impact of CDS contracts on the Reference Entity

Transcription

1 The effect of Credit Default Swaps trading on the bond market: impact of CDS contracts on the Reference Entity Irma Smaili Prof. Dr. N. Nicola Tilburg University March 18, 2018 Abstract There have been claims made by previous authors about credit derivatives. Some suggest, for example, that credit default swaps (CDSs) have lowered the cost of debt financing to reference firms by creating information for fixed-income investors. This paper examines the impact of CDS trading of the reference entities on relating bond spreads and credit risk. The trading of CDS negatively affects the cost of debt financing for the average borrower. This result is strongest during the peak of global financial crisis. In the pre-crisis period bond issuers with a traded CDS experienced a drop in the cost of debt. Next, I find that risky and opaque firms have benefit from a small reduction in bond spreads at issue. Furthermore, previous studies have shown that CDS trading reduces credit risk. However, I do not find evidence to support this claim. More safe bonds without a CDS trading have a higher credit rating compared to CDS trading investment grade bonds. On the other hand, risky bonds with a CDS contract in place have a much lower credit rating than their non-cds trading counterpart. The difference in credit ratings is small for investment grade bonds, and significantly large for highyield bond issues. This suggests that CDS trading is not beneficial to the reference firm, unlike previous studies claim. In fact, if trading of a CDS contract takes place, reference entities that are already considered more risky seem to experience a deterioration of their perceived credit risk.

2

3 Contents 1 Introduction Contribution and main hypotheses Data & Research Methodology Results Literature Review Institutional Details CDS market development Benefits of CDS trading: Cost of debt Cost of CDS trading: Bankruptcy Risk Contributing to the existing literature Research Question and Hypotheses 11 4 Data Bond Data CDS Data Fundamentals Data Descriptive Statistics Conceptual Framework Main variables explained Methodology and Empirical Results Credit Spread Changes Caused by the CDS Market Credit Spreads varying across Time Credit Spread Changes on types of Bond Issuers Credit Risk Sensitivity Caused by the CDS Market Credit Risk Sensitivity varying across Time Conclusion & suggestions for Further Research 33 2

4 1 Introduction In this study I attempt to investigate the empirical linkages between the CDS and bond markets in US during the period January 1997 to December I address the following three questions. First, what is the impact of CDS trading on the bonds market in terms of issuance costs? I am here observing how the bond spread changes for issued bonds with a traded CDS. Second, did the impact of CDS trading on the bond market show new characteristics during the financial crisis? Since my data covers the global financial crisis, I am able to observe how strong the effect of CDS trading is on the bond market. Third, which subset of bond issuers is most likely to benefit from the trading in the CDS markets? I distinguish my sample in investment-grade bonds and high yield bonds. Here, I compare the results between these two and to non-cds traded types. As last, does having a CDS traded on a bond affect reference entities credit risk and which type benefits the most from it? A Credit Default Swap is portrayed as an insurance contract that provides the buyer with a notional payment amount made by the seller in case of a default or any other certain credit event written in the contract of an underlying reference entity. Similar as in insurances, the buyer needs to pay premium for the protection seller offers and the buyer will be compensated for the loss incurred resulting from the credit event. CDSs were originally developed to help banks transfer credit risk maintain relationship with their borrowers and develop new businesses. This market allows for better risk allocation by allowing lenders to transfer credit risk to those who are willing to bear it. During the years CDSs became popular in the United States. Midway through 2007 the notional value surged to $58,- trillion. In this period the financial turmoil began and there was a significant large amount of CDS contracts dropping affected by regulators. 1.1 Contribution and main hypotheses I aim to explore the relation between benefits and costs of CDS trading followed by the suggestion of Subrahmanyam et al. (2014). They propose to research the overall impact of CDS trading on allocative efficiency by examining the potential benefits and credit risk vulnerability. Moreover, they claim that the overall effect of CDS trading on the reference entity credit risk depends on the trade-off of the costs and benefits.with this study, I attempt to explore their relationship and fill the gaps in the literature. CDSs are characterized for their damaging contribution to the financial crises and certain types were banned for their destabilizing nature, although this view is not supported by empirical research (International Monetary Fund, 2013). Despite this, few studies proved that CDS trading has beneficial effects on many aspects of the reference entities. These benefits include a drop in the cost of issuing bonds for firms raising funding, an increase in credit supply and debt maturity. The last two have already been examined and each study proved that the results remain robust. However, Ashcraft and Santos (2009) were the first one who tried to investigate whether a drop in the cost of bond spread does exist. Unfortunately, they did not find this evidence and claim that an average firm did not benefit from CDS trading in terms of cost of bond. A possible explanation for this could be due to the short period they have used in their sample ( ). Hence, I can improve this model by extending the time frame from 1997 to According to Ashcraft and Santos (2009) a reduction in cost of debt may occur through a price mechanism, which transfers the information by the CDS market. This theory is based on the findings of Longstaff et al. (2004), Norden and Weber (2004), and Blanco et al. (2005) who all find that CDS market plays a more important role in the price discovery process than the bond market. The information premium investors demand on firms bond can be measured by the bond spread, which is the difference between yield to maturity at 3

5 the issue date of a corporate bond and the yield to maturity of a U.S. treasury bond with a similar maturity. Therefore, Ashcraft and Santos (2009) suggest that this information role of the CDS market could lead to a reduction in the cost of issuing debt by reducing the information premium investors demand on firms bonds (bond spread). Based on this, I am aiming to find a significant effect on the bond spreads arising from CDS trading on the issued bond. Since the development of CDS market provides banks the ability to hedge or lay off their risk exposures to firm, I apply the same logic of banks as lenders to fixed income investors, and test my main hypothesis whether CDS firms pay a lower credit spread on their bonds. Hereby, I differentiate from the authors by (1) obtaining a greater sample period, (2) include variables tangibility and dividends, and (3) compare results between CDS and non-cds bonds, while they examine results based on the prior- and post onset of CDS trading. After testing the benefit of a possible reduction in bond spreads, I analyse cost of CDS trading of reference entities. For this purpose I observe whether CDS trading has a significant effect on credit rating, which simultaneously impacts the credit risk. In line with Saretto and Tookes (2013), who find that CDS trading firm s bankruptcy risk increases, Subrahmanyam, et al. (2014) confirms that reference firms credit risk increases after the onset of CDS trading resulting from a down-grading. Therefore, I test the next hypothesis whether credit risk increases for issued bonds with a CDS trading. Besides these two hypotheses, I observe whether a different pattern occurs during the global crisis and which subset of bond issuer benefit from CDS trading. To test these hypotheses, the variables that sets the foundation of my model are "TRADING" and "CDS_FIRM", which are both binary variables and used to control for fundamental differences between CDS firms and non-cds firms. Additionally, I also control for a set of firm-specific variables, bond features and other market variables. 1.2 Data & Research Methodology My sample exists of 1168 reference entities from Russell 3000 (RAY), who have in total 7334 bond issued during the period Of these firms 691 had a CDS trading during at any point in time during the sample period. The data that consists of bonds that were issued during the period 1997 to 2016 is retrieved from Thomson One banker. Next, I obtain single-name CDS data from Bloomberg Terminal (Credit Market Analysis). From this source I retrieve monthly CDS contracts of 5 years, because these are the most liquid ones and I need to determine a starting date. I use RAY as my leading sample and separate these firms into two samples just like authors Ashcraft and Santos (2009) and Subrahmanyam, et al. (2014) did. RAY is an index that tracks the performance of the largest US traded stocks that represent about 98% of all US equity securities. One sample exists of firms with a traded CDS and other with non-traded CDS. The period starts from 1997 to 2016, however RAY firms with a traded CDS had starting dates between 2001 to This means that either there were no CDS contracts issued on RAY firms in and or I have eliminated them due to several missing points when matching BB CDS Tickers with their fundamentals. Other annual financial data on the firms and S&P500 market returns were obtained from the Centre of Research in Security Prices (CRSP) and Compustat. My sample contains a vast majority of investment-grade bonds. Most of these bonds have credit ratings such as A+, A, A-, BBB+ and BBB, but also few high yield bonds. Overall, firms in my sample are of relative good credit quality at the given time, as measured by their credit ratings. All of these variables are winsorized at 1% to reduce the effect of spurious outliers. My research contains two parts: (1) benefit arising from CDS trading and (2) cost of CDS trading. First, I examine the effect of CDS trading on cost of debt by looking at the ex-ante credit spreads that firms pay to issue in the bond market. Here I deal with a repeated cross- 4

6 sections data, because I have bonds in my sample that are issued in different time periods. These bonds are issued either once or few times a year by a firm, so the observational unit is more than one time period and varies across different time periods.second, I study whether credit risk may get affected arising from CDS trading. This is a panel data, because the information that has been collected is from the same RAY firms varying over a period of time. Since the credit spreads on bonds may vary across firms, I estimate my models with a pooled OLS and firm fixed effects. Simultaneously, I use robust SE and cluster these by firm to correct for correlation across obervations of a given firm. Additionally, I use several controls for S&P numerical value, industry- and time FE. 1.3 Results In my attempt to formulate a trade-off between the cost and benefits of CDS trading in the bond market, I find in line with Ashcraft and Santos (2009) that there is no significant effect on an average firm benefitting from a reduced bond spread arising from CDS trading. Ergo, I have been unable to conclusively prove the theory/advice suggested by Subrahamanyam et al. (2014). Therefore, I changed the whole view by examining how CDS trading affects reference entities. According to my findings, the average firm does not benefit from a reduction on the credit spreads it has to offer investors in the primary market in order to raise bond financing once CDS trading starts. Despite those claims made by the economists and authors about the benefits of credit derivatives, the average corporate borrower has not yet seen a lower cost of debt capital. When I take a simple OLS regression and do not control for any FE, there is a small significant positive effect of 0.19% on bond spreads. This means that in general fixed income investors require to get more compensated for firms that raise bond financing with CDS trading in comparison to non-cds firms. When I split my sample in two periods (pre- and post-crisis), I observe that all issued bonds in the post-crisis were charged with a higher bond spread. However, an average firm with a traded CDS was charged relatively more with 0.35%, while for non-cds bonds 0.28% during this period, all else equal. When I control for firm FE, this effect even doubles. In other words, CDS traded bonds seem to more negatively impact by paying a higher credit spread than non-cds traded bonds. In other words, CDS traded bonds are negatively impacted due to a higher credit spread than non-cds traded bonds. One can here infer that before the crisis, investors did not react to CDSs on reference entities, because the economy was performing well. When the crisis hit off, CDSs were the ones to blame for causing crisis. Therefore, an average firm with a traded CDS had to pay more when issuing their bonds compared to non-cds firms. Next, when I interact CDS trading with several years to observe changes during my timeframe, I find that bond spreads for CDS traded bonds were influenced by the crisis period. While at both the peak and the end of crisis, the bond issuers with CDSs were charged with 1.39% (2008) and 2.03% (2009) more compared to non-cds traded bonds, all else equal. Before and after the crisis, a drop in the cost of bond spread occurred. These results remain robust even after controlling for firm FE. More interestingly, the types of firms that one expects would naturally benefit the most, which is consistent with prior findings safe and transparent firms. It appears to be adversely affected by the CDS market, since risky and opaque firms have benefitted from a small reduction in bond spreads at issue. These results appear to be explained by the monitoring channel. As Datta et al. (1999) points out riskier firms should benefit more from the monitoring process, since they are riskier and need to be monitored more closely. I find that high yield bonds pay 0.48% less than investment-grade bonds, all else equal. What appears to be even more surprisingly is that non-cds traded high-yield bonds were facing 14.76% on 5

7 bonds spreads, all else equal. Therefore, high yield bonds certainly benefit from CDS trading. An explanation for this might be that fixed income investors believe banks monitor active riskier firms with high yields bonds and prevent them from defaulting. For this reason, they do not charge a higher premium on bond spreads like for other non-cds traded high yield bonds. Additionally, I fail to find evidence for an average firms credit risk arising from CDS trading. Instead, I find an incremental positive change on the credit rating of CDS traded bonds. Who are the winners and losers from CDS trading on their credit risk? I find that non- CDS traded safe bonds receive a greater upgrade and hence, their credit risk improves while for CDS traded safe bonds the change is extremely small (0.49 rating points in comparison to 3.65 ratings). In this case, safe bonds are better off without a CDS contract. However, riskier CDS traded bonds experience a greater downgrate and its credit risk deteriorates. Lastly, I did not find evidence of a stronger negative effect of CDS trading on the credit rating after the start of financial crisis. Instead, I found a positive incremental change contributing to credit rating. However, this is less than 1 and does not impact the credit risk of reference entities. In addition to this introduction, this paper is organised as follows. Section 2 provides a brief literature review while in section 3 I discuss the main hypothesis and testable propositions. In section 4 I provide a description of my sample data and explain the main variables used in my model. Section 5 shows the emperical results and finally section 6 concludes. 2 Literature Review 2.1 Institutional Details A credit default swap (CDS) is a derivative 1 insurance contract that transfers risk from one party to another. It functions as an insurance for the lender against the risk of default by the borrower on its debt obligation and are traded over-the-counter. The borrower is called the reference entity whose debt is referenced by CDS contracts and this can be either a private- or public company. In return for sellers protection, the buyer makes regular 2 payments (premium), so called credit spread. This can be seen as the cost of hedging against the credit risk of the underlying firm. In the contract the legal definition of a credit event 3 is outlined and in case this occurs, the borrower is compensated for the loss incurred as a result of the credit event. When this occurs, the seller pays the principal amount of the bond (or a similar payment 4 depending on the precise contract details) to the buyer of the CDS. If the reference entity does not default during the agreed upon timeframe, the borrower loses its premium claims. Figure 1 demonstrates the framework of CDS concept. The purpose of a CDS is either to mitigate or hedge the credit risk by buying single-name CDS (debt of reference entity), or a multi-name CDS index (portfolio of reference entities). In this study borrower represents a reference entity where they have CDS contracts traded on them, whereas the creditor refers to a lender/bondholder. A CDS can be seen as an insurance 1 A financial derivative is a contract of an underlying assets such as currencies, commodities, stocks or bonds. 2 Expressed as a percentage of the face value of the debt against which insurance is bought. 3 If a reference entity fails to meet its payment obligations in case of a bankruptcy, repudiation or moratorium (for sovereign entities), material adverse restructuring of debt, acceleration or default obligation. 4 Default swap can be settled at par against physical delivery, of a reference asset (physical settlement) or the notional amount minus the post-default market value of the reference asset determined by a dealer poll (cash settlement). 6

8 contract against the default of the reference entity. They can easily be compared with each other, but there are key differences between them. Although buyers of insurance contracts are required to hold an insurable interest, buyers of CDSs are not. Hence, it is common for the total notional principal of a reference entity s CDS to exceed the value of that entity s outstanding debt. Furthermore, CDS contracts are traded on the secondary market, whereas insurance contracts cannot be traded. Therefore, this market allows for better risk allocation by allowing lenders to transfer credit risk to those most willing to bear it. In other words, corporate lenders do not need to hold credit risk associated with these loans, while financial institutions increase their credit risk if desired without funding the underlying loan. Figure 1. Framework of CDS contracts 2.2 CDS market development It all started when J.P. Morgan was lending money to Exxon Mobile in the aftermath of the Exxon Valdez oil spill in 1994 (Tett 2009). They bought protection from the European Bank for Reconstruction and Development against the default of Exxon Mobile. They introduced CDS as a plain villa in 1994, while the International Swaps and Derivatives Association presented a standardized contract in 1998 (Hull, Predescu & White, 2004). From there on, CDSs became popular. Before the financial turmoil that started mid-2007, the use of CDSs as an instrument to trade credit risk increased exponentially. However, since 2008 there is a significant downward trend in the number of CDS traded contracts and this reflects the strictly constrained credit markets. The notional amount of CDS was estimated $29 trillion in 2006 and this amount surged to $58 trillion in the second half of 2007 with a gross market value of 2 trillion. However, nine years later, this amount of outstanding CDS contracts dropped to 11.2 trillion, while the gross market value of CDS dropped to $292 billion (BIC, 2016). Figure 2 exhibits a time line of the CDS market size. CDS contracts have been appointed as being among the main causes of the US subprime crisis in , and have worsened the financial turmoil of 2008 and the Eurozone sovereign debt crisis in George Soros published an article in the WSJ 5 that CDS are toxic instruments and need to be strictly regulated. To support George s view and policy makers, President Obama signed the Dodd-Frank Wall Street Reform and Consumer Protection act in 2010, while the European Union banned certain sovereign CDS in 2012 due to their destabilizing influence 5 One Way to Stop Bear Raids CDS need much stricter regulations. Published in The Wall Street Journal (March 24, 2009). 7

9 on financial markets. The role of Securities Exchange Commission (SEC) and Commodity Futures Trade Commission (CFTC) is enhanced to regulate derivatives trading. Figure 2. Global CDS markets: Notional amounts outstanding. Data is retrieved from Bank for International Settlements. These amounts are estimates of the actual size of the credit derivatives market. As CDSs are OTC, data providers use different methods of estimation of the market size. BIS uses triennial surveys of national banks globally as data source. 2.3 Benefits of CDS trading: Cost of debt There are two approaches in the literature. One studies both the bond and CDS markets and compares them by looking at their active roles in price discovery. The other one approach focuses on the bond market and studies whether bond market development is affected by the introduction of CDS market. In this study, I adopt the second approach and leave the first untouched. Previous empirical work on the impact of CDS trading on bond market is limited. Empirical findings suggest that CDS have a leading role in price discoveries, but there is no arbitrage relationship between these two markets in the long run (Hull et al. (2004), Zhu (2006) and Blanco et al. (2005). Moreover, their contribution plays a more important role in the price discovery process than the bond market (Norden and Weber (2004), and Blanco et al. (2005)). Based on a theoretical perspective, we assume that the bond market is inefficient, because the liquidity is low, has restrictions in funding and short sale, and there exists asymmetric information between borrowers and lenders. Therefore, CDS market can mitigate or worsen some aspects of market inefficiencies in the bond market. In terms of benefits, CDS trading may (a) lower the cost of bond issuance, (b) improve liquidity in the bond market and (c) reveal new information about firms. Duffie (2008) argues that CDS increases liquidity of credit markets, lowers credit risk premia and offers hedging opportunities to investors. CDS contracts contain features that could eventually lower the cost of debt to the borrower. This instrument creates new hedging opportunities and information that was before CDS contract not even available to fixed income investors. Liquidity is very limited in the corporate bond market, because many investors hold their bonds until maturity, making it costly to trade large amounts of credit risk in that market (Alexander, Edwards and Ferri 1998). Even though the secondary market for loans has grown rapidly, bank loans still remain illiquid (Kamstra, Roberts and Shao (2006). Due to these circumstances, CDS market provided banks and fixed income investors with a new and cheap 8

10 way to hedge or pass the credit exposure (Billet, Flannery and Garfinkel (1995)). Duffie (2007) explains that reference entities give their banks and bondholders opportunities to diversify their credit exposures. These opportunities provide banks savings and efficient use of capital, while passing on to borrowers. In this case, reference entities with a CDS could indeed borrow from banks and issue bonds at lower interest rate. Empirical papers tried to study the different channel through which CDS may affect the bond market. Hirtle (2007) studies the relationship between credit derivatives and credit supply. She was aiming to understand if banks use these instruments to reduce the risk of their lending activities, or whether the diversification is illuminated by the increase of supply of credit. The sample existed of 250 U.S. banks between and a panel regression is used to measure this relationship. In addition to this study, the relationship between credit derivative trading activity, debt maturity and credit spreads is also researched. The results indicated that banks indeed increase credit supply after taking CDS and that the maturity increase while spreads decreased once the bank hedges away the risk of credit default. Regarding the information channel for banks, many empirical studies have evidence that the CDS market is a source of information on firms. Norden and Wagner (2008) find that changes in CDS spreads explain about 25% of following monthly changes in loan spreads for syndicated loans to US corporates in the period of 2000 to The study most related to my research is from Ashcraft and Santos (2009). They test whether the impact of onset of CDS trading impacts the credit spreads at bond issuance and loan origination in the US. According to them, a reduction in cost of debt can occur through a price mechanism, which transfers the information by the CDS market. In addition, they find two channels that reduce loan credit spread. The first one is a diversification/ hedging channel, which refers to a situation in where CDS firms offer their creditors more opportunities to hedge their risk exposures resulting in lower spreads when issuing subsequent bonds. Meaning, CDS offers hedging opportunities that lead to diversification of risk. As a result, borrower may issue bond at a lower interest rate. The second channel refers to the possibility of CDS revealing new information about firms. The secondary bond market is a poor source of information on firm, since the liquidity in the market is extremely limited due to a reduced number of loans being traded in the market. However, CDSs are seen as a good proxy to measure spreads that investors require to take on certain credit risk. CDS spreads could for that reason reduce information premium that investors require, especially for riskier firms. They find that an average non-financial firm has not benefited from CDS trading in terms of cost of bond, only in term of reduced interest rate spreads on loans. They use two approaches to come to their findings. The first approach identifies a sample of firms that become traded in the CDS market, referred to the traded sample, and compare the average bond spread before and after the CDS trading. So they exploit differences in timing of the CDS contract on entities. The latter approach identifies a sample of firms that are never traded but have similar characteristics to those that do, referred to matched sample, and compare the traded firms with the matched firms. Furthermore, they find a reduction in cost of debt for safe transparent firms. On the other hand, riskier and opaque firms have a slight increased cost of debt. A reason for this can be that syndicate participants may demand higher compensation to extend loans to these firms, because banks will rely on CDS and do less monitoring of these firms. This effect is driven by reduced incentives for banks to monitor risky entities. This reduced incentive to monitor them is due to CDS being used by the bank, because this reduces bank s exposure to its borrowers. Therefore, anticipating this effect, new creditors who would free-ride on monitoring of banks, may demand higher compensation for riskier and opaque firms. These results mentioned complement prior findings of Datta et al. (1999) where they examine whether the existence of a bank/firm relationship lowers the cost of public debt financing. By using a sample of first public straight debt offers, they test the 9

11 monitoring effect of bank debt. They find that firms having a loan at the bank while issuing bonds, experience benefits from bank monitoring. These firms pay a lower credit spread on their bonds, because investors assume that monitoring from the bank diminishes the risk of default. 2.4 Cost of CDS trading: Bankruptcy Risk CDS are often perceived as side bets that do not impact the fundamental value of underlying asset (Subrahmanyam, et al, 2014). CDS contracts are traded OTC by financial institutions and bank creditors of the reference entities. As a result, if creditors selectively trade CDS linked to their borrowers, this can distort the creditor and borrower relationship and affect the borrowers credit risk, which determines the CDS payoff. More investors feel secure with a CDS, since their risk is hedged by a CDS. Therefore, creditors may increase supply of credit to the underlying entities. Such improved access to capital may increase borrowers financial flexibility and resilience to financial distress. Hence, lenders incentive to monitor their borrowers may decrease and reduce credit supply frictions (Saretto and Tookes, 2013). When these firms are not monitored, they can take riskier projects than if they were monitored more closely. For this reason, the use of CDS may have an impact on the credit risk of the reference firms. Credit risk of a firm increases when they tend to be vulnerable during financial distress. This vulnerability results from CDS protected creditors, who seem to be tougher during debt renegotiations and refusing debt workouts. Hu and Black (2008) refer these CDS protected creditors as empty creditors, since they maintain negotiating rights. The empty creditor of a distressed firm who holds a CDS contract can be better if its total payoff (including CDS payments) would be larger in the default event. When the distressed reference entity approaches the creditor suggesting a restructuring on the debt, the creditor might decline this offer. If the firm defaults under the credit events specified in the CDS contract and restructuring is not included, only bankruptcy will ensure that the creditor receives the notional amount back. Bolton and Oehmke (2011) model the empty creditor problem by extending the research of Hu and Black (2008). They propose a two-period model in which a firm undertakes a project financed with debt. If credit event occurs at a time period 1, the creditor can choose (a) to receive its insurance payment, or (b) it can renegotiate debt contract. If the proceeds from the renegotiation were larger than the notional amount stated in the CDS contract, they would choose for the latter option. Consistent with the empty creditor theory, conditioning on financial distress, firms with CDS are more likely to file for bankruptcy. Furthermore, the findings show that CDS contracts make the creditor-borrower relationship more efficient by enhancing the borrowers more committed. This occurs due to the increasing bargaining power of the creditor, which decreases the chance borrowers choose for a strategic default. However, creditors tend to be over-insured when it comes down to the case if they can determine freely to what extend their credit portfolio is insured. As a result, debt renegotiations rarely succeed and borrowers are more likely to default. Moreover, Hull et al. (2004) and Norden and Weber (2004) find strong evidence that the CDS market anticipates credit rating announcements, particularly negative rating events. This is confirmed by Subrahmanyam et al. (2014) who focuses solely on the credit risk of the reference entity. In addition to this, I am partially replicating his model. Subrahmanyam et al. (2014) builds on the results of Bolton and Oehmke (2011) and the empty creditor problem introduced by Hu and Black (2008). They analyse if CDS trading indeed is a cause for a higher chance of bankruptcy for firms with a traded CDS. Subrahmanyam et al. (2014) say that creditors have no incentive to monitor the debtors once they have hedged their exposure. The creditors incentive to monitor their borrowers is reduced, because they do not bear the 10

12 credit risk anymore. Hence, this triggers the borrowing firms to take on riskier projects than if they were monitored more closely. This may alter the firms value and increases the chance of default. To test their hypotheses, they retrieve sample data of bankruptcies for the period An event study was conducted to estimate the impact of in the inception (starting date) of a traded CDS on a firms credit rating, as rating downgrades is the first step to bankruptcy. Furthermore, they use a proportional hazard model to estimate what the marginal bankruptcy probabilities and ratings downgrades are. The model consists of primary determinants of default risk and two dummy variables e.g. if the traded CDS was introduced for this firm equals 1 and another one that indicates if a firm has a traded CDS anytime in the sample period. These two approaches were first introduced in the models of Saretto and Tookes (2013) and Ashcraft and Santos (2009). With their findings, Subrahmanyam et al. (2014) proves that it is very likely that the risk of bankruptcy rises followed by a credit downgrade for firms after a traded CDS is introduced, which according to Merton (1974) in turn is driven by asset volatility and leverage. They even found out that the bankruptcy chances are more than two times higher after a CDS introduction. To examine whether credit spreads develop new characteristics during the crisis, only one paper investigated the impact of CDS trading on the bond market after the originators Ashcraft and Santos (2009). Shim and Zhu (2014) study the Asian bond market and find that CDS trading lowered the cost of issuing bonds and enhanced the liquidity in the Asian bond market. The positive impact is stronger for smaller firms, non-financial firms and those firms with higher liquidity in the CDS market. They also find that there is a strong evidence at the peak of global crisis, where CDS firms included in CDS indices faced higher bond spreads than those who did were not included. 2.5 Contributing to the existing literature This study builds upon the recent empirical findings and aims at helping to increase the understanding of the effects of CDS trading. It may be interesting for policymakers and investors to understand the overall effect of CDS trading to the reference entities. CDSs are characterized for their damaging contribution to the financial crises and certain types were banned for their destabilizing nature. Despite this, studies proved that CDS trading has beneficial effects on few aspects of reference entities. These benefits include a rise in credit supply, a drop in borrowing costs and higher debt maturity. Yet, this comes at the cost of a potential increase in bankruptcy risk. Following-up the advice of Subrahmanyam et al (2014) for future work, they suggest to research the overall impact of CDS trading on allocative efficiency by examining the potential benefits and bankruptcy vulnerability. Therefore, the overall effect of CDS trading on the reference entity credit risk depends on the trade-off of the costs and benefits. With this study, I am aiming to explore their relationship and fill the gap in the literature. 3 Research Question and Hypotheses Subrahmanyam et al. (2014) propose to research the overall impact of CDS trading on allocative efficiency by examining the potential benefits and credit risk vulnerability. The overall effect of CDS trading on the reference entity credit risk depends on the trade-off of the costs and benefits. Therefore, I propose the following research question: What is the effect of CDS trading on the reference entity? My main hypothesis tests whether CDS firms pay a lower cost of debt, which is estimated by the credit spreads on the bonds. Authors argue that the development of the CDS market could lead to a reduction in the cost of debt by the feature of new information revealing on firms. Based on this theory, Ashcraft 11

13 and Santos (2009) were exploring this relationship but did not find a statistical significant effect on the bond spread. A possible explanation for not finding this relation refers to their short sample period starting from 2001 to If CDS trading plays an important role for banks as lenders, then why should not the same logic apply to fixed income investors? The development of CDS market provided banks and investors the ability to hedge or lay off their risk exposures to firms. Reference entities give their banks and bondholders extra opportunities to diversify their credit exposures (Duffie (2007)). Some of these savings arising from these diversification opportunities could be passed on to borrowers. In this case, firms with CDSs could indeed be able to issue bonds at a lower credit spread when raising funding in the corporate bond. Hypothesis 1: CDS firms pay a lower credit spread on their bonds. Another question of interest is whether the impact of CDS trading varies over time, especially during the crisis period. Disruptive events in the derivatives market may spill over to bond market, hence causing the whole credit market malfunctions. Financial markets experienced great disruptions during the financial crisis. Credit spreads across all asset classes and rating categories extended to unknown levels. 6 Information flows from CDS market to the equity market during times of financial stress (Acharya and Johnson (2005)). According to Shim and Zhu (2014) there is a strong evidence that Asian firms included in CDS indices faced higher bond spreads than those who were not included. Since the bond spread works as an information premium investor s demand on firms bonds, during financial stress is it logic to assume that they want to get more compensated for bearing extra risk. Based on this, I am going to test my second hypothesis discussing whether CDS trading has a stronger positive effect on the credit spreads during financial stress. Hypothesis 2: The positive effect of CDS trading on the credit spread is stronger in time-periods after the start of the financial crisis. Ashcraft and Santos (2009) found that investment-grade credit rating experience a reduction in the cost of debt issue relative to high-yield firms. As these are safe and transparent firms, an explanation could be in bank monitoring with both bank and public debt as examined before by Datta et al. (1999). Firms that issue bonds, but also have debt from banks on their balance sheets, experience benefits from bank monitoring. As investors expect that the monitoring of banks reduce the risk of going bankrupted, these firms pay a lower credit spread on their bonds. Besides, as supported by the research of Ashcraft and Santos (2008), riskier and opaque firms experience a slight increase in the credit spread. Datta et al. (1999) points out that riskier firms benefit more from the monitoring process, since they are riskier and need to be monitored closely. Nevertheless, banks use CDS market to hedge the risk of loans, reducing the need to monitor these risky firms. The incentives from banks to monitor riskier firms reduces, because banks rely solely on CDS and do less monitoring on these firms. Therefore, investors may demand higher compensation to extend loans to these firms, resulting in a higher credit spread for high yield bonds. Hypothesis 3: High yield bonds pay a greater credit spread on their bonds. Since the benefits of CDS trading are explained and will be tested, CDS trading comes also at the cost of a potential increase in credit risk. The second stage examines what CDS trading does to the credit risk of reference entities. In line with Saretto and Tookes (2013), who find that CDS trading firm s bankruptcy risk increases, Subrahmanyam, et al. (2014) confirms that reference firms credit risk increases after the onset of CDS trading. 6 Few examples are that the swap spread in the interest rate market became negative, while Bai and Collin-Dufresne (2011) prove that in credit markets the CDS-bond basis turned negative. 12

14 Hypothesis 4: Credit risk increases for issued bonds with a CDS trading. Ashcraft and Santos (2009) found that the beneficial effects of a traded CDS was more for investment-grade firms. Subrahmanyam et al. (2014) did not include in their study which subunit was strongly affected. They only find that larger firms and firms with higher stock returns were less likely downgraded. Here, I examine how the credit ratings change between investment-grade and high yield bonds. Investment-grade bonds are considered as safer, transparent firms and bear minimal risk. Hypothesis 5: Investment graded bonds benefit the most of CDS trading by receiving an upgrade in their credit rating. The global financial crisis that occurred during my sample period offers a good case study to examine the behavior of CDS and bond markets under distress and their linkages. Therefore, I am going to examine whether issued bonds having a CDS traded received a stronger downgrade during the crisis. I examine here if issued bonds having a CDS traded during the crisis received a stronger (possible) downgrade after the start of the financial crisis in comparison to non-cds bonds. After all, firms were more vulnerable during the crisis and especially when having a traded CDS on them. To see how credit rating changes, I create several time year dummies and interact them with CDS trading. Hypothesis 6: The negative effect of CDS trading on the credit rating is stronger in time-periods after the start of the financial crisis. 4 Data To study the effect of CDS trading on bond spreads and credit risk, I use three types of data: (1) bond, (2) CDS spreads and (3) firms fundamentals. All financial firms with SICcodes from are excluded, because these are regulated differently. They cannot be compared to non-financial firms, because I need to split the raw samle in two different samples having same characteristics. When I match the bond dataset with RAY firms, this results in a total of 7344 bond observations. Of the matched bond sample, 2777 observations had Bloomberg tickers available. Of these firms 691 had a CDS trading during at any point in time during the sample period. Table 1 demonstrates how I constructed my sample. Table 1. Sample construction Bond sample - Thomson One Banker # of observations North U.S. firms 35,624 Less dupicates and financial firms 18,904 Number of bonds firms 16,720 CDS sample - Credit Market Analysis RAY observations matched with bond data 7334 RAY observations matched with Bloomberg Tickers 2777 RAY unique firms 1158 RAY bonds having a traded CDS

15 4.1 Bond Data This data consists of all US bonds that were issued during the period from 1997 to I retrieve this from Thomson One banker, which exists of a comprehensive database on bonds. Besides issued bonds in this database, I also obtain information about credit spreads, the issuing date, maturity and the size of the principal amount, the type of bond and whether the bond has a put, call or a sinking fund. The raw bond data exists of 35,624 observations, where 18,904 firms are name duplicates and financial firms. After deleting these, it comes down to 16,720 observations remaining. 4.2 CDS Data Many researchers use Markit as their database to retrieve CDS information. Datastream offers Markit data as well, but the data is only available from 2007 onward. Therefore, I obtain CDS data from Bloomberg Terminal: Credit Market Analysis. Data from CMA is said to lead price discovery and is seen as a reliable source of CDS information (Mayordomo, Peña, and Schwartz, 2014). I use Russell 3000 (RAY) as my leading sample since it contains firms with necessary firm size for controlling my sample. The purpose of this is to split my sample in two groups, one with traded CDSs and other with non-traded CDSs. The pricing source is CBIN and all monthly bid prices of CDS contracts were retrieved. I used monthly database to find out the onset date of CDS trading. The period starts from 1997 to 2017, however RAY firms had a CDS traded from 2001 onwards with 2002 as the majority of interceptions dates. When matching the US bond data with RAY firms, there remain 7334 bonds observed. From here, I match Bloomberg tickers resulting in 2777 bonds (1158 RAY firms had a CDS). Figure 4 shows how many CDSs were issued for certain years. This shows that my sample contains only CDS contracts from 2001 to In the following years up to 2016, here were no 5-year CDS contracts found on the bonds issued by RAY firms. Figure 3. Distribution of the interception CDS trading 4.3 Fundamentals Data Lastly, the financial data on the firms and S&P500 market returns is obtained from the Centre of Research in Security Prices (CRSP) and Compustat. These fundamentals can only be downloaded annually and I maintain the period 1997 to With a text file including permno and cusip RAY firms, I upload all the necessary fundamental characteristics such as total assets, power plant and equipment (PPE), book value, total liabilities, R&D, dividends, sales, EBITDA, net income/loss, share price and shares outstanding. Additionally, to proxy for the state of economy, the monthly returns and volatility of the S&P500 index is gathered. 14

16 4.4 Descriptive Statistics Table 3 below reports the descriptive statistics of all my variables used in the model. All of these variables are winsorized at 1% to reduce outliers. Hence, I find that my data set contains low kurtosis, which have light tails and represent few outliers. The majority of my variables have a skewness near zero, representing a normal distribution. The average volume on credit spread is 229 basis points, having the lowest at 0.28 basis points and the highest at 884 basis points. Of the total number of bond issues, on average 35% of these bond issuers were firms that had a CDS trading at any point during the sample period. Moreover, 9% of these bonds had a CDS trading on that time. Additionally, of all the bond issues there was on average 80% who paid out a dividend. RAY firms contains large firms and based on my CDS traded sample the average size is a 12. This indicates that on average, issued bonds have a credit rating of BBB. The vast majority of these bonds are specified as investment-grade bonds. Most of these bonds contain credit ratings such as A+, A, A-, BBB+ and BBB, but also fewer non-investment-grade bonds. Overall, firms in my sample are of relatively good credit quality at the given time, as measured by their credit ratings. Lastly, the S&P volatility during my sample period was 36%, while the market return was 43%. In the appendix a list of these categories is found, with two histograms that show the distribution of credit ratings and credit spreads. Table 2. Descriptive statistics of full sample. Credit spreads are mentioned in basis points. All of my firms control variables are ratios and measured in percentages. Sales and size of firms are in millions and I took natural logarithm. CDS trading, CDS firms and dividends are binomial variables. Variables Mean SD Med 95th % Min Max Skew Kurtosis Credit Spreads Trading CDS firms Sales Cashflow Profitmargin Leverage R&D Credit rating Market to book Collateral Tangability Size Dividends Market return Market Volatility Conceptual Framework To maintain a helicopter view on this research, I have created a framework that can guide me. Figure 4 demonstrates the concept about this framework. It shows how my key elements are related to each other including prior literature on them. The framework starts with CDS TRADING, referring to firms that have a CDS. There are two directions that a reference entity can obtain from having a CDS traded on them. The first creates a benefit on 15

17 the CREDIT SPREAD 7 and increases the chance of obtaining a cheaper debt by examining whether the issuing bond spreads got reduced thanks to CDS trading. The second displays CREDIT RISK, which refers to credit risk measured by credit agencies reflecting either in a down or high grade change due to the CDS trading. Besides examining these effects, I also look at the results before/after and during the crisis to contribute to the debate about their impact on the financial turmoil and which subset of bond type is impacted the most from CDS trading. The negative and plus sign indicate what the likely coefficients are, which represent the effects arising from CDS trading. Figure 4. Conceptual framework Main variables explained CDS- trading and firms CDS trading sets the foundation of my research and is a dummy variable. With this variable I can test the effects of CDS trading on the firms cost of debt and the credit risk vulnerability. Replicated from Ashcraft and Santos (2009) model, TRADING takes the value 1 if at time t (the issue date of the bond) a traded CDS was available. This will tell whether the credit spreads on the bonds issued arising from CDS firms have an effect on the bond spread. The construction of this variable is as follows. I create two columns where the first one provides start date of a traded CDS as t and the second is the last date traded (maturity 5 years as fixed) as t+5. Next to these, I put a column of bond issue date and check whether CDS trading occurred during that period of issuing. Based on these results, a column is created saying if trading happened during that period of time, result in number 1 or 0 otherwise. While Ashcraft and Santos (2009) and Subrahmanyam et al. (2014) check for the onset of CDS trading and compare their results with post trading, I focus on what the influence of CDS trading is on my dependent variables such as bond spread and credit risk. The latter TRADING variable is used to measure the relation between credit risk sensitivity (based on credit ratings) arising from CDS trading followed from Subrahmanyam (2014) model. It has the same meaning as described above. Since my sample exists of 7 The difference in yield between a US Treasury bond and a debt security that are the same in all aspects except for the credit rating. Debt issued by the US Treasury is used as the benchmark in the financial industry due to its risk-free status, being backed by the full faith and credit of the government. As the default risk of the issuer increases, its spread widens Investopedia. 16

18 members from RAY firms, I separate a sample of firms that had during my sample period a traded CDS in the market, referred to as the CDS_FIRM with a sample of firms that are never traded referred to as NON-CDS firms. Similar as how Ashcraft and Santos (2008) and Subrahmanyam et al. (2014) use their matches sample (non-cds firms), I use them as my control group. Crisis Variable CRISIS is a dummy variable and represents the financial crisis period. Nordon, Buston and Wagner (2014) use 2007 as the starting year of the crisis in their analysis of bank behavior in the presence of credit derivative. After the failure of Lehman Brothers on September 2008, the global crisis intensified. Therefore, I took the peak crisis period as the year This dummy variable takes value 1 if a traded CDS was during the financial crisis and 0 otherwise for pre- and post-crisis. Investment grade In order to examine which of the subset bond issuers has a greater impact of the CDS trading, I compare investment grade bonds to high yield bonds. Variable IGRADE will be an interaction term with TRADING. According to S&P credit rating, an investment grade bonds is higher than BBB-. Therefore, the dummy variable will be equal to 1 when the credit rating is above BBB- during any point in time a traded CDS was available and 0 for high yield bonds. CDS Spread Benefits Following the model of Ashcraft and Santos (2009), this variable CREDIT SPREAD is measured as the difference between the yield to maturity at the issue date of a bond and the yield to maturity of a US treasury bond with a similar maturity. It is regressed on the TRADING and CDS_FIRMS variables with a set of control variables that influence the credit spread. To control for market conditions, they include market return and market volatility. Furthermore, to control for certain bond characteristics, they control for callable and putable bonds, bond maturity and size amount. For controlling firm characteristics, they include leverage, R&D, credit rating, dividends, cash flow, tangibility of assets, collateral and firm size. Credit risk CREDIT RISK is measured based on S&P credit ratings. Here I analyse whether the effect resulting either in a positive or negative point is affected due to CDS trading. I converted S&P credit ratings to a numerical value (AAA=20, AA=19, A=18... C=1). 5 Methodology and Empirical Results This section characterizes the impact of CDS trading on the bond spreads and credit risk vulnerability. There are two stages where I examine the impact of CDS trading on the reference entity. The first stage focuses on the potential benefit that arise from CDS trading by looking at the cost of debt. I start my investigation of the effect of CDS trading on the cost of debt by looking at the ex-ante credit spreads that firms pay to issue in the bond market. Here I deal with a repreated cross-sections data, because my sample contains bonds that are issued at the same point of time varying across years. For instance, I have several bonds that were issued by RAY firms few times in a year. Meaning, the observational unit is more than one time period and varies across different time periods. Due to this type of data, I need to estimate my models with a pooled OLS regression. Next, the second stage studies the potential credit risk that arises from CDS trading. This is a panel data 8, because 8 Panel data differs from pooled cross section data across time, because it deals with the observations on the same subjects in different times whereas the latter observes different subjects in different time periods. 17

19 the information that has been collected is from the same RAY firms varying over a period of time. Additionally, I control for several fixed effects such as the issue year, industry and firm characteristics. Since I have converted the credit ratings into a nummerical value, I also have to control for these as well to ensure that they do not bias the results. As last, I estimate my models with robust standard errors and cluster these to correct for correlation across obervations of a given firm. 5.1 Credit Spread Changes Caused by the CDS Market My tests to investigate the effect of CDS trading on ex ante bond credit spreads builds on the model of Ashcraft and Santos (2008). For each bond that the firm has issued during my sample period, equation 1 will be regressed using robust standard errors. Bondspread it = c + β 1 T rading + β 2 CDSfirms + β 3 X it 1 + β 4 Y B it + β 5 Z it 1 + ε it (1) Bond spread represents credit spread (over treasury with same maturity of the bond) and Trading is a dummy variable that takes 1 if a traded CDS was available and CDSfirms controls for fundamental differences between CDS firms and non-cds firms. It has a value of 1 if at any point in time a traded CDS was available for this firm. I investigate whether the onset of CDS trading lowers bond ex ante credit spreads controlling for a set of firm-specific variable as Xit-1, a set of bond features as YBi and a set of other market variables as Zit-1. They are unrelated to the firm or bond characteristics, but vary over time and are likely to affect bond credit spreads. Discussing these controls next, Xit-1 exists of Sales (log of firms sales) which measures the risk of firm s overall risk. Furthermore, Profit Margin (net income over sales), Leverage (book value of debt over TA), firms Credit Rating provided by S&P (converted as AAA=20, AA=19, A=18 etc.), R&D (firms expenses with R&D over sales), and Market to Book (share price over book value). Following Binsbergen et al. (2010), I also add in my model Tangibility (tangible assets divided by TA), and Dividends (dummy that equals 1 if firm pay dividends at t-1 and 0 otherwise). These are lagged by one month calculated prior to the issue of the bond. These variables represent the standard measures of debt costs that are extensively used in the literature (Frank and Goyal (2009)). I include them in my model because profitable firms are better in refinancing their debt and I expect these to pay lower credit spreads, while riskier firms with high leverage are more likely to default on their debt. Therefore, I expect these variables to have a positive sign on ex ante credit spreads. To control for the quality of asset base that lenders can draw on in default, R&D expenses and tangibility is added. Myers (1977) argues that the opportunity cost of debt for growth firms is high because debt can restrict a firms ability to exercise future growth opportunities due to debt overhang. My next set of controls is YBi, which captures bond features that are likely to affect credit spreads. This set includes Issue Amount (log of issue amount), Bond Maturity (log of the bond maturity in years) and a set of dummy variables to identify Callable and Putable bond. Larger issuing amounts and longer maturities have uncertain effect on ex ante credit spreads. These bonds represent more credit risk, demanding higher yield, but also have economies of scale. Callable bonds will carry higher spreads, because they can be called back prior to maturity. Putable bonds are suitable to bondholders, so I expect these bonds to carry lower credit spreads. These funds are more often created by riskier issuers and may carry higher credit spreads. My final set of controls, Zit-1, controls for the variation in the overall economy that may influence credit spreads. Collin-Dufresne et al. (2001) proves that MVolatility (market volatility to control for market turbulence) and MReturn (market return to proxy for the state of economy) affect credit spreads. Market volatility was estimated as a change from the standard deviation of daily returns of the S&P500 index. Market returns 18

20 were calculated from the S&P500 index. These are also lagged by one month calculated prior to the issue of the bond. Table 3. CDS trading on bond spreads. Results are in basis points and based on model 1. The dependent variable is BOND SPREAD: Ex ante credit spread over Treasury with the same maturity of the bond. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS_FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. POST CRISIS: Dummy variable that takes value 1 for period SALES: Log of sales. PROFITMARGIN: Net income over sales. LEVERAGE: Total debt over assets. R&D: Research and development expenses over sales. CREDIT RATING: Equals the number associated with the firm credit and controlled for credit rating. MARKET TO BOOK: Price per share over equity book value. TANGIBILITY: Tangible assets over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. PRINCIPAL AMOUNT: log of issue amount. BOND MATURITY: log of the bond maturity in years. CALLABLE: variable that is equal to 1 for bonds having callable. PUTABLE: variable that is equal to 1 for bonds having putable. Controlling for market turbulence and proxy for economy state: MARKET VOL and MARKET RET. CDS trading x Post Crisis: Interaction variable with CDS trading for period Robust standard errors in parentheses. Statistically significant: *** p<0.01, ** p<0.05, * p<0.1 19

21 Variables Model 1-A Model 1-B Post Crisis -A Post Crisis -B Trading 18.58** * *** (7.595) (10.86) (13.71) (14.18) CDS firms ** 32.82*** (5.529) (11.56) (6.089) (7.640) Post Crisis 27.51*** 35.43*** (6.142) (7.586) Trading x Post Crisis 35.35** 70.37** (15.20) (16.83) Sales *** (3.449) (8.128) (5.456) (63.65) Profitmargin * (6.719) (14.13) (7.505) (386.3) Leverage * (13.86) (28.94) (21.26) (378.3) R&D ** ** -1,034** (50.81) (96.58) (100.7) (455.7) Market to Book * (0.033) (0.241) (0.276) (26.35) Tangibility 45.89*** *** ** (11.14) (28.50) (23.13) (443.8) Dividends *** *** (6.525) (13.40) (8.150) (243.9) Principal Amount 6.236** ** (3.063) (6.508) (3.725) (99.25) Bont Maturity * ** 905.7** (3.553) (4.010) (5.041) (411.6) Callable *** *** *** (5.495) (6.639) (6.503) (148.5) Putable (44.39) (34.23) (55.14) Market Return * *** (0.0306) (0.0411) (0.0347) (0.711) Market Volatility 2.712*** 1.437*** 2.046*** 33.38** (0.257) (0.471) (0.292) (15.32) Credit ratings *** (0.875) Constant 615.7*** 786.8*** 786.1*** -3,231* (26.43) (58.04) (68.07) (1,761) Observations 3,294 3,294 3,294 1,957 R-Squared Industry FE NO YES YES NO Year FE NO YES NO YES Credit rating FE NO YES YES YES Firm FE NO NO NO YES 20

22 Table 3 shows the results of my multivariate analysis on the effect of CDS trading on the bond credit spreads. I start with a simple pooled OLS model on the full sample, where I find a statistical significant positive effect at 5% level of 0.19%. Model l-a shows that bond spreads rise with a relative small effect of 0.19% if there is CDS trading on these bonds, all else equal. This contradicts my main hypothesis and I do not find evidence to support that an average CDS traded firm issues at a lower credit spread. Additionally, I add two columns in table 4 where I show how the spreads developed in the post-crisis 2007 to 2016 and name these post crisis. Both show a positive statistical significant effect at 5% of the interaction of CDS trading during this period. This means that bond spreads with a traded CDS faced an increase of 0.35%, all else equal. Even for non-cds traded bonds during this period had to issue at a higher credit spread at 0.28%, all else equal. Moreover. When I control for firm FE, the results become even stronger. The coefficient of the interaction between CDS trading and Post Crisis doubles. This means that bond spreads increase with 0.70% for CDS traded bonds in the post crisis time, all else equal. Non-CDS traded bonds do also pay a higher credit spread at 0.35% all else equal. It shows that CDS traded bonds are negatively impacted due to a higher credit spread than non-cds traded bonds. When I control for firm fixed effects, the variable Putable gets ommited due to perfect collinearity. A graphical representation of the model is provided in figure 5 below. This two-way interaction plots the effect between CDS trading in the post crisis. To test for this two-way interaction, I first generated a dummy variable for the period and regressed the interaction variables between CDS trading and this period. This two-way graph exhibits for pre-crisis a downward slope, while after the crisis the slope becomes positive. Results show that CDS trading had a relative impact on the bond spreads influenced by the crisis period. The result makes economic sense, because fixed income investors reacted while being influenced by the crisis. One can infer that before the crisis, investors did not react to CDSs on reference entities, because the economy was performing well. When the crisis hit off, CDSs were the ones to blame for causing crisis. Therefore, an average firm with a traded CDS had to pay more when issuing their bonds compared to non-cds firms. Before the crisis, CDS bonds could issue for lower bond spread, while after the crisis they have to pay more. With respect to the firm controls that I use in these regressions, those that are statistically significant are contrary with the discussions given in the Methodology section. The only consistent are the profitable firms that are able to pay out dividends pay lower bond spreads. Figure 5. Slope movement of the interaction between CDS trading on bonds spreads split into preand post-crisis. 21

23 5.2 Credit Spreads varying across Time To test whether CDS trading has a stronger positive effect on the credit spreads during the financial crisis, I introduce CRISIS variable, which represents a dummy variables and takes value 1 for crisis and zero otherwise. After the summer of 2008, the crisis that started in 2007 as a subprime crisis and developed into a liquidity crisis or credit crunch, became global financial crisis. After the failure of Lehman Brothers on September 2008, the global crisis intensified. Therefore, I took the peak crisis period as the year To test this theory, I create an interaction variable between CRISIS with TRADING and check what kind of effect this has on the bond spread. To see how credit spreads develop, I create time year dummies and interact them with CDS trading. Bondspread it = c + β 1 T rading + β 2 CDSfirms it + β 3 Crisis it + β 4 Crisis it T rading it + β 5 X it 1 + β 6 Y B it + β 7 Z it 1 + ε it (2) Table 4a. CDS trading interacted with YEARS on bond spreads. Results are in basis points and based on model 2. The dependent variable is BOND SPREAD: Ex ante credit spread over Treasury with the same maturity of the bond. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS_FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. CDS trading x YEAR: Interaction variable with CDS trading for a year. Log of sales. PROFITMARGIN: Net income over sales. LEVERAGE: Total debt over assets. R&D: Research and development expenses over sales. CREDIT RATING: Equals the number associated with the firm credit and controlled for credit rating. MARKET TO BOOK: Price per share over equity book value. TANGIBILITY: Tangible assets over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. PRINCIPAL AMOUNT: log of issue amount. BOND MATURITY: log of the bond maturity in years. CALLABLE: variable that is equal to 1 for bonds having callable. PUTABLE: variable that is equal to 1 for bonds having putable. Controlling for market turbulence and proxy for economy state: MARKET VOL and MARKET RET. Robust standard errors in parentheses. Statistically significant: *** p<0.01, ** p<0.05, * p<0.1 22

24 Model 2: CDS trading interacted with years on credit bonds Variables 2013Y 2012Y 2011Y 2010Y 2009Y 2008Y 2007Y 2006Y Trading 11.08* 11.21* 11.31* 11.12* 10.63* 10.80* 11.06* 10.74* (6.254) (6.253) (6.250) (6.253) (6.198) (6.221) (6.239) (6.235) CDS firms (6.678) (6.681) (6.687) (6.681) (6.504) (6.655) (6.677) (6.688) Trading x *** (19.01) Trading x ** (16.30) Trading x (12.92) Trading x (17.18) Trading x *** (30.00) Trading x *** (17.89) Trading x *** (12.34) Trading x *** (11.78) Sales (5.482) (5.481) (5.485) (5.483) (5.373) (5.432) (5.474) (5.465) Profitmargin (10.78) (10.75) (10.76) (10.73) (10.56) (11.04) (10.65) (10.87) Leverage (25.91) (25.91) (25.94) (25.93) (25.88) (25.86) (25.98) (25.99) R&D *** *** *** *** *** ** *** *** (90.53) (90.89) (90.63) (90.73) (90.39) (90.82) (90.46) (90.59) Market to Book (0.327) (0.328) (0.328) (0.328) (0.330) (0.326) (0.328) (0.329) Tangibility ** ** ** ** ** *** ** ** (27.83) (27.79) (27.82) (27.80) (27.67) (27.82) (27.76) (27.77) Dividends (10.68) (10.68) (10.69) (10.70) (10.55) (10.67) (10.68) (10.71) Principal Amount (4.480) (4.482) (4.481) (4.479) (4.383) (4.464) (4.468) (7.632) Bont Maturity ** ** ** ** ** ** ** ** (5.372) (5.342) (5.368) (5.354) (5.202) (5.281) (5.358) (5.356) Callable ** ** ** ** *** ** ** ** (7.338) (7.345) (7.345) (7.338) (7.258) (7.354) (7.336) (7.342) Putable 102.0** (49.84) (58.59) (58.57) (58.51) (31.12) (58.70) (58.44) (57.99) Market Return (0.0428) (0.0429) (0.0428) (0.0429) (0.0425) (0.0429) (0.0429) (0.0429) Market Volatility 2.278*** 2.338*** 2.287*** 2.316*** 2.329*** 2.382*** 2.248*** 2.193*** (0.381) (0.383) (0.381) (0.380) (0.378) (0.379) (0.383) (0.385) Constant 782.6*** 781.4*** 782.3*** 780.8*** 753.0*** 783.9*** 777.7*** 2.193*** (80.99) (81.00) (81.02) (80.98) (80.75) (80.91) (80.98) (0.385) Observations 3,294 3,294 3,294 3,294 3,294 3,294 3, *** R-Squared (81.06) 23 Industry FE YES YES YES YES YES YES YES YES Credit rating FE YES YES YES YES YES YES YES YES

25 Table 4b. CDS trading interacted with YEARS on bond spreads controlling for FIRM FE. Results are in basis points and based on model 2. The dependent variable is BOND SPREAD: Ex ante credit spread over Treasury with the same maturity of the bond. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS_FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. CDS trading x YEAR: Interaction variable with CDS trading for a year. Log of sales. PROFITMARGIN: Net income over sales. LEVERAGE: Total debt over assets. R&D: Research and development expenses over sales. CREDIT RATING: Equals the number associated with the firm credit and controlled for credit rating. MARKET TO BOOK: Price per share over equity book value. TANGIBILITY: Tangible assets over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. PRINCIPAL AMOUNT: log of issue amount. BOND MATURITY: log of the bond maturity in years. CALLABLE: variable that is equal to 1 for bonds having callable. PUTABLE: variable that is equal to 1 for bonds having putable. Controlling for market turbulence and proxy for economy state: MARKET VOL and MARKET RET. Robust standard errors in parentheses. Statistically significant: *** p<0.01, ** p<0.05, * p<0.1 24

26 Model 2: CDS trading interacted with years on credit bonds Variables 2013Y 2012Y 2011Y 2010Y 2009Y 2008Y 2007Y 2006Y Trading ** * ** (11.47) (11.50) (11.85) (11.72) (10.56) (11.11) (11.55) (12.65) CDS firms 28.45*** 28.88*** 28.80*** 28.95*** 26.40*** 27.78*** 29.18*** 28.10*** Trading x ** (7.800) (7.782) (7.789) (7.773) (7.559) (7.734) (7.755) (7.674) (22.98) Trading x (21.90) Trading x (16.67) Trading x (26.47) Trading x *** (38.85) Trading x *** (28.01) Trading x *** Trading x 2006 (22.85) *** (15.98) Sales * * * * ** ** * ** (65.75) (65.71) (65.76) (65.63) (65.07) (65.23) (65.65) (64.83) Profitmargin (392.4) (392.6) (392.6) (392.4) (391.5) (391.6) (392.0) (391.2) Leverage * (389.7) (389.7) (389.8) (389.5) (387.3) (386.7) (388.1) (385.6) R&D -1,046** -1,052** -1,051** -1,055** -1,051** -1,078** -1,054** -1,062** (468.1) (468.0) (468.1) (467.8) (464.1) (465.5) (468.0) (462.4) Market to Book * (27.16) (27.19) (27.19) (27.17) (27.01) (27.08) (27.12) (26.91) Tangibility * * * * * * * * (462.7) (463.1) (463.1) (462.7) (458.7) (460.4) (461.5) (457.9) Dividends 721.4*** 722.8*** 722.3*** 725.3*** 729.5*** 749.7*** 717.5*** 779.7*** (248.9) (248.9) (248.9) (248.7) (247.5) (247.7) (248.5) (247.1) Principal Amount 195.3* 196.3* 195.8* 196.6* 196.6* 202.3** 196.1* 209.1** (102.2) (102.3) (102.3) (102.3) (101.9) (101.9) (102.1) (101.8) Bont Maturity 746.4* 749.1* 747.5* 749.2* 737.0* 751.6* 742.1* 799.2* (422.8) (423.2) (423.2) (423.0) (421.0) (421.8) (422.3) (420.6) Callable (149.3) (149.3) (149.4) (149.3) (149.3) (149.4) (149.2) (149.2) Market Return 1.673** 1.679** 1.676** 1.677** 1.644** 1.679** 1.669** 1.738** (0.732) (0.733) (0.733) (0.733) (0.728) (0.729) (0.731) (0.728) Market Volatility 28.88* 29.05* 28.98* 29.03* 28.82* 29.58* 28.97* 30.64** (15.48) (15.48) (15.48) (15.48) (15.45) (15.46) (15.47) (15.44) Constant -2,594-2,612-2,603-2,611-2,609-2,661-2,599-2,879 (1,789) (1,790) (1,790) (1,789) (1,785) (1,786) (1,788) (1,786) Observations 1,957 1,957 1,957 1,957 1,957 1,957 1,957 1,957 R-Squared Industry FE NO NO NO NO NO NO NO NO Credit rating FE YES YES YES YES YES YES YES YES 25

27 Table 4a shows the results of a regression of CDS trading on bond spreads of years as a moderator. All years are based on the issue years and interacted with CDS trading, while controlling for industry- and credit rating FE. Both years 2006 and 2007 show negative statistical significance effects at 1%. Financial markets experienced great disruptions during the financial crisis and for these years there is a positive statistical significant at 5% on bond spreads. It makes sense, because after the failure of Lehman Brothers on September 2008, the global crisis intensified. This indicates that bond spreads increase with 1.39% if there is CDS trading during this year, all else equal. This effect becomes even stronger in 2009, resulting in an increase of 2.03% in bond spreads for CDS trading bonds, all else equal. After the crisis, there is a negative statistical significant effect found at 5% for years 2012 and The crisis is behind, the regulators are active on derivative instruments and ergo, bond spreads with CDS trading issue bonds on lower rate. This results remain robust even after controlling for firm FE. These results are shown in table 4b. The bond spreads in 2008 and 2009 become even stronger, resulting for 1.65% in 2008 and 2.38% in 2009, all else equal. While for the crisis the coefficients in 2006 and 2007 remain negative. This table captures a new characteristic developed on the bond spreads during the crisis. Furthermore, an overview of the number of bonds issued, their corresponding credit rates and issue years is shown in appendix. I was aiming here to see what type of bonds were traded during the crisis period and especially at the peak of crisis. Lots of A s and BBB s were traded with CDS bonds and only 1 or two junk bonds. The findings confirm my hypothesis by showing that positive effect of CDS trading on the bond spread is stronger in time-periods after the start of the financial crisis. 5.3 Credit Spread Changes on types of Bond Issuers To test whether high yield bonds pay a greater credit spread on their bonds, I regress equation 3 using robust standard errors. I introduce HYIELD variable, which represents a dummy variable that takes value 1 for high yield bonds and zero otherwise. Bondspread it = c + β 1 T rading it + β 2 CDSfirms it + β 3 HY IELD it + β 4 HY IELD it T rading it + β 5 X it 1 + β 6 Y B it + β 7 Z it 1 + ε it (3) Table 5. CDS trading on bond spreads interaction between CDS High Yield bonds. Results are in basis points and based on model 3. TThe dependent variable is BOND SPREAD: Ex ante credit spread over Treasury with the same maturity of the bond. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS_FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. HYIELD: Dummy variable that is equal to 1 for high yield bonds. SALES: Log of sales. PROFITMARGIN: Net income over sales. LEVERAGE: Total debt over assets. R&D: Research and development expenses over sales. CREDIT RATING: Equals the number associated with the firm credit and controlled for credit rating. MARKET TO BOOK: Price per share over equity book value. TANGIBILITY: Tangible assets over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. PRINCIPAL AMOUNT: log of issue amount. BOND MATURITY: log of the bond maturity in years. CALLABLE: variable that is equal to 1 for bonds having callable. PUTABLE: variable that is equal to 1 for bonds having putable. Controlling for market turbulence and proxy for economy state: MARKET VOL and MARKET RET. Robust standard errors in parentheses. Statistically significant: *** p<0.01, ** p<0.05, * p<0.1 26

28 Model 3: CDS trading on bonds spreads of high yield bonds Variables Model 3-A Model 3-B Model 3-FE(1) Model 3-FE(2) Trading * 17.36* (7.019) (6.828) (9.166) (8.988) HYIELD 75.95*** ,908*** 1,476*** (12.20) (12.22) (280.4) (308.3) Trading x HYIELD * (22.08) (20.65) (26.29) (29.12) FIRMSCDS *** (9.252) (8.748) (17.75) (19.67) Sales *** *** (6.099) (5.859) (64.14) (72.05) Profitmargin ,126*** -2,281*** (10.72) (9.710) (614.7) (653.9) Leverage * -11,738*** -9,164*** (23.90) (22.93) (1,716) (1,870) R&D *** *** 9,671*** 7,659*** (70.27) (72.38) (1,665) (1,805) Market to Book *** *** (0.380) (0.309) (72.82) (78.61) Tangibility ,633*** -6,763*** (24.52) (24.39) (1,274) (1,378) Dividends ,444*** 2,733*** (10.02) (9.429) (468.5) (507.5) Principal Amount ,390*** 1,090*** (4.357) (3.990) (219.1) (232.6) Bond Maturity *** ,266*** 4,133*** (4.578) (4.326) (809.6) (865.4) Callable *** *** Putable (26.23) (27.01) (6.563) (6.251) (40.21) (40.91) Market Return * * 7.938*** 6.304*** (0.0430) (0.0386) (1.185) (1.253) Market Volatility 1.180*** 1.427*** 68.64*** 52.66*** (0.376) (0.359) (12.70) (13.19) Credit Rating *** *** (1.709) (2.818) Constant 503.0*** 687.0*** -13,727*** -10,659*** (50.43) (83.32) (2,377) (2,490) Observations 3,294 3,294 1,957 1,957 R-squared Industry FE YES YES NO NO Year FE YES YES YES YES Credit rating FE NO YES NO YES Firm FE NO NO YES YES 27

29 Who are the winners and who are the losers? In table 5 the results are reported for model 3 where I test which subset of issuers gets influenced by CDS trading. In this section I either control for credit ratings or time fixed effects to capture the effect on CDS trading and the high yield rated bonds. Model 3-A controls only for industry and time FE, while keeping credit rating as a continuous variable. I find here no statistical significance effect on the interaction variable. Model 3-B controls for all FE except firm-fe, but the results remain robust. I add firm FE in the last two models and observe that the models improve but reduce from 3294 observations to 1957 observations. When I keep credit rating as a continuous variable, I do not see statistical significance results in model 3-FE(1). However, when I do control for them, I see a negative statistical significant effect at 10% of 0.48% in model 3-FE(2). This indicates that the bond spreads decrease with 0.48% controlling for firm FE for CDS traded high yield bonds. Additionally, the model of firm FE explains much more variation in credit spreads as the R-square increases to 0.85 (explains 85% of the variation in credit spreads in contrast to 0.80% to the model when not controlling form firm FE). Contrary to my assumption, high yield bonds benefit from CDS trading resulting in lower bond spreads compared to investment graded bonds. Therefore, I cannot reject my hypothesis and need to conclude that High yield bonds do not pay a greater credit spread on their bonds. Other high yield bonds with no CDS trading on them had to pay a significant high issue amount when controlling for firm FE. The coefficients are larger and statistical significant at 1%. The bond spreads increase with 14.76% for non-cds traded high yield bonds compared to non-cds traded investment-grade bonds. This shows that high yield CDS traded bonds were benefitting more from CDS trading. One may infer that this comes at the cost of monitoring. Fixed income investors believe that banks actively monitor risky and opaque information firms who issue bonds with a credit rating equal to high yields. For this reason, the do not demand a higher premium on bond spreads. 5.4 Credit Risk Sensitivity Caused by the CDS Market Credit risk is measured by S&P credit rating and converted as AAA=20, AA=19, C=1 etc. Trading is a dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. This is my second main variable of interest, because it tells whether the credit risk on bonds issued after the firms CDS trade have an effect on the credit risk. The variable CDSfirms is used to control for differences between CDS firms and non-cds firms. It has a value of 1 if at any point in time a traded CDS was available for this firm. Firm-specific variables Xit-1, that determine credit risk are used here as well, such as Collateral: the degree to which assets can be collateralized, measured as the sum of property, plant and equipment divided by total assets. Size: measured as the natural logarithm of total assets. Tangibility: measured as tangible assets divided by total assets. Cashflow: measured as ebitda over total assets. Dividends: binary variable that indicates if the firm pays out dividends, being 1 for firms that paid dividends at t-1 and 0 otherwise. Sales (log of firms sales) and Leverage (total debt over total assets). Additionally, I left out Zit-1, controls since these do not have any impact on the credit risk of the firm. Creditrisk it = c + β 1 T rading it + β 2 CDSfirms it + β 3 X it 1 + ε it (4) To test whether investment-grade bonds pay a greater credit spread on their bonds, I regress equation 5 using robust standard errors. Creditrisk it = c + β 1 T rading it + β 2 CDSfirms it + β 3 IGRADE it + β 4 IGRADE it T rading it + β 5 X it 1 + ε it (5) 28

30 Table 6. CDS trading on credit risk. Results are in rating categories. The dependent variable is CREDIT RISK: measured as credit ratings converted to numerical values. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. IGRADE: Dummy variable that is equal to 1 for investment-grade bonds. IGRADE x Trading: Interaction variable of CDS traded investment-grade bonds. LEVERAGE: Total debt over assets. COLLATERAL: Sum of power plant and equipment over total assets. SIZE: natural logarithm of total assets. CASH FLOW: measured as EBITDA over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. LN SALES: Log of sales. Robust standard errors in parentheses. Statistically significant at: *** p<0.01, ** p<0.05, * p<0.1 Variables Model 4 Model 5 Model 4-A Model 4-B Model 5-A Model 5-B Trading ** *** *** (0.112) (0.0832) (0.328) (0.323) IGRADE 3.640*** 3.638*** (0.190) (0.184) IGRADE x Trading 0.485* 0.460* (0.248) (0.277) CDS firms 0.198* *** *** (0.106) (0.157) (0.0884) (0.156) Leverage *** *** (0.264) (2.403) (0.222) (2.149) Collateral 1.404*** 7.735*** 0.611*** 6.520*** (0.259) (2.697) (0.164) (2.146) Size 0.776*** *** (0.0424) (0.241) (0.0328) (0.222) Tangibility 0.820** *** (0.359) (3.232) (0.265) (2.762) Cashflow 6.290*** *** (1.078) (2.030) (0.695) (1.385) Dividends 1.624*** 1.280** 0.893*** 1.285*** (0.103) (0.508) (0.0865) (0.480) Constant 2.172*** *** (0.731) (6.627) (0.734) (5.836) Observations 6,386 4,291 6,386 4,291 R-squared Industry FE YES YES YES YES Time FE YES YES YES YES Firm FE NO YES NO YES Table 6 reports the results of my multivariate analysis on the effect of CDS trading on the credit risk. First column represents model 4 where I do not control for firm FE, while the second column does. Only the second column shows when adjusting for firm FE, there is a small positive statistical significant effect of rating at 5%. This contradicts my assumption and hence, I fail to find evidence that credit risk increases for issued CDS traded bonds. Additionally, the effect is only shown when controlling for firm FE and is an incremental change in credit rating. For a credit rating to change, the coefficient needs to be bigger than 1. An incremental change like this does not impact the credit risk. With respect to the firm controls that I use in these regressions, all seem to be statistically significant. Profitable firms with high cash flows seem to experience the highest upward credit rating. One unit increase in credit ratings, increases cash flow with 6.3 rating categories, all else equal. On 29

31 other hand, firms with high leverage experience a downgrade of 7.22 rating categories. This is a radical change on the credit rating category and deteriorates credit risk tremendously. Who are the winners and who are the losers? Model 5 exhibits the interaction term between investment-grade bonds and CDS trading. Model 5-A does not control for firm FE, while Model 5-B does. CDS traded investment-grade bonds have a positive statistical significant effect of 0.49 at 10%. This indicates that the credit rating for a safe bond with a CDS increases with rating, all else equal. Here again, it is an incremental change and does not impact credit risk. Furthermore, non-cds traded safe bonds experience a credit upward of 3.64 points. It means that bonds without a CDS trading receive a much greater rating improvement than CDS safe bonds. The change is larger and shows it has a greater impact on credit risk. In this case, my findings show that non-cds traded safe bonds enhance their credit risk, while for CDS traded safe bonds it does not. When controlling for firm fixed effects, the result remain robust. The following two-way graph plots the interaction effect between CDS trading of investment-grade bonds. Figure 6 exhibits for investment-grade bonds a small upward slope, while for high-yield bonds a negative slope. For riskier bond the change is greater and hence, this deteriorates its credit risk. Therefore, I find that investmentgrade bonds do benefit more of CDS trading than high yield bonds. But, in comparison to non-cds traded bonds their credit risk remains the same and it shows that these types of bonds are better off without CDS trading. Figure 6. Slope movement of the interaction between CDS trading on type of issued bonds spreads. 5.5 Credit Risk Sensitivity varying across Time To examine whether negative effect of CDS trading on the credit rating is stronger in timeperiods after the start of the financial crisis, I regress equation 6 with robust standard errors. In order to see how credit rating changes, I create several time year dummies and interact these with CDS trading. Creditrisk it = c + β 1 T rading it + β 2 CDSfirms it + βcrisis it + βcrisis it T rading it + β 5 X it 1 + ε it (6) 30

32 Table 7. CDS trading interacted with YEARS on credit rating. Results are in rating categories and based on model 6. The dependent variable is CREDIT RISK: measured as credit ratings converted to numerical values. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. CDS trading x YEAR: Interaction variable with CDS trading for a year. LEVERAGE: Total debt over assets. COLLATERAL: Sum of power plant and equipment over total assets. Size: natural logarithm of total assets. CASH FLOW: measured as EBITDA over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. LN SALES: Log of sales. Robust standard errors in parentheses. Statistically significant at: *** p<0.01, ** p<0.05, * p<0.1 Model 6: CDS trading interacted with years on credit risk Variables 2013Y 2012Y 2011Y 2010Y 2009Y 2008Y 2007Y 2006Y Trading 0.313** 0.319** 0.305** 0.347** 0.297** *** (0.135) (0.139) (0.140) (0.137) (0.141) (0.143) (0.144) (0.147) CDS firms *** *** *** *** *** *** *** *** (0.0980) (0.0981) (0.0981) (0.0980) (0.0981) (0.0980) (0.0981) (0.0982) Trading x (0.909) Trading x (0.302) Trading x (0.272) Trading x (0.424) Trading x (0.317) Trading x *** (0.236) Trading x ** (0.267) Trading x ** (0.286) Leverage *** *** *** *** *** *** *** *** (0.430) (0.431) (0.431) (0.431) (0.431) (0.430) (0.430) (0.430) Collateral 1.739*** 1.744*** 1.738*** 1.736*** 1.737*** 1.745*** 1.743*** 1.706*** (0.353) (0.353) (0.353) (0.353) (0.353) (0.352) (0.352) (0.352) Size 1.070*** 1.068*** 1.068*** 1.069*** 1.069*** 1.072*** 1.069*** 1.074*** (0.0650) (0.0651) (0.0651) (0.0651) (0.0652) (0.0649) (0.0650) (0.0652) Tangibility 1.325*** 1.315*** 1.322*** 1.326*** 1.320*** 1.302*** 1.338*** 1.316*** (0.464) (0.464) (0.464) (0.464) (0.464) (0.464) (0.463) (0.462) Cashflow 3.825*** 3.837*** 3.821*** 3.816*** 3.824*** 3.777*** 3.770*** 3.836*** (1.169) (1.170) (1.170) (1.168) (1.170) (1.165) (1.165) (1.178) Dividends 1.472*** 1.472*** 1.473*** 1.475*** 1.473*** 1.472*** 1.468*** 1.459*** (0.139) (0.139) (0.139) (0.139) (0.139) (0.139) (0.139) (0.139) Constant *** *** *** *** *** *** *** *** (0.740) (0.741) (0.741) (0.740) (0.741) (0.740) (0.739) (0.742) Observations 3,294 3,294 3,294 3,294 3,294 3,294 3,294 3,294 R-square Industry FE YES YES YES YES YES YES YES YES Time FE NO NO NO NO NO NO NO NO Firm FE NO NO NO NO NO NO NO NO 31

33 Table 7b. CDS trading interacted with YEARS on credit rating controlling for firm FE. Results are in rating categories and based on model 6. The dependent variable is CREDIT RISK: measured as credit ratings converted to numerical values. TRADING: Dummy variable that takes value 1 if at time t (the issue date of the bond) a traded CDS was available. CDS FIRMS: Dummy variable that is equal to 1 for firms that ever have a traded CDS. CDS trading x YEAR: Interaction variable with CDS trading for a year. LEVERAGE: Total debt over assets. COLLATERAL: Sum of power plant and equipment over total assets. Size: natural logarithm of total assets. CASH FLOW: measured as EBITDA over total assets. DIVIDENDS: variable that is equal to 1 for firms paying dividends. LN SALES: Log of sales. Robust standard errors in parentheses. Statistically significant at: *** p<0.01, ** p<0.05, * p<0.1 Model 6: CDS trading interacted with years on credit rating Variables 2013Y 2012Y 2011Y 2010Y 2009Y 2008Y 2007Y 2006Y Trading (0.0794) (0.0818) (0.0825) (0.0813) (0.0813) (0.0821) (0.0830) (0.0821) CDS firms *** *** *** *** *** *** *** *** Trading x (0.0704) (0.0704) (0.0704) (0.0704) (0.0704) (0.0704) (0.0705) (0.0703) (0.220) Trading x (0.184) Trading x (0.154) Trading x (0.169) Trading x * (0.178) Trading x Trading x 2007 Trading x 2006 (0.175) (0.187) (0.207) Leverage * (2.433) (2.444) (2.435) (2.434) (2.441) (2.436) (2.435) (2.442) Collateral 8.991*** 8.873*** 8.983*** 8.982*** 8.947*** 8.972*** 8.991*** 8.979*** (2.727) (2.704) (2.728) (2.727) (2.728) (2.728) (2.729) (2.734) Size (0.242) (0.240) (0.242) (0.242) (0.242) (0.242) (0.242) (0.242) Tangibility (3.256) (3.235) (3.257) (3.257) (3.259) (3.258) (3.258) (3.267) Cashflow (2.207) (2.214) (2.208) (2.209) (2.212) (2.209) (2.207) (2.210) Dividends 1.353*** 1.406*** 1.354*** 1.355*** 1.362*** 1.357*** 1.350*** 1.347*** (0.517) (0.524) (0.518) (0.517) (0.519) (0.518) (0.517) (0.519) Constant 144.5*** 145.0*** 144.4*** 145.2*** 138.6*** 145.1*** 143.3*** 141.8*** (52.41) (52.45) (52.98) (53.30) (50.84) (53.31) (52.57) (52.62) Observations 4,291 4,291 4,291 4,291 4,291 4,291 4,291 4,291 R-square Industry FE NO NO NO NO NO NO NO NO Time FE NO NO NO NO NO NO NO NO Firm FE YES YES YES YES YES YES YES YES 32

34 Table 7 exhibits findings on a regression of CDS trading on credit risk of years a moderator. All years are based on the issue years and interacted with CDS trading, while controlling for several FE. What immediately draws my attention is the positive coefficients for years 2007 and Before the crisis, credit rating for CDS bonds received downgrades of 0.65 points in 2006 compared to non-cds trading bonds, all else equal. While at beginning of the crisis, credit rating increased to 0.59 all else equal. The effect becomes even stronger at the peak of crisis, resulting in 0.62 points more for CDS traded bonds in comparison to non-cds traded bonds. When I control for firm FE, results do not remain robust. A full rating to change implies a coefficient of bigger than 1. This means that the credit rating during this period did not have a great impact on CDS traded bonds and their credit risk remains the same. Nevertheless, my findings in this model contradict my assumption. I did not find evidence of a stronger negative effect of CDS trading on the credit rating in time-periods after the start of the financial crisis. 6 Conclusion & suggestions for Further Research Despite the claims made by the economists and authors about the benefits of credit derivatives, the average corporate borrower has not yet seen a lower cost of debt capital. When I run a OLS regression and do not control for any FE, I observe a small positive effect of 0.19% on bond spreads. It means that fixed income investors require to get more compensated for taking extra risk. This finding does not support the diversification theory where creditors have extra opportunity to hedge their risk exposure, because there is no drop in the cost of bond spread found. However, it does support the information theory where CDSs could reveal new information about firms. Researchers suggest that information role of CDS market can lead to a reduction in the cost of issuing debt by reducing the information premium fixed income investors demand. This is measured by taking the firms bond on a bond adjusted for the yield on a US treasury bond (risk free) at the time of issue (bond spreads). In this case, I found the opposite where fixed income investors were demanding a higher credit spread. When I split my sample in two periods pre- and post-crisis, I first observe that all bonds during post-crisis had to pay a greater bond spread. According to my findings, an average firm with a CDS traded bonds had to pay 0.35% more in comparison to non-cds traded bond. When I control for firm FE, this effect becomes significant stronger and doubles to 0.70%. This means that CDS traded bonds are negatively impacted due to a higher credit spread than non-cds traded bonds. One can infer that before the crisis, investors did not react to CDSs on reference entities, because the economy was performing well. When the crisis hit off, CDSs were the ones to blame for causing crisis. Therefore, an average firm with a traded CDS had to pay more when issuing their bonds compared to non-cds firms. Additionally, I fail to find evidence for an average firms credit risk arising from CDS trading. Instead, I find a small incremental positive change on the credit rating of CDS traded bonds, however this does not impact the credit risk. Furthermore, when I interact CDS trading with several years to observe the changes during pre-, at- and post- crisis, I find that CDS trading had an effect on the bond spreads influenced by the crisis period. During pre-crisis bond issuers with a traded CDS had to pay less bond spreads. While at the peak and in 2009 the bond issuers with CDSs were charged with 1.39% (2008) and 2.03% (2009), all else equal. After the crisis, I found that the bond spreads became negative again. It appears when the crisis is behind, the regulators are active on derivative instruments and ergo, bond spreads with CDS trading issue bonds on lower rate. This results remain robust even after controlling for firm FE and become stronger resulting in 1.65% (2008) and 2.38% in 2009, all else equal. Moreover, I did not find evidence of a stronger negative effect of CDS trading on the credit rating after the 33

35 start of financial crisis. Instead, I found a positive incremental change contributing to credit rating. Credit risk increases with 0.59 ratings for CDS bonds during 2007 and with 0.62 points during 2008, all else equal. However, this is less than 1 and does not impact the credit risk of reference entities. More interestingly, the types of firms that one expects would naturally benefit the most, which is consisted with prior findings safe and transparent firms. It appears to be adversely affected by the CDS market, since risky and information opaque firms have benefitted from a small reduction in bond spreads at issue. These results appear to be explained by the monitoring channel. As Datta et al. (1999) points out that riskier firms benefit more from the monitoring process, since they are riskier and need to be monitored closely. It seems that high-yield bonds need to pay 0.48% on bond spreads less than investment-grade bonds, all else equal. What appears to be even more surprisingly is that a non-cds traded high-yield bonds was facing 14.76% on bonds spreads, all else equal. Therefore, high yield bonds definitely benefitted from CDS trading. An explanation for this might be that fixed income investors believe banks monitor these CDS traded bonds and for this reason they do not charge them with a higher premium on bond spreads like for others. Who are the winners and losers from CDS trading on their credit risk? I find that non-cds traded safe bonds receive a greater upgrade and therefore, their credit risk improves while for CDS traded safe bonds the change is extremely small (0.49 rating points in comparison to 3.65 ratings). Risky CDS traded bonds experience a greater negative change in their credit rating and hence, its credit risk deteriorates. Coincidentally, I discovered that the benefit arising from cheaper bond spread and the disadvantages resulting from a deteriorate credit risk by CDS trading have changed. During the crisis, investors demanded a higher premium for CDS traded bonds, but the credit risk remained the same. Risky and information opaque firms with a CDS trading have benefitted from a small reduction in bond spreads at issue, however their credit rating deteriorated from a downgrade. Non-CDS traded riskier bonds did had to pay full price on bond spreads, so in this case CDS trading seems to be only beneficial for them. Together, these results suggest that the impact of CDSs on borrowers is not as positive as some economists and researchers have suggested. Limitations to this study are twofold. The first issue deals with the fact that CDSs are traded OTC, which makes determining the exact date of the onset of CDS trading for a reference entity difficult. Therefore, it would be correct to use multiple data sources for CDS data to validate the accuracy. Unfortunately, there is no CDS data provider who offers a complete historical data for the whole period of analysis. Since Bloomberg Terminal lacks of this, my model might be confound due to some missing data points. Therefore, I eliminated several firms for which it was not clear when the onset date was for CDS trading. As a result, this led to a relatively small sample of firms that may cause for biases and inconsistent estimators. But with controlling for several FE, this could lead to robustness checks. Next, when I attempted to model CDS trading on credit ratings, I used credit ratings as a continuous variable. To test some hypothesis, the results appeared to be smaller than 1. For a credit rating to change the coefficients needs to be bigger than 1. An incremental change of 0.30 e.g. does not imply a rating change. This is a consequence of including credit ratings as a continuous variable in my model. Nevertheless, the small incremental changes in credit ratings remain significant because they contribute to the ratings categories. Another issue deals with the possibility that CDS trading is endogenous. One method to deal with endogeneity is introducing an instrumental variable, which allows for consistent estimation when my independent variables are correlated with the error terms of regressions. This correlation occurs when the dependent variable causes at least one of the independent variables (omitted variable bias or independent variables are subject to measurement error). To control for this endogeneity, one needs to find an instrumental variable that accounts 34

36 for CDS trading and is exogenous to the dependent variables bond spreads and credit risk. One example for a potential instrumental variable (IV) is the study of Minton, Stulz and Williamson (2008). The authors find that banks who use foreign exchange hedging instruments buy more CDS contracts to hedge their exposure to credit risk. Linking firms lenders as well as bond underwriters to their respective usage of derivatives, one can then expect that the firms relationships with banks to capture important changes in credit supply (Saretto and Tookes, 2013). Given that this approach is extremely time intensive and the scope for writing my thesis is limited, seeking for an instrumental variable approach is not feasible. Nevertheless, further work could be done by introducing an IV such as foreign exchange rates and use Hausman test programmed on STATA. 9 Another method to control for endogeneity is to take lags (previous trends in rate changes and bond spreads). However, the format of my data set prevents me from taking such actions. To deal with the endogeneity problem, my data needs time series of CDS spreads, bond spreads and time series of credit ratings for each firm across period. Since my data is a repeated cross-sectional, this would require an exuberant amount of computational power as well advanced programming skills. 9 Stata 14 offers eteffects, which obtains treatment effects when unobserved variables affect both treatment assignment and outcomes. This command estimates the correlation between unobservables and if the correlation is 0, there is no endogeneity. Commands are estat endogenous. 35

37 References [1] Acharya, Viral and Johnson, Timothy. (2007). Insider Trading in Credit Derivatives, forthcoming in the Journal of Financial Economics. [2] Ashcraft, A. B., & Santos, J. A. (2009). Has the CDS market lowered the cost of corporate debt?. Journal of Monetary Economics, 56(4), [3] Bai, J.& Collin-Dufresne, P. (2011). The determinants of the cds-bond basis during the financial crisis of , NETSPAR, Discussion Papers. [4] Binsbergen, V., Jules, H., Graham, J. R., & Yang, J. (2010). The cost of debt. The Journal of Finance, 65(6), [5] BIS. (2016). Statistical Release: OTC Derivatives Statistics. [6] Blanco, R., S. Brennan, and I.W. Marsh. (2003) An Empirical Analysis of the Dynamic Relationship Between Investment Grade Bonds and Credit Default Swaps, forthcoming in the Journal of Finance. [7] Bolton, P., & Oehmke, M. (2011). Credit default swaps and the empty creditor problem. Review of Financial Studies, 24(8), [8] Bris, A., Welch, I., & Zhu, N. (2006). The costs of bankruptcy: Chapter 7 liquidation versus Chapter 11 reorganization. The Journal of Finance, 61(3), [9] Collin-Dufresne, P., Goldstein, R. S., & Martin, J. S. (2001). The determinants of credit spread changes. The Journal of Finance, 56(6), [10] Datta, S., Iskandar-Datta, M., & Patel, A. (1999). Bank monitoring and the pricing of corporate public debt. Journal of Financial Economics, 51(3), [11] Danis, Andras, and Andrea Gamba, 2015, The real effects of credit default swaps, Georgia Tech Scheller College of Business Research Paper No [12] Frank, Murray Z., and Vidhan K. Goyal, 2009, Capital structure decisions: Which factors are reliably important? Financial Management 38, [13] Gruber, M. J., & Warner, J. B. (1977). Bankruptcy costs: Some evidence. The journal of Finance, 32(2), [14] Hirtle, B. (2009). Credit derivatives and bank credit supply. Journal of Financial Intermediation, 18(2), [15] Hu, H. T., & Black, B. S. (2008). Equity and debt decoupling and empty voting II: importance and extensions. as published in University of Pennsylvania Law Review, 156, [16] Hull, J., Predescu, M., White, A., The relationship between credit default swap spreads, bond yields, and credit rating announcements. Journal of Banking and Finance 28, [17] Jarrow, R. A. (2010). The economics of credit default swaps (CDS). Johnson School research paper series, ( ). 36

38 [18] Jorion, P., & Zhang, G. (2007). Good and bad credit contagion: Evidence from credit default swaps. Journal of Financial Economics, 84(3), [19] Longstaff, F.A., S. Mithal, and E. Neis. (2004). Corporate yield spreads: Default risk or liquidity, forthcoming in the Journal of Finance. [20] Mayordomo, Sergio, Juan Ignacio Peña, and Eduardo S. Schwartz, 2014, Are all credit default swap databases equal?, European Financial Management, 20, [21] Minton, Bernadette A., René Stulz, and Rohan Williamson, 2008, How much do banks use credit derivatives to hedge loans?, Journal of Financial Services Research 35, [22] Myers, Stewart C., 1977, Determinants of corporate borrowing, Journal of Financial Economics 5, [23] Norden, L. and W. Wagner. (2007). Credit derivatives and loan pricing, Working paper, University of Mannheim. [24] Norden, L. and M. Weber. (2004). Informational E.ciency of Credit Default Swap and Stock Markets: The Impact of Credit Rating Announcements, Working paper, University of Mannheim. [25] Oehmke, M., & Zawadowski, A. (2014). The anatomy of the CDS market. [26] Saretto, A., & Tookes, H. E. (2013). Corporate leverage, debt maturity, and credit supply: The role of credit default swaps. Review of Financial Studies, 26(5), [27] Shim, I. & Zhu, H. (2014). The impact of CDS trading on the bond market: Evidence from Asia. The Journal of Banking & Finance, 40, [28] Shumway, T. (2001). Forecasting bankruptcy more accurately: A simple hazard model*. The Journal of Business, 74(1), [29] Subrahmanyam, M. G., Tang, D. Y., & Wang, S. Q. (2014). Does the Tail Wag the Dog?: The Effect of Credit Default Swaps on Credit Risk. Review of Financial Studies, 27(10), [30] Wallison, P. J. (2009). Everything you wanted to know about credit default swaps: but were never told. Journal of Structured Finance, 15(2), 20. [31] Zhu, H., An empirical comparison of credit spreads between the bond market and the credit default swap market. Journal of Financial Services Research 29,

39 Appendix Credit ratings categories S&P ratings Category AAA 20 AA+ 19 AA 18 AA- 17 A+ 16 A 15 A- 14 BBB+ 13 BBB 12 BBB- 11 BB+ 10 BB 9 BB- 8 B+ 7 B 6 B- 5 CCC+ 4 CCC 3 CCC- 2 C 1 38

40 Histograms 39