Credit Default Swaps, Fire Sale Risk and the Liquidity Provision in the Bond Market

Transcription

1 Credit Default Swaps, Fire Sale Risk and the Liquidity Provision in the Bond Market Massimo Massa* Lei Zhang** Abstract We study the effect of credit default swaps (CDSs) on the corporate bond market. We argue that CDSs reduce fire sale risk for investment grade bonds and test this hypothesis using a comprehensive sample of US corporate bonds with information on CDS contracts availability and institutional bond ownership over the period. We start with two events triggering forced sales by bond investors: bond rating downgrades from investment grade to high yield, and the selling pressure from property insurance companies following Hurricane Katrina. The presence of CDS contracts lowers fire sales and their impact by reducing the drop in bond liquidity and the rise in yield spreads. Then, we show that in general the presence of CDSs significantly reduces yield spreads and increases liquidity for investment grade bonds. This effect is stronger during the financial crisis period. The presence of CDS contracts reduces bond yield spreads by 24 bps, with an economic significance higher than the one related to one notch improvement in credit rating (17bps). The results are confirmed by an instrumental variable identification that pins down the need for CDS contracts by exploiting the level of loan concentration of the banks that the bond issuer borrows from. Our results have important normative implications, as they suggest that at least for the class of investment grade bonds CDSs may actually help to reduce risk contagion around financial crises.. JEL Classification: G10, G15, G21 Keywords: CDS, bond liquidity, yield spreads, fire sales, crisis * We would like to thank Joshua Coval, Madhu Kalimipalli, Martin Oehmke and conference participants at the Western Finance Association 2013 Meeting for their helpful comments and discussions. All remaining errors and omissions are our own. Massimo Massa: INSEAD, Finance Department, Bd de Constance, Fontainebleau Cedex, France, tel. +33( , massimo.massa@insead.edu. ** Lei Zhang: Nanyang Business School, Nanyang Technological University, Division of Banking and Finance, 50 Nanyang Avenue, Singapore , tel , zhangl@ntu.edu.sg. Please send all the correspondence to Lei Zhang (the corresponding author).

2 Introduction The recent financial crisis has brought to the fore the role of credit derivatives and their implications for the financial markets. The existence of credit derivatives and, in particular, of credit default swaps (CDSs) has been considered instrumental to the spread of the financial crisis (Allen and Carletti, 2006, Stulz, 2010). Chris Wolf, a hedge fund manager, stated that CDS has become essentially the dark matter of the financial universe. 1 George Soros has called for most or all trading in credit default swaps to be banned or strictly regulated. 2 The rapid rise of such a market has made the problem even more acute: the total notional amount of the CDS market has grown from $6 trillion to $41 trillion from 2004 to The indictment is based on the way credit insurance would affect the debtor-creditor relation in the case of distress of the borrower. CDS contracts, by protecting the lenders in the case of distress, would reduce their incentives to restructure the debt and create the empty creditor problem (Bolton and Oehmke, 2011). 4 The lower risk would also reduce the lenders incentives to effectively monitor the borrowers (Thompson, 2010, Parlour and Winton, 2012). Regardless of this almost consensual theoretical dark view, the empirical evidence is mixed (e.g., Ashcraft and Santos, 2009, Hirtle, 2009, Purnanandam, 2011, Subrahmanyam et al., 2011, Saretto and Tookes, 2011, Bedendo et al., 2013). In this paper, we take a new approach and focus on one important yet undocumented role of CDSs: the reduction in forced sale (fire sale) risk in the underlying bonds. Fire sales are related to the need of financial intermediaries e.g., mutual funds to meet investor withdrawals when facing sudden liquidity outflows, or to meet suddenly deteriorated capital ratios e.g., insurance companies. If a shock induces some financial intermediaries to liquidate an asset, the forecast of such a sale will lead the other players in the market to try to preempt it by selling as well. The ensuing massive sales can 1 The $55 trillion question, by Nicholas Varchaver, Fortune Magazine, September 30, One way to stop bear raids, Wall Street Journal, March 24, This figure is reported by the Bank of International Settlements. The total notional value of outstanding CDSs was over 60 trillion dollars at the 2007 peak. The number dropped to 30 trillion dollars in The reduction was not due to decreased trading activity in CDSs (Jarrow, 2010). It occurred through the portfolio compression as demanded by regulators where long and short positions on the same underlying credit entity held by the same institution are netted (Duffie et. al, 2010). 4 The use of derivatives can in general allow a decoupling of voting and cash-flow rights in common equity through the judicious use of derivatives to hedge cash-flow risk (Hu and Black, 2006, 2007, Kahan and Rock, 2007). 1

3 drastically negatively affect the asset price and its liquidity. This phenomenon has been termed as fire sales (Coval and Stafford, 2007, Shleifer and Vishny, 2011), financial runs (Bernardo and Welch, 2004), or sales due to regulatory pressure (Ellul et al., 2011). Fire sales are especially relevant in the bond market, given that most corporate bonds are traded over-the-counter with high search costs, and liquidity relies on the bond dealers committing risk capital to market making. More importantly, regulatory constraints may force some large institutional bond investors to sell in the presence of a drop in market value or a downgrade in bond ratings, and thus create significant fire sale or liquidity risk for the bonds. For example, consider the largest class of investors in the corporate bond market: the insurance companies. 5 By regulation, insurance companies can only hold investment grade bonds, and the amount of risk-based capital required by the state regulator is based on the credit quality of their asset holdings (Ellul et al., 2011). A negative shock to the bonds they hold, such as a rating downgrade, will require them to post additional equity capital, unless they choose to sell the bonds. However, if such insurance companies were to buy CDS protection, the need to find new capital would be largely reduced. For example, the NAIC Capital Markets Special Report (2007) states that purchasing CDS helps insurers reduce the credit risk of the bond portfolio, which is one of the major investment assets held by insurance companies. The purchase of CDS can also help insurers save their risk-based capital (RBC) and meet regulatory requirements. And indeed, insurance companies use derivatives extensively to hedge risk (e.g., Cummins and Sommer, 1996, Cumming and Hirtle 2001, Cummins et al., 2001, Cummins et al., 2006) and ameliorate their performances (Goldfried, 2003). 6 5 According to the NAIC report (2007), life insurers invested 76.1% of their total assets in bonds, whereas property/casualty insurers invested in bonds 62.1% of their assets. As reported in the Lipper/eMAXX fixed income database, at the end of the second quarter of 2005, insurance companies held $574 billion of publicly issued corporate bonds, among which $484 billion were held by life insurance companies and $90 billion by property and reinsurance companies. 6 Over the period , the usage of CDS by life and property/casualty insurance companies has grown significantly. As of Dec. 31, 2010, a total of 223 insurance companies participated in the derivatives market. Of this number, 140 were life insurance companies, 63 were property/casualty insurance companies, 14 were health insurance companies and 6 were fraternal insurance companies. The insurance companies with derivatives exposure were domiciled in 39 states, with New York, Connecticut, Michigan and Iowa having the largest exposures. Furthermore, there were approximately 48,600 individual derivative positions across the insurance industry. The average position size was $18 million in notional value, and the largest single position noted was $6 billion in notional value. For example, MetLife uses CDSs to hedge against creditrelated changes in the value of its investments and to synthetically create investments that are either more expensive to acquire or otherwise unavailable in the cash markets (NAIC Capital Markets Special Report , 2007). 2

4 Therefore, CDS contracts allow investors who are required by regulation to hold high quality bonds to defer the sales in the case the bonds they hold are downgraded or lose the coveted investment grade status. Importantly, this should also lower the other players preemptive incentives to front-run such investors, and therefore reduce the need to rush to sell the bonds in the market. Moreover, the presence of CDSs may induce liquidity providers to provide liquidity at the very moment when bonds are subject to such fire sale related liquidity shocks. Indeed, the availability of CDSs makes it possible to enact a CDS-bond basis arbitrage strategy i.e., buy bonds and buy CDS protection. This facilitates liquidity provision (Choudhry, 2006), effectively reducing the limits of arbitrage (Shleifer and Vishny, 1997) in the bond market. Overall, these considerations suggest that bonds issued by firms with CDS contracts would suffer less fire sale risk. The lower risk of fire sales should reduce bond yield spreads and increase bond liquidity. This intuition provides a novel angle unexplored in the literature till now on the role of CDS in the corporate bond market. More specifically, we hypothesize that the presence of CDS contracts lowers the yield spreads of the bonds and increases their liquidity. We argue that such effect should concentrate among investment grade bonds the ones that experience fire sale risk due to regulatory pressures. In contrast, high yield bonds, already held by investors who are not subject to regulatory constraints in terms of the quality of the assets they hold e.g., hedge funds and high yield mutual funds should be less vulnerable to fire sale risk. We test this hypothesis using data on a comprehensive sample of US corporate bonds with CDS contracts information (monthly frequency) over the 2001 to 2009 period. 7 We also rely on a novel dataset of institutional investors quarterly bond holdings during the sample period. The sample is not only large one of the largest used in the literature, but the availability for the first time of information on corporate bonds, institutional bond holdings and CDS contracts is critical, as it 7 Our base sample include all the bonds included in the Bank of America Merrill Lynch Coporate Master Investment Grade and High Yield Bond Index, one of the most commonly used index in the corporate bond market. 3

5 provides us with the power to identify fire sales effects and to distinguish them from firm-specific and bond-specific factors. We start with some univariate results that link bond yield spreads and bond illiquidity 8 to the availability of CDS contracts. We find opposite results for investment grade bonds and for high yield bonds i.e., the presence of CDS contracts reduces bond yield spreads and bond illiquidity for investment grade bonds while it increases them for high yield bonds. 9 Investment grade bonds with CDS contracts outstanding have on average 17 bps lower yield spreads (12% lower relative to the sample average) than the ones without CDSs. This is equivalent to the effect of a one notch improvement in credit ratings. The results on bond illiquidity are consistent. For investment grade bonds, the ones with CDS contracts outstanding have 7% lower illiquidity than those without CDSs. Moreover, the effects of lower yield spreads and lower illiquidity are concentrated among BBB and A rated bonds rather than AAA and AA rated bonds i.e., the ones most subject to fire sale risk. These findings suggest that the channel of impact of CDSs on the underlying bonds, if there is any, works very differently for investment grade bonds and for high yield bonds. It points in the direction of a link between the economic function of CDSs and the salient market segmentation in the bond market: investment grade bonds mostly held by institutional investors subject to risk-based capital requirements are more vulnerable to fire sale risk, while high yield bonds held by investors not facing such constraints are less subject to fire sale risk. Next, we focus on two events that have been shown to trigger forced sales by bond institutional investors: bond rating downgrades from investment grade to high yield grade, and the selling pressure from property insurance companies following Hurricane Katrina. The downgrade from investment grade to high yield triggers forced sales as insurance companies liquidate their holdings in the downgraded bonds ( fallen angels ) to comply with the NAIC s 8 We measure bond yield spread as the option-adjusted (OA) spread, defined as the number of percentage points that the treasury spot curve must be shifted in order to match discounted cash flows to the bond s price. We follow Bao et al. (2011) and measure bond illiquidity as the implied bid-ask spread based on the auto-covariances of bond price changes. 9 The effect on high yield bonds can be explained by the empty creditor problem (Bolton and Oehmke, 2011) or a loss of bank montioring (Parlour and Winton, 2012). However, high yield bonds only represent around 20% of the corporate bond universe. This suggests that the documentation of the distinct function of CDSs on investment grade bonds is of particular importance given their majority representation in the bond market as well as a general lack of empirical evidence in the literature. 4

6 regulatory constraints (Ellul et al., 2011). 10 We argue that the presence of CDS contracts lowers the impact of the forced selling pressure from insurance companies upon such downgrade. And indeed, we find that bonds without CDS contracts experience a 150% higher drop in institutional ownership than the bonds with CDS contracts. This provides a direct evidence in support of our hypothesis, as the lower drop in institutional ownership can only be attributed to the lower need to sell assets generated by the presence of CDSs. Moreover, in line with our hypothesis, the presence of CDS contracts lowers the impact of fallen angels in terms of both yield spreads and bond liquidity. Bonds without CDS contracts experience a 200% (150%) higher increase in yield spreads (bond illiquidity) than bonds with CDS contracts. Next, we examine a purely exogenous event: the shock to property insurance companies provoked by Hurricane Katrina. Hurricane Katrina (August 23-30, 2005) is the costliest natural disaster in the history of the United States with an insured damage of over $40 billion. The property insurance companies exposed to Katrina, driven by the need to meet redemption claims, generated a selling pressure of the bonds held by those investors (Massa and Zhang, 2011). These forced sales can only be attributed to supply side shocks i.e., shocks to the Katrina-exposed property insurance companies as opposed to firm-specific shocks such as rating downgrades. 11 We document that the presence of CDS contracts reduced the impact of forced sales in terms of both the drop in bond prices as well as the drop in liquidity. Specifically, in the face of the selling pressure from Katrina-exposed property insurance companies, the bonds without CDS contracts experienced a 45% higher increase in yield spreads than the bonds with CDS contracts. In line with the previous results, the increase on bond illiquidity concentrates among bonds without CDS contracts. The univariate results and the event-based analyses provide a first evidence that CDSs reduce fire sales, increase bond liquidity and reduce yield spreads for investment grade bonds. Next, we perform a 10 Ellul et al. (2011) show that the selling pressure from insurance companies generates significant price drops on the bonds downgraded from investment grade to high yield. In line with them, we find a significant 4% decrease in bond institutional ownership around the quarter of such downgrade, which is much higher than that of an average 1% decrease in bond institutional ownership around other rating downgrades not crossing the investment grade/high yield threshold. 11 In this case, the concern of a potentially spurious correlation between the presence of CDS contracts and firm-specific shocks in the proximity of rating downgrades can be ruled out. 5

7 multivariate analysis to examine the impact of CDSs on the underlying bonds. In this setting, we control for major bond and firm characteristics such as maturity, coupon rates, bond issue outstanding, firm size, leverage, market-to-book and equity volatility etc. In particular, we control extensively for default risk, by considering specifications with detailed credit rating fixed effects or even credit rating times time fixed effects. Our results are robust to all these specifications. We document a strongly negative relationship between yield spreads and the availability of CDS contracts in the case of investment grade firms. The presence of CDS contracts reduces bond yield spreads by 24 bps. The economic significance is higher than the one related to one notch improvement in credit rating (17bps). The link between CDS contracts and yield spreads is stronger during periods of tight liquidity conditions (the dotcom crash and the subprime crisis), in line with our hypothesis of a liquidity provision role of CDSs. There is instead no significant relationship for high yield bonds. Then, we focus on the impact of CDSs on bond liquidity. We find that, in the case of investment grade bonds, the presence of CDS contracts lowers bond illiquidity by 9% relative to the sample average. In contrast, for high yield bonds, the availability of CDS contracts actually increases bond illiquidity. These effects are also stronger during periods of tight liquidity conditions. That is, CDS contracts increase liquidity for investment grade bonds and reduce it for high yield bonds, and the relationship is stronger when liquidity conditions deteriorate in the market. One potential concern is that the presence of CDS contracts may be endogenously determined. Therefore, to provide a more causal interpretation of our analysis, we consider an instrumental variable identification that exogenously pins down the need for CDS contracts in the market. We use as instrument the level of loan concentration of the banks lending money to the firm on which the CDSs are written. The intuition is that banks use CDSs to hedge their loan positions. The less diversified their loan portfolio is, the higher their incentives are to purchase CDSs for protection. 12 We first provide evidence of this claim, by explicitly relating the degree of concentration of the loan 12 Jarrow (2010) argues that For financial institutions that originate a large quantity of loans with a particular geographic or industry concentration, the ability to hedge the credit risk of these loans by purchasing a CDS enables the financial institution to eliminate the geographic or industry concentration from their portfolio. 6

8 portfolios of the banks across different industries and geographical regions to their use of credit derivatives for hedging purposes. We find a strongly positive relationship between the use of credit derivatives and the degree of concentration of the loan portfolios, for both the notional amount of credit derivatives and the notional amount standardized by the assets of the bank. 13 A one standard deviation increase in loan portfolio concentration is related to 59% higher use of credit derivatives for hedging purposes. The instrumental variable analysis confirms the previous findings, displaying a significantly negative relationship between the presence of CDS contracts and both yield spreads and bond illiquidity for investment grade bonds. A one standard deviation increase in the instrumented presence of CDS contracts lowers yield spreads by 26 bps and reduces illiquidity by 8% relative to the sample average. These results are similar in terms of economic significance to the non-instrumented ones. Again, no effect is there for high yield bonds. Several robustness checks confirm our results. We start by exploiting a proxy of depth in the CDS market based on the number of CDS quotes provided by the CDS dealers (Qiu and Yu, 2012). Consistently, we find that the CDS depth reduces bond yield spreads and increases bond liquidity for investment grade bonds. No effect is there for high yield bonds. Then, we further address the potential concern that the presence of CDS contracts may be driven by some unobserved characteristic of firm risk. First, we focus on the CDS initiation and investigate the impact of CDS initiation on yield spreads using a difference-in-difference approach. A univariate matching sample comparison shows that the introduction of CDS contracts reduces yield spreads by 21 bps for investment grade bonds. No effect is observed for high yield bonds. A multivariate difference-in-difference approach confirms the univariate results and shows that the CDS initiation reduces yield spreads by 15 bps for investment grade bonds, while no significant effect is there for high yield bonds. 13 There is instead no relationship between loan concentration and the bank s use of derivatives to hedge interest rate risk and foreign exchange risk. This implies that the degree of loan concentration is directly related to the bank s hedging of credit risk but not some general hedging purposes. 7

9 Second, we exploit the fact that in general firms have many bonds outstanding with different maturities. This allows us to focus on the component of yield spreads driven by bond liquidity rather than default risk, by linking the differences in yield spreads between long-term bonds and short-term bonds for the same bond issuer, to the difference in the availability between long-term maturity CDS contracts and short-term maturity CDS contracts. We find that in the case of investment grade bonds, the presence of long-term CDS contracts in excess of short-term CDS contracts reduces the difference in yield spreads between long-term bonds and short-term bonds by 7 bps. This represents a significant 35% reduction if compared to the sample average. No significant effect is there for high yield bonds. This evidence further supports our argument that CDSs reduce yield spreads by increasing bond liquidity. Our study contributes to several strands of literature. First, our paper contributes to the emerging literature on the impact of CDS contracts on the underlying firm. The theoretical literature has mostly focused on the effects of CDSs on renegotiation between debtors and creditors, and the associated costs and benefits (e.g., Arping, 2004, Hu and Black, 2008a, b, Yavorsky, 2009, Bolton and Oehmke, 2011, Gormley et al., 2011), as well as the ensuing managerial incentive in risk taking (Thompson, 2010, Parlour and Winton, 2012). On the bright side, Duffee and Zhou (2001) argue that CDS allows for the decomposition of credit risk into components that have different sensitivities to information, thus potentially helping banks overcome the lemon problem by hedging credit risk. Jarrow (2010) argues that the trading of CDSs is welfare increasing because it facilitates a better allocation of risks. The empirical literature does not provide unequivocal evidence (e.g., Hull et al., 2004, Longstaff et al., 2005, Norden and Wagner, 2008, Norden and Weber, 2009, Chen et al., 2010, Ismailescu and Phillips, 2011, Kim, 2011, Nashikkar et al., 2011). On the one hand, CDSs are found to reduce loan quality, increase bankruptcy risk and lower the efficiency of the bond market (Ashcraft and Santos, 2007, Peristiani and Savino, 2011, Purnanandam, 2011, Subrahmanyam et al., 2011, Das et al., 2012). On the other hand, Bedendo et al. (2013) find that CDS contracts do not significantly increase the probability of bankruptcy when the firm is already in distress. They show that the access to credit 8

10 insurance does not favor bankruptcy over a debt workout. CDSs are also found to stimulate bank credit supply and improve borrowing terms, and enable firms to maintain higher leverage and borrow at longer maturities (Hirtle, 2009, Saretto and Tookes, 2011). One potential reason why the empirical findings in the literature are inconsistent is that the samples are often very different and in general relatively small. For example, Das et al. (2012), using a sample of 82 bonds to examine the impact of CDS initiation on bond liquidity, conclude that there is no definite evidence that CDS introduction improved the liquidity of the bonds underlying the CDS entity. From this perspective, to the best of our knowledge, our paper is the first one to provide largesample and consistent evidence that CDSs reduce fire sale risk, increase bond liquidity and reduce bond yield spreads for investment grade bonds. We further contribute to this literature by providing direct evidence of the bright side of CDS contracts on the bond market: CDSs facilitate liquidity provision and reduce fire sale risk. We show that this effect concentrates among investment grade bonds, in line with the fact that the bond market is heavily segmented and major investment grade bond investors are subject to the prudent-man rule or the risk-based capital requirement. From this perspective, our paper sheds a new light on the apparent contradiction between the dramatic growth in the CDS market over the last decade and the overwhelmingly gloomy view of the impact of CDS on the underlying firm. Second, we contribute to the literature on fire sales and financial crisis (Shleifer and Vishny, 2011, Bernardo and Welch, 2004, Coval and Stafford, 2007) and on fire sales in the bond market in particular (Da and Gao, 2009, Ellul et al., 2011). We contribute by focusing on the way CDS contracts reduces the market impact of such selling pressures on the bond market, improving liquidity and reducing yield spreads. Third, we contribute to the literature on the impact of bond illiquidity on corporate yield spreads. Bao et al. (2011) find that in the cross-section, bond illiquidity explains the individual bond yield spreads with large economic significance. Friewald et al. (2012) and Nielsen et al. (2012) confirm that illiquidity explains a large part of the variation in yield spreads across bonds after accounting for credit 9

11 risk, and the yield spread contribution from bond illiquidity increased dramatically during the period of the subprime crisis. We contribute by showing that the presence of CDS contracts facilitates the liquidity provision to investment grade bonds by reducing fire sale risk, which directly translates into significantly lower yield spreads. Our findings have important normative implications as well. Indeed, they show that the presence of CDS contracts at least for the class of investment grade bonds does in fact reduce the effect of fire sales and this benefit is especially stronger during the period of financial crisis. This evidence suggests that, contrary to the general media perception, CDS may actually help to reduce credit risk transfer and contagion around financial crises (Stulz, 2010, Jarrow, 2010). The remainder of the paper is organized as follows. In Section II, we describe the data, the construction of the main variables and present the univariate results. In Section III, we examine the role of CDSs in two fire sale related events. In Section IV, we present evidence on the link between the presence of CDS contracts and both yield spreads and bond liquidity. In Section V, we use an instrumental variable specification to establish a causal relationship. In Section VI, we perform additional tests on the impact of CDSs on yield spreads. A short conclusion follows. II. Data and Univariate Results A. Data We use data from multiple sources. The data on monthly bond yield spreads come from Bank of America-Merrill Lynch Corporate Master Index Compositions. The BofA-Merrill data have been used in previous studies (Schaefer and Strebulaev, 2008, Acharya et al., 2010). These data cover most rated US publicly issued corporate bonds (Acharya et al., 2010) and provide bond-level information on the option-adjusted yield spread, coupon rate, duration, face value and credit ratings. We require each bond to be included in the index for over 24 months. We obtain information on a number of bond characteristics such as the offering date, the maturity date, offering amount, seniority, callability, fungibility and credit enhancement from the 10

12 Mergent Fixed Income Securities Database (FISD). 14 This database reports the characteristics of US fixed income securities. We derive the data on quarterly institutional holdings of corporate bonds from Lipper s emaxx fixed income database from the first quarter of 2001 to the second quarter of It contains details of fixed income holdings for U.S. and European insurance companies, U.S., Canadian and European mutual funds and leading U.S. public pension funds. We merge the BofA- Merrill data with the Mergent FISD and the Lipper/eMAXX data using bond CUSIPs. We obtain information on the tick-by-tick bond transactions from the Trade Reporting and Compliance Engine database (TRACE) from 2002 to TRACE is the Financial Industry Regulatory Authority (FIRA) s over-the-counter (OTC) corporate bond market real-time price dissemination service. TRACE consolidates transaction data for all eligible corporate bonds - investment grade, high yield and convertible debt. It provides detailed records on the time of trade execution, price, yield and some information on trading volume. We get information on CDS contracts from the Markit CDS database. 15 This dataset is the main source of information used in many existing research on CDSs (e.g., Ashcraft and Santos, 2009, Qiu and Yu, 2012). The Markit CDS data are recently offered through WRDS and are expected to be widely used by academic researchers. It provides daily firm-level data on CDS spreads for the period from 2001 through The CDS spread is the periodic fee that the protection buyer pays to the protection seller in a credit default swap contract until the contract matures or a credit event occurs. 16 The Restructuring Clause of a CDS contract specifies the credit events that trigger the settlement. Typically, Markit reports a composite daily CDS spread, which is an average across all the quotes provided by market makers after a series of data cleaning tests. The Markit database also provides identifying information on the reference entity (such as firm name and ticker), the number of dealers 14 The FISD data used in our analysis are based on the 2009 edition of the FISD database. 15 Markit was founded in 2001 as the first independent source of credit derivative pricing and has rapidly grown both organically and through strategic acquisitions. Today, their data, valuations and trade processing services are regarded as a market standard in the credit markets. In 2009, Markit teamed with DTCC (The Depository Trust & Clearing Corporation) to create MarkitSERV which combines the DTCC Deriv/SERV and Markit Wire trade confirmation platforms to cover all major asset classes including credit, interest rate, equity and commodity derivatives. 16 In the latter case, either the protection buyer delivers defaulted bonds to the seller in exchange for the face value of the issue in cash (physical settlement) or the protection seller directly pays the difference between the market value and face value of the issue to the protection buyer (cash settlement). 11

13 providing CDS quotes, and the terms of the CDS contract (maturity, currency denomination and restructuring clauses). 17 We focus on the spreads of all the CDS contracts written on US firms and denominated in US dollars. Our final combined sample includes 158,122 bond-month observations (3,468 firm-year observations) from January 2001 to December We consider major bond characteristics such as the amount of bond issue outstanding, bond duration, coupon rate, callability, fungibility, credit enhancement, and detailed bond-level credit rating. 18 We consider major firm characteristics including equity volatility, equity beta, book size, market-to-book, book leverage, profitability, cash holding and dividend payments. The detailed definitions of each variable can be found in the Appendix. We focus on the effect of CDS contracts on bond yield spreads and bond illiquidity. We define CDS presence as a dummy variable equal to 1 if the issuing firm has quoted CDS contracts on its bonds in the previous month and 0 otherwise. We measure yield spread as the option-adjusted spread (OAS), defined as the number of percentage points that the fair value of the treasury spot curve is shifted to match the present value of the discounted cash flows to the bond s price. Following Bao et al. (2011), we define bond illiquidity as the implied bid-ask spread based on the auto-covariances of bond price changes: Bond Illiquity 2 γ (0 if γ 0, where γ Cov p, p, and p is the log price at time t. 19 Bao et al. (2011) show that the implied bid-ask spread γ is the most effective liquidity measures in explaining corporate yield spreads in the cross-section. They also show that the variation in aggregate bond liquidity is the dominant factor in explaining the time variation in bond indices for different ratings, and the relationship is stronger during the crisis period. 17 Specifically the maturity of CDS contracts ranges from 6 months up to 30 years, including 6-month, 1-year, 2-year, 3-year, 4-year, 5-year, 7-year, 10-year, 15-year, 20-year and 30-year maturity contracts. There are four major restructuring clauses (full restructuring, modified restructuring, modified-modified restructuring and no-restructuring). A detailed discussion of different restructuring clauses can be found in Packer and Zhu (2005). 18 In our study, to better control for the default risk of the bonds, rather than using the Standard and Poor s long-term credit rating from Compustat, we use the monthly updated bond specific composite rating directly obtained from the BofA-Merrill Lynch bond index database. The composite rating is calculated as the simple averages of ratings from Moody s, S&P and Fitch, with 21 levels ranging from AAA to C. We provide a detailed rating correspondence in the Appendix. In our sample, 3% of the bonds are AAA-rated, 7% of the bonds are AA-rated (AA1, AA2, AA3), 33% of the bonds are A-rated (A1, A2, A3), 38% of the bonds are BBB-rated (BBB1, BBB2, BBB3), 11% of the bonds are BB-rated (BB1, BB2, BB3), 5% of the bonds are B-rated (B1, B2, B3), and the rest 3% the bonds are rated below (CCC1, CCC2, CCC3, CC, C). Investment grade bonds refer to the bonds with bond composite rating above or equal to BBB3, while high yield bonds refer to the bonds with the bond composite rating below BBB3. 19 We use the tick-by-tick transaction data from TRACE to calculate the changes in bond prices. 12

14 There are other liquidity proxies proposed in the equity market, such as the Amihud illiquidity measure on the price impact of trading, the trading volume, or the zero-return measure based on the number of days with no price changes (Goyenko et al., 2008). However, many of them cannot be precisely constructed in the bond markets, as it is impossible to observe the exact amount of trading volume given that the information on trade size is truncated in TRACE. 20 Moreover, liquidity proxies such as trading volumes and non-zero return days are not suitable to study fire sale risk or liquidity characteristics upon fire sale events. For instance, during late 2008 when Lehman Brothers filed for bankruptcy, there were huge trading volumes and fewer zero-return days in individual stocks. Clearly, this was not an indication that the market or the stock was more liquid. 21 We provide the descriptive statistics of variables in Table I. For each variable, we report the data frequency, source, number of observations, mean and standard deviation. In our sample, the average bond yield spread is 2.12%, 1.54% for investment grade bonds and 4.92% for high yield bonds. The average bond illiquidity is 1.71, 1.62 for investment grade bonds and 2.27 for high yield bonds. On average, among the bond issuers, 74% have CDS contracts outstanding during the sample period. This fraction is higher among investment grade issuers (78%) than among high yield issuers (61%). Among other variables, the average amount of bond issue outstanding is $495 million, the average bond duration is 5.8 years, the average bond age since issuances is 4.5 years, 64% of the sample are callable bonds, 44% of the bonds are fungible and 9% of the bonds have credit enhancement. B. Univariate Results We now present some univariate results on yield spreads and bond illiquidity by bond rating and by presence of CDS contracts. We report the results in Table II, Panel A and Panel B. Panel A reports the results on bond yield spreads. We consider both the raw spread and the industry-adjusted spread calculated as the difference between the raw spread and the industry average 20 TRACE does not completely report the information on trade size. For investment grade bonds, trade size is reported if the par value of the transaction is less than $5 million; otherwise, an indicator denotes a trade of greater than $5 million. For below investment grade bonds, the reported trade size is truncated at $1 million. 21 Consistent with this argument, Bao et. al (2011) show that there is no meaningful connection between bond yield spreads and the percentage of non-trading days. 13

15 spread at the two-digit SIC level. We proceed as follows. First, for every rating level (21 rating levels, from AAA to C) in each month (107 months), we calculate the median bond yield spread (industryadjusted yield spread) for bonds with CDS contracts outstanding ( CDS presence subsample ) and for bonds without CDSs ( No-CDS presence subsample ). Then, we broadly classify the sample into Investment Grade subsample and High Yield subsample. 22 Under the investment grade category, we further classify the ratings into Strong (AAA, AA1, AA2 or AA3), Above average (A1, A2, or A3), and Average (BBB1, BBB2, or BBB3). Under the high yield category, we classify the ratings into Below Average (BB1, BB2, or BB3), and Weak (B1, B2, B3, CCC1, CCC2, CCC3, CC, or C). Next, for each of these categories, we provide t-tests to compare the differences in raw yield spreads (industry-adjusted yield spreads) between the CDS presence subsample and the no-cds presence subsample. We proceed in a similar way for bond illiquidity and report the results in Panel B. The results are interesting. We find opposite results for investment grade bonds and for high yield bonds: the presence of CDSs reduces bond yield spreads and bond illiquidity for investment grade bonds while it increases them for high yield bonds. The results are robust whether we focus on raw yield spreads (illiquidity) or on industry-adjusted yield spreads (industry-adjusted illiquidity). Investment grade bonds with CDS presence have on average 17 bps less yield spreads (12% less) than the ones without CDSs. The economic magnitude is equivalent to the effect of one notch improvement in credit rating. We find consistent results on bond illiquidity. Among investment grade bonds, bonds with CDS presence have 7% lower illiquidity than bonds without CDSs. For high yield bonds, the presence of CDSs increases the raw yield spreads by 38 bps (9% increase relative to the sample average). The effect shrinks to 10 bps and turns insignificant for the industry-adjusted yield spreads. These results suggests that the channel of impact of CDSs on the bond market, if there is any, works very differently for investment grade bonds and for high yield bonds, indicating that bond market segmentation 23 may be playing a crucial role with regards to the economic function of CDSs. 22 Investment grade refers to the rating levels from AAA to BBB3, while high yield refers to the rating levels from BB1 to C (the investment grade/high yield threshold is BBB3). 23 Here we refere to the fact that unlike high yield bonds, investment grade bonds are mostly held by institutions subject to risk-based capital requirements, and therefore are more subject to fire sale risk. 14

16 Moreover, within the investment grade category, the effects of lower yield spreads and lower illiquidity due to the presence of CDSs are concentrated among BBB and A rated firms rather than AAA and AA rated firms i.e., the ones that are most subject to fire sale risk. The univariate results imply that the economic function of CDSs in reducing fire sale risk and increasing bond liquidity for investment grade bonds can be a first-order effect and effectively dominates any potential negative effects related to the empty creditor problem or the reduction of debt holder monitoring. We next provide more direct evidence 24 by examining two events that have been shown to trigger forced sales by bond institutional investors. III. Event-based Analysis We now focus on two events that affect the behavior of institutional investors holding corporate bonds: the rating downgrade from investment grade status to high yield, and the selling pressure from Katrina-exposed property insurance companies following Hurricane Katrina. A. Fallen Angels We begin by focusing on the bonds that are downgraded from investment grade to high yield ( fallen angels ). Ellul et al. (2011) show that such downgrade triggers the forced sales of insurance companies, and generates large liquidity-driven negative effect on bond prices. And indeed, in line with their findings, we find that in our sample there is a significant 4% decrease in bond institutional ownership around the quarter of such downgrade (9% decrease relative to the ownership before the downgrade). In contrast, the average drop in bond institutional ownership around downgrades without crossing the investment grade/high yield threshold is only 1%. Our hypothesis predicts that downgrades from investment grade to high yield will induce a greater drop in institutional bond ownership, a higher increase in bond yield spreads and bond illiquidity for bonds without CDS contracts outstanding than for bonds with CDSs. We verify this prediction by 24 It is worth noticing that, a direct test of the channel of liquidity provision related to the CDS-bond basis arbitrage, is not feasible as the value of the basis is not defined in the absence of CDS. 15

17 regressing the changes in bond ownership, the changes in bond yield spreads and the changes in bond illiquidity around the month of rating changes on a fallen angel indicator, its interaction with a no CDS indicator and a set of control variables. For a given bond-month (bond-quarter), 25 we define the fallen angel indicator as a dummy equal to 1 in the month (quarter) in which the bond is downgraded from investment grade to high yield and zero otherwise. The no CDS indicator equals 1 if the bond issuer has no CDS contracts in the previous month (quarter) and 0 otherwise. We report the results in Table III. In Panel A, we focus on the changes in bond institutional ownership around rating changes, and in Panel B and Panel C on the changes in bond yield spreads and the changes in bond illiquidity, respectively. Columns (1)-(3) are based on the full sample of rating changes including both rating downgrades and rating upgrades. In column (1), we only interact the fallen angel dummy with the no CDS dummy. In column (2), we add the interaction terms of fallen angel dummy with bond characteristics including bond duration, offering amount and bond age. In column (3), we add additional interaction terms of fallen angel with risk characteristics such as equity volatility and equity beta. 26 Columns (4)-(6) follow the same specifications as columns (1)-(3), except that they are only based on the subsample of rating downgrades. The variable of focus is the interaction term between the no CDS dummy and the fallen angel dummy. We expect it to be negatively related to the changes in bond institutional ownership and positively related to the changes in bond yield spreads and the changes in bond illiquidity. And indeed, the interaction term is negative and significant for the changes in bond institutional ownership, and positive and significant for both the changes in yield spreads and the changes in bond illiquidity. This holds across different specifications. Bonds without CDS contracts experience a 150% higher drop in institutional ownership than bonds with CDS presence. A similar effect is there in the case of yield spreads and bond illiquidity. Bonds without CDS contracts experience a 200% (150%) higher increase in yield spread (bond illiquidity) than bonds with CDS presence. These findings strongly support our 25 The test on the change in bond ownership is at the bond-quarter level given the quarterly frequency in the institutional holdings data. The tests on the change in bond yield spreads and the change in bond liquidity are at the bond-month level. 26 The purpose of including these additional interaction terms is to eliminate concerns that the result on the interaction of the fallen angel dummy and the no-cds dummy may be driven by other bond or firm characteristics. 16

18 hypothesis that the presence of CDS contracts reduces the fire sale effect for investment grade bonds in the event of rating downgrades. B. Hurricane Katrina The second experiment is based on Hurricane Katrina (August 23-30, 2005) and the Katrina-exposed property-casualty insurance companies. Hurricane Katrina is the costliest natural disaster in the history of the United States, with total property damage estimated at $81 billion (2005USD) and almost $40.6 billion of insured losses (Knabb et al., 2005). 27 It represents a large exogenous shock to the property insurance and reinsurance industry, especially for the insurance companies with large business exposure to Katrina. 28 Given that insurance companies are the largest corporate bond holders, this provides an ideal experiment, in which the selling pressure of the bonds held by exposed insurance companies is not related to firm specific characteristics such as credit risk, but is driven by the market concerns on the forced sales by the affected insurance companies to meet redemption claims. Previous studies have shown that insurers stock prices decline in response to the losses due to hurricanes and the effect is particularly strong for insurers with more regional exposure (e.g., Lamb, 1995, 1998). Massa and Zhang (2011) show that the selling pressure from Katrina-exposed insurance companies induced an increase in the short-selling on the bonds held by those investors, and led to significant price drops up to seven months after the hurricane. In this context, we test how the presence of CDS contracts may help to reduce such impact on bond yield spreads and bond illiquidity. We use the pre-katrina exposed insurance bond ownership to proxy for the selling pressure of the bonds after the hurricane. 29 We proceed as follows. 27 A special report by Towers Perrin Co. (2005) studying the impact of Hurricane Katrina on the insurance industry estimates the range of privately insured loss to be between 40$ and 55$ billion. 28 Here by large exposure, we mean those insurance companies that have large market share of insurance business in the Gulf region (state of Mississippi, Alabama, and Louisiana). For example, State Farm Insurance, which has the largest market share in the Gulf region (26.62%), states the following words on its website: In a typical year, State Farm receives between 600, ,000 catastrophe claims. In 2005, we received that number in a six week period immediately following Katrina. Since then, nearly 100 percent of all claims have been resolved. In total, State Farm has paid more than $3.1 billion in claims as a result of Katrina, which does not include payments to policy holders from the National Flood Insurance Program. 29 We focus on the pre-katrina property insurance ownership for the following reasons. First, it is exogenous with respect to the changes in bond liquidity and yield spreads given the total unexpectedness of insured damages. Second, the economic rationale can be explained with a simple example. Suppose that Start Farm Insurance has 10 billion bond holdings before Katrina, invested in two bonds, 8 billion in bond A and 2 billion in bond B. For each bond, the total issue outstanding is

19 First, we identify the set of property and casualty insurance and reinsurance companies that are considered to have high exposure to Hurricane Katrina, using data from the Holborn Corporation (2005) s Hurricane Katrina report. 30 The Holborn Report lists the names of property & casualty (reinsurance) companies along with their 2004 market shares in the states of Louisiana, Mississippi, and Alabama, and whether they have rating or outlook changes immediately after the hurricane. We include the top ten property insurance companies by their market shares (including both personal and commercial lines) and eight reinsurance companies with negative rating outlook changes. The names of those insurance companies are provided in the Appendix. 31 Then, we define the pre-katrina exposed insurance bond ownership as the par amounts held by property and reinsurance companies with high exposure to hurricane Katrina at the end of the second quarter of 2005 divided by the amount of bond issue outstanding. Non-exposed bond ownership is defined as the difference between total institutional ownership minus the exposed insurance ownership. In our sample, the pre-katrina exposed property insurance ownership ranges from 0% (1-percentile) to 12% (99-percentile) of bond issue outstanding, with a mean of 1.3% and a standard deviation of 2.4%. Finally, we regress the changes in bond yield spreads and the changes in bond illiquidity around Katrina, on the pre-katrina exposed insurance ownership, a no CDS dummy as defined before, and the interaction term between them. Our variable of interest is the interaction term. Our hypothesis predicts a positive relationship between the interaction term and both the changes in yield spreads as well as the changes in bond illiquidity around Katrina. We report the results in Table IV. In columns (1)-(3), the dependent variable is the change in bond yield spreads from Aug 23, 2005 to Sep 9, 2005 (the two weeks during which Hurricane Katrina formed and fully dissipated). In columns (4)-(6), the dependent variable is the difference of bond billion. Therefore, before Katrina, State Farm s ownership in bond A is 8% and ownership in bond B is 2%. After Katrina, State Farm needs to immediately liquidate 5 billion to deal with insurance claims. Ideally, it would want to liquidate its bonds across the boards and keep the portfolio balanced. In this case, it should sell 4 billion in bond A and 1 billion in bond B. As a result, State Farm's ownership in bond A would drop from 8% to 4% and the ownership in bond B would drop from 2% to 1%. Therefore, the forced liquidation will have a much bigger impact on bond A than on bond B because of higher pre- Katrina ownership. In other words, higher exposed property ownership implies higher forced selling pressure after Katrina. 30 The Holborn report is publicly available at the URL: 31 It is worth mentioning that we exclude those bond issuers that may be directly affected by the hurricane, which include life, property insurance and reinsurance companies, and firms headquartered in the states of Louisiana, Mississippi, and Alabama. 18