Within-Syndicate Conflicts and Financial Contracts: Evidence from Bank Loans

Transcription

1 Within-Syndicate Conflicts and Financial Contracts: Evidence from Bank Loans Nishant Dass, Vikram Nanda, Qinghai Wang First Draft: October 2010 This Draft: June 2012 Abstract Using a sample of bank loans, we study how financial contracts are different when there are conflicts of interest and information problems not only between the contracting parties but within one of the contracting parties. Such conflicts can arise in lending syndicates since lead arrangers tend to be better informed; they may also differ from other syndicate participants in their willingness to support a poorly performing borrower. We hypothesize that contractual devices and participant selection can reduce within-syndicate conflicts. Our empirical findings are strongly supportive: When the potential for conflicts is greater, contractual features such as accounting-based quantitative covenants and sweeps are more prevalent as they lower the risk of the loan as well as give relatively more decision rights to syndicate participants. Incentives of banks that (indirectly) hold equity of the borrower may be better aligned with those of the lead arranger and borrower. This reduces the need for contractual restrictions such as covenants and sweeps. Therefore, banks with an (indirect) equity stake in the borrower are more likely to be brought in as syndicate-participants. Keywords: Bank Loans, Conflicts of Interest, Covenants, Lending Syndicate, Monitoring JEL Codes: G20, G21, G32 We are grateful to Sudheer Chava, Robert Hansen, Robert Hauswald, Ron Masulis, and Peter Swan, as well as to seminar participants at Georgia Institute of Technology, Tulane University, and the University of New South Wales; and conference participants at the 2011 Australasian Finance and Banking Conference (Sydney) and 2012 European Winter Finance Summit (Davos) for their comments on the paper. We thank Michael Roberts for sharing his Dealscan-Compustat link with us. We are responsible for all remaining errors. The authors are at the College of Management, Georgia Institute of Technology, 800 West Peachtree Street NW, Atlanta, GA They can be reached by at: nishant.dass@mgt.gatech.edu, vikram.nanda@mgt.gatech.edu, and qinghai.wang@mgt.gatech.edu, respectively.

2 Within-Syndicate Conflicts and Financial Contracts: Evidence from Bank Loans Abstract Using a sample of bank loans, we study how financial contracts are different when there are conflicts of interest and information problems not only between the contracting parties but within one of the contracting parties. Such conflicts can arise in lending syndicates since lead arrangers tend to be better informed; they may also differ from other syndicate participants in their willingness to support a poorly performing borrower. We hypothesize that contractual devices and participant selection can reduce within-syndicate conflicts. Our empirical findings are strongly supportive: When the potential for conflicts is greater, contractual features such as accounting-based quantitative covenants and sweeps are more prevalent as they lower the risk of the loan as well as give relatively more decision rights to syndicate participants. Incentives of banks that (indirectly) hold equity of the borrower may be better aligned with those of the lead arranger and borrower. This reduces the need for contractual restrictions such as covenants and sweeps. Therefore, banks with an (indirect) equity stake in the borrower are more likely to be brought in as syndicate-participants. Keywords: Bank Loans, Conflicts of Interest, Covenants, Lending Syndicate, Monitoring JEL Codes: G20, G21, G32

3 1 Introduction Many financial services are delivered by financial intermediaries organized as syndicates. These intermediaries can, however, have divergent interests and information leading to potential conflicts of interest within the syndicate. How does the presence of these conflicts of interest affect the financial contracts written by syndicates? While there is ample research on the role of contracts in addressing the conflicts of interest between any two contracting parties (e.g., Harris and Raviv, 1992), in this paper we ask how the contract might be different in the presence of conflicts of interest within one of the contracting parties. We address this question in the context of bank loans by developing and testing hypotheses about the specific type of conflicts within a lending syndicate. Consistent with our predictions, we find that the potential for within-syndicate conflicts affects features of the loan contract (such as covenants and sweeps) as well as the choice of syndicate members. Information asymmetries and conflicts of interest arise naturally between a lead arranger and participant banks in a lending syndicate. Lead arrangers, given their role in initiating the loan, are usually regarded as being better informed about the borrower than participating lenders; they may also have incentives that diverge from those of other syndicate members. For instance, if the borrower is in financial distress, the lead arranger may prefer to renegotiate a loan either because the lead arranger expects to have an ongoing relationship with the borrower that would lead to future rents (Boot and Thakor, 2000), or because default by the borrower can have adverse effects on the lead arranger s reputation (Gopalan, Nanda, and Yerramilli, 2010). As a consequence, the lead arranger may favor the survival of the borrower even when it is sub-optimal from the participants perspective. 1 The situation is further aggravated when the lead arranger retains a relatively small fraction of the loan, while the participants are saddled with the rest. Therefore, 1 This is exemplified in the controversy that surrounded a syndicated loan made to Enron just before its bankruptcy filing. In this case, the participants accused the lead arranger (J.P. Morgan Chase & Co.) of deliberately concealing Enron s perilous financial condition and of using part of the loan proceeds to lower its own exposure to Enron. For details, see Enron Ties May Haunt J.P. Morgan Anew Finance Firm Could Face Action By Banks That Joined in Loan To Failed Houston Energy Trader in The Wall Street Journal, 21 February

4 the lead arranger plays a sort of dual role that of a debt-holder but also akin to an equity-holder, which distinguishes it from other syndicate participants who may be pure debt-holders. This results in a conflict similar to that between equity- and debt-holders (henceforth, equity-debt type conflict ), except that it is among a group of debt-holders (the lending syndicate). We first contend that loan contracts written between syndicates and borrowers can be structured in ways that reduce within-syndicate conflicts between the lead arranger (pseudo equity-holder) and other banks (pure debt-holders). 2 The resulting contract may contain features that assuage the concerns of participant-banks, and give them more control and/or cashflow rights with the aim of limiting lead arranger s (as well as borrower s) moral hazard. 3 Our second contention is that the formation of the syndicate itself can be affected by the desire to moderate within-syndicate conflicts that are expected to arise. So, for each loan, the lead arranger will seek out those banks that are likely to result in less severe equity-debt type conflicts within the syndicate (such as, banks that have an equity stake in the borrower). We thus hypothesize that syndicated loans are likely to have more restrictive contractual features. Moreover, participant lenders will seek greater restrictions on the loan when they are less informed than the lead arranger and there is a greater potential for within-syndicate conflicts. We examine two broad types of loan restrictions: accounting-based quantitative financial covenants and mandatory restrictions on the borrower s cashflows (which include cashflow sweeps for loanrepayment and restrictions on dividends). All loan contracts are expected to include some boilerplate covenants; these are typically qualitative or positive in nature. 4 However, our argument is about including additional accounting-based quantitative covenants and sweeps/restrictions that are specific to the loan (i.e., not boilerplate). A corollary of our hypothesis is that when interests of syndicate members are better aligned and equity-debt type conflicts are mitigated, then such 2 Although occasionally bank loans can have multiple lead arrangers, a single lead arranger is most typical, and as such, we refer to the lead-arranger in the singular throughout the paper. 3 In principle, the syndicate members could write separate contracts amongst themselves. However, these contracts would be prohibitively costly and difficult to enforce. Therefore, the syndicate converges on a single contract with which to mitigate the conflicts both between and within the contracting parties. 4 For example, these boilerplate covenants are intended to ensure that a company maintains its operations in a responsible manner, and obtains an unqualified audit report, etc. 2

5 covenants and sweeps/restrictions are less likely to be included in the loan. These loan restrictions are determined after negotiations between the lenders and the borrower, with considerable variation across loans. The syndication process functions as follows: While the preliminary term sheet of the loan includes many contractual features (such as pricing, structure, and collateral), the quantitative covenants and sweeps/restrictions are only finalized after the lead arranger has obtained feedback from potential participants. 5 This sequence is consistent with the view that these restrictions are not solely determined by the borrower s and lead-arranger s attributes. These state-contingent provisions give the participant banks (i.e., pure debt-holders) the ability to intervene and limit risk-shifting by the lead arranger (i.e., the pseudo equity-holder) as well as the borrower. For example, the violation of a covenant can allow the participant banks to call the loan even when the lead arranger might prefer to renegotiate instead. 6 Cashflow sweeps and dividend restrictions are also useful in that they safeguard the participants loan-recovery. We would like to emphasize that our conflicts-of-interest hypothesis does not preclude the conventional explanation such as borrower s quality or moral hazard for the contractual features of the loan; we denote the latter as borrower-quality hypothesis. 7 However, our hypothesis predicts that even after controlling for borrower risk and information attributes, the contractual features of the loan will be affected by the potential for conflicts within the syndicate. Although these contractual restrictions are beneficial for the lenders in controlling the borrower s moral hazard, they can impose costs on the lead arranger due to the possibility of intervention and/or dissipative renegotiation by syndicate members. Hence, our hypothesis also has the novel prediction that the characteristics of potential participants, such as their ownership in the borrower s equity, will influence their selection in the syndicate. The empirical findings are strongly supportive of our hypotheses. The data for our tests are drawn from a 2009 extract of the Thomson Reuters Dealscan database. We start with evidence 5 See, for example, Taylor and Sansone, 2006, and Standard & Poor s, Some syndicated loans require unanimity among the lenders for loan modification, giving considerable voice to each participating bank; for other loans it may only be necessary for banks with a majority of the loan to agree on a modification. 7 See, for instance, Greenbaum and Thakor (2007) and the literature cited therein. 3

6 on the greater prevalence of quantitative covenants in syndicated loans, when compared with nonsyndicated loans. Our estimates show that the odds of a covenant are 56% higher in syndicated loans. Similarly, the odds of a sweep/restriction in a syndicated loan are more than twice as high as those in non-syndicated loans. This holds even after controlling for borrower fixed-effects and comparing syndicated versus non-syndicated loans of the same borrower, thereby controlling for any time-invariant borrower characteristics that may govern the presence of these restrictive features. 8 Our results are also robust to controlling for lead-arranger fixed-effects quantitative-covenants are more likely to be included in a syndicated loan than non-syndicated loans issued by the same lead arranger. In fact, we show that these results hold even after controlling for the joint fixed-effects of the borrower and the lead arranger specifically, within the sample of repeated interactions between a pair of borrower and lead arranger, syndicated loans are twice as likely to have a covenant or a sweep. Furthermore, we show that covenants are more likely to be present when there is greater potential for within-syndicate conflicts, e.g., when the lead arranger has a past lending relationship with the borrower while none of the participants do. These findings are not easily explained by the borrower-quality hypothesis for the use of quantitative covenants. Next, we argue that potential conflicts of interest can be mitigated if the participating lenders have an equity stake in the borrower (through affiliated institutional investors or held in a fiduciary capacity). Specifically, greater participant equity holdings in the borrower can help align their incentives with those of the lead arranger and the borrower (Santos and Wilson, 2008; Jiang, Li, and Shao, 2010). As a result, participants would be more inclined to renegotiate the loan and avoid the adverse impact of a default on equity value. Consistent with this conjecture, we find that the odds of including a quantitative covenant (sweep) are 22% (39%) lower when the participant banks have a large equity stake in the borrower. Our results are robust to alternative definitions of the dependent variable. Given the special role of equity-holdings in this setting of equity-debt type 8 Bradley and Roberts (2004) have also suggested that syndicated loans are more likely to include covenants. However, their argument for covenants is based on the notion that a greater number of lenders in syndicated loans is reflective of the greater riskiness of syndicated loans. In our empirical analysis, we address this alternative hypothesis by controlling for borrower and lead arranger fixed effects as well as borrower and loan characteristics. 4

7 conflicts, we also analyze restrictions on dividend payments as a separate dependent variable. We find similar results dividend restrictions are more likely when conflicts are anticipated but much less likely when participants have an equity stake in the borrower. The larger equity holdings of the participants in the borrower as well as fewer restrictive covenants/sweeps in the loan may both be driven by a third factor such as the borrower s quality a better-quality borrower could attract equity investment and would also need fewer covenants. We conduct a series of robustness tests to rule out this possibility as best as the data allow. We further address endogeneity concerns using instrumental variable regressions and find that our results continue to hold. We also test for the effect of within-syndicate conflicts on covenant-tightness. Our results show that covenants are tighter in syndicated loans but less so when the participants have an equity stake in the borrower. From recent studies (such as Sufi, 2007), we know that a greater allocation of the loan to the lead arranger can also mitigate intra-syndicate conflicts of interest. We show that the presence of covenants and sweeps is a substitute for the lead arranger s loan allocation. As covenants/sweeps and loan allocation are endogenously determined, we instrument for the lead arranger s allocation. We also find that after controlling for the lead arranger s loan allocation, the participants equity stake continues to have a negative effect on the likelihood of quantitative covenants and sweeps/restrictions. The findings support our conflicts-of-interest hypothesis and are not, in general, predicted by the borrower-quality hypothesis for the use of such covenants/sweeps. While the presence of covenants/sweeps and the lead arranger s loan allocation can mitigate the conflicts of interest, they impose dissipative costs on the contracting parties. Enforcing accountingbased covenants may, for instance, be inefficient in comparison with renegotiation with the borrower. Also, a greater allocation may increase the credit risk on the lead arranger s balance sheet. Given that participants with a greater equity stake in the borrower have their incentives better aligned with those of the lead arranger, does the lead arranger seek out such participants when forming the syndicate? To answer this, we form a pool of potential participants by grouping all those lenders 5

8 that are participants in at least one loan deal arranged by that lead in the given year. We find that, given this pool of potential participant lenders, the lead arranger matches potential participants with borrowers by their equity stake (through affiliated institutional investors or held in a fiduciary capacity) in the borrower. Thus, the lead arranger seeks to minimize the inefficiencies in loan contracting by picking the participants that are most compatible for a given deal. Our paper makes several contributions to the finance literature. First, our paper shows that conflicts of interest within a syndicate can affect the nature of the contract with the borrower. This provides novel evidence of a financial contract between two parties being affected by conflicts of interest within one of the contracting parties. Hence, the paper contributes to the extensive literature analyzing the effects of conflicts of interest on the design of financial contracts. For instance, Townsend (1979), Grossman and Hart (1983), Hart and Moore (1988), and Aghion and Bolton (1992), all study the conflict of interest between the borrower and the investor, and its impact on the financial contract between the two. 9 Although we focus on bank loans, our arguments are applicable to other syndicates of financial intermediaries as well as to more general settings where conflicts arise within a contracting party. Second, our paper provides greater insight into the use of covenants and sweeps in bank loan contracts. The banking literature over the past few decades has built on the premise that the raison d être of banks is their expertise in monitoring (Diamond, 1984). Covenants have been seen as facilitating the lender s monitoring of the borrower (e.g., Smith and Warner, 1979; Berlin and Loeys, 1988; Gorton and Kahn, 2000). Loan covenants facilitate such monitoring and limit moral hazard by requiring the borrower to periodically provide the lender with accounting information that reflects its financial health. Covenants can strengthen the incentives of the financial intermediary to monitor and collect private information on the borrower (Rajan and Winton, 1995). Chava and Roberts (2008) show that the violation of covenants can adversely affect the borrower s investment. Overall, 9 The literature is too vast to describe here but the general theme has been covered in many excellent surveys of the literature as well as books over the years, such as Harris and Raviv (1992), Allen and Winton (1995), Hart (1995), Bolton and Dewatripont (2005), Tirole (2006), and Roberts and Sufi (2009). 6

9 the literature sees covenants as contractual features that protect the creditors from borrower s moral hazard (e.g., Dichev and Skinner, 2002; Greenbaum and Thakor, 2007; Nini, Smith, and Sufi, 2009; Chava, Kumar, and Warga, 2010). Although sweeps have not received as much attention in the literature, they are commonly used by banks and venture capitalists in financial contracts. Our evidence suggests that sweeps play a role similar to those of covenants in mitigating the conflicts of interest within the syndicate. We argue that both quantitative covenants and sweeps restrain moral hazard in more ways than has been recognized: by limiting the lead arranger s moral hazard along with the borrower s. Third, our paper contributes to the literature on the role and formation of syndicates. In general, the literature has argued that syndicates offer the benefit of risk-sharing (Wilson, 1968) or allow for the capture of quasi-rents when syndicate membership serves as a barrier to entry (e.g., Pichler and Wilhelm, 2001). More specifically, in the context of lending syndicates, Bolton and Scharfstein (1996) suggest that a syndicate that leads to inefficient renegotiation of the loan upon default may be useful ex-ante in deterring default, though costly if the default occurs. Syndicate size is ultimately a function of this trade-off as well as borrower characteristics. Lee and Mullineaux (2004) and Sufi (2007) examine the size and composition of the lending syndicate and show that syndicates are smaller when the loan is risky. They conclude that syndicates are formed so as to improve the lenders monitoring. We show that the lead arranger forms the syndicate such that it lowers the potential for within-syndicate conflicts of interest. Hence, our findings can shed light on the formation of different syndicates across loan deals arranged by the same lead bank. The rest of the paper is structured as follows. Section 2 discusses the hypotheses tested in this paper and Section 3 describes the data sample as well as the variables used to empirically test these hypotheses. Section 4 presents our main results on the use of covenants and sweeps in syndicated loans. Section 5 documents the role of lenders equity stake in aligning the interests of syndicate members. Section 6 compares the role of the lead arranger s loan-allocation in mitigating the within-syndicate conflicts and Section 7 addresses the effect of potential conflicts of interest on 7

10 the syndicate formed by the lead arranger. Concluding remarks are presented in Section 8. 2 Conflicts of Interest Within the Syndicate: Hypotheses and Empirical Predictions In this section, we discuss our hypotheses in more detail and develop the empirical predictions. As noted above, conflicts of interest and information concerns can arise within lending syndicates. The lead arranger originates the loan and is usually perceived as being better informed about the borrower than participating lenders, possibly due to a prior lending relationship with the current borrower. Participant banks may also have concerns about the incentives of the lead arranger. At the time of loan origination, the lead bank may exercise less than optimal due diligence in screening the borrower. If the borrower is subsequently in financial distress, the lead arranger may want to be more accommodating than other lenders in terms of renegotiating and modifying the loan. This may be either because of the possibility of future rents from an ongoing relationship with the borrower (Boot and Thakor, 2000) or because the borrower s default can have adverse effects on the lead arranger s reputation (Gopalan, Nanda, and Yerramilli, 2010). Therefore, the lead arranger may favor the survival of the borrower while the participant banks interest primarily lies in loan recovery. The lead arranger could also have more precise information about whether a loan modification or covenant enforcement is more desirable in a given situation. However, because of differing incentives, the participating banks may not necessarily be able to rely on the lead arranger to choose their preferred alternative. The literature argues that mechanisms such as reputation concerns and skin in the game of syndicate members can help counter these problems in loan syndicates (e.g., Sufi 2007, Ivashina 2009, and Gopalan et al. 2010). Our hypothesis is that, in addition to relying on the financial stakes and reputation concerns of members, syndicates will also utilize contractual devices and participant selection to reduce internal frictions. Specifically, we propose that accounting-based quantitative covenants and sweeps (mandatory repayment provisions), despite some inefficiencies, will emerge in syndicated loans in response to potential conflicts between lenders. These types of restrictions 8

11 tend to be loan-specific and are negotiated with the participant banks and the borrower. These covenants are useful in mitigating conflicts because they award control rights to participant banks in the event of a default. We confirm our intuition with the aid of a simple model, shown in the Appendix. The model, based on conflicts of interest and asymmetric information between syndicate members, shows that participant banks will demand covenants so as to limit lead arranger s (as well as borrower s) moral hazard. By extension, these arguments should also hold for the tightness of covenants, i.e., we expect covenants to be tighter when conflicts of interest are likely to be greater, and vice versa. Therefore, our first hypothesis is: Hypothesis 1 (H1): Syndicated loans are more likely to have quantitative covenants, cashflow sweeps, and other loan restrictions than non-syndicated loans. Covenants are expected to be tighter in syndicated loans. Among syndicated loans, these contractual restrictions will be more prevalent when there is a greater potential for within-syndicate conflicts. The differences in information and incentives create a wedge between the lead arranger and participant banks, which results in the former behaving like a pseudo equity-holder while the latter are pure debt-holders. Given that these within-syndicate conflicts can be characterized as being similar to those between equity- and debt-holders, we hypothesize that the conflicts can be mitigated if the participant banks have an equity stake in the borrower. Although banks in the US cannot hold a direct equity stake in firms, they can hold sway over a borrower s equity through the holdings of affiliated institutional investors or through shares held in a fiduciary capacity. By mitigating the within-syndicate conflicts, an equity stake of the participants in the borrower will reduce the need for quantitative covenants, cashflow sweeps, and other restrictions in the loan. 10 This leads to our second hypothesis: Hypothesis 2 (H2): Among syndicated loans, the likelihood of quantitative covenants, cashflow sweeps, and other loan restrictions is lower when participating lenders have a significant equity 10 Santos and Wilson (2008) provide evidence that a bank holding a firm s equity in a fiduciary capacity charges lower interest rates on loans to that firm. This is consistent with our claim that an equity stake of the participants in the borrower can make them more favorable toward the borrower. 9

12 stake in the borrower. Covenants are also expected to be less tight when participants have equity holdings in the borrower. Sufi (2007) argues that a higher loan allocation to the lead can also mitigate these intrasyndicate conflicts of interest. However, an increase in allocation may not be desirable for the lead bank because it requires both more capital and increases its exposure to borrower risk. A smaller lead allocation may imply more conflicts, and thus, would be associated with a greater use of quantitative covenants and sweeps/restrictions. Therefore, lead bank s loan allocation and the types of loan restrictions described above are expected to be substitutes in terms of their effect on mitigating within-syndicate conflicts. The model (in the Appendix) confirms this prediction and shows that participants are more likely to demand covenants when the lead arranger s loan allocation is smaller. Since, both the lead s allocation and the use of covenants are endogenous, our empirical tests will rely on instrumental-variable regressions. Our third hypothesis can be described as follows: Hypothesis 3 (H3): Among syndicated loans, a small allocation of the loan to the lead arranger will be associated with a greater use of quantitative covenants and cashflow sweeps. Hence, lead allocation and these contractual restrictions will be substitutes in terms of their effect on mitigating the conflicts of interest within the syndicate. Budget constraints and portfolio-diversification incentives imply that the lead arranger would prefer to keep its allocation of a loan smaller. However, as described above, this would result in more restrictive loan contracts that can lead to inefficiencies and costly renegotiation. Therefore, we argue that the lead arranger will try to minimize these frictions by seeking specific types of participants that will result in fewer within-syndicate conflicts. Specifically, our fourth hypothesis is that: Hypothesis 4 (H4): The lead arranger is more likely to form a syndicate with lenders that have an equity stake in the borrower as this will reduce equity-debt type conflicts. In our empirical tests, we seek to distinguish between our hypothesis and existing quality-based 10

13 explanations for the use of financial covenants. As per the quality-based hypothesis, covenants are primarily used to counter concerns about the borrower s moral hazard. In our tests we aim to show that covenant use is explained not only by borrower-quality issues but also by the presence of potential conflicts within the syndicate, as predicted by our conflict hypothesis. 3 Data and Description of Variables 3.1 Data We draw our data from four main sources Thomson Reuters Dealscan and 13F Institutional Holdings, S&P s Compustat (Fundamentals Quarterly), and CRSP s Monthly Stock databases and merge them to construct our final sample. We start by collecting information on bank loans from a December 2009 extract of Thomson Reuters Dealscan database. We only include completed loan deals that are syndicated in the U.S. and are denominated in US$. We exclude loans made to financial firms (that have SIC codes ). We further screen the data to ensure that we have the necessary accounting information for the borrower. This is facilitated by the Dealscan- Compustat link used in Chava and Roberts (2008), an updated version of which was provided to us by Michael Roberts. We use the loan package as the unit of analysis since covenants and sweeps, which are the focus of this paper, are determined at the level of a package and not facility. For information such as the loan spread that is unique at the facility level but may vary across facilities within the package, we only retain the facility with the largest loan amount in the package. To obtain equity holdings of lenders in the borrowing firms, we begin by searching the 13F institutional holdings database for investment affiliates of the lending banks. Using an algorithmic match that is further improved by manually searching for lender names in the institutional holdings database, we are able to find a matching manager number in 13F for about four hundred lenders. Based on the holdings of the lender-affiliated investor thus identified and the borrower s shares outstanding in CRSP, we calculate the percentage equity ownership of the lead arranger as well as the participant banks in the borrowing firm. The sample thus constructed with information on 11

14 loans, borrower characteristics, and lenders equity stake in the borrower forms the basis of our analysis. Our final sample period starts in 1994 due to limited and unreliable data on covenants prior to that year. We next provide a brief discussion of the variables used in our analysis. 3.2 Main Dependent Variables The main dependent variables, Covenant Dummy and Sweep Dummy, are indicator variables for the presence of an accounting-based quantitative covenant or a cashflow-sweep/restriction, respectively, in the loan contract. 11 Panels A and B of Table 1 list the seventeen quantitative covenants and eight cashflow-sweeps/restrictions, respectively, that are found across all the loans in Dealscan. 12 If any one of the covenants (sweeps/restrictions) listed in that table is found in a given loan, then Covenant Dummy (Sweep Dummy) equals one for that package; it is zero otherwise. Two alternative dependent variables Number of Covenants and Number of Sweeps are employed as well. The former counts the number of quantitative covenants that are used in a loan package, and ranges between zero (for no quantitative covenants) and eight (the maximum in the sample). The latter is defined similarly, and ranges between zero (for no cashflow sweeps or other restrictions) and six (its maximum in the sample). We also employ a dependent variable based on classifying covenants into four types, as indicated in Table 1, Panel A. The dependent variable Types of Covenants counts the number of different types of covenants present in a given loan, and ranges between zero (for no covenants) and four (the maximum number of covenant types in a given loan). Lastly, we define a dummy variable Dividend Restriction, which equals one when this restriction (listed in Table 1, Panel B) is present in a loan; it is zero otherwise. This restriction is of special interest because lenders that hold equity may be less likely to impose it on the borrower. In addition to these loan characteristics, we also analyze the tightness of covenants, the allocation of the loan to the lead arranger, and the choice of specific syndicate members by the lead arranger. The corresponding dependent variables for these tests are described later. 11 Recall that this Covenant Dummy does not account for the positive or qualitative covenants, which are part of the boilerplate language of all loan contracts. 12 The characteristics of our sample shown in Panel A are similar to those of the covenantsample used in Chava and Roberts (2008). 12

15 3.3 Main Independent Variables We start our analysis with a comparison of syndicated loans with non-syndicated loans because, by definition, a non-syndicated loan does not have within-syndicate conflicts. The independent variable of interest in this case is Syndicate, which equals one for a syndicated loan, and zero otherwise. It is plausible that the potential for conflicts and the difficulty in their resolution both increase with the size of the syndicate. Hence, we employ Number of Lenders and Large Syndicate to reflect the potential for conflicts. The former counts the number of lenders (and equals one if the loan is not syndicated) while the latter is a dummy variable that is defined only for syndicated loans and equals one if the number of lenders is greater than the sample mean (which is six). Equity ownership in the borrower can align interests of the lenders. We measure the Lead s Equity Holdings as the equity stake held by the lead arranger in the borrower, calculated as a percentage of the borrower s total shares outstanding. For syndicated loans, Participants Equity Holdings are measured similarly and aggregated across all the participants in the syndicate. Both the lead s and the participants equity holdings are measured four quarters before the loan s startdate. The equity holdings of a bank are assumed to be zero if it is found not to have any affiliation in 13F. The lenders equity holdings may be naturally higher in firms that have greater institutional ownership. Therefore, we control for the overall institutional ownership in the firm s equity with Institutional Equity Holdings, which is the percentage equity ownership of all the 13F institutions in the borrower, also measured four quarters before the loan s start-date. 13 We also employ alternative versions of the lead s and participants equity holdings to counter the effect of their skewness on our inferences. Lead s Equity Holdings are Large is a dummy variable that equals one if the lead arranger s equity stake in the borrower is above its mean value of 2%; it is zero otherwise. The average number of lenders in a syndicate is six, i.e., on average there are five participants in a syndicate. Correspondingly, we define Participants Equity Holdings are Large as a dummy variable equal to one if the aggregate equity holdings of the participants in the 13 The three equity-holdings variables that of lead, participants, and all institutions as well as the number of lenders are winsorized at the 99th percentile so as to avoid the effect of outliers on our estimation. 13

16 borrower are above 10% (i.e., 5 2%); it is zero otherwise. In these specifications, we also measure the overall institutional ownership with a dummy variable, Institutional Equity Holdings are Large; it equals one if the overall institutional holdings in the borrower are above its mean value, 40%, and it is zero otherwise. 3.4 Loan-specific and Borrower-specific Control Variables The regression specifications include several loan-specific control variables. Loan Size is the logarithm of the loan amount. Loan s Maturity is the logarithm of the loan s maturity counted in number of months and encompasses all facilities within the package. LIBOR (Drawn) is the yield spread over LIBOR that is paid on drawn funds. 14 Secured and Senior are dummy variables that equal one if the loan is secured or senior, respectively; they are zero otherwise. 15 In most of the specifications, we include various borrower characteristics in addition to the loan-specific control variables. 16 Book Leverage is the sum of long term debt and debt in current liabilities, calculated as a fraction of assets. Cash Holdings are cash holdings and short-term investments as a fraction of lagged assets. Return on Assets is income before extraordinary items as a percentage of lagged assets. Tobin s Q is the ratio of the sum of assets and market equity less book equity and deferred taxes to assets. Tangibility Ratio is the ratio of net property, plant, and equipment to assets. Capital Expenditures is the ratio of capital expenditures to net property, plant, and equipment. KZ Index is (Book Leverage) (Tobin s Q) (Cashflow) (Dividends) (Cash Holdings). 17 Here, Cashflow is the sum of income before extraordinary items, depreciation and amortization, calculated as a fraction of lagged assets, and Dividends is total dividends calculated as a fraction of lagged assets. The summary statistics for all the above variables are reported in Panel A of Table 2 while a comparison of the same across 14 Loan Size, Loan s Maturity and LIBOR (Drawn) are winsorized at the 99th percentile. 15 As mentioned above, for loan-specific information that is only available at the facility level and not the package level, we pick the largest facility in the package. As such, LIBOR (Drawn), Secured, and Senior correspond to the largest facility. 16 Borrower Size is not included because it is highly correlated with Loan Size and causes multicollinearity. 17 Our results are robust to using alternative measures of financial constraints, as proposed by Whited and Wu (2006) and Hadlock and Pierce (2010). 14

17 syndicated and non-syndicated subsamples is reported in Panel B of Table 2. We also include industry, year, and borrower-ratings dummies in all the regressions. Industry dummies are based on the 48 Fama and French industries, except we exclude financial firms belonging to the four Fama-French industries covering SIC codes For the credit ratings dummies, we classify all borrowers into seven groups. Borrowers without a long-term S&P credit rating are the benchmark group. All borrowers with a long-term S&P credit rating are classified into the following six groups: Group 1 is for ratings CCC and below. Groups 2, 3, and 4 are for borrowers rated B, BB, and BBB, respectively. Group 5 is for borrowers rated A and Group 6 is for those rated higher than A. All these groups also include the and + variations of the above ratings. 4 Contractual Restrictions and Conflicts of Interest Within the Syndicate 4.1 Covenants, Sweeps, and Restrictions in Bank Loans We start by presenting empirical results from the test of our first prediction that syndicated loans are more likely to include contractual features such as quantitative covenants, cashflow sweeps, and other restrictions, when compared with loans that are not syndicated. Our empirical test for this prediction involves estimating the odds for the presence of at least one quantitative covenant or sweep/restriction in a syndicated loan. We estimate a Logit regression where the dependent variable is either Covenant Dummy or Sweep Dummy, and the explanatory variable of interest is Syndicate. In addition, guided by the recent literature on bank loans, we include several variables to control for other potential determinants of the presence of covenants/sweeps in a loan contract. Specifically, we start by estimating the following Logit model: Covenant Dummy = α + β 1 Syndicate + γ 1 LOAN + γ 2 F IRM + δ i + φ j + ψ t + ɛ (1) Here, LOAN and FIRM denote the various loan- and borrower-specific control variables that are described above in Section 3.4, while δ i, φ j, and ψ t represent industry, credit rating, and year 15

18 dummies, respectively. As the number of lenders increases or the syndicate grows larger, the lead arranger s loan allocation is likely to be smaller. This accentuates the conflict of interest between the lead arranger (the pseudo equity-holder) and the other participants (the pure debt-holders). Therefore, we also estimate the model by replacing Syndicate with either the Number of Lenders or the dummy variable Large Syndicate. Results obtained from equation (1) using the dependent variable Covenant Dummy are reported in Panel A of Table The results in column (1) strongly support our main prediction that syndicated loans are more likely to include covenants. In fact, the odds of a covenant are nearly 60% higher in a syndicated loan, and these higher odds are statistically significant at the 1% level. We also find that the odds of a covenant are lower in larger loans (denoted by Loan Size), higher when a loan has longer maturity (Loan s Maturity), lower when the borrower pays a higher yield spread (LIBOR (Drawn)), and lower when the loan is secured (Secured). All these coefficients are economically meaningful as well as statistically significant and generally conform with the extant literature. Including borrowerspecific control variables in column (2) does not change any of the column (1) results discussed above. For the tests in columns (3) and (4) of Panel A, the independent variable of interest is Number of Lenders. We find that the odds of a covenant increase by 5% for every additional lender in the syndicate. This result holds even in column (5) when we limit the analysis to the subsample of syndicated loans. Further, within this subsample of syndicated loans, larger syndicates are also more likely than smaller syndicates to have covenants. In fact, the estimates in columns (6) and (7) show that the odds of using covenants are more than twice as high in larger syndicates as in smaller syndicates. We re-estimate the model (1) with Number of Covenants or Types of Covenants as alternative dependent variables; results using these are reported in Panels B and C, respectively, of Table 3. As described in Section 3.2 above, these are cardinal variables that count the number and types of 18 We report odds ratios, which means that the coefficient estimates that are larger (smaller) than one indicate that odds are higher (lower) due to the corresponding independent variables. 16

19 covenants, respectively. Accordingly, these alternative versions of the model are estimated using an Ordered Logit and test whether syndicated loans are likely to have a larger number and more types of covenants. In column (1) of Panel B, we find that the odds of having more covenants are almost doubled in syndicated loans, when compared with non-syndicated loans. Given the proportional slopes assumption underlying the Ordered Logit model, this implies that the odds of a borrower having eight covenants (the maximum number of covenants in our sample) versus less than eight covenants as well as the odds of, say, four or more covenants versus less than four covenants are both almost twice as large in syndicated loans. This interpretation is equally valid for the odds of a borrower having at least one covenant versus having no covenant at all. We find similar results in columns (2)-(7) of Panel B, and as above, the coefficients on the main independent variable are significant at the 1% level throughout the Panel. Although loan- and borrower-specific control variables are included in the regressions, we do not report their estimated coefficients for brevity. The estimated cut-offs for the eight categories of Number of Covenants are statistically significant, thereby justifying the use of this cardinal variable instead of collapsing the categories. However, collapsing the higher categories of four or more covenants into a single fourth category does not alter our results in any way; these are left unreported. For the tests in Panel C, the dependent variable is Types of Covenants. The estimates are again obtained from an Ordered Logit model and we find that the odds of having more types of covenants are also significantly higher in syndicated loans, when compared with non-syndicated loans. As described above, this implies that the odds of having four types versus less than four types of covenants as well as the odds of having, say, two or more types versus less than two types are nearly double in syndicated loans (as per columns (1) and (2)). Also, the estimated cut-offs for the four categories of Types of Covenants are significant, which justifies the use of this cardinal variable instead of one where the higher categories are collapsed. 19 The estimated coefficients in 19 However, collapsing the categories of two or more types of covenants into a single second category does not affect our results; these are left unreported. Our results (unreported) are also robust to using an OLS model instead of the Ordered Logit. 17

20 columns (3)-(7) are all significant at the 1% level, and suggest that the odds of including more types of covenants increase with the number of lenders or in larger syndicates. Again, the loan- and borrower-specific control variables are included but their coefficients are left unreported for brevity. In Table 4, we report the results obtained from equation (1) but now using the dependent variables based on the presence of cashflow sweeps and other restrictions. In Panel A, the dependent variable is Sweep Dummy and in Panel B, the dependent variable is Number of Sweeps. The independent variables of interest are the same as before. Overall, the results are similar to (though, somewhat weaker than) those reported in Table 3 above. In Panel A of Table 4, we find (in columns (1)-(2)) that the odds of a sweep/restriction are more than twice as high in syndicated loans than in non-syndicated loans; these estimates are significant at the 1% level. While an increase in the number of lenders (in columns (3)-(4)) does not seem to affect the odds of a sweep/restriction, the odds are higher by 40% in columns (6)-(7) for large syndicates (these are significant at the 1% level). We find similar results in Panel B, estimated using an Ordered Logit model with the dependent variable Number of Sweeps. The estimated coefficients in columns (1)-(2) are significant at the 1% level and suggest that the odds of a borrower having six (the maximum number in our sample) versus less than six sweeps/restrictions are twice as large in syndicated loans. Coefficients in columns (3)-(7) are positive but not always statistically significant. In Panel C of Table 4, we separately analyze the Dividend Restriction, which is one of the restrictions used in the loans. We especially focus on it because, as argued above, the lead arranger s information about and incentives vis-à-vis the borrower make it a pseudo equity-holder. We find that the odds of a Dividend Restriction are more than twice as high in syndicated loans; the odds increase with the number of lenders, and are more than 50% higher when the syndicate is large. 4.2 Controlling for Borrower and Lead-Arranger Fixed-Effects Even though the results in Table 3 and 4 are robust to controlling for an array of loan- and borrower-specific variables as well as dummies for year, industry, and credit ratings, the presence of covenants/sweeps may be due to some omitted borrower-characteristics. Therefore, we re-estimate 18

21 equation (1) after controlling for borrower fixed-effects (as well as all the loan- and borrowercharacteristics used in Tables 3 and 4). This automatically limits our data to only those borrowers that have at least one syndicated as well as one non-syndicated loan over the sample period. For brevity, we only report the results using dependent variables Covenant Dummy and Sweep Dummy, and show the main coefficients. These are reported in columns (1)-(3) and columns (4)- (6), respectively, of Panel A, Table 5. We find that our results continue to hold after controlling for borrower fixed-effects, the odds of a quantitative covenant (sweep/restriction) are nearly 70% (60%) higher in syndicated loans, when compared with non-syndicated loans of the same borrower. These odds increase with each additional lender in the syndicate and are also substantially higher in loans issued by large syndicates. All these estimates are significant at the 1% level. These results confirm that the use of restrictive loan features cannot be fully explained by borrower characteristics. Are loan contract features specific to the bank that arranges the loan? To address this concern, we estimate equation (1) with lead-arranger fixed-effects, which limits the data to only those banks that have arranged at least one syndicated loan as well as issued one non-syndicated loan. We present the estimated results in Panel B of Table 5; in all these regressions, we control for all the loan- and borrower-characteristics used earlier in Tables 3 and 4. We find that our conclusions are unchanged a given lead arranger is much more likely to include covenants and sweeps/restrictions in syndicated loans, when compared with its own non-syndicated loans. The odds of the presence of a covenant/sweep are also higher with large syndicates, when compared with other syndicated loans. To further test for the robustness of these results, in Panel C of Table 5, we re-estimate equation (1) with joint fixed effects of the borrower and lead arranger. This limits the sample to multiple loans across the same pair of borrower and lead-arranger. Even with this strict set of controls, we find that syndicated loans are about twice as likely to include quantitative covenants and sweeps/restrictions as non-syndicated loans. 19