Bank Loan Renegotiation and Credit Default Swaps *

Transcription

1 Bank Loan Renegotiation and Credit Default Swaps * Brian Clark 1,2 clarkb2@rpi.edu James Donato 1 james.donato@liu.edu Bill Francis 1 francb@rpi.edu Thomas Shohfi 1 shohft@rpi.edu September 2017 ABSTRACT We find that the inception of CDS trading on reference firms debt is associated with a decreased number and lower probability of amendments, restatements, and rollovers to existing lenders of bank loans. Reference firms are also less likely to terminate loans prematurely or refinance with different lenders after the inception of CDS trading and tend to experience shorter loan renegotiations. Evidence is consistent with the empty creditor problem arising from CDS trading and the resulting decrease in the negotiation power of borrowers. Our research contributes to understanding how financial innovations alter bank lending relationships. 1 Lally School of Management, Rensselaer Polytechnic Institute, th Street, Troy, NY Office of the Comptroller of the Currency, 400 E St. SW, Washington, DC * The views expressed in this paper are those of the authors alone and do not necessarily reflect those of the Office of the Comptroller of the Currency or the U.S. Treasury Department.

2 1 Introduction There is a large literature devoted to studying how financial innovations strengthen the financial sector by providing benefits such as diversification of risk, increased liquidity, reduced transactions costs, and improvements in risk management. At the same time, financial innovations have also been shown to play a role in the de-stabilization of the financial sector, which was a key factor in the financial crisis. For example, innovative products were misvalued, contributed to excessive risk taking in the financial sector, and in some cases allowed financial institutions to skirt regulations put in place to safeguard the financial system. In this paper, we examine how financial innovations affect stakeholders outside of the financial sector. In particular, we focus on how the presence of credit default swap (CDS) contracts alters bank lending relationships in the context of bank loan contract renegotiations. The bank loan market is of primary importance to firm borrowing. According to the Shared National Credits Program Review, $1.986 trillion of total outstanding loan commitments were reported as of the first quarter of Similarly, the CDS market has grown from the early 2000 s into a $5.58 trillion market. 2 Given the size and interplay between these markets, it is important to understand how they interact. Prior work examining how the existence of CDS contracts have affected the bank loan market has focused on loan monitoring (see, e.g., Hu and Black (2008), Shan, Tang, and Winton (2016a)), the availability and cost of credit (see, e.g., Subrahmanyam, Tang, and Wang (2014), Saretto and Tookes (2013)), and the structure of debt contracts (Shan, Tang, and Winton (2016b)). However, less work exists that examines how the presence of CDS trading affects the dynamics of creditor-debtor relationships That figure is the notional exposure for single-name instruments as of the second half of 2016 which is the latest data available at 2

3 As such, we study how the inception of CDS trading on reference firms impacts the renegotiation of their bank loans. Understanding this relationship takes on additional importance given that, as pointed out by Roberts (2015), Danis and Wang (2014) and Nini et al. (2012), the vast majority of bank loans are renegotiated with renegotiations starting within the first 3 months of the life of the loan. This phenomenon carries implications for our understanding of financial contracting and bank debt. Specifically, we ask whether the onset of CDS trading on a firm s debt alters the creditor-debtor relationship so as to increase the relative negotiating power of lenders over borrowers. Our hypothesis is motivated by the empty creditor problem whereby CDS contracts are said to fundamentally change the relationship between creditors and debtors in favor of the lender by providing an outside option by which lenders can hedge default risk. And by doing so, the lender separates itself from economic exposure to default. Simply put, we hypothesize that lenders become disinterested in renegotiating bank loans since the probability of bankruptcy is no longer threatening. That is, agreeing to an out-of-court debt restructuring or renegotiation is no longer necessary as a way to salvage value when the CDS will make the creditor whole when triggered in bankruptcy -- in fact, the payoff could even be greater (Bolton and Oehmke, 2011). While we focus on bank loan renegotiation, it should be noted that the majority of CDS contracts trade on senior corporate bonds not bank debt. However, default on a senior unsecured bond is no doubt highly correlated with default on a bank loan. Additionally, crossacceleration and cross-default clauses help to strengthen this relationship as default on one type of debt triggers any CDS trading on the firm. From a bank s perspective, under the Basel Accords, banks may use these CDSs in order to obtain capital relief on existing bank loans (Shan et al., 2016). In other words, banks hedge loan exposure through CDS contracts as a way to 3

4 mitigate credit risk even though these instruments are not directly insuring bank debt against default. Roberts (2015) points out that [r]enegotiation occurs when the parties to a contract are unable to commit to the terms of their agreement (p. 66). Theory suggests (see, e.g., Hart and Moore, 1988) that this occurs because of the inherent difficulty in verifying important information that could have significant impact on the value of debt. Nevertheless, the question still exists as to why would both parties engage in a contract ex ante, that has a significant chance of running afoul ex post? In other words, why would two parties design an agreement where renegotiation is expected to occur given that renegotiation is inefficient and costly for both parties to the transaction? Roberts suggests two major reasons. The first is that contracts are written with an eye toward punishing the borrower for risky risk-shifting ex-post behavior. The second, as is well known, financial contracts are inherently incomplete because of the significant costs that both parties would incur to write contracts that would account for all states of nature ex-post (Hart and Moore, 1988). Thus, as suggested by the extant literature (see, e.g., Hart and Moore, 1988; Aghion and Bolton, 1992; Dewatripont and Tirole, 1994) renegotiation occurs to correct ex-post inefficiencies that occur due to the need to impose ex-ante disciple on borrowers. By writing loan contracts with covenants and imposing strict penalties for deviations from covenants, lenders intend to discourage risky behavior ex ante. In this way, creditors seek to provide a deterrent to behavior they deem risky. However, contract violations and subsequent renegotiations could occur precisely because lenders have inserted penalties into the debt contract, which leads to costly and inefficient outcomes. Furthermore, if renegotiation is possible, then the strength of the deterrent is questionable. That is, if borrowers realize that they can renegotiate bank loans, then they could be dis-incentivized to adhere to the strict conditions 4

5 of the contractual agreement. For the purposes of this paper, we ask how these incentives change due to the inception of CDS trading. As we discuss below, CDSs strengthen the creditor s relative negotiating power and provide a disincentive to engage in costly renegotiations, which could result in fewer renegotiations. Traditionally, a creditor has three basic rights when lending: economic rights to be repaid principal and interest, contractual control rights to monitor the borrower, and legal rights to pursue litigation in the event of contractual breach or default (see, Hu and Black (2008)). Furthermore, lenders are usually quite concerned about being paid back their loan. However, CDSs fundamentally change this relationship between borrower and lender. Because creditors now have the ability to hedge exposure to default risk through use of what is essentially an insurance product, they become disentangled or decoupled from their economic rights. Although creditors still retain these traditional rights noted above, they are likely to become disinterested in exercising them. If the borrower defaults and enters bankruptcy, then this event triggers the swap, which makes the creditor whole. As a result of this phenomenon, Hu and Black label this behavior as empty crediting (p. 665). Creditors are empty in the sense that while they remain debtholders they may no longer be interested in out-of-court debt restructuring or renegotiation; instead, they opt for bankruptcy and a possible higher payoff. Additionally, empty creditors may have the upper-hand in any debt negotiation with borrowers simply because they may lack the need or desire to avoid bankruptcy. There is ample empirical evidence in favor of the empty creditor problem. Subrahmanyam et al. (2014b) demonstrate that CDSs cause an increase in the probability of bankruptcy for firms due to empty crediting as the likelihood of bankruptcy more than doubles following the onset of CDS trading. Peristiani and Savino (2011) and Danis (2016) find that 5

6 bondholders are less likely to participate in out-of-court debt exchanges with financially distressed firms after the onset of CDS trading. However, Bedendo et al. (2016) examine a large sample of distressed firms facing either out-of-court restructuring or bankruptcy and find no evidence that CDS trading impacts the probability of a firm filing for bankruptcy as opposed to engaging in a debt exchange. Because of empty crediting, firms with CDSs traded on their debt ( CDS firms ) are at a disadvantage when renegotiating bank loans compared to firms without traded CDSs ( non-cds firms ). After the introduction of CDS trading, these borrowers face creditors who may not be welcome to renegotiation of any kind, instead preferring bankruptcy for the defaulted firm so as to trigger the payout from the swap contract. In terms of incentives, bank loan debtors now face the prospect of adhering to ex ante covenants designed to prevent risky behavior lest suffer stiff penalties or the real possibility of entering bankruptcy. As a result, CDS firms should exhibit lower numbers of renegotiations as well as a decreased probability of renegotiating a bank loan. CDS trading should result in the ability of bank lenders to drive a hard bargain in any renegotiation to the extent that renegotiation may be impossible. Using a sample of 101 individual firms of which 22 are CDS firms and the remainder non-cds firms, we find that the beginning of CDS trading is associated with a decrease in the likelihood of bank loan renegotiations. In addition, we find that after the inception of CDS trading, firms are more likely to end bank loans on the stated maturity date and that event durations are shorter in terms of months. Our findings are robust to a variety of robustness tests meant to control for endogeneity and self-selection issues. Our paper contributes to the literature by showing how financial innovations such as CDS contracts affect traditional creditor-debtor relationships. We show how the existence of CDS 6

7 contracts affect the relative bargaining power between borrowers and lenders and ultimately result in fewer loan renegotiations which can be costly to firms that could otherwise restructure their debt out of court. Our paper is closely related to the literature that studies the relation between CDSs and bank loans. For example, Shan, Tang, and Yan (2016) show that after the inception of CDS trading, firms are likely to switch lenders, face higher a cost of bank debt, and increase their reliance on bond debt. Shan, Tang, and Winton (2016) similarly show that the inception of CDS trading influences the structure of bank loans in that new loans post CDS trading tend to have less creditor protection provisions. For example, new loans tend to have fewer net worth requirements and are less likely to be secured. Our paper differs in that we are the first, to the best of our knowledge, to show how CDS trading affects existing creditor-debtor relationships in outstanding loans. 2 Related Literature The issue of CDSs and empty creditors offers new insight into the extant literature on creditor-debtor conflict as well as shareholder-debtholder conflict, which have received substantial prior coverage and discussion. Berlin and Loeys (1988) model the tradeoff of financial contracting with bonds containing covenants versus bank loans with monitoring and conclude that the optimal resides between balancing the restrictiveness and inefficiency of debt covenants with efficient but costly bank monitoring. Likewise, Smith and Warner (1979) argue that there exists a unique optimal set of financial contracts which maximizes the value of the firm (p. 121). They theorize that dividend and financing (e.g. debt issuance or seniority structure) covenants are less costly to monitor while value maximizing for shareholders. By including these covenants in debt agreements, lenders align managerial incentives with their own and mitigate agency problems stemming from shareholders. Rajan and Winton (1995) ask if 7

8 creditors are concerned with control rights why not simply lend on a short-term basis instead of relying on long-term debt contracts replete with covenants. They answer that covenants and collateral help incentivize lenders to acquire information about the borrower through increased monitoring. Beneish and Press (1993) turn to an empirical examination of the effect of accounting covenant violations on firm financing and find these firms face higher refinancing and restructuring costs as well as greater creditor control through the addition of financing and investing covenants. Similarly, Chen and Wei (1993) investigate accounting-based covenant violations but from the lenders side. The authors show that for less risky and lower leveraged firms creditors are more likely to forego punitive action, opting instead to waive the violation as a way to ensure the survivability of the firm. More recently, Chava and Roberts (2008) examine how debt covenant violation affects investment. They show that when a firm breaks a financial covenant (for example, a liquidity ratio), this allows creditors to interfere in managerial decisions such as capital expenditures. As the authors note, following a covenant violation, the threat of accelerating the loan (which may force a firm into bankruptcy) and the ability to extract concessions from the borrower provides lenders with enormous clout in which they can exert influence over a firm s financing decisions (p. 2086). Nini, Smith, and Sufi (2009) collect a novel dataset on the terms of sole-lender and syndicated bank loans by reviewing documentation on private credit agreements for a sample of public firms. They find that a significant portion of their sample contains covenants pertaining to capital expenditures, which become more restrictive as a firm s credit rating declines. Roberts and Sufi (2009) extend these results by examining the frequency, outcomes, and determinative factors of bank loan renegotiation. Using a sub-sample of the Nini, Smith, and Sufi (2009) data, 8

9 the authors demonstrate that the vast majority of bank loans are renegotiated at some point following origination, leading to significant modifications to loan terms such as maturity, amount, and interest rate spread. Furthermore, Roberts and Sufi show that as assets increase and leverage and the cost of equity decrease firms renegotiate more favorable terms with creditors. Roberts (2015) provides further granularity to our understanding of creditor-debtor conflict by analyzing not just bank loan renegotiation but what kind of renegotiation events occur. Consistent with the previous papers, he finds that loans are renegotiated on average 3.5 times per loan (p. 64). Bradley and Roberts (2015) posit that the Agency Theory of Contracts (ATC) may explain the theoretical underpinnings of why debt covenants exist. Under ATC, shareholderdebtholder conflicts may occur when managerial incentives align with shareholders at the cost of debtholders who face agency problems. For example, they note that managers may seek to enrich stockholders (at bondholders expense) by distributing assets in the form of dividends or stock buybacks or modifying the seniority structure to dilute current debtholders (p. 5). Bradley and Roberts argue that covenants help to mitigate these agency costs by incentivizing managers toward the interests of debtholders. Following Gompers et al. (2003) who form a Governance Index to measure restrictions to shareholders rights, Bradley and Roberts (2015) construct a similar index for covenant intensity, which they model as a way to gauge the strictness of covenants over managerial activity. They find that as covenant rigidity increases bond yields decrease, which indicates a tradeoff between the benefits of lower agency costs (for debtholders) and lower borrowing costs (for the firm). Murfin (2012) extends this line of research by constructing a proxy of covenant strictness, which applies a probabilistic approach to measuring covenant violation. Interestingly, 9

10 he examines covenants from the supply-side perspective of creditors to illustrate how credit shocks influence bank lending and finds an increase in covenant strictness. This result is consistent with Gopalan et al. (2011) who show how large bankruptcies affect bank lending to other unrelated firms and Chava and Purnanandam (2011) who demonstrate the impact of the Russian debt crisis on creditor portfolios. Shan and Tang (2013) tie the bank loan & debt covenant literature to credit default swap trading by investigating how CDSs influence the strictness of tangible net worth covenants at the time of loan origination. In fact, the authors find that the beginning of CDS trading is associated with covenant loosening, which they attribute to three possible explanations: 1) given the availability of an outside hedging option via CDSs, creditors may no longer need as rigid control rights; 2) CDSs add to information efficiency by providing an explicit measure of default risk, which may manifest in non-price benefits to borrowers (Ashcraft and Santos, 2009, find no decrease in credit spreads for CDS firms while Saretto and Tookes (2013), argue that non-price benefits include lengthened maturity and increased leverage); and 3) lead syndicate arrangers who can hedge exposure may no longer need to signal their commitment to monitoring of the debtor. In this paper, we similarly examine how CDSs influence bank loans but focus on withinloan renegotiation events compared to features present at initiation. 3 Data & Methodology Quarterly financial and accounting control variables are from Compustat, CDS start dates are from the Bloomberg terminal, macroeconomic variables are from FRED (Federal Reserve Economic Data), syndicated loan characteristics are from Dealscan, and bank loan renegotiation 10

11 data is publicly available from Michael Roberts s website. 34 We drop utilities (SIC ), financials (SIC ), and firms incorporated outside of the U.S. We remove missing, negative or zero value observations for total assets and total long term debt and set missing observations for capital expenditures and acquisition costs to zero. The bank loan data are structured at the loan-level not firm-level and is over events not time. In order to work with the renegotiation data, we sum loan-events into quarterly periods and then merge with the remaining data producing firm-quarter observations. Using the link file provided by Chava and Roberts (2008), we merge Dealscan and Compustat data. As noted previously, this paper is built upon hand-collected data from Michael Roberts (Roberts, 2015), which requires further elaboration here. Roberts selects at random 114 public firms with data available both in Compustat and Dealscan. He then manually examines EDGAR (Electronic Data Gathering, Analysis, and Retrieval) filings for syndicated bank loan details including origination, renegotiation (if any), and terminal events discussed anywhere in the text, footnotes, or management discussion and analysis (MD&A) section. His dataset spans from 1994 to 2011, contains 501 loans (and, therefore, 501 loan paths), and 2,775 loan-event observations including 1,773 renegotiation events (as well as 501 origination and 501 terminal events). Roberts defines a renegotiation event as either amendments (loan modification; 1,354 obs.), amended and restated agreements (loan modification but entailing a new agreement; 303 obs.) or rollovers (new agreement but with same creditor; 116 obs.; see pp ). Interestingly, Roberts s numbers show that on average there are 3.5 renegotiations per loan. While each loan begins with an origination event, Roberts identifies four terminal events that signal the end of the bank loan: maturity (loan ceases on stated maturity date; 229 obs.),

12 termination (loan ends prematurely either because of pre-payment or default; 28 obs.), refinanced (loan is replaced with a new loan from new lenders; 141 obs.), and censored (unavailable data; 103 obs.). Our final dataset consists of 3,451 observations for years , which we chose because of the correspondence with the rise of CDS trading in the early 2000 s while Roberts s data ends in Our sample includes 101 individual firms of which 22 are CDS firms and 79 are non-cds firms. Only 8 firms have no renegotiation events while 93 have one or more. For firms that engage in renegotiations, there is an average of 11.8 events per firm. Of the total 3,451 firm-quarter observations, 2,693 are for non-cds firms and 758 are for CDS firms (392 before and 366 after the inception of CDS trading). Lastly, the sample contains a total of 1,059 renegotiation events (903 for non-cds and 156 for CDS firms). One potential issue with using a small dataset like this one is the likelihood of oversampling a handful of certain industries, which may spuriously drive results. Roberts s data contains 114 randomly selected firms, which reduces to 101 in our sample after removing years prior to CDS trading. Of these remaining firms, there are 37 different industry groups represented by 2-digit SIC code (average component weight of 2.7%), although some industries stand out. For example, chemical and allied product manufacturing firms (2-digit SIC 28) constitute approximately 9.61% of the overall sample. For robustness purposes, we re-run our main results after first removing the top five industries by percentage of total (1,507 obs. or 38.7%) and achieve stronger estimates compared to those reported below. The main dependent variables are Event, Event1, MatEvent, and EventDuration. Event is a non-negative count variable equal to the number of renegotiations in a given quarter. Event1 is 5 We run baseline regressions on the entire sample period provided by Roberts ( ) for robustness and see similar results as for the time period. 12

13 a dummy variable equal to one if a firm engages in one or more renegotiation events in a quarter and zero otherwise. MatEvent is a dummy variable equal to one if the firm has one or more maturity events in a given quarter and zero otherwise. EventDuration is loan renegotiation event duration in months and represents the time between events (for example, from origination to amendment or from the second amendment to the third). Event and EventDuration are constructed when collapsing Roberts s data from high frequency loan-events to firm-quarters by summing variables while Event1 and MatEvent are formed into indicator variables after collapsing the bank loan data. The main independent variable of interest is CDS_Active, which is a dummy variable equal to one when the CDS contract begins trading on the firm s debt and thereafter. For non- CDS firms, CDS_Active always equals zero. We require that CDS firms have at least one observation before the onset of CDS trading and one after. CDS_Firm is an indicator variable equal to one if the firm has a CDS traded at any point in the sample period. In prior literature, CDS_Firm has been used much like a fixed effect to control for time invariant unobservable differences between CDS and non-cds firms (see Ashcraft and Santos (2009) and Saretto and Tookes (2013)). However, this usage of CDS_Firm is at odds with what we know about the effects of CDS trading, which are time varying. For example, Subrahmanyam et al. (2014b) show that credit risk increases following the beginning of CDS trading and Donato (2017) finds that firms concentrate debt structure after the onset CDS trading. Furthermore, CDS_Active and CDS_Firm variables are highly correlated (appox. 0.63), which presents a problem with multicollinearity. Because of these issues, we drop CDS_Firm from the regression models, although we use the variable for sub-sampling the dataset into CDS and non-cds firms as well 13

14 as propensity score matching. Indeed, if CDS trading is driven by unique firm characteristics, then matching techniques should reveal this effect. 6 We also include variables to control for borrower characteristics. Tangibility is total net property, plant, and equipment scaled by total assets, which serves as a proxy for bankruptcy costs (Titman and Wessels (1988)). MktBk is total assets minus the sum of book value of equity and deferred taxes and investment tax credit plus market capitalization all divided by total assets. Zscore is defined as Altman s Z-Score. lngdp_real is the log of real U.S. gross domestic product. Profitability is operating income before depreciation scaled by total assets. CFvol (cash flow volatility) is the twelve month rolling quarterly standard deviation of Profitability. CurrRatio is current assets divided by total assets. All continuous controls for borrower characteristics are winsorized at the 1% and 99% levels. All models include Fama-French 48 industry and year fixed effects. Finally, we include controls for loan characteristics: loan type (term loan, revolver, 364-day facility, other) and loan purpose (corporate purposes, debt repayment, leveraged buyout/management buyout, takeover, working capital, commercial paper backup, acquisition line, other) variables, which are non-negative count variables formed after collapsing (by summing) syndicated bank loan data merged into the dataset from Dealscan. 4 Empirical Results In this section, we outline the empirical results. Section 4.1 provides univariate analysis of the dataset, Section 4.2 details the baseline results, Section 4.3 covers propensity score 6 We don t use a difference-in-differences (DID) approach as an alternative method to gauge the treatment effect of CDS trading because of the endogeneity of CDS trading with bank loan renegotiation. Additionally, DID relies on a single defining event with distinct pre- and post-periods for both treatment and control groups, which is not the case with this dataset. CDS_Active corresponds to the onset of CDS trading for different firms at different times; as such, while we have pre- and post-periods for the treatment group, the control group of non-cds firms lacks a before and after period for comparison. Of course, DID can be constructed using matching techniques. However, in Section 4.3, we prefer to use propensity score matching to create a counter-factual non-cds firm control group to measure the average treatment effect for the treated. 14

15 matching results, Section 4.4 offers extensions to the baseline results, and Section 4.5 outlines the effects of CDS trading on bank loans and the cost of borrowing. 4.1 Univariate Analysis Table I outlines descriptive statistics for variables used in the paper. Of the 3,451 firmquarter observations, 22% pertain to firms with CDS contracts traded on their debt at some point in the data while 10.6% correspond to observations occurring after the inception of CDS trading for these firms. 15.3% of the sample are observations where there is one or more renegotiation events in a given quarter per firm. On average, there are renegotiation events for reiteration, defined as amendments (loan modification), amended and restated agreements (loan modification but entailing a new agreement) and rollovers (new agreement but with same creditor). In the sub-sample of renegotiation events only, 1.7 events occur on average per quarter. [Table I] Untabulated pairwise correlations for CDS_Active, Event, Event1, MatEvent, and EventDuration variables show, as expected, coefficients that are negative for correlations between CDS_Active and Event ( ), Event1 ( ), and EventDuration ( ) while Event and Event1 are statistically significant at the 1% level. Intuition suggests that MatEvent should exhibit a positive correlation (since CDSs act as a commitment device compelling firms to adhere to the original stated maturity date in the bank loan agreement), although the results show a small, statistically insignificant negative coefficient. In Chart I, we display the average number of renegotiation events per year by CDS and non-cds firm. As noted above, the average number of events for the entire dataset is Over the time series for the sample, however, the number varies considerably from in

16 to in In the aftermath of the 2008 financial crisis, the average number of events drops from in 2007 to in 2009, which follows a longer-term decline from the beginning of the sample period. For robustness purposes, we run the baseline results on these periods separately and obtain similar results to regressions performed on the entire time period. Additionally, we construct an indicator variable for the crisis and interact it with our main variable of interest, CDS_Active, but do not obtain statistical significance on the interaction term (see Tables A1 and A2 in the Appendix). Although Chart I indicates a decline in events especially around the crisis, the observed effect does not appear to translate empirically in regression form. [Chart I] We further divide the dataset between CDS and non-cds firms in Chart I. On average, non-cds firms have more bank loan renegotiation events than CDS firms; for example, in 2001, non-cds firms had events versus for CDS firms. However, these differences are not statistically significant for all years; in fact, only for years 2000, 2004, and 2006, is the difference significant at the 5% level and at the 1% level in year 2007 while all other years are not statistically significant. Non-CDS firms have a greater number of events on average but in many years (for example, 2002, 2003, and 2010) the split appears to be evenly divided. Our data also include information on covenant changes. Per renegotiation event, we can observe whether an event involves a change to a covenant; however, not all renegotiation events entail covenant changes. Of the total 1,059 events in our sample, 866 are associated with changes to covenants. There are five categories of covenants identified: accounting ratios (e.g. liquidity or leverage ratios), distribution (e.g. dividends or stock buybacks), investment (e.g. capital expenditures), collateral (e.g. secured debt), and financing (e.g. debt issuance or changes 16

17 to debt seniority structure). Changes to accounting ratios (37.1%) followed by investment and collateral (17.7% and 12.1%, respectively) constitute the bulk of covenant changes. Table II illustrates covenant changes by these five categories for CDS and non-cds firms with paired t- tests. On average, CDS firms exhibit less covenant changes than non-cds firms. There is a statistically significant difference at the 10% level for accounting ratio and distribution covenants and significance at the 5% level for changes to investment covenants between the two sets of firms. We also include covenant changes by category for the sub-sample of CDS firms; however, given the small sample size (97 covenant changes out of 156 total events or 62.2%), it is not surprising that the results lack statistical significance with the exception of accounting ratio and collateral covenants, which show an increase in changes (significant at the 10% level) following the inception of CDS trading. [Table II] We further explore characteristics of CDS firms and how they differ from non-cds firms. Table III sub-samples the data into the two sets of firms and provides descriptive statistics for each. Out of 3,451 total firm-quarter observations, 2,693 observations correspond to non- CDS firms and 758 for CDS firms. With the exception of Zscore, paired t-tests show a statistically significant difference and most at the 1% level. On average, CDS firms exhibit a lower number of renegotiation events, are larger in terms of real book assets ($6.2 versus $1 billion), have greater cash flow volatility (1.7% versus 1.2%), and have a greater percentage of tangible assets (39.8% versus 27.6%) than non-cds firms. Consistent with Subrahmanyam et al. (2014a) who show that CDS firms use increased cash holdings to mitigate rollover and refinancing risk, the sample reveals higher Cash for CDS firms relative to non-cds ones (9.2% versus 6.9%); however, our sample also indicates lower CurrRatio (1.84 versus 1.99) and 17

18 IntCoverRatio (13.4% versus 20.4%). As noted, Zscore is not statistically different between the two sub-samples. But Zscore is significantly different for CDS firms pre- and post-cds trading. Prior to the beginning of CDS trading, these firms exhibit a z-score of 2.41 compared to 2.05 after CDS trading commences, which represents a statistically significant difference of 0.36 at the 5% level. The decrease in z-score indicates a higher probability of bankruptcy, which is consistent with the intuition of Bolton and Oehmke (2011) and the empirics of Subrahmanyam et al. (2014b). Lastly, Table III suggests that the results may be driven by differing firm characteristics between CDS and non-cds firms and not the advent of CDS trading. This potential problem motivates the reasoning for why we explore propensity score matching in Section 4.3 as a means to hold firm characteristics constant while isolating the treatment effect of CDS_Active. [Table III] 4.2 Baseline Results Following the inception of CDS trading on firm debt, intuition suggests that the probability of renegotiating a bank loan should decrease given the empty creditor problem discussed above. Table IV provides support for this claim. We run probit models for Event1, which is a dummy variable equal to one if a firm engages in one or more renegotiation events in a quarter and zero otherwise. All five regression results reveal a statistically significant (at the 5% level for models 1 4 and 1% level for model 5) average marginal effect of approximately - 5% on CDS_Active, which suggests that after the onset of CDS trading the probability of a renegotiation event decreases a result robust to various specifications. Pseudo R-squared s range from 7.9% (model 1) to 9.8% (model 5). We include lngdp_real as a macroeconomic variable to control for fluctuations in the business cycle that may affect bank loans. Tangibility, 18

19 Zscore, and CFvol are proxies for bankruptcy risk (see Titman and Wessels (1988)). Profitability and MktBk are included as measures of operating cash flow and market value, respectively, and CurrRatio proxies for firm liquidity. [Table IV] We run Poisson regressions in Table V in order to test the effect of CDS trading on the number of renegotiation events in a given quarter, modeled as a non-negative count variable, Event. Consistent with the previous table s results, Table V provides evidence that CDS trading is associated with decreased bank loan renegotiation. All five coefficients on CDS_Active indicate a negative and statistically significant (at the 10% level) relationship with Event with pseudo R-squared s ranging from 10.3% (model 1) to 13.9% (model 5). Exponentiating the coefficient on model 5 provides a value of 0.7, which suggests that firms with active CDS contracts have 0.7 times the number of renegotiation events compared with the index group of non-cds firms (and CDS firms that don t yet have an active traded contract). In short, following the inception of CDS trading on firm debt, the number of renegotiation events decreases. [Table V] In Table VI, we extend the results from the previous table by scaling Event by the number of outstanding bank loans per firm-quarter. The reason for this change is that the number of events per quarter is measured at the firm level so firms with more outstanding loans will naturally have more renegotiations. Since CDS firms tend to be larger they are also likely to have more loans so, if anything, our prior results should bias us against finding support for our hypothesis. However, after scaling the dependent variable, we show consistent estimates with 19

20 Table V. CDS_Active is statistically significant at the 5% level throughout each specification albeit with small adjusted R-squareds. [Table VI] 4.3 Propensity Score Matching (PSM) Which firms become CDS firms is, of course, not governed by random selection; instead, these firms are selected based on various characteristics. This inherent selection bias can be examined and controlled through propensity score matching. PSM offers the ability to create a counter-factual sample of CDS firms if they had never had an active CDS contract in the first place. By matching on determinants of borrower characteristics, we generate matches that are similar to the sample of CDS firms in terms of firm characteristics but are not subject to the treatment effect (i.e. CDS trading). By comparing treated (CDS) with untreated (non-cds) firms, we test for the average treatment effect on the treated (ATET), which isolates the effect of CDS trading on bank loan renegotiation. [Table VII] In Table VII, we present summary results for our PSM methodology, which are based on similar approaches by Subrahmanyam et al. (2014b) and Uzmanoglu (2015). Model 1 details the regression used to estimate propensity scores. We regress CDS_Active on lngdp_real, Tangibility, Zscore, Profitability, CFvol, MktBk, CurrRatio, industry and year fixed effects, and loan type and loan purpose controls as a way to predict the probability of CDS trading. Column 2 displays marginal effects on borrower characteristics from model 1. Zscore, MktBk, and CurrRatio are statistically significant at the 1% level while Profitability is significant at the 5% level with a model goodness-of-fit of 40.2%. 20

21 Models 3 6 represent regressions on the matched sample, which we obtain by matching CDS firms with non-cds firms based on the propensity scores estimated in model 1. In models 3 and 4, we re-run the regressions from Table IV on the matched sample, which tests the onset of CDS trading on the probability of a renegotiation event. Compared to those previous results, these estimates are statistically stronger (significant at the 1% level) with larger marginal effects (-8.78% and -7.5% for Table VII, models 3 and 4 versus -5.27% and -4.89% for Table IV, models 1 and 5); likewise, pseudo R-squareds are larger as well (12.9% and 18.3% versus 7.9% and 9.8%). In Table VII, models 5 and 6, I re-run the Poisson regressions from Table V on the matched sample. As noted above, Event is a non-negative count variable, which we use to test the effect of the inception of CDS trading on the number of bank loan renegotiation events. The results in these models are considerably stronger in terms of economic and statistical significance. After exponentiating the coefficients, the estimates show a partial effect of approximately 0.59 compared to 0.70 for models 1 and 5 in Table V, which suggests that firms have 0.59 times the number of events following the beginning of CDS trading. Also, pseudo R- squareds are larger as well (14.6% and 21.5% versus 10.3% and 13.9%). In summary, we introduce PSM methodology in this section as a means to mitigate concerns that firm or borrower characteristics drive our baseline results shown in Section 4.2. As discussed previously, CDS and non-cds firms differ considerably for a range of variables as evidenced by Table III. However, after matching on these attributes via propensity scores, Table VII offers support for the argument that CDS trading impacts both the probability of bank loan renegotiation events as well as the number even after controlling for firm characteristics. The ATET indicates an economically and statistically significant difference in the number of events 21

22 between CDS and non-cds firms. Theoretically, if an unobservable variable (such as a confounding borrower characteristic not controlled for in the regression models) were to be driving the results instead of the treatment effect of CDS trading, then we would expect that that latent variable should also affect the number of renegotiations of the non-cds matched firms; after all, the treated and untreated matched observations should be very similar based on propensity score. Instead, the findings suggest that the inception of CDS trading is the probable cause of the decrease in loan renegotiation events due to the empty-creditor problem weakening the debtor s negotiating capacity. Although regression on matched samples is often used in the literature, Abadie and Imbens (2016) argue that the standard errors for these models are not only biased but biased in an uncertain direction. In effect, this method occurs in two stages: first, an estimation stage through propensity scores; and, second, regression on the matched sample. The authors argue that the standard errors for the second stage do not account for the initial estimation phase and, thus, are incorrect. For robustness purposes, we use Stata s teffects nnmatch command set at 1:1 matching and match CDS and non-cds firms on the same variables discussed above. We start with a sample consisting of 3,152 firm-quarter observations and obtain a matched sample using non-cds observations with replacement in order to calculate the difference in means between the two groups for variable Event. The difference of is statistically significant at the 1% level with Abadie-Imbens robust standard errors (0.0655), which suggests that the beginning of CDS trading is the reason for the decrease in bank loan renegotiations even after matching on various borrower characteristics. 22

23 4.4 Extensions to Baseline Results As an extension of the above analysis, we construct CumEvent, which is a lead (1, 2, 3, 4, and 5) variable representing the cumulative count of renegotiations per firm. By doing so, we are testing whether the effect of CDS trading on renegotiation persists over time or dissipates. Intuition suggests that the onset of CDS trading acts as a permanent shock by which the creditordebtor relationship is altered in favor of bank lenders, leading to a lower number of loan renegotiations. Table VIII presents the results from regressing CumEvent on CDS_Active and controls using Poisson models similar to those in the previous table. Interestingly, the coefficients show a consistent negative effect in the number of loan renegotiations not declining or increasing in a significant way, which argues in favor of a permanent effect on firm bargaining power with creditors resulting from the inception of CDS trading. In exponentiated form, the coefficients on CDS_Active range from (lead 1) to (lead 5), which indicate that on average firms with a credit default swap traded on their debt engage in approximately 0.6 times the number of loan renegotiations compared with the index group on a cumulative basis. Although the pseudo- R-squared declines from model 1 at 44.8% (one quarter forward) to model 5 at 41.3% (five quarters forward), the betas remain statistically significant at the 1% level. [Table VIII] As mentioned in the introduction, CDSs fundamentally alter the lender-borrower relationship in favor of the creditor. Although bank lenders retain economic rights to principal and interest, contractual control rights to monitor the firm, and legal rights to pursue litigation against defaulted borrowers, they may simply be disinterested (i.e. empty creditors ) or incentivized to allow bankruptcy to progress following a default on a bank loan. Prior to the advent of CDS trading, creditors may have instead chosen to agree to an out-of-court 23

24 restructuring of debt as a way to avoid liquidation and losses in bankruptcy court; following the beginning of CDS trading, the opposite may be true: empty creditor banks may see no benefit to such deals. As a result, one might expect to see a higher probability of bankruptcy for CDS firms since their negotiating leverage declines (Subrahmanyam et al., 2014b). However, because banks hedged with CDSs have the upper-hand in any bank loan renegotiation and can drive a tougher deal with debtors, swap contracts may act as a commitment device whereby borrowers are locked in to the terms of bank loan agreement and face dismal prospects of successful renegotiation as Bolton and Oehmke (2011) theorized. Our results provide evidence in support of this line of reasoning. As the above intuition suggests, bank loan renegotiation decreases after the inception of CDS trading and continues to do so on a cumulative basis as evidenced by the estimates in Table VIII. 4.5 Effects of CDS Trading on Bank Loans and the Cost of Borrowing Given the Bolton and Oehmke (2011) framework of CDSs as commitment devices, we expect to see a higher probability of maturity terminal events than instances of either termination or refinancing/replacement. A CDS firm that terminates a bank loan early because of default faces the higher prospect of entering bankruptcy given the empty creditor problem while a firm seeking to refinance and replace an existing loan may face bank lenders offering tougher deals with more restrictive covenants. In other words, following the inception of CDS trading, firms become committed to the originally stated terms of the loan agreement and are, therefore, more likely to continue through to the initial maturity date. In Table IX, we test whether the onset of CDS trading impacts the probability of a maturity terminal event. MatEvent is a dummy variable equal to one if the firm has one or more maturity events in a given quarter and zero otherwise. We run probit regressions on the sub- 24

25 sample of terminal events (excluding censored), which leaves a dataset of only 111 observations. Our results indicate that the probability of a firm with an active CDS contract engaging in a maturity event increases considerably an average marginal effect ranging from 38.37% to 44.67%, which is both economically and statistically significant. These regression estimates appear to show evidence of the effect of CDS trading incentivizing debtors to commit to loan agreements. However, given the small sample size (n = 111) and large number of right-hand side variables (over one hundred including controls and fixed effects), these results are obtained with few degrees of freedom and, as such, should be taken with some measure of caution. [Table IX] We also examine how CDS trading affects the duration of bank loan renegotiations by constructing the variable, EventDuration, which is discussed in the Data & Methodology section. EventDuration represents the duration of loan renegotiation events (in months) for the time between events (for example, from the second amendment to the third or from origination to the first renegotiation event). EventDuration is constructed by collapsing Roberts s data from high frequency loan-events to firm-quarters by summing the variable per firm per quarter. Intuitively, we expect that the beginning of CDS trading leads to shorter duration times because of the higher probability of bankruptcy associated with empty crediting. We argue that these firms seek to avoid prolonged renegotiation as a way to lessen the likelihood of entering bankruptcy and facing the prospect of liquidation. Table X s results show that following the inception of CDS trading the average duration of a renegotiation event decreases by approximately two months. Admittedly, the adjusted R-squareds are low for each model, although the coefficients on CDS_Active are significant at the 5% level. However, Table X 25

26 supports the argument that CDS trading is associated with decreased event duration due to the increased probability of bankruptcy. [Table X] Finally, Table XI illustrates the effect of CDS trading on the cost of borrowing for firms. I construct the variable, lnallindrawn, which is the natural logarithm of the all-in-drawn spread from Dealscan. This measure of loan pricing represents the spread over a benchmark (e.g. LIBOR, the prime rate, etc.) plus any additional fees paid to bank loan syndicate members and is on the drawn portion of credit. 7 In the table, models 1 and 2 are based on the full sample, models 3 and 4 on the CDS sub-sample, and models 5 and 6 on the full dataset conditioned on a loan renegotiation event occurring. I also re-run all results on the loan start date only but do not obtain results, which is most likely due to the small sample size and lack of degrees of freedom. [Table XI] In the prior CDS literature, Ashcraft and Santos (2009) demonstrate that the onset of CDS trading has no discernible effect on corporate borrowing costs in terms of bond spreads. Saretto and Tookes (2013) show the same result but provide evidence that firms benefit in terms of non-price measures such as an increase in debt maturity and leverage. Shan and Tang (2013) examine how the inception of CDS trading prior to loan origination affects net worth covenants. They find that these covenants are less restrictive when a CDS contract begins trading ex ante. In Table XI, our results indicate that the beginning of CDS trading is associated with a decrease in borrowing costs for firms in terms of the all-in-drawn spread. For models 1 4, the regression estimates are statistically significant at the 1% level with adjusted R-squareds ranging from 60.3% to 81.2% for the full sample and CDS sub-sample, respectively. In models 5 and 6 conditioned on Event, we find negative coefficients, which are significant at the 10% level. 7 For a discussion of how fees affect bank loans, please see Berg et al. (2016). 26