Are syndicated loans really cheaper?

Transcription

1 Are syndicated loans really cheaper? Abstract In this paper we compare the spreads of single lender loans with multiple lenders in syndicated loans of similar characteristics. There are two countervailing effects at work. A risk sharing effect, which points to a negative relationship between the number of lenders and the loan spreads. A collusive effect, which suggests a positive connection given that borrowers have less opportunities outside the syndicate when lenders in the syndicate collude among themselves. We use data on large corporate loans in the US market for the period to provide evidence that, for comparable size, the second effect is more important than the first one once we control for borrower characteristics and potential selection biases. There is also evidence that as the deal amount increases the differential effect on spreads between syndicated and non-syndicated loans is reduced. Finally, the analysis shows two additional results. First, the differential in the spreads between syndicated and non-syndicated loans is reduced in recessions. Second, there is a non-lineal relationship between the number of members and leaders in the syndicate and the spreads. Keywords: syndicated loans, loan spreads JEL Classification: G20, G30

2 1. Introduction A traditional loan involves a relationship between the borrower and a single lender, which screens and monitors the borrower. The spread of such a loan is determined by the market conditions and the specific characteristics of the borrower. In syndicated lending, lead arrangers negotiate the contract with the borrower and organize a syndicate with participating lenders, with each member holding a fraction of the loan. This type of contract allows lenders to share credit risk. Reductions in committed capital and greater flexibility in reallocating this limited capital may be reflected in lower financial transaction costs for borrowers (Rajan, 2005). However, credit risk sharing creates disincentives for monitoring, which may result in asymmetric information problems between the informed lead arranger and the participant lenders. Such asymmetric problems may also have an impact on loan spread (Ivashina, 2009). Besides, the structure of the loan is important given that lenders in a syndicate may take advantage of their bargaining position and fix highly demanding financial conditions in comparison to single lender loans. In this papers we study the question of whether syndicated loans are cheaper than single lender ones. In this analysis, we recognize the key role that the size of the loan plays as a determinant on the lender s decision to syndicate a loan (Simmons, 1993). When lenders have limited capacity, syndication is natural. But sometimes the size of the loan is not a restriction. In our sample, at the lower range of the deal size, we observe syndicated loans such that some of its members had the capacity of taking the loan in full by themselves. In these cases, syndication is a strategic choice as a way to increase lenders bargaining power in front of a borrower by reducing this latter outside opportunities. Hence, our basic claim is that syndicated loans, particularly for small loans, may induce higher loan spreads than non-syndicated ones once the comparison is made between loans of similar size. There is an extensive empirical literature that studies the relationship between syndicated loan structure and its loan spread. Ivashina (2009) observes that loan prices on syndicated loans are lower than sole-lender loans, other things equal. Angbazo et al. (1998) investigate highly leveraged loans and find that syndicated loans have lower spreads than other similar loans. Dennis et al. (2000) find similar results for revolving loans. However, the empirical design in these papers does not address the selection concerns with regard to the endogeneity of the decision to syndicate. In our sample of US loans to public firms during , we estimate that single loans are, on average, 63 basis points cheaper than syndicated loans when the deal amount is in the lower 60 th intra-year

3 percentile. Our empirical design addresses potential problems of endogeneity in the syndication decision, and selection bias that may arise. We use a two-step linear regression method with endogenous treatment. Similar results are obtained when we use a non-parametric nearest-neighbor matching estimator and a propensity-score matching estimator. We provide three different contributions to the literature (i) We take into account the tranched structure of loans. Most of the empirical papers in the literature treat different tranches as different loans without taking into account the total size of the deal from which they come. However, the decision to syndicate the loan is made at deal level, and the pricing and other general contract terms (maturity, covenants, collateral, etc.) are made at tranche level. We compare syndicated versus non-syndicated loans for the same size level. To do so, we divide our sample into deciles of deal amount within each year, in order to study the impact of the size of the deal on the spread of its tranches. The top four deciles are composed mostly of syndicated loans, which tranches are on average cheaper that the tranches of the few sole-lenders loans with comparable characteristics. However, in the lower six deciles we find similar numbers of sole-lender and syndicated loans and, remarkably, the result is the opposite, that is, syndicated loans are more expensive than comparable sole-lender loans. Furthermore, we have found that this difference in spreads gets higher on average as we lower the loan size decile. (ii) We explore the impact of the number of lender in a syndicate on the loan spreads considering different loan sizes. We are able to reconcile different results in the literature by proposing a quadratic inverted U-shape relationship between the number of leaders in the syndicate and the spread. Ivashina (2009) explores this issue by studying the relationship between the lead share, inversely connected to the number of leaders, and the syndicated loan spread. On one hand, she argues that adverse selection and moral hazard problems imply that participants are exposed to the risk of wrongdoing by the leaders, and as a result they will demand a higher loan spread. An increase of the lead share will reduce these problems of asymmetric information, and thus the spread will be reduced. On the other hand, an increase in the lead share will also increase the lead credit risk exposure; hence the spread will increase. She was able to isolate both effects and found that the asymmetric information effect is more important, suggesting a positive connection between the number of leaders (negatively related

4 to the stake of the lead bank) and loan spreads. Other papers like Bae et al. (2014) found that the percentage of lead arrangers relative to the total number of leaders in a syndicate is significantly negatively related to the loan risk premium. In this paper, we find that the relationship between the number of leaders and syndicated loans spreads is quadratic with an inverted U-shape. In our econometric model, for 4 or more leaders, our results are consistent with Bae et al. (2014) findings. However, for 1 to 3 leaders we find that the predicted spread increases, as can be extrapolated from Ivashina (2009). (iii) Once we have separated the analysis in terms of expansive and recessive periods, we have found that the differences in the spreads between syndicated and non-syndicated loans are less important in recessive periods. In these periods, risk considerations are more important than strategic/collusive ones. The rest of the paper is organized as follows. Section II covers the related theoretical and empirical literature and the hypotheses to be tested. Section III discusses data. Section IV covers methodological issues and presents the results as well as some extensions/robustness analysis. Finally, Section V concludes. 2. Literature review and Hypotheses Empirical finance research has paid attention to the costs of syndicated lending. This literature has developed mainly along three lines of research: (i) the influence of the syndicate structure on the loan pricing; (ii) its relationship with other financial instruments; and (iii) the evolution of these costs of lending through the business cycle. (i) Syndicate structure and the pricing Several papers have explored how the structure of the lending syndicate affects the pricing. Angbazo et al. (1998) show that the spread on highly leveraged syndicated transaction loans is smaller when the arranger retains the largest share of any provider of funds. Hence, an increase in the number of leaders in the syndicate, which is inversely related to the stake of the lead bank, is connected to larger spreads. A similar result is presented in Bae, Chong and Kim (2014). They use an international dataset and find that the proportion of loan amount retained by lead arrangers in syndication is significantly negatively related to loan risk premiums after controlling for contract and country characteristics. Ivashina (2010) explains that the observable relation

5 between the lead share and the loan spread is endogenous. There are two opposing effects that simultaneously influence the loan spread. On the one hand, an increase in the lead share would reduce the asymmetric information between the lead lender and the participants, thus decreasing the premium demanded by the participants. On the other hand, an increase in the lead share would increase the lead risk exposure and increasing the premium demanded by the lead lender. She finds that the first effect, what she calls the asymmetric information effect is the most relevant one. Panyagometh and Roberts (2010), Ross (2010), and Godlewski et al. (2012) evaluate the sensitivity of loan pricing to lenders' experience and reputation, proxied by network centrality. They rely on three most widely used centrality measures: betweenness, closeness and degree to show that loan pricing is smaller as the syndicate is more central. In a similar vein, Wu et al. (2013) propose a theoretical model to analyze how interactions among potential lenders may influence the contract terms of the syndicated loan. They find that the relational distance, inversely connected to network centrality and that is higher in syndicated loans in comparison to non-syndicated loans, is positively related to the loan spread and the requirements for collateral. In a related paper, Wasan, Vijayakumar, and Daniels (2013) investigate the role of accrual quality in the syndicated loan costs. They show that lower accrual quality can exacerbate conditions of information asymmetry and lead to higher borrowing costs. Their results suggest that this may be more predominant in loans with multiple-arrangers. Hence the previous set of papers suggests that loan spreads may be larger in syndicated loans in comparison to non-syndicated ones. Another set of papers study the relationship between the cost of lending and the type of lenders that form syndicates. Ferreira and Matos (2012) examine cases where the bank is an insider of the borrower firm by representation on the board of directors or by holding equity stakes directly or indirectly. They find that banks charge higher interest rate spreads and face less credit risk after origination when they lend to firms where the bank is an insider. Harjoto, Mullineaux, and Yi (2006) find that investment banks charge higher spreads and are more generous to borrowers than commercial banks in pricing credit risk on the margin. They attributed these differences to the previous relationships that commercial banks are able to build with borrowers. Such lending relationship, which lead to a reduction in the spreads, are more likely to appear in sole-lender loans rather than in syndicated loans, thus, also suggesting lower spreads sole-lender loans in comparison to syndicated ones.

6 (ii) Linkages between securities markets and syndicated loan spreads This line of research studies the reaction of loan spreads to the information provided through other markets and financial instruments. Angbazo, Mei and Saunders (1998) investigate whether there is positive syndicated loan spread sensitivity to changes in spreads in the corporate bond market. They find that pricing in the syndicated loan and corporate bond markets diverge. Santos and Winton (2008) analyze whether there are differences in the pricing of loans for bank-dependent borrowers with the pricing of loans for borrowers with access to public debt markets. They find that firms with public debt market access pay lower spreads and their spreads rise significantly less in recessions. Relatedly, to this result, Focarelli, Pozzolo, and Casolaro (2008) show that arranger with a large stake (inversely related to the number of leaders in the syndicate) send a powerful signal to the financial markets over the quality of the borrower. Such certification signal leads to positive stock reactions. In this context banks can anticipate the positive borrower s stock reaction and charge lower spreads. Finally, in Schenone (2010) the borrower's initial public offering (IPO) is used as an informationreleasing event. After the IPO, for those firms with close relationship lending, which is more likely in solelender loans rather than in syndicated loans, interest rates decrease more. Hence, in this line of research there is also some evidence that loan spreads among non-syndicated loans should be lower than syndicated ones. The model of Adamuz and Hernández (2014) allows advancing in this idea and provide a theoretical setting to propose the existence of a positive connection between loan spreads and loan syndication. The key element in order to justify this result is that the comparison between the spreads of multiple versus single creditor loans has to be made for a fixed loan size. In case of multiple creditors, lender may collude among themselves and force borrowers to pay higher loan spreads in comparison to single lenders where borrowers have the option to borrow from other single lenders. In a similar vein, Carey and Nini (2007) show that among those loans in which the leader has a large stake, which are expected to be negatively related to the number of leaders, there is a negative association between banks stake and loan spread, and this effect is not explained by the characteristics of borrowers and lenders. These authors connect large number of leaders to reduced bank bundling, which give banks less leeway for reducing spreads. Angbazo, Mei, and Saunders (1998) as well as Ivashina (2009) also find a negative connection between the stake of the leading bank in a syndicate, inversely connected to the number of leaders in a syndicate, and loan spreads, particularly for highly leveraged syndicated transaction loans.

7 Keeping in mind the previous arguments, we posit the following Hypothesis to be tested: H1: Loan spreads in syndicated loans are larger than in non-syndicated ones once the loan size is fixed. (iii) The evolution of the loan spreads through the business cycle Santos and Winton (2008) shows that loan spreads rise in recessions, but firms with negotiated debt pay lower spreads given that their spreads rise significantly less in these periods. Mattes, Steffen and Wahrenburg (2013) seek empirical evidence for information rents in loan spreads by analyzing a sample of UK syndicated loan contracts for the period from 1996 to They find that undercapitalized banks charge higher loan spreads for loans to opaque borrowers and that this effect prevails only during recessions. With a macroeconomic perspective, Ivashina (2010) discusses the question whether syndication of loans increases the cyclicality of credit supply. She argues that loan syndication can amplify credit cycles. The process of syndication leads to an expansion of credit because it enables banks to share risk, which lowers their risk threshold for originating a loan. Syndication also expands credit by making it possible for institutional investors to participate directly in funding the types of loans that banks originate, rather than indirectly by funding the bank. Then, the possibility to manage risks more efficiently in syndicated loans in comparison to non-syndicated ones, which is pivotal in recessions, leads to a lower increase in loan spreads in syndicated loans with regards to non-syndicated ones during recessions. Hence, we can state: H2: The difference in spreads between syndicated loans and non-syndicated ones with a similar size is reduced in recessions. There are two loan characteristics that play a relevant role in the definition of loan spreads: loan size and the number of leaders in a syndicated loan. The size of a loan affects the balance between risk and strategic considerations in the determination of the loan spreads. For larger loans, the stake of the leader in the syndicated loans tends to decrease (Angbazo et al., 1998), which, in turn, reduces its risk exposure in comparison to non-syndicated loans. Consistently Bae, Chong and Kim (2014) finds, using an international database, that risk decreases with reduction in the stake of the lead arranger. Finally, Ivashina (2010) highlights that, apart from the decrease in risks when the stake of the leading bank decreases, there is also another effect that moves in the opposite direction. Information quality worsens as lead bank with lower stake have less incentives to monitor. However, this opposite effect is reduced

8 as the other banks in the syndicate, that may also become lead banks, may increasingly adopt a monitoring role. Hence, we posit that for larger loans, lenders risk considerations that favor the decrease in spreads among syndicated loans are more important than information asymmetries ones related to monitoring. As a consequence, and taking into consideration the statement of Hypothesis 1, the difference between the larger spread in syndicated loans in comparison to non-syndicated ones should be reduced as loan size increases. H3: The differences in spreads between syndicated loan spreads and non-syndicated ones with similar size is reduced as the size of the loan increases. Once we analyze the effect of the number of leaders in the syndicate, we expect that as the number of leaders in the syndicate increases, the strategic collusive effect becomes the most relevant one, which lead to an increase in the loan spreads. However, beyond some level, collusion among leaders of the syndicate is more difficult to achieve and borrowers can play with certain lead banks in the syndicate in order to get better financing conditions from them. Also, when there is a large number of leaders in the syndicate, their stake is reduced significantly, and the risk among lenders is more balanced, which lead in equilibrium to lower spreads. Finally, when the number of leaders in the syndicate is high, they can complement the monitoring tasks, which reduces information asymmetries with borrowers. The result is a reduction in the risk borne by bank leaders in a syndicate which, in turn, decreases loan spreads. Such tension between the previous effects in the relationship between the number of leaders and loan spread, eventually leading to a non-lineal relationship, is also captured in Ivashina (2010). Hence, we state as a last hypothesis: loan spreads H4: There is an inverted U-shape relationship between the number of leaders in the syndicate and the 3. Data and descriptive analysis 3.1. The Data The test of the previous hypothesis relies on the Dealscan database, which is compiled by Reuters Loan Pricing Corporation (LPC). This database contains information on large corporate and middle market commercial loans filed with the Securities and Exchange Commission, or obtained through other reliable public sources. Compustat database complements the former database by providing information on borrowers. The merge of both databases uses the link table provided by Roberts and Chava (2008). Each contract listed in Dealscan, referred as a deal, consists of one or several facilities or tranches. Maskara (2010) gives a theoretical explanation for the economic value of this structure. Dealscan provides a unique identification number for each

9 deal allowing an easy identification of all the tranches belonging to the same deal. Different tranches in a deal can show heterogeneous characteristics in terms of price, amount, currency, maturity, and default probability, among others. In particular, not all the lenders that are members in a syndicated loan participate in every tranche of the deal. Our initial information was composed of 134,791 tranches corresponding to 84,760 contracts of loans to 30,496 different borrowers originated in the U.S. market, over the period (inclusive). In the empirical analysis, there are several measures related to the syndicate structure and its previous relationships with the deal borrower that requires using the information of the four years previous to each deal active date. This implies that deals from years are only used for that purpose, and therefore the actual sample only considers loans issued between 1990 and From the period, there are some loans that are excluded. First, those loans to borrowers that are government entities, banks, or financial institutions, and also regulated borrowers such as transportation and public utilities (industries identified as SIC 91-99, and 40-49). Second, those deals where at least one tranche is either non Senior debt, or is a denominated in currencies other than the US dollar. For the 90.22% percent of the remaining deals, the base rate is LIBOR. Therefore, deals with base rate other than LIBOR were also excluded for the sake of homogeneity. After this cleaning process, the final sample included 59,951 tranches corresponding to 36,338 deals to 14,855 different borrowers, over the period However, only 5,257 of these borrowers have accounting information available in Compustat, since the rest are mostly companies that have never issued publicly traded securities. This implies that when we restrict our econometric models to contracts for which the information from both DealScan and Compustat is available, there are only 32,352 tranches corresponding to 21,187 deals (about one fourth of the original data) Construction and description of the variables The dependent variable of the analysis is Dealscan s all-in-spread drawn, which we will simply call spread. It is defined by DealScan as the total annual cost, including a set of fees and fixed spread, paid over LIBOR for each dollar used under the loan commitment, and it is expressed in basis points. This spread represents the price of each tranche of the loan.

10 Among the independent variables, there are two key groups of variables for our analysis, namely, the number of members in syndicated loans, decomposed into leaders plus participants, and the dollar size of the loan, represented both by the deal and its tranches amounts. These variables will allow us to contrast Hypotheses 3 and 4. The definition of the number of members in a syndicated loan relies on the definition of what a member is. We define as member any lender that figures in the contract. Some of these members are classified as lead arrangers since they underwrite the contract and negotiate loan terms with the borrower, retaining primary administrative, monitoring, and contract enforcement responsibilities. Typically, lead arrangers retain the largest stake in the loan, but will also find other lenders to share the total amount of the deal. Other members perform administrative oversight duties, although these are relatively minor, and their share ownership in the loan is smaller, on average, than lead arrangers. These lenders are generally referred to as loan managers. Both groups are jointly defined in the present paper as the leaders of the deal, given the difficulty to discriminate lead arrangers from loan managers from the information provided by Dealscan. A third group, which we will refer to as participants, do not perform special functions other than being signatories to the loan agreement. This structure is further complicated by the tranched structure of a deal. For instance, a deal leader can act as a participant in one or more tranches. In order to avoid language verbosity, we will assume that a non-syndicated deal is a contract with exactly one leader and zero participants. DealScan provides detailed information about the syndicate structure at tranche level. Specifically the All Lenders variable list the names of all the members of every tranche. The percentage of the tranche amount retained by some or all lenders is also reported, but only for 40.95% of the tranches in our sample. DealScan also provides us with the variables Agent, Arranger, Bookrunner, Co-Arranger, Lead Arranger, Lead Manager and Top Tier Arranger, that lists all the members with specific roles within the tranche. We will assume that a lender is a tranche leader if it appears in at least one of these seven lists, and a tranche participant otherwise. We further assume that a lender is a deal leader if it is a leader in at least one tranche of the deal, and a deal participant otherwise. The second most relevant explanatory variables are both the dollar deal amount and dollar tranche amount. 1 If we look at the distribution of deal amounts within every year in our study period , there is a clear increasing trend in mean and dispersion as time advances. This effect is due to inflation and other 1 For any given deal, the sum of all its tranche amounts should be equal to the deal amount. However, this consistency condition is not satisfied by a significant part of the sample. In particular, between 2007 and 2013 there is a high percentage of deals in which the sum of tranches is considerably higher than the reported deal amount. This percentage ranges from a minimum of 9.6% in 2007 to a maximum of 32.9% attained in 2010 (these deals account for the 55% of all tranches in 2010). Before 2006 this percentage never exceeds 3.6%. We do not know the reason for this increasing level of inconsistencies. In order to save about half of the tranches of the most recent years in our sample, we decided to ignore the reported deal amount and use the sum of tranches instead.

11 historical factors. To account for this source of inhomogeneity in our econometrical models, we sort all deals (syndicated or not) within a given year by increasing value in deal amount. We then assign to every deal its percentile, ranging from 0 to 100, according to this order. We repeat this same procedure assigning intra-year deciles labels to each deal. Our main hypothesis (H1) is clearly stated in term of these new variables: all other factors equal, syndicated loans are more expensive than non-syndicated loans in the lower quintiles of deal amount. Following Sufi (2007) there is an aggregation of all lenders to their parent company in the present work. This implies, for example, that a tranche with two reported lenders which are subsidiaries from the same bank is taken as a single lender tranche. 2 In order to compare the size of single-lender credits with multiple-lender credits, ideally, the comparable quantity in a syndicated tranche should be the biggest dollar share retained by any member (usually a tranche leader), or the alternatively the total dollar share retained by all the leaders. This information is regrettably very scarce in our sample, it can only be computed for about one fifth of all syndicated tranches. A simple proxy is the deal amount divided by the number of members in the deal. This proxy was also computed at tranche level. The rest of the explanatory and control variables included in our econometrical models are the usual variables found in studies similar to ours. In particular, we drew inspiration from Ivashina (2009), Maskara (2010) and Sufi (2007), adding some adjustments and refinements. We will explain now how we constructed the variables for which the definition is not trivial. The complete list of variables can be found in Appendix A. Several characteristics about the borrower and the lenders were computed using information from the four years previous to the deal active date. This is what is defined as the previous period. In particular, Dealscan information from the period is only used for creating the corresponding variables for loans in 1990, and therefore that initial period does not directly appear in our analysis. The approach to lenders reputation ranks all lenders that appears in the sample in any given year by sorting in descending value the total dollar amount of all deals in the previous period (syndicated or not) where this lender participated. Since this is a reputation measure we used the total deal amount, and not the share 2 Previous studies that use DealScan data at tranche level consider that a tranche belongs to a syndicated deal whenever its number of lenders is reported as one. This is not accurate. A multi-tranched syndicated deal can have one or more tranches with only one participating lender. This can happen for example when the lead arranger retains the riskier part of the loan for itself in one tranche. There are 794 of those single-lender tranches belonging to a syndicated loan in our sample. In order to correct this and other inaccuracies, a deal is taken as syndicated if all its tranches have the same sole-lender.

12 retained by the lender. For instance, lender in position 1 had the biggest participation in the sample in the previous four year. All lenders that had no participation during that time were ranked last (in a position number around the three thousands). In the case of syndicated deals, we use the best ranked leader, that is, the one with lowest position number. Finally, we take the decimal logarithm of the obtained position number. Notice that this is an inverse reputation measure, in the sense that its lowest value is 0 for the deals with the best reputed leaders, and its highest value is around 3 for deals with leaders with little or no participation in this market during the previous period. Following Ivashina (2009), there is the definition of the lending limit at tranche level for every lender in the sample in any given year. It is defined as the 75th size percentile of the lender dollar share in tranches from loans issued in the previous period. According to Ivashina (2009), the lending limit is a proxy for the lender s loan portfolio diversification. It is important to stress two points here. First, recall that we always aggregate lenders to their parent company in this work, which tend to increase the lending limit. Second, the dollar share in non-syndicated tranches is simply the tranche amount, however, it can only be computed in those syndicated tranches that provides the share percentage. The scarcity of this information varies from one lender to another, but we expect the 75th percentile to be a consistent measure, at least for the best ranked lenders. We assign the sole-lender lending limit to every non-syndicated tranche. In the case of syndicated tranches, we use the maximum lending limit among all tranche leaders. We also compute this variable at deal level, and take decimal logarithms in both cases. There are another two measures taken from Ivashina (2009) that are specific to syndicated loans, Lead Reputation and Lead Ratio. Only the first one requires further clarification here. As explained before, the member lenders in a syndicated loan are divided into two disjoint groups: leaders and participants. For each pair leader-participant (L-P), we count the number of deals in the previous four years in which L was also a leader, and P was a member, regardless of P s role. This is what we refer to as number of links between L and P. This number of links is then scaled by the total number of deals during the previous period where L was a leader. In this way we obtain the percentage of deals lead by L where P was a member. Finally, we take the maximum of this percentage among all possible pairs L-P within the deal. The construction of the rest of the variable should be clear from the descriptions provided in Appendix A. Notice that there is only partial availability of the data for several of the variables in the analysis. For that reason we will always report the number of observations considered in each of our econometric models.

13 4. Empirical analysis 4.1. Methodology We use a two-step linear regression with endogenous treatment effects model to correct for selection bias (Heckman, 1978): Tranche Spread ij = α 0 + α 1 Syndicated i + α 2 Syndicated i DealSizeQuantile i + α 3 log(numleaders ij ) + α 4 [log(numleaders ij )] 2 + α 5 log(numparticipants ij ) + α 6 TrancheControls ij + α 7 DealControls i + α 8 SectorControls i + α 9 YearControls i + ε ij. (1) Y i = β 0 + β 1 DealSizeQuantile i + β 2 log(dealsize i /LeadersLendingLimit i ) + β 3 DealControls i + β 4 SectorControls i + β 5 YearControls i + η ij. (2) Where Syndicated i = 1 when Y i > 0 and Syndicated i = 0 otherwise, and corr(ε ij, η ij )=ρ. Subscripts i and j index deals and tranches within the deal respectively. To correct for the correlation between the error terms of different tranches in the same deal, we cluster observations (tranches) by deal, and use the bootstrap method with 10,000 sub-samples in order to compute the standard errors Results Descriptive Analysis There is an overall increasing tendency in the number of leaders per deal during our period of study. The intra-year median remained at 1 from 1990 to 1998, and then at 2 from 1999 to 2013, with the exception of 2011 that displays a median of 3. During that time the intra-year average number of leaders oscillated between 1.47 in 1990 and 3.16 in These statistics are higher if we restrict our sample to public firms, i.e. the borrowers that we were able to link to Compustat, although the trend is the same. In this case the intra-year median number of leaders per deal ranges from 1 to 4, whereas the average oscillates between 1.45 and In Figure 1, we show initial evidence of the comparison between the spread of syndicated loans and non-syndicated ones. We also show the distribution of the syndicated loans in terms of the number of members

14 in the syndicate. There is initial evidence that the spread of the syndicated loan is larger than the non-syndicated ones when the number of members is low. This conforms to Hypothesis 1. However, this relationship is reversed when the number of members in a tranche is large, which is not in contradiction with Hypothesis Insert Figure 1 about here In what follows we restrict the sample to contracts for which the information from both DealScan and Compustat is available, consisting all in loans to public firms. Figure 2 shows the intra-year distribution of deal amount (in log units) for both types of deals. Clearly, syndicated loans are bigger. However, each year both distributions have an important overlap, implying that size alone cannot explain the decision to syndicate deals of low and medium size. There must be other factors at play such as risk sharing, or strategic pricing decisions by the lenders Insert Figure 2 about here We further explore this point in Table 1. Within each year, we group all loans by deciles of deal amount. On each row we show the number of non-syndicated and syndicated deals. The percentages in parenthesis below each number are cumulative, including all the deals in the given decile or lower. For example, in our sample there are 294 non-syndicated and 1,850 syndicated deals in the 6 th decile by deal size, whereas 40.3% of the deals from deciles 1 to 6 are non-syndicated, and 59.7% are syndicated Insert Table 1 about here From this point on we will discard all deals in deciles 7 to 10 from our sample, since this size range is dominated by the syndicated loans. Thus, we will focus on the region where the decision of syndication versus non syndication is not driven mechanically by a question of loan size. We argue that for these bigger deals the size is the main determinant in the decision to syndicate. Instead, we would like to focus on deciles 1 to 6, where there are comparable numbers of deals of both types, and where strategic collusion decisions cannot be discarded a priori. Figure 3 depicts the distribution of deals by type over the years ( ), and Tables 2A and 2B shows the descriptive statistics of all the variables. In both cases we restrict the sample to deals in the lower

15 intra-year 60 th percentile in size. Therefore, in all further analysis our sample will consist in 4,229 nonsyndicated deals that spans a total of 5,357 tranches, and 6,255 syndicated loans with a total 8,661 tranches Insert Figure 3 about here This figure shows that in the recessive year of 2008, there has been an increase in the non-syndicated loans compared to the syndicated ones. Such result conforms to Hypothesis 2 on the smoothing effect in recessions of the spread differential between syndicated and non-syndicated loans. The descriptive results of Table 2A (that focuses on the 60% lowest deciles in terms of loan size) confirm that syndicated loans are larger and they are lent to larger and more profitable firms. Also, in terms of spreads, Table 2B shows that the mean spread of syndicated loans is higher (241,8 bp) in comparison to nonsyndicated loans (223,1 bp) in the 60% lowest size deciles Insert Table 2A and Table 2B about here Tests of Hypotheses Table 3 tests the main hypothesis of the paper (H1) as well as H3. Column 1 shows the results on loans spreads, while column 2 shows the results of the probability of syndication Insert Table 3 about here Table 3 shows that, independently of the loan size decile considered, the spread of syndicated loans are larger than those of the non-syndicated ones. In particular, the syndicated loans in the first decile show bases point more than non-syndicated ones in the same decile. Such a result holds for the different loan size deciles up to the sixth decile given that the sum of the coefficients of Synd + Deal_Amount_Deciles x Synd is negative for all deciles. On average, syndicated loans show spreads that are 63 basis points higher than nonsyndicated ones when the loan size is in the lower 60 th intra-year percentile. This conforms to H1 and provides empirical support to the main contention of the paper.

16 Remarkably, as there is an increase in deciles (loans become larger), such differential in the spreads is reduced (coefficient of Deal_Amount_Deciles x Synd becomes more negative as we move upwards in the loan size deciles). This last result conforms to H3. To test H4, Table 4 introduces as additional explanatory variables Log_Deal_Leaders and Log_Deal_Leaders2 (columns 1) as well as Log_Deal_Members and Log_Deal_Members2 (columns 2). The result indicates that the linear coefficient is positive and the quadratic is negative. Such evidence is also consistent with the inverted U-shape relationship shown in Figure 4, hence, confirming H Insert Table 4 and Figure 4 about here Finally, to test H2, Figure 5 provides evidence that the differential effect of loan spreads once we compare syndicated versus non-syndicated loans is much lower (and even disappears) once we focus in a recessive period ( ). This evidence is consistent to the result shown in Table 5 once we focus on the most robust estimation of column 3 (linear regression with an endogenous treatment of the decision to syndicate through a two-step approach). In column 3, we have found that the differential effects in the spreads once we compare syndicated versus non-syndicated loans disappears in the period , which is fully consistent to H Insert Table 5 and Figure 5 about here Robustness Contingency Analysis In Table 6 we have conducted an analysis for the period (previous to the recession) of the differences in spreads between syndicated and non-syndicated loans in different scenarios. We have found that such differences are larger for smaller loans (consistently to H3) and short-maturity loans. In terms of borrower characteristics, these differences are larger when borrowers are smaller and they have sound financial conditions. Finally, in terms of lenders, the differential in spreads are larger when there is no previous relationship between borrowers and lenders (high information asymmetry).

17 Insert Table 6 about here Endogeneity Analysis One problem in the estimation is that the key explanatory variable (Synd) is endogenous with the determination of the spreads in the loan contracts. We have tackled this issue to some extend by following a two-step approach as explained in the methodological section. With such approach we have used as instrument of the Synd variable, the prediction of this variable according to the specification (2) shown in section 4.1. Alternatively, we have taken advantage of a normative shock (market flex provision) that was introduced after 1998: Such change allowed the lead arranger to change the pricing terms of the loan during the syndication process. Before such normative change, while the pricing terms were set prior to the award of the mandate, loans were underwritten under the best-effort criteria. In that sense, if at a given spread the deal wasn t fully subscribed, the loan wouldn t go through. Hence, such shock clearly affected the syndication probability and it is much less connected to the loan spread. Making use of this shock, in Table 7 we show the differential effects in the spread between syndicated and non-syndicated loans once we compare the three years before the introduction of the market flex provision (1995, 1996 and 1997), with those after such provision (1998, 1999 and 2000). We have found that such differences are larger in the latter period ( =23.2) rather in the former one (20). This result is also shown in Table 5 (last column) when we compare the period with the period The differences in spreads are only significant in the latter period, which is a signal of the robustness of the result found Insert Table 7 about here / Discussion and Conclusions In this paper we examine a dataset of large corporate loans in the US market for the period and estimate the relative loan cost disadvantage of sole lender loans with respect to syndicated ones. Contrary to some of the literature, we posit that the spreads are larger for syndicated loans in comparison to non-syndicated ones once the comparison is made with credits of similar size. We also analyze whether the size of the loan moderates the previous difference in spreads. We find that single loans are, on average, 63 basis points cheaper than syndicated loans when the deal amount is in the lower

18 60 th intra-year percentile. We acknowledge the potential endogeneity of the decision to syndicate by using several econometric methodologies like endogenous treatment models, nearest-neighbor matching and a propensity-score matching methods. All tests support our findings that there is a loan cost disadvantage of syndicated loans versus traditional loans, particularly for smaller loans. We run some robustness checks by analyzing the average treatment effects for different sub-samples and found that borrower s size and borrower s risk decreases the differential in the spreads between syndicated and non-syndicated loans of similar characteristics. Besides, we have analyzed how the number of members of a syndicate influences the syndicated spread. We differentiate the impact on the loan spread of different type of membership and we have found that there is an inverted U-shape relationship between the number of leaders and members of the syndicate and the spreads. Last, we study the evolution of the differential in spreads through time. We consider three temporal subsamples, and find that this differential decreased during the period , that comprises the last financial crisis. Also, we have taken advantage of a normative shock that appeared in 1998 (market flex provision) to show the robustness of the results found. There are several avenues for further research. First, we can introduce in the analysis the characteristics of the ownership structure of the borrower as well as their corporate governance. Second, it may be worth studying the effect on the bank loan spreads when borrowers issue in financial markets different financial instruments like debt or SEOs. This phenomenon is increasingly more important as firms use more market mechanism to raise funds. Finally, an evaluation of the economic effect in terms of growth of such differential in loan spreads would contribute to the ongoing debate of the best way to provide funds to firms in order to stimulate the economic growth. This is left for future research.

19 References Adamuz, M.M and Hernández, J. (2015). Endogenous screening and the formation of loan syndicates. International Review of Economics & Finance, 37, pp Angbazo, L., Mei, J., & Saunders, A. (1998). Credit spreads in the market for highly leveraged transaction loans. Journal of Banking & Finance, 22 (10), Bae S.C.,B. Chong and Y.Kim (2014). Informational frictions, syndicate structure and loan pricing: new evidence from international lending. Asia-Pacific Journal of Financial Studies, 43, Carey, M.,and Nini, G. (2007). Is the corporate loan market globally integrated? A pricing puzzle. The Journal of Finance, 62(6), Dennis,S.,D.Nandy and I.G. Sharpe (2000). The determinants of contract terms in bank revolving credit agreements. Journal of Financial and Quantitative Analysis. 35 (1), Ferreira, M.A. and P.Matos (2012). Universal banks and corporate control: evidence from the global syndicated loan market. Review of Financial Studies, 25(9), Focarelli, D.,A.F. Pozzolo and L. Casolaro (2008). The pricing effect of certification on syndicated loans. Journal of Monetary Economics, 55, Godlewski, C.J.,B. Sanditov and T.B. Helmchen (2012). Bank lending networks, experience, reputation, and borrowing costs: empirical evidence from the french syndicated lending market. Journal of Business, Finance & Accounting, 39(1), Harjoto, M. D.J. Mullineaux and H.C. Yin (2006). A comparison of syndicated loan pricing at investment and commercial banks. Financial Management, 35(4), Ivashina, V., (2009). Asymmetric information effects on loan spreads. Journal of Financial Economics 92 (2), Ivashina, V. and D. Scharfstein (2010). Loan syndication and credit cycles. American Economic Review: Papers &Proceedings. 100, Heckman, J. (1978). Dummy endogenous variables in a simultaneous equation system. Econometrica 46:

20 Maskara, P., (2010). Economic value in tranching of syndicated loans. Journal of Banking & Finance 34 (5), Mattes, J.A., Steffen, S. and M. J. Wahrenburg (2013). Do Information Rents in Loan Spreads Persist over the Business Cycles? Journal of Financial Services Research l 43 (2), Panaygometh K. and G.S. Roberts (2010). Do banks exploit syndicate participants?. Evidence from ex-post risk. Financial Management, 39(1), Roberts, M., Chava, S., (2008). How does Financing Impact Investment? The Role of Debt Covenants. The Journal of Finance 63 (5). Ross, D.G. (2010). The dominant bank effect : how high lender reputation affects the information content and terms of bank loans. Review of Financial Studies, 23(7), Santos, J.A.C. and A.Winton (2008). Bank loans, bonds, and information monopolies across the business cycle. Journal of Finance, LXIII (3), Schenone, C. (2010). Lending relationships and information rents: do banks exploit their information advantages?. Review of Financial Studies, 23(3), Simmons (1993). Why do banks syndicate loans?. New England Economic Review of the Federal Reserve Bank of Boston, pp Sufi, A., (2007). Information asymmetry and financing arrangements: Evidence from syndicated loans. The Journal of Finance 62 (2), Wasan, S. J. Vijayakumar and K.N. Daniels (2013). Accrual quality and borrowing costs in the syndicated loan market. Journal of Accounting and Finance, 13(6), Wu et al. (2013). The cascade effect on lending conditions: evidence from the syndicated loan market. Journal of Business, Finance & Accounting, 40(9),

21 Appendix A - Description of the Variables Deal Level Variables Endogenous variables: synd Dummy equal to 1 when all the tranches in the same deal have the same sole-lender. * DealScan Borrower characteristics: zscore Altman's Z-Score. *** Compustat AltamanClass Altman's Z-score classification: (1="distress" if zscore<1.81, 2="grey" if 1.81 zscore 2.99, 3="safe" if 2.99<zscore). Compustat log_assets log10 of Total Firm Assets Compustat Tangibility Percentage of tangible assets = 100*(Total Assets - Intangible Assets) / Total Assets. Compustat ROA Net Income & Losses / Total Assets. *** Compustat Revenue Net Sales / Total Assets. *** Compustat Leverage Total Debt / Total Assets. *** Compustat MajorExchange Dummy equal to 1 if the company is traded on major exchange, and 0 if publicly traded, but not on major exchange. Compustat Sector Industry sector of the borrower identified by its SIC code. Discrete variable with codes: DealScan 1 for SIC 01-09: Agriculture, Forestry, and Fishing 2 for SIC 15-17: Construction 4 for SIC 20-39: Manufacturing 5 for SIC 10-14: Mining 7 for SIC 52-59: Retail Trade 8 for SIC 70-89: Services 10 for SIC 50-51: Wholesale Trade PrevSynd Dummy equal to 1 if the borrower was issued a syndicated loan in the previous period. ** DealScan Contract characteristics: DealAmount Deal Amount. This is the total amount lent in the current contract, in Millions USD. It is the sum of the amounts of all tranches constituting the deal. DealScan deal_amount_p Percentile of deal amount within the same year (continuous, ranging from 0 to 100). DealScan deal_amount_d Decile of deal amount within the same year (discrete variable ranging from 1 to 10). DealScan avgdealamount Average deal amount per lender (Deal Amount / number of leaders in this deal). * DealScan deal_lendlimit Leaders lending limit (75th size percentile of leader share in loans issued in the previous period, taking the maximum in case of several leaders). ** DealScan deal_excess Ratio of total Deal Amount to deal Leaders Lending Limit. DealScan Tranches_per_Deal Total number of tranches in this deal. DealScan log_deal_members log10 of the number of lenders that are members in this deal. * DealScan log_deal_leaders log10 of the number of leaders in this deal. * DealScan log_deal_participants log10 of one plus the number of participants in this deal. * DealScan year Year obtained from the deal active date (discrete variable ranging from 1990 to 2013). DealScan Leader/sole-lender characteristics: PrevLending BorrowerRel Dummy equal to 1 if any member lender of the present deal had issued a loan to the borrower in the previous period, either as a sole lender or as syndicate leader. ** Percentage of lenders in borrower s network. Is the number of member lenders in this this deal that has participated in a loan to the borrower (syndicated or not) during the previous period, normalized by the number of lenders in this deal. ** DealScan DealScan LeadRank log10 of the rank of the lowest ranked leader in this deal. The rank is based in the position when the total dollar value of all deals where this lender participated in the previous period, sorted by descending value. ** DealScan