080) 0.378** [0.097] (0.158) 0.317** [0.081] (0.134) 0.486*** [0.124] (0.155) 0.105 [0.027] (0.187) Yes Yes Yes Yes Yes Yes Yes 0.181 38,537

(3) 0.205*** [0.057] (0.047)

Yes Yes Yes Yes Yes Yes Yes 0.172 27,605

(4)

0.100** [0.027] (0.041) 0.088* [0.024] (0.051) 0.449*** [0.123] (0.174) 0.440*** [0.121] (0.096) 0.692*** [0.190] (0.226) 0.469*** [0.129] (0.172) 0.877*** [0.241] (0.248) 0.322 [0.089] (0.253) Yes Yes Yes Yes Yes Yes Yes 0.175 27,605

This table presents estimated coefficients from a Probit regression where the dependent variable is whether the lender exits before the first renegotiation round is completed. Marginal effects are reported in brackets, and standard errors are reported in parentheses. The analysis is conducted at the loan-lender level. In Columns (1) and (2), we use all the loans from our sample. In Columns (3) and (4), we use the sub-sample of loans without a secondary loan market. Columns (1) and (3) use the general classification of nonbank lender and Columns (2) and (4) distinguish nonbank lenders into 8 different types. The base case is commercial bank. We also include controls for the loan arranger indicator, controls for loan characteristics before renegotiation, including the loan amount, maturity, spread, secured indicator, covenant indicator, borrowing base indicator, performance pricing indicator, and initial size of the syndicate, controls for the levels of firm characteristics including the log of assets, coverage ratio, profitability, leverage, and earnings volatility at loan origination, as well as the changes in these firm characteristics between origination and the first renegotiation, and controls for market factors including changes in gross domestic product growth, stock market return, aggregate credit spread, and aggregate banking sector leverage. All specifications include year fixed effects, industry fixed effects, and credit rating fixed effects. Standard errors are clustered by loan. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

indicator, covenant indicator, borrowing base indicator, performance pricing indicator, and initial size of the syndicate. We also control for the borrower's size (the log of assets), cash flow to debt, profitability, leverage, and earnings volatility at loan origination, as well as the changes in these firm characteristics between origination and the first renegotiation. Similar to firm characteristics, we include changes in market factors including gross domestic product growth, stock market returns, aggregate credit spread, and aggregate banking sector leverage, collectively represented by ∆Market ConditionsB,C in our set of control variables. Detailed descriptions of all variables are provided in Table 2. Table 4 presents the coefficients, marginal effects in brackets, and standard errors in parentheses, of variables based on the Probit estimation of Model (1). Specification (1) in Table 4 reports a general setting in which lenders are divided into either commercial banks or nonbanks. In Specification (2) we distinguish between each type of nonbank institution by using indicator variables for the specific lender types. Specifications (3) and (4) are similar to Specifications (1) and (2); however, instead of using all sample loans, we use a subset of loans that do not have a secondary loan market. The standard errors are clustered by borrowing firms. The analysis contains an extensive set of fixed effects including year, industry, and the credit rating of the borrowing firm. Consistent with our Hypothesis I, the results show that controlling for other factors, a nonbank institution is more likely to exit a 492

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

lending syndicate than a bank lender, and exiting nonbank institutions are most frequently replaced by commercial banks (see Table 9 below). The additional probability of exit for a nonbank in Specification (1) is 4.1 percentage point, and this coefficient is significant at the 1% level. In additional untabulated results, we find that being a loan arranger decreases the probability of exit by around 50%, consistent with the notion that because of reputation and contractual obligations, these lenders are more committed to their loan arrangements than an average lender. In Specification (2) of Table 4, we distinguish between each type of nonbank institution by using indicator variables. The results stay qualitatively similar when general identification of nonbank institutions is decomposed into specific types. Focusing on the marginal effects of the institutional types, we observe that all types of nonbank institutions are more likely to exit than commercial banks, although these results are not significant for finance companies, insurance companies, or other lenders. The results show that CLOs, closed-end funds, and open-end mutual funds have respectively 12.4, 9.7, and 8.4 percentage point higher chances of exiting a syndicate than commercial banks. The likelihood of exit is 8.1 percentage points greater for hedge funds/private equities and 2.2 percentage points greater for investment banks. While we do not observe the funding liquidity for these different types of nonbanks directly, we interpret these results as being consistent with our Hypothesis II. In Specifications (3) and (4) of Table 4, we restrict our analysis to the subsample of loans without a secondary loan market. From the DealScan data, we can see whether a lender has exited a syndicate before the loan renegotiation is complete, but we cannot know exactly at what point the lender has exited, or whether the lender has exited knowing that the loan would be renegotiated. The practitioner literature (Taylor and Sansone, 2006) sheds light on the possible timing of a given lender's exit. This literature indicates that typically a lender can exit a syndicate by assigning (or reassigning) all or part of its loan to current or new syndicate members during loan renegotiations. The other possibility is that the lender sells its loan on the secondary loan market. To control for this possibility, we obtain secondary loan market data from the Loan Syndication and Trading Association (LSTA) and the LPC mark-tomarket pricing service. The secondary loan data provides detailed dealer-based information on loans that are “available for sale.” We match our loans to secondary loan quotes using the procedure described in the loan sale literature (Drucker and Puri, 2009; Gande and Saunders, 2012; and Beyhaghi and Ehsani, 2017).16 In Specifications (3) and (4) we repeat the analysis in Specifications (1) and (2) after excluding the loans that matched with the LSTA data. This reduces our sample by roughly one quarter. We find that the results are qualitatively similar to the main tests in Table 4. Conditioning on not selling the loan on the secondary loan market before renegotiation, we show that nonbank institutions as a whole and individually, compared to commercial banks, are still more likely to choose exit over participating in renegotiation. In Table 5, we investigate whether there is heterogeneity in how nonbanks react to a change in borrower risk by including interaction variables between the lender type indicator and the changes in borrower credit rating before renegotiation (Specifications (1) and (2)). An improvement in borrower credit rating is measured by the number of levels S&P ratings improved; e.g. a change from BBB- to BBB+ is considered a + 2 change. Moreover, as an additional measure, we obtain quarterly data of financial covenant violations as in Nini et al. (2009).17,18 We also interact lender types with a dummy variable that equals 1 if a borrowing firm violates a financial covenant in the period before renegotiation (Specifications (3) and (4)). The results in Specifications (1) and (2) of Table 5 show that while the coefficients for lender types stay qualitatively the same as in our results in Table 4 (except for the insurance companies that are now significantly positive), the interaction coefficients show that nonbanks are generally more likely to exit when borrowers become less risky. This finding is significant for all types of nonbanks with the exceptions of investment banks and finance companies. For example, a one unit increase in credit rating increases the likelihood of exit for a hedge fund/private equity by an average of 10.7 percentage points. These results show that nonbanks prefer riskier, higher-yielding investments. In Specifications (3) and (4) of Table 5, we repeat Specifications (1) and (2) but this time we include interactions between actual covenant violations by the borrowing firm prior to a loan renegotiation and lender types. Consistent with our prior findings, the results suggest that, all else equal, nonbank institutions are more likely to exit than commercial banks. Interestingly, we find that the likelihood of exiting for hedge funds, private equity firms, and CLOs decreases when the borrowing firm has violated a financial covenant prior to renegotiation. This finding in consistent with the general finding in Columns (1) and (2) that nonbanks are less likely to exit when the borrowing firm's credit rating deteriorates. This finding is also consistent with the notion that hedge funds, private equity firms, and CLOs welcome risk and the opportunity to obtain higher yields on their investments (Taylor and Sansone, 2006; Lim et al., 2014). We next consider whether the probability of exit for open-end mutual funds varies with respect to net fund inflows. We focus on open-end mutual funds because they are the most regulated nonbank institutions, and because quarterly data for fund flows is available. There are 149 unique open-end mutual funds in our sample and we manually match 88 of these funds to CRSP data. The net fund inflow is computed as the net growth in fund assets beyond reinvested dividends as in Sirri and Tufano (1998):

Net fund inflowi, t =

mtnai, t

mtnai, t 1 (1 + mret i, t ) mtnai, t 1

100

16

If there is a record of a loan in the LSTA data, it indicates that the loan has been available for sale; however, it does not necessarily mean that the loan was traded. But if a loan does not have any record in the LSTA data, we believe that the loan has, in all likelihood, remained unsold up to the renegotiation. 17 Data are available on Professor Sufi's website. 18 In unreported tests, we use the ex-ante probability of covenant violation, as proposed by Demerjian and Owens (2016), as a measure of borrower risk. We find qualitatively similar results. The data on probability of covenant violation are available on Professor Demerjian's website. 493

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 5 Probability of Exit and Change in Borrower Risk. (1) Nonbank Institution Nonbank Institution × Improvement in Credit Rating Nonbank Institution × Covenant Violation

(2)

0.183*** [0.047] (0.045) 0.134*** [0.034] (0.012)

Investment Bank

Insurance Company Open-end Mutual Fund Closed-end Fund Hedge Fund/Private Equity CLO Other Investment Bank × Improvement in Credit Rating Finance Company × Improvement in Credit Rating Insurance Company × Improvement in Credit Rating Open-end Mutual Fund × Improvement in Credit Rating Closed-end Fund × Improvement in Credit Rating Hedge Fund/PE × Improvement in Credit Rating CLO × Improvement in Credit Rating Other × Improvement in Credit Rating Investment Bank × Covenant Violation Finance Company × Covenant Violation Insurance Company × Covenant Violation

(4)

0.122** [0.029] (0.057)

0.091** [0.023] (0.035) 0.071 [0.018] (0.046) 0.245* [0.062] (0.133) 0.380*** [0.097] (0.020) 0.423** [0.108] (0.167) 0.401*** [0.102] (0.139) 0.573*** [0.146] (0.162) 0.012 [0.003] (0.163) 0.022 [0.006] (0.038) 0.004 [0.001] (0.039) 0.172*** [0.044] (0.084) 0.193** [0.049] (0.077) 0.650* [0.165] (0.351) 0.421*** [0.107] (0.133) 0.894** [0.277] (0.386) 1.266** [0.322] (0.552)

Finance Company

(3)

−0.325 [−0.078] (0.052)

0.067 [0.016] (0.050) −0.034 [0.008] (0.054) 0.140 [0.033] (0.153) 0.229** [0.055] (0.098) 0.315* [0.076] (0.190) 0.404** [0.097] (0.178) 0.464*** [0.111] (0.173) 0.213 [0.051] (0.203)

−0.047 [0.011] (0.255) −0.125 [−0.030] (0.186) −0.357 [−0.086] (0.439)

(continued on next page)

494

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 5 (continued) (1)

(2)

(3)

Open-end Mutual Fund × Covenant Violation

−0.132 [−0.032] (0.307) −0.272 [−0.065] (0.401) −0.853** [−0.204] (0.381) −0.809** [−0.194] (0.405) −0.685 [−0.164] (0.472)

Closed-end Fund × Covenant Violation Hedge Fund/PE × Covenant Violation CLO × Covenant Violation Other × Covenant Violation Improvement in Credit Rating Covenant Violation Controls for Loan Arrangers Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Pseudo R2 Observations

−0.049 [−0.013] (0.032)

−0.051 [−0.013] (0.032)

Yes Yes Yes Yes Yes Yes Yes 0.181 38,537

Yes Yes Yes Yes Yes Yes Yes 0.185 38,537

(4)

−0.253 [−0.061] (0.128) Yes Yes Yes Yes Yes Yes Yes 0.176 22,770

−0.244 [−0.058] (0.128) Yes Yes Yes Yes Yes Yes Yes 0.178 22,770

This table presents estimated coefficients from a Probit regression where the dependent variable is whether the lender exits before the first renegotiation round is completed. Marginal effects are reported in brackets, and standard errors are reported in parentheses. The analysis is conducted at the loan-lender level. Columns (1) and (3) use the general classification of nonbank lender and Columns (2) and (4) distinguish nonbank lenders into 8 different types. Interactions of lender type with improvement in credit rating are used in (1) and (2). Interactions of lender type with a dummy variable that indicates whether the borrowing firm has violated a financial covenant prior to renegotiation are used in (3) and (4). The base case is commercial bank. All specifications include year fixed effects, industry fixed effects, and credit rating fixed effects, as well as the additional controls detailed in Table 4. Standard errors are clustered by loan. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

where mtnai, t is the monthly total net assets of fund i at time t, and mreti, t are the total returns of fund i at the end of month t. We add the net fund inflows in the 3 months, 6 months, 9 months, and 12 months prior to the renegotiation data as additional explanatory variables and the results are presented in Table 6. Consistent with our Hypothesis II, that the exit of mutual fund lenders is partly driven by fund flows, we find that greater net fund inflows in the prior 6 or 9 months imply significantly lower likelihoods of exiting the lending syndicate. A 1% increase in the net fund inflows in the prior 6 months implies a 0.71 percentage point decrease in the probability of exit (significant at the 1% level), while a 1% increase in the net fund inflow in the prior 9 months implies a 1.45 percentage point decrease in the probability of exit (significant at the 5% level). These findings are also related to the work of Manconi et al. (2012), who show that mutual funds with greater fund outflows were more likely to sell corporate bonds and retain illiquid securitized bonds, thus propagating the financial crisis. In untabulated analysis, we examine whether the specific time period affects nonbank institutions' behavior, especially since the financial crisis occurs during our sample period. We compare the probability of exit across nonbank institutions over three time periods: January 1987 to December 2000, January 2001 to July 2007, and August 2007 to December 2013. Ivashina and Sun (2011) document a significant increase in institutional fund inflow in the syndicated loan market during the period 2001 to the first half of 2007. Ivashina and Scharfstein (2010) mark August 2007 as the start of the crisis, and according to them loan originations were significantly reduced after this time. The results show that nonbank institutions in general are more likely to exit in the 2001–2007 and 2007–2013 periods than in the earlier period. Consistent with our other findings, we find that open-end mutual funds have significantly higher likelihoods of withdrawal than commercial banks in the period that starts with the financial crisis, while insurance companies, closed-end funds, and CLOs are more likely to exit during the 2001–2007 period. The results of our exit analyses generally support the hypothesis that nonbank institutions are more likely to exit the lending syndicate than banks. The exit behavior is different for different types of institutions. We find that each institution reacts differently to a change in borrower risk. We also find partial support for the notion that institutions with higher liquidity risk, such as those dependent on redeemable capital or security issuance, are more likely to exit. As hedge funds/private equities are also more likely to exit, this does not support the notion that the amount of regulation faced by an institution affects the decision to exit. For the full time 495

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 6 Probability of Exit of Open-end Mutual Funds. (1) Net Fund Inflows in the Prior 3 Months Net Fund Inflows in the Prior 6 Months

−0.0206 [−0.0035] (0.0526)

Net Fund Inflows in the Prior 9 Months

(2)

−0.0440*** [−0.0071] (0.0159)

Net Fund Inflows in the Prior 12 Months Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Pseudo R2 Observations

Yes Yes Yes Yes Yes Yes 0.539 215

Yes Yes Yes Yes Yes Yes 0.565 210

(3)

−0.0888** [−0.0145] (0.0427)

Yes Yes Yes Yes Yes Yes 0.556 210

(4)

0.0238 [0.00359] (0.0674) Yes Yes Yes Yes Yes Yes 0.597 204

This table presents estimated coefficients from a Probit regression where the dependent variable is whether an open-end mutual fund, which is a member of a lending syndicate, exits before the first round of loan renegotiation between the lending syndicate and the borrower is completed. Marginal effects are reported in brackets, and standard errors are reported in parentheses. The variables of interest are the mutual fund's net fund inflows from 3 months, 6 months, 9 months, and 12 months prior to the renegotiation date. The analysis is conducted at the loan-lender level. All specifications include year fixed effects, industry fixed effects, and credit rating fixed effects as well as the additional controls detailed in Table 4. Standard errors are clustered by loan. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

period, we are unable to reject the hypothesis that investment banks, finance companies, and insurance companies exit in a way that is similar to commercial banks. Lastly, we do not find strong evidence that all nonbank institutions change their exit strategy systematically after the crisis. 5.3. Institutional types and renegotiation results: descriptive statistics Our second set of analyses focuses on the relation between nonbank participation and loan terms. We conduct these analyses at the renegotiation round level. Table 7 reports the descriptive statistics of the variables used in the analyses. Panel A of Table 7 shows that a loan in our sample is on average renegotiated 2.53 times during its life.19 Because we restrict the sample to loans that are renegotiated, the minimum number of renegotiations is 1 while the maximum number of renegotiations for a loan in our sample is 13. Similar to Table 3, Table 7 also provides descriptive statistics on prior loan characteristics (Panel B), which provide the base from which we measure the change in the contract variables following a renegotiation. Panel C of Table 7 describes the distribution of renegotiation results. On average, following a renegotiation, the size of a loan increases by 16% (or about $56 million for the average loan of $352.81 million as reported in Panel B) and the maturity of the loan is extended by 25% (equivalent to 14 and half months for an average loan). Loan spread as a measure of cost of debt for the borrower and as a measure of required rate of return for the lender rises by an average of 18% (a 37 basis point increase for the average spread over LIBOR). We also consider various measures of covenant tightness.20 If a lender is actively engaged in monitoring the borrower, it is more likely that the lender will modify the covenants during renegotiations. We build two measures of covenant modifications. The first measure is the overall change in the number of covenants. The second measure is a covenant tightening index that considers both the overall change in the number of covenants as well as the tightening of each covenant. We infer the tightening of covenants by comparing new minimum or maximum allowable levels and by reading all the comments on the covenant section of the contracts in DealScan. This index is set to 1 if the covenants are looser relative to pre-renegotiation value, to 2 if they are the same, and to 3 if the covenants are tighter. On average the number of covenants increased by about 0.02 covenants (an average loan has 2.6 covenants 19 Loans in Roberts' (2015) sample have an average of 3.5 renegotiation rounds. The reason for the difference is that a large number of renegotiations in Roberts' (2015) sample are changes to the contracts' definitions section. On the other hand, we follow Roberts and Sufi (2009) and consider only renegotiations which also include changes in loan amount, maturity, spread, and covenants. One concern is that this selection may bias our sample toward larger or more significant changes in syndicates. To address this sample selection issue, in unreported tables, we repeat our analyses using Roberts' (2015) sample and find qualitatively similar results. 20 The two most popular covenants are restrictions based on debt to EBITDA ratios and the interest coverage ratio. While amount, maturity, and spread are determined at the facility level, covenants are typically determined at the package level. Hence, we use only the largest facility from each package to assess the change in covenant tightness. This reduces our sample from 7408 to 3991 observations.

496

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 7 Summary Statistics for Loan Renegotiation.

Panel A: Renegotiation Rounds Number of Renegotiation Rounds Panel B: Prior Level of Loan Characteristics Amount (mil) Maturity (month) Spread (bps) Secured Borrowing Base Performance Pricing Covenant Number of Covenants Panel C: Renegotiation Result ΔAmount (%) ΔMaturity (%) ΔSpread (%) Change in Number of Covenants Covenant Index Panel D: Lender Characteristics Nonbank to Nonbank All-bank to Nonbank Nonbank to All-bank Change in No of Commercial Banks Change in No of Nonbank Institutions Change in No of Investment Banks Change in No of Finance Companies Change in No of Insurance Companies Change in No of Open-end Mutual Funds Change in No of Closed-end Mutual Funds Change in No of Hedge Funds/Private Equities Change in No of CLO Change in No of Other Prior No of Lenders Change in No of Lenders Change in Bank Loan Holdings Change in Nonbank Loan Holdings Prior Bank Loan Holdings Prior Nonbank Loan Holdings Panel E: Prior Level of Firm Characteristics Assets (bil) EBITDA/Debt Leverage Tobin's Q Profitability Earnings Volatility Panel F: Change in Firm Characteristics Log(Assets) EBITDA/Debt Leverage Tobin's Q Profitability Earnings Volatility S&P Credit Rating Panel G: Change in Macroeconomic factors GDP Market Return Banking Sector Leverage Aggregate Credit Spread

N

Mean

SD

Min

Median

Max

7408

2.529

0.496

1.000

2.000

13.000

7408 7408 7408 7408 7408 7408 7408 3991

352.813 57.922 204.113 0.613 0.156 0.611 0.753 2.601

0.496 0.496 0.496 0.496 0.496 0.496 0.496 1.372

1.000 9.000 15.000

165.000 60.000 200.000

24,000 156.000 700.000

0.000

3.000

8.000

7408 7408 7408 3991 3991

0.157 0.252 0.175 0.020 1.990

0.589 0.410 0.686 1.262 0.573

−0.821 −0.567 −0.667 −6.000 1.000

0.000 0.009 0.000 0.000 2.000

4.000 2.167 5.000 7.000 3.000

7408 7408 7408 7408 7408 7408 7408 7408 7408 7408 7408 7408 7408 7408 7408 1486 1486 1486 1486

0.459 0.060 0.050 −0.023 0.010 0.011 −0.004 −0.000 0.001 −0.001 −0.001 0.001 0.006 9.082 −0.017 −0.319 0.319 90.058 9.942

0.498 0.238 0.218 4.014 0.956 0.473 0.510 0.108 0.204 0.094 0.122 0.111 0.077 8.305 4.523 4.502 4.502 20.611 20.611

−23.000 −5.000 −2.000 −3.000 −1.000 −1.000 −1.000 −1.000 −1.000 0.000 1.000 −25.000 −33.333 −23.407 0.000 0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 7.000 0.000 0.000 0.000 100.500 0.000

19.000 5.000 2.000 3.000 1.000 1.000 1.000 1.000 1.000 1.000 45.000 22.000 23.407 33.333 100.000 100.000

7408 7408 7408 7408 7408 7408

3.616 0.558 0.351 1.338 0.032 0.019

0.496 0.496 0.496 0.496 0.496 0.496

0.012 −0.999 0.001 0.229 −0.103 0.001

0.819 0.092 0.323 1.109 0.032 0.012

781.818 38.893 1.376 6.193 0.133 0.242

7408 7408 7408 7408 7408 7408 7408

0.139 −0.211 0.024 −0.089 −0.003 −0.000 −0.129

0.496 0.496 0.496 0.496 0.496 0.496 0.496

−0.898 −29.676 −0.432 −3.396 −0.154 −0.076 −5.000

0.061 −0.002 0.005 −0.016 −0.001 0.000 0.000

1.566 10.842 0.678 2.161 0.129 0.074 2.000

7408 7408 7408 7408

−0.001 −0.005 −0.002 0.084

0.496 0.496 0.496 0.496

−0.028 −0.322 −0.019 −2.020

−0.000 −0.003 −0.001 0.010

0.024 0.335 0.011 2.470

This table presents the summary statistics for the sample of 7408 renegotiation rounds between 1987 and 2013. All continuous variables are winsorized at the 0.5% and 99.5% levels. Variable definitions are provided in Table 2.

with the median loan having exactly three covenants). The results show that while the number of covenants slightly increased, the covenant tightening index has a mean slightly under 2, implying that on average covenants were relaxed slightly during renegotiations. This observation is consistent with Denis and Wang's (2014) finding that renegotiations are typically associated with a loosening of existing restrictions. Panel C of Table 7 reports the summary statistics for changing participation of bank and nonbank institutions. A nonbank syndicate structure is a syndicate with at least one nonbank institution, whereas an all-bank syndicate is a syndicate that only consists of 497

Syndicate Structure at t

498

Only Nonbanks

Only Commercial Banks and

Only Commercial Banks

Only nonbanks

0

88

More than one type

0

0

2

Hedge Funds/Private Equities

11

1

CLOs

28

Mutual Funds Closed-end Funds

Others

0

4

3

90

1

0

7

0

100

24

1,028

Finance Companies

213

Investment Banks

166

Investment Banks

Insurance Companies

3,188

Only Commercial Banks

Only Commercial Banks

Finance Companies 2

49

0

0

0

0

1

0

594

20

70

0

0

0

0

0

0

0

6

0

0

3

Insurance Companies Mutual Funds 0

3

0

0

0

0

60

0

2

13

30

0

1

0

0

0

7

0

0

0

0

4

Closed-end Funds

Only Banks and

Syndicate Structure at t−1

0

1

0

0

4

0

0

0

0

0

3

Hedge Funds/Private Equities

Table 8 Lending syndicate transition matrix.

CLOs 0

2

0

0

0

0

0

0

0

0

0

Other 0

1

9

0

0

1

0

0

0

1

1

More than one type 5

1,009

0

0

3

1

5

0

50

127

95

245

13

0

0

0

0

0

0

6

7

0

Only nonbanks

M. Beyhaghi, et al.

Journal of Corporate Finance 56 (2019) 482–505

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

commercial banks. Of the original lending syndicates with at least one nonbank, 46% retain their lending structure during renegotiations. Changing from an all-bank to nonbank syndicate (6% of renegotiations) or vice versa (5% of renegotiations) is considered an extreme case. Panel D of Table 7 reports details in the change in the number of each nonbank institution type during a renegotiation round. The net change in the number of commercial banks is −2.3%. The net change in the number of other types is < 1%. The average number of lenders in a syndicate before renegotiation is 9 and this decreases slightly in the renegotiation. Nonbank loan holdings are approximately 10% prior to renegotiation, and this increases by, on average, 3.19 percentage points in the renegotiation. Similar to Table 3, Table 7 also provides descriptive statistics on prior levels of firm characteristics (Panel E), changes in firm characteristics from the previous renegotiation to the current (Panel F), and changes in macroeconomic factors (Panel G). One key difference with Table 3 is that whereas Table 3 only covers the first renegotiation round for each loan, Table 7 covers all the renegotiation rounds. Moreover, Table 3 is at the loan-lender level while Table 7 is at the loan-renegotiation round level. Table 8 provides a schematic representation of lending syndicate transformation during each renegotiation round. Each column in the matrix of Table 8 represents possible lending syndicate structure at the beginning of the renegotiation round and each row represents the possible outcome at the end of the renegotiation round. The possible structures include only commercial banks, or some combination of banks with various types of nonbank institutions, or only nonbank institutions. For example, Table 8 shows that of the syndicates that were originally only commercial banks, 3188 remain only commercial banks, while 213 become commercial banks plus investment banks. This table shows that while the most common structure is commercial banks only, a nontrivial number of syndicates include nonbank institutions. Additionally, no change in the makeup of the syndicate structure is the most common outcome, and adding and removing other types of lenders occurs with sufficient frequency to provide an adequate sample for analysis. 5.4. Changes in the syndicate structure and subsequent changes in loan terms We examine the relation between changes in the lending syndicate structure and changes in loan features while controlling for prior loan terms, changes in firm characteristics, and market conditions. We take three separate approaches in measuring the change in the lending syndicate structure. These approaches are described in Models (2), (3), and (4) below. In Model (2) the change in a lending syndicate can be one of four general forms: a transition from a bank-only syndicate to a syndicate with at least one nonbank institution (which we call a nonbank syndicate), a transition from a nonbank to a bank-only syndicate, a transition from a bank-only syndicate to another bank-only syndicate, or a transition from a nonbank syndicate to another nonbank syndicate. We use all-bank to all-bank as the base case and use three dummies to measure the other transitions.

Renegotiation resultt , B, C, r =

11

+

(Nonbank to Nonbank )B, C , r + 2 Loan

CharacteristicsB, C, r +

+ Other Controls +

12

(All

bank to Nonbank )B, C, r +

Firm CharacteristicsB, C , r +

3

4

13

(Nonbank to all

bank )B, C , r

Market ConditionsB, C , r (2)

B, C, r

Subscript r represents renegotiation round, where round 1 starts from loan initiation and ends in the first renegotiation. t represents one of the loan terms including spread, amount, maturity, and covenant tightness. Renegotiation result t, B, C, r is the change in loan term t as a result of the renegotiation round r for borrower B's contract C. The coefficients of interest are β11, β12, and β13. They measure the marginal effect of each form of transition on the renegotiation outcome. Loan CharacteristicsB,C,r is the vector of Loan C initial terms at the beginning of round r. ∆Firm CharacteristicsB,C,r and ∆Market ConditionsB,C,r are changes in firm characteristics and market conditions from the beginning of round r until round r renegotiation is completed. To estimate Model (2), we employ an Ordinary Least Square regression where the dependent variables are the percentage changes in loan amount, maturity, spread, or change in the number of covenants. We also examine three measures of covenant tightness using an Ordered Probit regression (these ranks refer to whether a renegotiated covenant is looser (1), unchanged (2), or tighter (3)). In Model (3), we take a different approach in how we measure a change in a lending syndicate. For each of our nine types of lenders, we define a variable that shows the change in the number of that type of lender during a renegotiation. For example, if the number of commercial banks has gone up by two banks during a renegotiation, then the variable corresponding to commercial banks is assigned a value of 2. This model is as follow:

Renegotiation resultt , B, C, r =

1i

+

4

( Number of Lender Typei )B, C, r +

2 Loan

CharacteristicsB, C, r +

Market ConditionsB, C, r + Other Controls +

3

Firm CharacteristicsB, C , t (3)

L , B, C , r

where β 1i is the marginal effect of a one unit increase in the number of lender type i on the loan term t. Lastly, in Model (4) we take advantage of the information available in DealScan on the share of the loan held by each syndicate member. DealScan reports information on the share of each lender for only about 30% of all lenders. The loan observations that have complete lender share information for all lenders in their lending syndicate are about 20% of the total sample. To that end, in Model (4), we define a variable that shows the change in the total share of nonbank institutions during a renegotiation. For example, if the share of nonbank institutions in a loan changes from 20% to 35%, then this variable is assigned a value of 15. This model is as follow:

Renegotiation resultt , B, C, r =

1

+

( Share of Nonbank Institutions in the Loan)B, C , r + 3

Firm CharacteristicsB, C , t +

4

499

2 Loan

CharacteristicsB, C , r

Market ConditionsB, C, r + Other Controls +

L, B , C , r

(3)

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 9 Lending Structure Transition and Impact on Loan Terms.

Nonbank to Nonbank All-bank to Nonbank Nonbank to All-bank No of Months Since Last Controls for Prior Levels Controls for Prior Levels Controls for Prior Levels Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Observations

Renegotiation in Loan Characteristics and Changes in Borrower Characteristics and Changes in Macroeconomic Factors

(1)

(2)

(3)

ΔAmount (%)

ΔMaturity (%)

ΔSpread (%)

0.006 (0.016) 0.079* (0.041) 0.013 (0.031) 0.000 (0.001) Yes Yes Yes Yes Yes Yes Yes Yes 0.180 7408

−0.013 (0.010) 0.123*** (0.019) 0.066*** (0.019) 0.012*** (0.000) Yes Yes Yes Yes Yes Yes Yes Yes 0.466 7408

0.081*** (0.019) 0.066* (0.037) −0.002 (0.035) 0.007*** (0.001) Yes Yes Yes Yes Yes Yes Yes Yes 0.344 7408

This table presents the estimated coefficients from OLS regressions where the dependent variables are the percentage changes in amount, maturity, and spread, in Columns (1), (2), and (3), respectively. The variables of interest are indicators of transitions in lending structure: Nonbank to Nonbank, All-bank to Nonbank, and Nonbank to All-bank. The base case is All-bank to All-bank. The analysis is conducted at the loan renegotiation round level. We also include but do not report coefficients on loan characteristics before renegotiation, including the loan amount, maturity, spread, secured indicator, covenant indicator, borrowing base indicator, performance pricing indicator, and initial size of the syndicate, coefficients on the levels of firm characteristics including the log of assets, coverage ratio, profitability, leverage, and earnings volatility at loan origination, as well as the changes in these firm characteristics, and coefficients on market factors including changes in gross domestic product growth, stock market returns, aggregate credit spreads, and aggregate banking sector leverage. All specifications include year fixed effects, industry fixed effects, credit rating fixed effects, loan type fixed effects, and loan purpose fixed effects. Standard errors are calculated adjusting for clustering by borrowing firm and are reported in parentheses. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

where β 1 is the marginal effect of a 1% increase in the share of lender type i on the loan term t. In Tables 9 and 10, the change in the lending structure means a transition from an all-bank syndicate to an all-bank or a nonbank syndicate and vice versa. In Tables 11, 13, 14, and 15, the change in the lending structure is measured by the change in the number of each type of nonbank institution. In Table 12, the change in the lending structure is measured by the change in the total share of nonbank institutions in the loan. The base case in Table 9 is all-bank to all-bank, which refers to the case where the only-commercial bank syndicates remain with only-commercial banks after the renegotiation. Note that the number of commercial banks before and after the renegotiation might be different, but in this base case, there is no nonbank added to the syndicate. Consistent with our Hypotheses III and IV, the results in Table 9 show that when at least one nonbank institution joins the syndicate, the loan amount increases by 7.9% (equivalent to $28 Table 10 Lending structure transition and impact on loan covenants. Model

Nonbank to Nonbank All-bank to Nonbank Nonbank to All-bank Prior Number of Covenants Controls for Initial Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Pseudo R2 Observations

(1)

(2)

Change in number of covenants

Covenant tightening index

OLS

Ordered Probit

−0.074** (0.034) −0.185** (0.077) −0.036 (0.082) −0.246*** (0.021) Yes Yes Yes Yes Yes Yes Yes Yes 0.508

−0.003 (0.049) −0.121 (0.136) −0.124 (0.116)

3991

500

Yes Yes Yes Yes Yes Yes Yes Yes 0.263 3991

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 11 Changes in the Lending Syndicate and Impact on Loan Terms.

Change in No of Commercial Banks Change in No of Nonbank Institutions Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Observations

(1)

(2)

(3)

ΔAmount (%)

ΔMaturity (%)

ΔSpread (%)

0.024*** (0.003) 0.031*** (0.009) Yes Yes Yes Yes Yes Yes Yes Yes 0.179 7408

0.003*** (0.001) 0.011*** (0.004) Yes Yes Yes Yes Yes Yes Yes Yes 0.460 7408

−0.001 (0.002) 0.021*** 0.007) Yes Yes Yes Yes Yes Yes Yes Yes 0.343 7408

Table 12 Changes in Lender Share and Impact on Loan Terms.

Change in Nonbank Institutions' Loan Holdings Prior Level in Nonbank Institutions' Loan Holdings Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Observations

(1)

(2)

(3)

ΔAmount (%)

ΔMaturity (%)

ΔSpread (%)

−0.006 (0.005) 0.001 (0.001) Yes Yes Yes Yes Yes Yes Yes Yes 0.179 1486

0.003 (0.003) 0.000 (0.001) Yes Yes Yes Yes Yes Yes Yes Yes 0.460 1486

0.008** (0.003) 0.001* (0.001) Yes Yes Yes Yes Yes Yes Yes Yes 0.343 1486

This table presents estimated coefficients from OLS regressions where the dependent variables are the percentage changes in amount, maturity, and spread in Columns (1), (2), and (3), respectively. The variables of interest are the change in nonbank institution's share of the loan before and after renegotiation. Prior nonbank institutions' loan holdings are the level of loan allocation committed by nonbank institutions in a loan syndicate before the renegotiation takes place. The analysis is conducted at the renegotiation round level. All specifications include year fixed effects, industry fixed effects, credit rating fixed effects, loan type fixed effects, and loan purpose fixed effects as well as the additional controls detailed in Table 9. Standard errors are calculated adjusting for clustering by borrowing firm and are reported in parentheses. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

million for an average loan) and the loan maturity increases by 12.3% on average (equivalent to 7 months for an average loan).21 This transition comes at a higher cost for the borrower as the average loan spread increases by 6.6% (about 13 bps for an average loan). While it is not surprising that the amount or the maturity of a loan increases during renegotiations, focusing on the change in spread reveals that when the final syndicate structure includes at least one nonbank, the borrower is more likely to experience a higher loan spread. The unreported coefficients on our other firm and loan characteristics are consistent with expectations. Borrowers that experience asset growth, borrowers who become more profitable, and borrowers whose credit ratings are positively updated experience decreases in the cost of debt as well as increases in the loan amount and maturity. Borrowers that become more leveraged are more likely to experience reductions in the maturity of their loan and increases in their loan spreads. In Table 10, we examine the effect of syndicate structure transition on covenants. The coefficients of Nonbank to Nonbank and Allbank to Nonbank in Column (1) are −0.074 and − 0.185, respectively, and these coefficients are statistically significant at the 5% level. The results support our Hypothesis V that, relative to the case where there were only commercial banks in the syndicate before and after the renegotiation, the continuation of nonbank lenders or the addition of nonbank lenders to an all-bank lending syndicate

21 Less consistent with our hypothesis IV is the finding in Column (2) of Table 9 which shows that the coefficient on Nonbank to All-Bank is positive and significant, thus suggesting that syndicates which switch to All-Bank also lengthen their maturity. We test whether the coefficient on All-Bank to Nonbank is significantly larger than that on Nonbank to All-bank, and it is (p-value is 0.018). Thus both the transitions from an all-bank syndicate to a nonbank syndicate and vice versa are associated with increases in loan maturity. However, the magnitude of maturity extension is more pronounced for loans held by a nonbank syndicate than for those by an all-bank syndicate.

501

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 13 Changes in the Lending Syndicate and Impact on Loan Covenants. Model

Change in No of Commercial Banks Change in No of Nonbank Institutions Prior Number of Covenants Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Pseudo R2 Observations

(1)

(2)

Change in Number of Covenants

Covenant Tightening Index

OLS

Ordered Probit

0.026*** (0.006) 0.009 (0.020) −0.244*** (0.021) Yes Yes Yes Yes Yes Yes Yes Yes 0.507

0.030*** (0.009) 0.027 (0.027) Yes Yes Yes Yes Yes Yes Yes Yes 0.262 3991

3991

This table presents the estimated coefficients of the covenant tightness regressions. In Column (1), the dependent variable is the change in number of covenants and the estimates are obtained from an OLS regression. Columns (2) reports the results from an Ordered Probit model where the dependent variable is an index measuring whether the loan covenants loosen, stay the same, or become tighter. The variables of interest are the change in number of commercial banks and nonbank lenders. The analysis is conducted at the renegotiation round level. All specifications include year fixed effects, industry fixed effects, credit rating fixed effects, loan type fixed effects, and loan purpose fixed effects as well as the additional controls detailed in Table 9. Standard errors are calculated adjusting for clustering by borrowing firm and are reported in parentheses. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

is associated with fewer covenants. In Tables 11, 13, 14, and 15, instead of looking at extreme structural changes, we look at the change in the number of lenders from before the renegotiation starts to after the renegotiation is complete. In Table 11, we categorize lenders into two main groups of commercial banks and nonbank institutions and we estimate their marginal impact on the change in loan terms—amount, maturity, and spread— respectively in Specifications (1), (2), and (3). The results show that an increase in the number of both types of lenders Table 14 Loan Term Modification by Type of Lender.

Change in No of Commercial Banks Change in No of Investment Banks Change in No of Finance Companies Change in No of Insurance Companies Change in No of Open-end Mutual Funds Change in No of Closed-end Funds Change in No of Hedge Funds/Private Equities Change in No of CLO Change in No of Other Controls for Prior Levels in Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Observations

(1)

(2)

(3)

ΔAmount (%)

ΔMaturity (%)

ΔSpread (%)

0.024*** (0.003) 0.025 (0.018) 0.042** (0.017) −0.020 (0.130) 0.011 (0.042) 0.126 (0.101) 0.121 (0.075) −0.017 (0.100) −0.013 (0.135) Yes Yes Yes Yes Yes Yes Yes Yes 0.179 7408

0.003*** (0.001) 0.017** (0.007) 0.010 (0.008) −0.027 (0.043) 0.040* (0.021) 0.055 (0.045) −0.020 (0.033) −0.030 (0.038) 0.078 (0.058) Yes Yes Yes Yes Yes Yes Yes Yes 0.460 7408

−0.001 (0.002) 0.034** (0.014) 0.015 (0.015) 0.126** (0.052) 0.007 (0.046) −0.045 (0.048) −0.077 (0.071) 0.027 (0.060) 0.038 (0.083) Yes Yes Yes Yes Yes Yes Yes Yes 0.343 7408

This table presents estimated coefficients from OLS regressions where the dependent variables are the percentage changes in amount, maturity, and spread, corresponding to Columns (1), (2), and (3), respectively. The variables of interest are the change in number of commercial banks and different types of nonbank lenders. The analysis is conducted at the renegotiation round level. All specifications include year fixed effects, industry fixed effects, credit rating fixed effects, loan type fixed effects, and loan purpose fixed effects as well as the additional controls detailed in Table 9. Standard errors are calculated adjusting for clustering by borrowing firm and are reported in parentheses. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2. 502

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Table 15 Covenant Modification by Type of Lender. Model

Change in No of Commercial Banks Change in No of Investment Banks Change in No of Finance Companies Change in No of Insurance Companies Change in No of Open-end Mutual Funds Change in No of Closed-end Funds Change in No of Hedge Funds/Private Equities Change in No of CLO Change in No of Other Prior Number of Covenants Controls for Initial Loan Characteristics Controls for Prior Levels and Changes in Borrower Characteristics Controls for Prior Levels and Changes in Macroeconomic Factors Year FE Industry FE Credit Rating FE Loan Type FE Loan Purpose FE Adjusted R2 Pseudo R2 Observations

(1)

(2)

Change in Number of Covenants

Covenant Tightening Index

OLS

Ordered Probit

0.027*** (0.006) −0.061 (0.041) 0.057 (0.041) −0.055 (0.266) 0.002 (0.067) −0.086 (0.295) −0.197 (0.212) 0.542** (0.244) −0.484* (0.252) −0.242*** (0.021) Yes Yes Yes Yes Yes Yes Yes Yes 0.510

0.033*** (0.009) −0.044 (0.057) 0.044 (0.054) 0.572* (0.292) −0.121 (0.126) −0.213 (0.397) 0.245 (0.216) 0.256 (0.293) −0.788** (0.342)

3991

Yes Yes Yes Yes Yes Yes Yes Yes 0.265 3991

This table presents the estimated coefficients of the covenant tightness regressions. In Column (1), the dependent variable is the change in number of covenants and the estimates are obtained from an OLS regression. Columns (2) reports the results from an Ordered Probit model where the dependent variable is an index measuring whether the loan covenants loosen, stay the same, or become tighter. The variables of interest are the change in number of commercial banks and different types of nonbank lenders. The analysis is conducted at the renegotiation round level. All specifications include year fixed effects, industry fixed effects, credit rating fixed effects, loan type fixed effects, and loan purpose fixed effects as well as the additional controls detailed in Table 9. Standard errors are calculated adjusting for clustering by borrowing firm and are reported in parentheses. ***, **, and * correspond to statistical significance at 1%, 5%, and 10% level, respectively. Variable definitions are provided in Table 2.

is accompanied by an increase in loan amount and maturity. This finding is consistent with the notion that the addition of lenders directly affects the supply of credit to the borrower. The results also show that only the addition of nonbank institutions is accompanied by an increase in spreads. This finding is also consistent with Nandy and Shao (2010) and Lim et al. (2014) who show that nonbanks in general require a higher rate of return from their investments than banks. Alternatively, this finding could also reflect the lower diversification of most nonbank institutions relative to bank loan portfolios, and thus a required rate of return for nonbank investments. Yet another alternative is that the positive relation between nonbank participation and spreads reflects the higher required return for assets which have more variable funding liquidity as described by Brunnermeier and Pedersen (2008).22 In Table 12, we examine the robustness of the results in Table 11 by using the change in nonbank institutions' share in a loan as the explanatory variable of interest. The share data is only available for a small fraction of the loans in DealScan. The results in Table 12 confirm our previous findings for how spreads change. An increase in the share of nonbank institutions is significantly associated with an increase in loan spread. The estimated coefficients for the change in loan amount and the change in loan maturity are insignificant, which is not counter to our previous findings: a 1% increase in the share of nonbank institutions implies 1% decrease in the share of commercial banks. The overall effect of this change on the supply of credit is insignificant because, as shown earlier, the addition of both banks and nonbank institutions is associated with an increase in the supply of credit. In untabulated regressions, we examine whether increases in the number of bank or nonbank lenders are associated with different changes in spreads over different sub-periods. We find no evidence of a significant change on nonbank's impact on spreads as we move from one sub-period to another. This result differs from Ivashina and Sun (2011) who find that corporate credit expansion between 2001 and first half of 2007 by nonbanks led to lower spreads in corporate loans. Table 13 reports the results of regressing the changes in covenant variables on the changes in the number of lenders. An increase in the number of commercial banks is associated with an increase in the number of covenants, while an increase in nonbank lenders has no relation with covenants as shown in Column (1). Moreover, the results show that while an increase in the number of nonbanks has no significant effect on covenant tightness, an increase in the number of banks in the syndicate increases the tightness of the loan, 22 We also run tests where we include as additional explanatory variables, the interactions between the change in the number of each lender type and the change in borrower's credit rating. The results show that the interaction of change in the number of nonbanks and credit ratings is significantly negatively associated with changes in spreads. Thus nonbanks require a higher spread for a one unit decrease in borrower's credit rating than banks do. All else equal, one additional nonbank institution implies a 1.8% increase in loan spread on average (over 4 bps). Nonbanks require an additional 1.4% increase in loan spread than banks when borrowers' S&P credit rating is downgraded by one notch.

503

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

and this supports the notion that banks are more involved in covenant negotiations. In Table 14, we provide regressions which include separate indicators for different types of nonbank institutions on loan amount, loan maturity, and loan spread. Specifically, we consider the change in the number of commercial banks, investment banks, finance companies, insurance companies, open-end mutual funds, closed-end funds, hedge funds/private equities, CLOs, and others. Consistent with our prior findings, we find that adding commercial banks or finance companies is associated with an increase in both the loan amount. Further, supporting our prior findings, we find that the cost of debt is not significantly affected by a change in the number of commercial banks. The results are different for investment banks and insurance companies for which we find that a one unit increase is associated with a 3.4% and 12.6% increase in spread, respectively. Considering that an average loan spread over LIBOR is 204 bps for this sample, these changes imply an average 7 bps and 26 bps increases in spreads for investment banks and insurance companies, respectively. The marginal effects are insignificant across other types of lenders such as open-end mutual funds, closed-end funds, hedge funds/private equities, and CLOs.23 Table 15 presents the results of how covenants change with the number of different types of lenders. These results are consistent with the prior covenant analysis. We find that the addition of commercial banks is associated with both an increase in the number of covenants and an increase in covenant tightness (significant at the 1% level). We also find that an increase in the number of insurance companies is accompanied by tighter covenants (significant at the 10% level). While the results for banks are not surprising, we suspect that insurance companies require tighter covenants in renegotiations due to their higher lending standards when compared to other nonbank institutions. 6. Conclusion Nonbank institutions come from a variety of legal and regulatory backgrounds and they are exposed to different funding risks. Both nonbank institutions and commercial banks can be a part of one loan's lending syndicate. They can also join or exit a lending syndicate over the life of the loan and through loan renegotiations. Taking advantage of an extensive data set on loan renegotiations that includes the evolution of lending syndicates as well as detailed renegotiation outcomes, we find that nonbank institutions are more likely than commercial banks to exit the lending syndicate rather than to engage in loan renegotiations. In particular, institutional lenders with more funding liquidity risk, such as those that rely on redeemable capital or security issuance, are more likely to exit loan investments than engage in renegotiations. We directly show that mutual fund outflows imply an increase in the probability of exit by mutual fund syndicate members. These results corroborate the discussion in Stein (2013) that greater nonbank participation in the syndicated loan market may add to systematic risk because of the greater funding risk from nonbank participants. We also analyze how the presence and the change in the combination of nonbank institutional lenders in a syndicate are related to revisions in loan terms throughout the life of a loan. By examining the same loan over time, we reduce the selection bias from certain types of nonbank investors being more likely to participate in more risky loans. We find that the continuation or addition of nonbank investors, in general, is more likely to be associated with higher spreads on a particular loan. We also examine how the presence of nonbank investors is associated with differences in covenants. If a nonbank is added to or continues to be a part of the lending syndicate, the number of covenants is likely to decline or not tighten. However, if commercial banks are added, covenants are more likely to become tighter. These findings are consistent with the notion that commercial banks are in general more adept at information collection and monitoring in the loan market than nonbank institutions. Thus these findings are consistent with loan renegotiations being relatively less costly for commercial banks than nonbank lenders. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi.org/10.1016/j.jcorpfin.2019.03.005. References Aghion, P., Bolton, P., 1992. An incomplete contracts approach to financial contracting. Rev. Econ. Stud. 59 (3), 473–494. Back, K., Li, T., Ljungqvist, A., 2015. Liquidity and Governance. ECGI - Finance Working Paper No. 388. Berlin, M., Nini, G., Yu, E., 2017. Concentration of Control Rights in Leveraged Loan Syndicates. (Working paper). Beyhaghi, M., Ehsani, S., 2017. The cross-section of expected returns in the secondary corporate loan market. Rev. Asset Pricing Stud. 7 (2), 243–277. Bolton, P., Scharfstein, D.S., 1996. Optimal debt structure and the number of creditors. J. Polit. Econ. 104 (1), 1–25. Boot, A.W., 2000. Relationship banking: what do we know? J. Financ. Intermed. 9 (1), 7–25. Boot, A.W., Thakor, A.V., 2000. Can relationship banking survive competition? J. Financ. 55 (2), 679–713. Bradley, M., Roberts, M.R., 2015. The structure and pricing of corporate debt covenants. Q. J. Financ. 5 (2), 1–37. Brunnermeier, M.K., Pedersen, L.H., 2008. Market liquidity and funding liquidity. Rev. Financ. Stud. 22 (6), 2201–2238. Carey, M., Post, M., Sharpe, S.A., 1998. Does corporate lending by banks and finance companies differ? Evidence on specialization in private debt contracting. J. Financ. 53 (3), 845–878. Chava, S., Roberts, M.R., 2008. How does financing impact investment? The role of debt covenants. J. Financ. 63 (5), 2085–2121. Chemmanur, T.J., Fulghieri, P., 1994. Reputation, renegotiation, and the choice between bank loans and publicly traded debt. Rev. Financ. Stud. 7 (3), 475–506.

23 In additional unreported regressions, we consider an instrumental variable analysis for the effect of changes in mutual fund lenders only, where we use the net fund inflows as instruments. As with the OLS results reported in Table 15, the estimated coefficients from the second stage of the IV estimation show that the effect of a change in mutual fund lenders has no significant effect on debt cost, size, or maturity.

504

Journal of Corporate Finance 56 (2019) 482–505

M. Beyhaghi, et al.

Demerjian, P.R., Owens, E.L., 2016. Measuring the probability of financial covenant violation in private debt contracts. J. Account. Econ. 61 (2–3), 433–447. Demiroglu, C., James, C., 2015. Bank loans and troubled debt restructurings. J. Financ. Econ. 118 (1), 192–210. Denis, D.J., Wang, J., 2014. Debt covenant renegotiations and creditor control rights. J. Financ. Econ. 113 (3), 348–367. Diamond, D.W., 1984. Financial intermediation and delegated monitoring. Rev. Econ. Stud. 51 (3), 393–414. Dou, Y., 2016. The Debt-Contracting Value of Accounting Numbers and Financial Covenant Renegotiation. New York University working paper. Drucker, S., Puri, M., 2009. On loan sales, loan contracting, and lending relationships. Rev. Financ. Stud. 22 (7), 2835–2872. Fama, E.F., 1985. What's different about banks? J. Monet. Econ. 15 (1), 29–39. Gande, A., Saunders, A., 2012. Are banks still special when there is a secondary market for loans? J. Financ. 67 (5), 1649–1684. Garleanu, N., Zwiebel, J., 2009. Design and renegotiation of debt covenants. Rev. Financ. Stud. 22 (2), 749–781. Gilson, S.C., John, K., Lang, L.H., 1990. Troubled debt restructurings: an empirical study of private reorganization of firms in default. J. Financ. Econ. 27 (2), 315–353. Gorton, G., Kahn, J., 2000. The design of bank loan contracts. Rev. Financ. Stud. 13 (2), 331–364. Hart, O., Moore, J., 1988. Incomplete contracts and renegotiation. Econometrica 755–785. Irani, R., Mesenzahl, R., 2017. Loan sales and bank liquidity management: evidence from a U.S. credit register. Rev. Financ. Stud. 30 (10), 3455–3501. Ivashina, V., 2009. Asymmetric information effects on loan spreads. J. Financ. Econ. 92 (2), 300–319. Ivashina, V., Scharfstein, D., 2010. Bank lending during the financial crisis of 2008. J. Financ. Econ. 97 (3), 319–338. Ivashina, V., Sun, Z., 2011. Institutional demand pressure and the cost of corporate loans. J. Financ. Econ. 99 (3), 500–522. Ivashina, V., Iverson, B., Smith, D.C., 2016. The ownership and trading of debt claims in chapter 11 restructurings. J. Financ. Econ. 119 (2), 316–335. Lim, J., Minton, B. A., & Weisbach, M. S. (2014). Syndicated loan spreads and the composition of the syndicate. J. Financ. Econ., 111(1), 45–69. Manconi, A., Massa, M., Yasuda, A., 2012. The role of institutional investors in propagating the crisis of 2007-2008. J. Financ. Econ. 104, 491–518. McCahery, J.A., Sautner, Z., Starks, L.T., 2016. Behind the scenes: the corporate governance preferences of institutional investors. J. Financ. 71, 2905–2932. Nadauld, T.D., Weisbach, M.S., 2012. Did securitization affect the cost of corporate debt? J. Financ. Econ. 105 (2), 332–352. Nandy, D., Shao, P., 2010. Institutional Investment in Syndicated Loans. (Working Paper). Nikolaev, V.V., 2018. Scope of renegotiation in private debt contracts. J. Account. Econ. 65, 270–301. Nini, G., Smith, D.C., Sufi, A., 2009. Creditor control rights and firm investment policy. J. Financ. Econ. 92 (3), 400–420. Paligorova, T., Santos, J.A.C., 2016. Nonbank Investors and Loan Renegotiations. (Working paper). Roberts, M.R., 2015. The role of dynamic renegotiation and asymmetric information in financial contracting. J. Financ. Econ. 116 (1), 61–81. Roberts, M.R., Sufi, A., 2009. Renegotiation of financial contracts: evidence from private credit agreements. J. Financ. Econ. 93 (2), 159–184. Saavedra, D., 2018. Syndicate size and the choice of covenants in debt contracts. Account. Rev. 93 (6), 301–329. Shleifer, A., Vishny, R.W., 1992. Liquidation values and debt capacity: A market equilibrium approach. J. Financ. 47 (4), 1343–1366. Sirri, E., Tufano, P., 1998. Costly search and mutual fund flows. J. Financ. 53 (5), 1589–1622. Standard & Poor's, 2013. A Guide to the U.S. Loan Market. Standard & Poor's, New York. Stein, J.C., 2013. Overheating in Credit Markets: Origins, Measurement, and Policy Responses. (speech, February, 7). Sufi, A., 2007. Information asymmetry and financing arrangements: evidence from syndicated loans. J. Financ. 62 (2), 629–668. Taylor, A., Sansone, A., 2006. The Handbook of Loan Syndications and Trading. McGraw Hill Professional.

505