CREDIT RISK EFFECTS IN CORPORATE FINANCE AND INVESTMENTS By CORBITT STACE SIRMANS A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF D OCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2014
Â© 2014 Corbitt Stace Sirmans
To my dear est Kimberly, Julian, and Ella Grace; and to my loving parents, Stacy and Elaine
4 ACKNOWLEDGMENTS I cannot express enough gratitude to my advisor, Dr. Andy Naranjo, and my co author, Dr. Jongsub Lee, for the incalculable number of hours and energy spent completing this project. I also thank Dr. David Brown and other faculty members of the Department of Finance, Insura nce & Real Estate at the University of Florida for their guidance and support throughout my time in the doctoral program. This dissertation has also benefited from the financial support of the University of Florida. This dissertation would not have been po ssible without the unwavering support of my wife, Kimberly, and my beautiful children, Julian and Ella Grace. I am also thankful to my parents, Stacy and Elaine, for teaching me that success requires dedication, passion and hard work.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 8 ABSTRACT ................................ ................................ ................................ ................................ ..... 9 1 INTRODUCTION ................................ ................................ ................................ .................. 11 Overview of C DS Momentum ................................ ................................ ................................ 11 Overview of The Exodus from Sovereign Risk ................................ ................................ ...... 13 2 CDS MOMENTUM ................................ ................................ ................................ ............... 15 Data and CDS Return Computation ................................ ................................ ....................... 21 CDS Momentum ................................ ................................ ................................ ..................... 29 Risk Adjustments and Firm Characteristics ................................ ................................ .... 32 Credit Rating Changes ................................ ................................ ................................ ..... 37 CDS Momentum and Stock Returns ................................ ................................ ....................... 41 Credit Rating Changes ................................ ................................ ................................ ..... 41 Momentum Spillover from CDS to Stock ................................ ................................ ....... 44 Final Thoughts ................................ ................................ ................................ ........................ 48 3 THE EXODUS FROM SOVEREIGN RISK ................................ ................................ ......... 61 Institutional Background on the CDS Market ................................ ................................ ........ 67 Hypothesis Development ................................ ................................ ................................ ........ 70 Data and Variables ................................ ................................ ................................ .................. 72 Sovereign Level Variables ................................ ................................ .............................. 72 Firm Level Variables ................................ ................................ ................................ ....... 74 Scaled Exposure and Extra Disclosure Variables ................................ ........................... 75 Main Results ................................ ................................ ................................ ........................... 78 Local Institutional and Informational Factors in Sovereign CDS Spreads ..................... 78 Determinants of Sovereign Ceiling Violations: Firm minus Sovereign CDS Spread ..... 82 Determinants of Sovereign Ceiling Violations: S&P Ratings ................................ ......... 89 Determinants of Sovereign Ceiling Violations: Probit and Ordered Prob it Analysis ..... 91 Information Spillovers from CDS Markets to Credit Ratings ................................ ......... 93 Concluding Remarks ................................ ................................ ................................ .............. 96 4 CONCLUSION ................................ ................................ ................................ ..................... 111 LIST OF REFERENCES ................................ ................................ ................................ ............. 113
6 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 119
7 LIST OF TABLES Table page 2 1 Sample statistics ................................ ................................ ................................ ................. 50 2 2 CDS Mome ntum. ................................ ................................ ................................ ............... 51 2 3 CDS momentum returns: size, rating and depth ................................ ................................ 52 2 4 CDS momentum returns and credit rating changes. ................................ .......................... 53 2 5 CDS momentum and momentum elsewhere. ................................ ................................ ..... 54 2 6 Moment um spillover from CDS to stocks ................................ ................................ ......... 55 3 1 Summary statistics ................................ ................................ ................................ ............. 99 3 2 Determ inants of sovereign credit risk ................................ ................................ .............. 100 3 3 CDS spreads and multinational firms ................................ ................................ .............. 101 3 4 Firm sovereign CDS spread difference ................................ ................................ ........... 102 3 5 Firm sovereign CDS sprea d difference: robustness checks ................................ ............. 103 3 6 Firm sov ereign credit rating difference ................................ ................................ ........... 104 3 7 CDS sovereign ceiling vio lations and multinational firms ................................ .............. 105 3 8 CDS versus S&P rating: lead/lag predictions in sovereign ceiling violations ................. 106
8 LIST OF FIGURES Figure page 2 1 CDS spreads and credit rating changes ................................ ................................ .............. 56 2 2 CDS momentum returns ................................ ................................ ................................ .... 57 2 3 Sample composition ................................ ................................ ................................ ........... 58 2 4 Time series of CDS spreads, CDS returns and stock returns ................................ ............. 59 2 5 CDS to stock momentum ................................ ................................ ................................ .. 60 3 1 Global sovereign CDS spreads ................................ ................................ ........................ 107 3 2 Global sovereign ceiling violations (SCVs) in the CDS market ................................ ...... 108 3 3 Sovereign CDS spreads and country instit utional characteristics ................................ .... 109 3 4 Sovereign ceiling violations (SCVs): CDS precedes S&P ratings ................................ .. 110
9 Abstract of Dissertation Presented to the Graduate School o f the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy CREDIT RIS K EFFECTS IN CORPORATE FINANCE AND INVESTMENTS By Corbitt Stace Sirmans August 2014 Chair: Andy Naranjo Major: Busin ess Administration My research examines credit risk issues in two important topics in finance: corporate finance and investments. The purpose of this study is to empirically test determinants of credit default swap (CDS) spreads in both a temporal and cro ss sectional sense. In doing this, I address questions regarding the financing of multinational firms, the influence of credit rating s on asset prices and the relationship between equity and fixed income markets. In CDS Momentum , I examine how credit ratin g changes induce return momentum in the CDS market. Using a large sample of CDS contracts from 2003 2011, I find that an investment strategy based on momentum in CDS returns yields a 52 bps return that is highly statistically significant. Further, I find t hat these momentum returns are closely related to information induced trading of CDS contracts and credit rating changes. This CDS momentum effect spills over to the stock market and can be used to enhance traditional stock market momentum investment strategies. In T he Exodus from Sovereign Risk , using a large sample of CDS contracts, I empirically test how corporate credit risk is influenced by linkages to stronger or weaker external institutional a nd information al environments. Firm credit risk is generally considered to be capped at the sovereign level . However, I find that firms can
10 move beyond the level of its sovereign by increasing foreign exposure to strong er institutional and informational environments . That is, by moving assets to countries with stronger property rights protection and cross listing equity on high disclosure foreign stock exchanges, a firm can reduce its credit risk relative to its sovereig n. Further, I find that credit ratings strictly follow the sovereign ceiling rule and do not consistently take into account the importance of a firm exposure to stronger institutions.
11 CHAPTER 1 INTRODUCTION My dissertation utilizes credit default swap (CDS) data to address two important topics in corporate finance and investments. The CDS market has had extraordinary growth over the last two decades and played a controversial role in the recent credit crisis. CDS Momentum provides insight into determina nts of CDS contract returns , the impact of information and liquidity on CDS returns , and the relationship between returns in the CDS market and the stock market. The recent increase in global sovereign risk has triggered concerns over how sovereign risks w ill be transferred to the private sector. T he Exodus from Sovereign Risk outlines specific channels through sovereign risks are transferred to corporations and how firms use these channels to limit sovereign risk exposure. Overview of CDS Momentum Innova tion in financial products over the last several decades has sparked the creation of many large security markets, such as the CDS market, which has become one of the most popular among credit derivatives. Although the CDS market has reached an economically relevant $26 trillion dollars in gross notional value, much has yet to be discovered regarding its own unique return properties and, importantly, its relation and influence on other asset classes. Not only does the CDS serve as a risk management tool allo wing for re allocation of credit risk among a specific set of investors, it provides an information channel and signaling function to the entire market. CDS Mome ntum explores the asset pricing characteristics of the CDS market. Using a sample of 5 year CDS contracts of 1,247 U.S. firms from 2003 to 2011, I test for one of the most persistent and pervasive asset pricing anomalies: return momentum. Despite the role of more sophisticated institutional market participants in the CDS market, I find that recent p erformance
12 positively predicts future performance a three month formation and one month holding period CDS momentum strategy yields a monthly return of 52 basis points with a Sharpe ratio of 0.423. Performance is best among lower grade entities (83 bps per month) and, notably, higher depth CDS contracts (97 bps per month). High depth contracts are considered to be the most informative. Given that CDS momentum is driven by informed contracts, what is the nature of the information? I find that momentum profit s are almost entirely due to correct anticipation of future credit rating changes by the CDS market; that is, the winner (loser) portfolio is driven by firms undergoing rating upgrades (downgrades). There is significant return run up (drop off) leading up to a rating upgrade (downgrade), and then positive (negative) returns in the month of the announcement. This mechanism works particularly well among high depth contracts. Furthermore, the cross market tests show that this information channel is relevant to stock momentum. By incorporating CDS return information into the portfolio formation process, the traditional stock momentum strategy avoids abrupt losses during the financial crisis period and improves its performance by a net of 104 bps per month. In th is joint market momentum strategy, CDS returns, particularly those of high depth contracts, assist stocks in more accurately predicting important future rating change events. I make several important contributions to the credit risk/credit rating literatur e and the literature on the momentum anomaly. First, I am the first to document the existence of the momentum anomaly in the CDS market, extending the literature that documents that momentum exists everywhere. Second, I prove that the underlying mechanisms of CDS momentum and corporate bond momentum are completely different and illustrate a close relation between CDS momentum and significant predictability of past CDS returns on future rating changes in
13 anticipated di rections. By doing so, I contribute to t he literature on the informational relation between CDS markets and credit rating agencies that document a viability of market CDS rates as an alternative credit ris k benchmark to credit ratings. In addition, the channel of CDS momentum highlights how slug gish moves by major ratings agencies in assigning corporate credit ratings could deter the capital market efficiency even in the market that is perceived to be more informationally efficient. Third, I am the first to document a significant cross market mom market interaction in the momentum anomaly. Finally, I highlight the role of CDS depth in explaining significant predictability of CDS returns on future rating changes. This depth effe ct recalls the recent findings on endogenous liquidity in CDS markets more information flows from the CDS to other asset classes when the liquidity of a CDS contract is high. Overview of The Exodus from Sovereign Risk The role of government in the private sector is an important topic of debate and has historically been a great source of political contention. Both the established institutional and legal structure of a country as well as the ongoing government financing and regulatory decisions made by politi cians affect the risk environment of firms in the private sector. Firms in countries with relatively stronger property rights institutions worry less about whether financial contracts will be upheld or whether its government will expropriate private assets . Recent studies have In the aftermath of the financial crisis, global sovereign risk increased dramatically, exacerbating worries about how sovereign risk would disseminate to the private sector. T he Exodus from Sovereign Risk examines the extent to which firms delink themselves from their its home country. Using CDS data for a large sample of firms, I show that the geographic
14 Specifica lly, I find that a firm with foreign assets in a country with relatively stronger property rights than its home country has a lower CDS spread relative to the sovereign spread, ceteris paribus. Similarly, if a firm has equity listed on a foreign stock exch ange with relatively stricter disclosure requirements, then the firm has a lower CDS spread relative to the sovereign spread, ceteris paribus. The credit in the private sector, bu t it also identifies two important channels through which this relationship exists. In addition to the broad analysis described above, I take a closer look at firms that violate on more favorable terms than that of the government. The rationale behind this rule is that the government is considered the ultimate claimholder; that is, in theory, the government will extract wealth from the private sector until its debt obligations are met and, therefore, will be the very last one to default. The sovereign ceiling rule is widely followed by credit rating agencies and nearly perfectly held true in the CDS market before the financial crisis. However, when sovereign CDS spreads rose dramat ically during the crisis, there were a significant number of instances of wherein firm CDS spreads fell below the equivalent sovereign spread . In contrast, during this same time period, very few sovereign ceiling violations occur r ed in the credit ratings, and when one did occur, it was long anticipated by the CDS market. I find that firms that experienced a sovereign ceiling violation in the CDS market often had assets in countries with relatively stronger property rights and equity listed in foreign exchan ges with relatively stricter disclosure requirements.
15 CHAPTER 2 CDS MOMENTUM Related markets often reveal information non synchronously, with different frequencies, speeds, and content. An important question is to w hat extent does a more informa tionally e fficient market trading alongside a related sluggish, but material, information market generate market anomalies? A case in point is the credit default swap market (CDS) and credit ratings where several recent papers document informational advantages with CDS at market spreads being more timely and often predicting credit ratings (Hull, Predescu, and White, 2004; Flannery, Houston, and Partnoy, 2010; Chava, Ganduri, and Ornthanalai, 2012). In fact, Figure 1 shows that CDS spreads react at least 90 days prio r to subsequent rating changes. At the same time, credit rating changes have material impacts, including changes in accessibility to capital, investor bases, supply chain relationships, ability to retain key executives, and/or disclosure requirements (Klig er and Sarig, 2000; Dichev and Piotroski, 2001; Kisgen, 2006, 2007). unique structure with trades being brokered by bulge bracket investment banks who are well informe d on overall capital markets. Branching into the information content of CDS contracts, others have also recently examined the cross market informational linkages from CDS to equity markets (Acharya and Johnson, 2007; Ni and Pan, 2010; Han and Zhou, 2011; Q iu and Yu, 2012). These studies generally find significant information flowing from the CDS market to the stock market, particularly for entities with negative news, more informed insiders such as relationship banks, a nd high CDS contract liquidity. Severa l studies using CDS market data are also able to reconcile well known asset pricing puzzles such as a distress puzzle (Campbell, Hilscher, and Szilagyi, 2008) that is documented mostly with traditional default risk measures or corporate bond yields (Friewa ld, Wagner, and Zechner, 2012). These documented findings
16 suggest that CDS markets provide useful information for understanding various market inefficiencies. Motivated by the information content in CDS contracts, we examine the extent to which a more info rmationally efficient CDS market trad ing alongside sluggish, but ma terial, credit ratings generates market anomalies and cross market spillover effects. We focus our tests on the return momentum anomaly and address two important questions. First, does mome ntum exist in CDS returns? Second, are there spillover effects between CDS and stock return momentum? Despite the impor tant risk management and infor mation signaling functions played by the CDS market, no previous study has tested for the existence of CDS return momentum. Given the sophisticated participants in the CDS market, one might expect that CDS mo mentum profits would not exist. At the same time, ratings changes, though slow moving, have material impacts (Kisgen, 2006, 2007) and create potential mark et frictions (Manso, 2013) with a more informationally efficient CDS market. These rating change impacts can affect the information transmission effi ciency of both CDS and stock markets, creating significant market efficiency distortions and potential inve stment strategy opportunities. The pervasive finding that return momentum exists across a range of asset classes, markets, and over time is an important phenomenon that has received substantial atten tion by financial economists (Jegadeesh and Titman, 1993, 2001, 2011). Recent work by Asness, Moskowitz, and Pedersen (2013) further documents strong co movement in return momentum across various asset classes and markets. A puzzling result in the recent return momentum literature is that while stock and bond ma r kets each exhibit return momen tum, their contemporaneous co movement is surprisingly low for these related securities (Asness, Moskowitz, and Pedersen, 2013; Jostova, Nikolova, Philipov, and Stahel, 2013). An additional
17 aspect of this puzzle is that momen tum returns do not spill over from past bond to stock returns, suggesting that their time series behavior evolves differentially over time (Gebhardt, Hvidkjaer, and Swaminathan, 2005; Jostova, Nikolov a, Philipov, and Stahel, 2013). In contrast, Gebhardt, H vidkjaer, and Swaminathan (2005) find that return momen tum spills over from past stock to bond returns and provide evidence that such momentum spillovers stock r eturns predict bond rating upgrades (downgrades) in the future. They argue that both equity and bond returns unde rreact to common firm fundamen tals, but past equity returns being a better proxy for firm fundamentals than past bond returns predict future cr edit rating changes that could f urther introduce unforeseen ma terial impacts on firm fundamentals (Kisgen, 2006, 2007). Similar sluggish corporate rating responses to available market information have also been associated with various equity return anomali es such as return momentum, earnings recasts, idiosyncratic volatil ity, and capital investments (Avramov, Chordia, Jostova, and Philipov, 2012). Disentan gling the sourc es that lie behind these stylized return momentum findings is a formidable challenge that many empirical and theoretical researchers continue to investigate. Using 5 year CDS contracts on 1,247 US firms from January 2003 to December 2011, we document the f ollowing results. First, we find significant CDS return momentum. A three month formation and one month holding period CDS return momentum strategy yields 52 bps per month. The performance is better for entities with lower credit ratings (83 bps per month for junk grade entities) and high CDS depth (80 bps per month for highest depth quintile). As seen in Part A of Figure 2 , a $1,000 investment in this strategy starting in 2003 grows to approximately $1,700 in 2011, with a Sharpe ratio of 0.423. Given Sharp e ratios of 0.29 and 0.33, respectively,
18 for U.S. equity value and momentum strategies in the post 1972 period (see Table 1 of Asness, Moskowitz, and Pedersen, 2013), these CDS momentum profits are sub stantial even on a simple risk adjusted basis. The stra tegy also tends to perform better during the crisis period (97 bps per month for the period of July 2007 to April 2010; see Part B of Figure 2 ). We further show that CDS momentum returns are robust to using various formation and holding periods, controllin g for fundamental risk factors from both equity and bond markets, and are also profitable on a transaction cost adjusted basis. Finally, we show that CDS momentum profits are generated from incremental information in past CDS returns above and beyond infor mation possessed in past stock returns. What are the sources of these CDS momentum profi ts? We find that the CDS momen tum profits are almost entirely due to correct anticipation of future credit rating changes by the CDS market following the formation of t he momen tum strategy. That is, the win ner (loser) portfolio is driven by firms who undergo rating upgrades (downgrades) over a six month horizon. We find a significant cumulative return run up of 221 bps (run down of 324 bps) prior to the rating upgrade ( downgrade) over the six month horizon, which is further followed by a return of 44 bps ( 133 bps) in the month of the announcement of the rating upgrade (downgrade). This credit ratings and high CDS depth, which are arguably most demanded by informed traders in this market. This credit rating changes channel further distinguishes the sources of CDS momentum returns from those of corporate bond momentum returns. Jostova, Nik olova, Philipov, and Stahel (2013) find that corporate bond momentum returns exist even after they eliminate bond month observations from 12 months prior to 12 months after a credit rating change.
19 Second, we find significant cross market momentum spillover s from CDS markets to stock markets. We show that a CDS to stock cross market momentum strategy yields 52 bps per month and that the same strategy yields greater profits for entities with lower ratings (137 bps per month for junk grade entities) and entiti es with high CDS depth (81 bps per month). These results suggest that there is greater information flow from the CDS market to the stock market, particularly for entities that attract the highest hedging demand for underlying credit risk and also contracts with greater trade quotes. Importantly, we show that incremental information i n CDS returns significantly im proves the performance of traditional stock to stock mom entum strategies. That is, tra ditional stock to stock momentum strategies improve by net 10 4 bps per month when we create sharper signals by combining the information in past CDS and st ock returns. As seen in Figure 5 , this joint market momentum strategy generates persistent and robust profits even during the crisis period when traditional stock only signals incur abrupt and very significant losses (Daniel, 2011; Daniel, Jagannathan, and Kim, 2012). Moreover, our joint market momentum stock trading strategy is al so profitable and further mini mizes transaction cost adjusted return effects (Frazzin i, Israel, and Moskowitz, 2012). We again document a close relation between this enhanced performance of the joint market momentum strategy and the significant predictability of the marginal joint market momentum strategy, CDS returns assist stock momentum strategies in more accurately predicting future rating changes and ensuing stock momentum returns. Overall, our results suggest that greater information signal ing in the CDS market, together with sluggish updates on corporate credit ratings assigned by major rating agen cies,
20 creates anomalies such as return momentum within the CDS market and across CDS and stock market momentum effects. We make several important contributions to the return momentum and credit risk literature. In particular, we are the first to document the existence of CDS return mo mentum, which extends the literature documenting the existence of return momentum across various asset classes to an economically important and growing market (Okunev and White, 2003; Pirrong, 2005; Menkhoff, Sarno, Schmeling, and Schrimpf, 2012; Asness, Moskowitz, and Pedersen, 2013). Importantly, we show that the underlying mechanisms of CDS and corporate bond return momentum (Jostova, Nikolova, Philipov, and Stahel, 2013) are different and provide evidence of a link between CDS return momentum and the predictability of past CDS returns on future rating to the literature on the informational relation between CDS markets and credit rating agencies, which documents the effectiveness of market CDS rates as a potential alternative credit risk benchmark to credit ratings (Hull, Predescu, and White, 2004; Nord en and Weber, 2004; Acharya and Johnson, 2007; Flan nery, Hous ton, and Partnoy, 2010; Chava, Ganduri, and Ornt hanalai, 2012; Qiu and Yu, 2012 ). Relatedly, our CDS return momentum channel highlights how sluggish moves by major ratings agencies in assigning c orporate credit ratings influ ences capital market efficiency, even in a market that is arguably informationally efficient. We are also the first to document a significant cross market return momentum spillover from CDS to stock markets, extending the liter ature on relate d asset cross market in teraction effects in momentum returns (Gebhardt, Hvidkjaer, and Swaminathan, 2005). Finally, we highlight the role of CDS depth in explaining significant predictability of CDS returns on future rating changes and also the incremental information content in CDS returns over stock returns.
21 This depth effect is consistent with the recent findings on endogenous liquidity in CDS markets, whereby more information flows from the CDS to other asset classes when the liquidity of a CDS contract is high (Qiu and Yu, 2012). The remainder of this paper is organized as follows: Section 2 describes our data and CDS return construction process. We present the results of CDS momentum in Section 3 and the results of both cross and joint Section 5 concludes. Data and CDS Return Computation We obtain data from a variety of sources. CDS data are acquired from the Markit Group, a leading financial information services company. The sample covers 1,247 publicly held U.S. companies from January 2003 to December 2011 for which an active single name CDS contract is traded. There are 667 firms in the sample in 2003, and this number quickly rises to more than 1,000 firms by 2007. All CDS c ontracts have a five year maturity and are denominated in U.S. number of dimensions, including a move from Modified Restructuring (MR) to No Restructuring (XR) for N orth American CDS contracts. As such, our database consists of MR contracts prior to input from a variety of market makers and ensures each daily observation passes a rigorous cleaning test to ensure accuracy and reliability. Table 2 1 provides summary statistics and a correlation matrix of variables used in this study. The mean spread of CDS contracts in our sample is approximately 202 basis points. As a measure of li quidity, Markit reports on a daily used to construct the composite spread. Markit requires a minimum of two contributors. The mean depth of our sample is 6.396.
22 E quity data are from the Center for Research in Security Prices (CRSP). We require the NASDAQ. Monthly equity returns are compounded from daily returns provided by CR SP to match the exact holding period of the CDS contract. Delisting returns from CRSP are used in the event that a stock is delisted. The mean monthly stock return of our sample is 0.396%. Firm size is measured through equity market capitalization and is c omputed as the number of common time of portfolio formation are excluded. The mean market capitalization of our sample is $20.4 billion. S&P credit ratings are available through Compustat. We convert the rating into a represent s larger firms in the universe of stocks and is fairly normally distributed across credit ratings. We measure the relative size of our firms by computing decile ranges of equity market capitalization for all stocks in the NYSE/AMEX universe and assigning a corresponding size decile to each firm in our sample (e.g., a size decile value of 10 indicates the firm falls in the highest decile of firms in the NYSE/AMEX universe). Figure 3 portrays the size decile and credit rating distribution of firms in our sam ple. Part A is a histogram of the size decile distribution. More than 50% of our firms fall in the highest size decile, and less than 5% are in the smallest five deciles, suggesting that CDS contracts generally trade on large firms and that our results are unlikely to be driven by small firms. The mean size decile is 9.229. Part B of Figure 3 is a credit rating histogram that shows that our sample appears to be normally distributed across ratings. A little more than a third of the firms
23 are junk grade (i.e. , below BBB ). Furthermore, Part C of Figure 3 shows that our sample of firms has reasonable representation across industries. A histogram of the CDS depth distribution is also provided in Part D of the same figure. CDS Return Computation: A single name CD S contract is designed to transfer the c redit risk of an underlying en tity (often called reference entity) from a protection buyer to a protection seller. The protection buyer pays a periodic coupon to the seller, wh ich is also called the CDS pre mium, unti l the maturity of the contract or a credit event (also called a trigger event), whichever comes first.6 Coupons are usually paid quarterly on the 20th day of March, June, September, and December, following an Actual/360 day count convention. These quarterl y coupon payment dates of a CDS are called the IMM dates inspired by the IMM dates in the interest rate futures markets, though the IMM dates of a CDS are not al ways the same as the corresponding IMM dates in the interest futures markets the third Wednesda y of the month. If a credit event occurs during the life of a CDS, the protection seller compensates the buyer by therefore, the protection seller bears the full loss given default. To compute the CDS holding period (excess) return, we need to compute the prof its/losses (P&L) of a CDS over a given holding interval. The P&L of a CDS trading with a unit $1 notional is what we term the CDS holding period excess ret urn. This notion of a CDS holding period excess return is consistent with Berndt and Obreja (2010). They view the protection bonds issued by the same reference entity. Hence, the pro levered risky par bond position that is financed at the default free riskless rate, which serves as
24 the basis for We interchangeably use the terms CDS holding perio d excess return, CDS holding period return, and CDS return throughout this manuscript. We illustrate the computation steps of the P&L of a CDS as follows: First, we provide a standard CDS pricing model as in (2008). Then, through this pricing frame work, we define the mark to market value of a CDS with a unit $1 notional using at market spread quotes from Markit. The change of these mark to market values over a given holding period determines the CDS holding period return. We split the pricing of a CDS contract with a unit $1 notional into two legs, the premium leg and protection leg. To simplify our illustration, we assume that we are on the inception date of a 5 year CDS. This fresh 5 year contract matures on the first IMM date 5 years after the tr ade date. For example, a 5 year CDS contract trading on 1 2/20/2013 matures on 3/20/2018. The premium leg has two components. First, there are 21 scheduled premium payments on a quarterly cycle with the CDS IMM dates the 20th of March, June, September, and December until the maturity date as long as the reference entity survives. When there is a credit event, there is a payment of the premium that has accrued since the last quarterly premium payment date. This is the second component of the premium leg. Let us denote the quarterly premium payment dates over a 5 year horizon by , = 1, 2, . . . , 21, and let denote our valuation date. Given the quoted spread of at time , the present value of the first comp onent of the premium leg is displayed as , (2 1) where denotes the accrual factor for the time period, , and and , respectively, denote the survival probability of the reference entity and default free discounting factor for the time period,
25 Now, we consider the premium accrued at default for the th premium period, . Over an infinitesimal time interval, for , the expected p resent value of the premium accrued upon default is given by . (2 2) Then, the value of the premium accrued upon default for all 21 premium periods is given by . (2 3) By summing up Eq. (1) and (3), the present value of the premium leg becomes (2 4) where is given by (2 5) The integration in the second term in Eq. (5) can be approximated as (2 6) Thus, we have (2 7)
26 Assuming a constant loss given default, , together with the standard assumption of independence of interest rate and the default time, we can write the present value of the protection leg as (2 8) The second line shows that the integration in the first line is performed by discretizing the 5 year horizon by 21 intervals with each coupon payment date. Combining the present values of the premium and the protection legs gives the mark to market value of a 5 yea r short protection position of a CDS with a unit $1 notional as (2 9) where is as in Eq. (7). With the quoted spread, , as required. However, soon a fter the inception of trading, this requirement is no longer true since the market spread of the CDS reference entity moves from the spread that the protection seller/buyer are locked in. Finally, with this pricing framework, we can easily define the P&L o f a CDS with a unit $1 notional over a holding period, . For simplicity, we assume for a moment that this interval is short enough so that we can ignore any coupon flows and also potential credit event during this holding period. If we entered as a seller of a protection at time and unwind the position at time by buying a protection on the same reference entity and the same maturity date, then the CDS holding period excess return is given as
27 (2 10) in the above Eq. (10) denotes the time value of the protection we sold at time , and the time cost to purchase the protection on the same reference entity with the same maturity date. If there is a credit event over our holding period, then the realized return will be equal to where is a realized recovery rate upon the credit event. Eq. (10) does not take into account coupon flows during our holding period and the accrued premium that should be exchanged at each selling and buying transaction of the default protection. We carefully incorporate these factors when we implement this CDS return framework using the quoted spreads from Markit. The U.S. $Libor curve retri eved from DataStream is used to construct the default free discounting factor. After all these considerations, As illustrated above, we approach CDS returns from the perspective of the protection seller (i.e., a negative return corresponds to an increase in credit risk). We do monthly trading of 5 year contracts. To compute the CDS return, we need to construct the survival probability curve on a given valuation date , Q( , ), for = 1 , 2, ..., 21. Instead of bootstrapping this survival probability curve using the quoted spreads of CDS contracts across entire maturity groups, we assume a flat hazard rate, for , and calibrate the hazard from the quoted spread of a 5 year CDS contract. Our monthly trading of a CDS starts at midnight of the 19th of a given month and ends at midnight of the 19th of the next month, and therefore, this trading timeline e nsures us to trade only 5 year contracts with a fixed maturity date in and ou t of each trade. With this trading timeline, we do not have to be concerned about the violation of no arbitrage in contract prices across different maturity groups that could exis t due in part to our flat hazard rate assumption. We trade only 5 year contracts and, therefore, our
28 trading strategy returns are immune from this potential cross sectional pricing inconsistency in CDS prices. By trading only fresh 5 year contracts, we can further ensure that the liquidity concerns of our trading returns are minimal, which is another potential benefit of our trading timeline. There is no intermediate coupon income for this protection seller during a one month holding period. All the coupon values are fully embedded in It is possible that credit events occur during our monthly holding period. If a credit event occurs, we need to assign the realized loss given default to the CDS return for that holding period. We find that 40 firms in our sample exp erience credit events during our sample period. We use the realized recovery rate information that we compiled from the Creditex Group and make appropriate adjustments in our holding period excess returns. 15 In untabulated robustness checks, we further co nfirm the robustness of our results to these credit event concerns by exclusively focusing on the firms that survive throughout our whole sample period. These results are available on request from the authors. As explained in the previous Subsection 2.1.1, an accrued premium should be adjusted when the CDS trade occurs in between the quarterly coupon payment interval. There are two different ways the accrued premium payment is handled during our sample period. Before the premium accrued since the trade date was paid either on the next immediate coupon date (i.e., short stub) or the following coupon date (i.e., long stub), depending on the trade date. If the trade date fell within a 30 day window prior to the first upcoming coupon date, it would follow the long stub rule and the short stub otherwise. However, post now the single name CDS contracts trade just like the Markit CDS indices where the new protection seller will receive the full quarterly coupon on each coupon payment date, and any
29 paid premium to this seller by the protec tion buyer is rebated upfront. We follow this post m to get the clean P&L 100 bps/500 bps coupons with upfront adjustments. Since the default free rate during our sample period is relatively low, the potential erro rs in our treatment of the stub algorithms should be minimal for CDS returns in the pre Table 1 provides summary statistics on CDS returns for the firms in our sample over the years 2003 2011. The average CDS return over our time period is 0.018% with a standa rd deviation of 3.947%. Figure 4 displays the average CDS spread (top figure) as well as the time series of CDS and stock returns (bottom figure). Table 2 1 also shows that CDS returns are highly correlated with stock returns; howeve r, the CDS returns lead the stock returns, particularly during the recent financial crisis period. This pattern suggests a potential information advantage embedded in CDS returns under certain conditions. CDS Momentum Our CDS momentum trading strategy is c onstructed in the spirit of Jegadeesh and T it man (1993). Firms are sorted at the end of each month into five equally sized portfolios, P1 to P5, based on their CDS return over the past J months (referred to as the formation period). Formation periods longe r than one month are compounded from one month re turns. We write default protection on firms in P5 that have the highest returns (winners) and acquire default protection on firms in P1 that have the lowest returns (losers). All positions are unwound afte r K months (referred to as the holding per iod) and equally weighted returns are computed for each momentum portfolio. Similar to previous studies (Jegadeesh and Titman (1993), among others), returns greater than one month are con structed to avoid overlappi ng returns.
30 Table 2 summarizes momentum profits in the CDS market over the years 2003 2011. We first document the profits of a weekly momentum strategy. Though we primarily trade monthly, in this exercise we separately compute weekly excess returns of trad ing a 5 year CDS contract. The long/short return using a weekly formation period is 0.193% per week with a t statistic of 2.99, which implies short term momentum instead of a reversal. Therefore, unlike Jegadeesh and Titman (1993) who adjust the formation period to avoid short term reversals in the stock market, our CDS momentum strategy can be implemented without a one month gap between the formation and holding periods. For these reasons, we use contiguous formation and holding periods for the remaining m omentum strategies implemented in this paper. Table 2 also presents CDS momentum returns using various other formation periods J one month, three months, six months, nine mo nths, and 12 months. The three month formation period provides the largest momentum profits, producing a one month return of 52 bps (= 0.520% per month, or 6.42% on an annual basis) with a t statistic of 3.87 and a Sharpe ratio of 0.423. The loser portfolio P1 produces an equally weighted monthly return of 0.193%, and the winner portfol io P5 produces a return of 0.327%. While momentum profits are generated from both P1 and P5, the return on P5 (t statistic of 2.52) is statistically more significant than the return of P1 (t statistic of 0.86). CDS momentum profits are also significant f or holding periods K longer than one month. For example, the strategy using a three mont h formation period (J = 3m) gen erates future three month (K = 3m) and six month (K = 6m) average monthly returns of 0.467% and 0.310%, respectively. However, the magnit ude of the average monthly momentum return becomes weaker as the holding period K is extended.
31 Momentum returns exist for longer formation periods up to nine months. The monthly returns using formation periods of six (J = 6m) and nine (J = 9m) months are 0 .443% and 0.412%, respectively. They are statistically significant at least at the 5% level. When the formation period is extended to 12 months (J = 12m), the momentum return becomes relatively weak both economically (0.351% monthly return) and statistical ly (t statistic of 1.62). Also provided in Table 2 2 are the average size decile, rating, and depth characteristics for each portfolio. Recall that the size decile is measured relative to the equity market capitalization of all firms in the NYSE/AMEX unive rse, the reported S&P rating is a numerical representation of the rating, and CDS depth is a measure of CDS liquidity provided by Markit. The extreme portfolios (P1 and P5) generally contain smaller and less creditworthy firms. On average, the winner portf olio (P5) contains slightly larger firms than the loser portfolio (P1), significantly different at formation periods of six, nine, and 12 months. P5 contains riskier firms than P1, a difference that is statistically significant at all formation periods. Av erage depth shows that firms in P1 have more liquid CDS contracts than firms in P5. This difference is statistically significant at all formation periods except for the one month formation period (J = 1m). Depending on the formation period (J) length, each momentum portfolio contains between 130 and 150 firms, on average. In Subsection 3.1 later, we show that CDS momentum is robust to these size, rating, and depth characteristics, and we uncover an important credit rating and CDS depth interaction role in g enerating substantial momentum profits. In Table 2 2, we also present the performance of the three month momentum strategy (J = 3m) during different time periods. The pre crisis period spans January 2003 to June 2007, the crisis period spans July 2007 to A pril 2010, and the post crisis period spans May 2010 to December 2011. One month holding (K = 1m) long/short strategy returns are positive and
32 statistically significant a t the 5% level in all three sub periods. The magnitude is greatest during the crisis p eriod with average returns of 97 bps (= 0.971%) and moderate during the pre crisis and post crisis periods with returns of 0.228% and 0.640%, respectively. Although the magnitude varies, the performance in each time period is statistically significant at t he 5% level. We find similar CDS momentum return patterns for the other two holding periods K = 3m, 6m. Also in Table 2 2, we separately compute CDS momentum returns for investment grade (BBB and above) and junk grade entities (BB+ and below). Consistent with stock momentum evidence (Avramov, Chordia, Jostova, and Philipov, 2007), the CDS momentum long/short strategy is substantially higher for junk grade firms than investment grade firms. Junk grade firms produce a return of 83 bps (= 0.833%) with a t sta tistic of 3.81 and Sharpe ratio of 0.320. Although smaller in magnitude, the investment grade firms still produce a statistically significant momentum return of 27 bps (= 0.269%) with a t statistic of 2.88 and a Sharpe ratio of 0.310. This result contrasts with earlier research finding that corporate bond momentum does not exist among investment grade entities in the post 1991 time period as documented by Jostova, Nikolova, Philipov, and Stahel (2013). Although the magnitudes of the CDS momentum returns for the two groups are different, their Sharpe ratios are similar suggesting the return compensation per unit of volatility of each group is almost equivalent. Risk Adjustments and Firm Characteristics In this section, we discuss the robustness of CDS momentu m to risk factors and firm characteristics. We show that momentum returns are ro bust to common stock and bond based risk factors as well as characteristic adjusted returns, though momentum profits are greatest in smaller firms, less creditworthy firms, an d firms with higher CDS liquidity measured by contract depth. The momentum strategy analyzed in this subsection uses formation J = 3m and holding K = 1m, which was shown to yield the greatest momentum profits as illustrated in Table 2 2.
33 We first analyze w hether CDS momentum returns are explaine d by common risk factors. Table 2 3 also presents alpha coefficients from regressions on various risk factor models. Using OLS with Newey We st standard errors, we estimate where is the CDS return for momentum portfolio , is a vector of common risk factors from bond and stock markets. The first bond based facto r, brDEFAULT, captures the difference in returns between BBB rated and AAA rated bonds. The second bond based factor, brTERM, is the differ ence in returns between long term and short term bonds. These two factors are provided as end of month series, so the re is a slight timing m ismatch between these bond fac tors and our portfolio timing (roughly a week and a half difference, on average). The equity factors Mkt, SMB, HML, and UMD match the timing of the CDS momentum strategy. momentum factor, sMOM, using our sample of firms and the strategy J = 12m and K = 1m, skipping a month between formation and holding periods, which is standard in the st ock momentum literature (see Asness, Moskowitz, and Pedersen (2013), among others). As reported in Table 2 3, the long/short strategy monthly alphas are between 0.506% and 0.652%, which is similar to the long/short momentum returns (0.520% per month) prev iously reported in Table 2 2. Across the four different factor models, the alphas are all significant at the 1% level of statistical significance. We next examine whether the CDS momentum profits are robust to using characteristic adjusted returns accordi ng to firm size, S&P rating, and CDS depth. Motivated by our findings in Table 2 2 that the winner portfolio is comprised of larger and riskier firms with lower contract depth, we examine whether potential risks associated with these characteristics drive the CDS
34 momentum profits. To this end, we compute the characteristic adjusted momentum returns with firm size, S&P rating, and CDS depth, and we report the results in Table 2 3. The characteristic adjustment is made by removing the mean return at each deci le or group from th e actual return . Even after adjusting for these characteristics, the magnitude of CDS momentum re mains large and all long/short returns are statistically significant at the 1% level. The size adjusted, rating adjusted and depth adjuste d returns produce long/short returns of 0.408%, 0.344% and 0.363%, respectively. Adjusting for all three characteristics at the same time still produces a long/short return of 0.314%. Although we do not find that these firm and contract characteristics are key drivers of momentum profits in CDS returns, their relation to the performance of CDS momentum strategy is still an open and interesting question. An analysis of this relation could pro vide some insights on pinning down the potential mechanisms that drive CDS momentum profits. In Table 2 3, we present long/short strategy returns within quintiles (Q1 to Q5) of firm size, S&P rating, and CDS depth, resp ectively. We first create quin tiles based on each raw variable. Then, to better isolate s own effect, we orthogonalize it to the other two variables. We find in Table 2 3 that small firms produce greater momentum returns than large firms. The smallest size quintile produces a long/short return that is 0.682% greater than the largest quintile, a result that is statistically significant at the 5% level. However, this size effect is confounded with the two other characteristics S&P rating and CDS depth. After orthogonalizing to rating and depth, the size effect is reduced to 0.311% with much weak er statistical significance. Confirming the result in Table 2 2 that junk grade firms generate larger momentum returns than investment grade firms, we also find that the riskiest rating quintile produces a
35 long/short return that is 103 bps (=1.031%) greate r than the safest quintile (see S&P Rating results). This result is statistically significant at the 1% level. Even after orthogonalizing rating to size and depth, the difference is still 69 bps (=0.693%) and is statistically significant at the 1% level (s ee S&P Rating (Orthogonalized)), indicating a close relation between CDS momentum These CDS momentum results with corporate credit ratings are consistent with Avramov, Chordia, Jostova, and Philipov (2007), who document that stock momentum exists only within firms with poor credit ratings. Related to this point, Avramov, Chordia, Jostova, and Philipov (2012) further show that momentum returns are generated primarily in high credit risk firms that experience conti nuing financial distress. We find that rating changes, both upgrades and downgrades, are much more frequent in junk grade entities than in vestment grade entities there are 10.50% (5.91%) more upgrades (downgrades) among junk grade entities in our sample. I n the next Subsection 3.2, we explore the role of ratings upgrades (downgrades) following substantial run ups (run downs) in CDS returns in driving CDS momentum profits. Using our third characteristic, we also analyze the performance of CDS momentum across different contract depths. The relationship between depth and momentum can a priori take several forms. If CDS momentum is driven by illiquidity, then we might expect momentum to be concentrated within the least liquid CDS contracts (i.e., lowest CDS dept h). Relatedly, if depth is an indicator of the information environment of the firm (similar to analyst coverage), we might expect momentum to be concentrated within opaque firms, particularly if it is driven by slow information diffusion and underreaction (Hong and Stein, 1999; Hong, Lim, and Stein, 2000). In contrast to these first two hypotheses, we might expect momentum to be greatest among firms with relatively high depth if depth captures a glamour effect. Lee and Swaminathan
36 (2000) find that stock ret urn momentum is more pronounced among high volume stocks, an effect that is primarily driven by a quicker and strong rebound of low volume losers relative to high volume losers. They explain this quicker reversal of the low volume losers through a phenomen on they term as the momentum life cycle (MLC). A fourth hypothesis we consider, which places an emphasis on the endogenous aspect of CDS liquidity (Qiu and Yu, 2012), is that CDS depth is positively correlated with the level of informed trading/quote updat ing in the CDS market. In this case, the more informed high depth CDS contracts could have richer information on firm default risk, thereby more accurately predicting future credit rating changes. With this endogenous liquidity, more upgrades (downgrades) could exist in the winner (loser) CDS momentum portfolio. This rating change channel is also documented as a main driver of equity momentum returns (Avramov, Chordia, Jostova, and Philipov, 2012). The third and fourth hypotheses described above predict gre ater momentum profits in more actively trading entities i.e., entities with high CDS depth. In Table 2 3, we find support for the latter two hypotheses that CDS contracts with high depth generate higher momentum returns than those with low depth. The most liquid quintile based on CDS depth produces a long/short return that is 1.092% greater than the least liquid quintile, a result that is statistically significant at the 1% level. The depth effect is strong even after orthogonalization to both size and rati ng; the difference between Q5 and Q1 remains large at 0.925% and is statistically significant at the 1% level. Moreover, the CDS momentum returns in the highest depth quintile (80 bps=0.798%) are also robust to transaction cost adjustments. Using Datastrea m, we find that the average bid ask spread for the entities in the highest depth quintile is 9.5 bps.22 Even with an extreme 100% turnover assumption, our three month
37 formation and one month holding period CDS return momentum strategy yields 42 bps per mon th, net of transaction costs, for the high depth group. Credit Rating Changes Motivated by our findings on the effects of S&P rating and CDS depth on the profitability of the CDS momentum strategy, this subsection examines whether CDS momentum profits are upgrades following run ups and downgrades following run downs of cumulative CDS returns. If this is the channel through which CDS momentum profits are generated, it will effectively rule out the ML C hypothesis by Lee and Swaminathan (2000), which also predicts greater momentum profits for high depth CDS contracts. To analyze the relationship between CDS momentum and credit rating changes, we report separately the returns of firms that experience an upgrade, downgrade, or no rating change within six months after the momentum portfolios are formed. In a similar analysis conducted by Jostova, Nikolova, Philipov, and Stahel (2013), no relation is found between the momentum profits in corporate bond retur ns and the future rating change channels. Contrary to their findings, our results in Table 2 4 show that CDS momentum and ratings changes are closely related. In Table 2 4, we focus on the most profitable CDS momentum strategy that uses formation J = 3m an d holding K = 1m, similar to what we did in the previous Subsection 3.1. Confirming prior studies on the relationship between CDS and credit ratings, Table 2 4 shows that a firm has positive (negative) market adjusted CDS returns in the months prior to a r ating upgrade (downgrade) and a large positive (negative) return in the month of the announcement. In the month before the new rating announcement, upgraded (downgraded) firms already experience a six month cumulative CDS return of 221 bps ( 324 bps). In the month of the announcement, upgrades (downgrades) further experience a return of 44 bps ( 133 bps). To
38 the extent that our momentum portfolios effectively place firms to be upgraded (downgraded) in the winner (loser) portfolio, the long/short return wil l benefit from the divergence in spread movements between upgrades and downgrades. These rating upgrades (downgrades) following past cumulative CDS return run up (run down) are referred to future rating changes in In Table 2 4, w e show that CDS momentum profits are generated from these anticipated changes in future credit ratings over the six month horizon following the formation date of the momentum portfolio. For the firms that never undergo any rating changes over the six month horizon, we do not find statistically significant returns for the long/short momentum strategy (0.102% monthly return with a t statistics of 1.04), though only during the post crisis period does this group generate statistically significant positive long/ short returns (= 0.379% with a t statistics of 2.54). To provide more direct evidence on our proposed mechanism of momentum profits, we analyze the returns of the firms that experience rating changes in anticipated directions. Looking at the fraction of up grades and downgrades for the winner and loser portfolios in Table 2 4, we find that P5 has significantly more upgrades (10.2%) than P1 (5.00%), whereas P1 has significantly more downgrades (20.6%) than P5 (11.0%). These wedges between P5 and P1 portfolios are statistically significant at the 1% level for both upgrades and downgrades, and the firms that are upgraded (downgraded) in P5 (P1) in the anticipated directions (i.e., upgraded firms in P5 and downgraded firms in P1) generate statistically significan t monthly momentum profits of 2.044% at the 1% level. Strikingly, firms that go through unanticipated rating changes over the same six month horizon show return reversal ( 0.587%) and are statistically significant
39 at the 5% level. From these results, it is clear that when the rating changes occur in the From the same table , we find that the difference in the wedges between the fractions of upgraded firms and downgraded firms in the long/short por tfolio is at its maximum during the significant long/short portfolio monthly return of 3.395% at the 1% level during that time period. Therefore, the future ratin g change channels we propose here appear to consistently explain the greater performance of the CDS momentum strategy we documented earlier for the crisis sub period. How are these findings related to the S&P rating and CDS depth characteristics that we fo und earlier to be positively correlated with the CDS momentum performance? In Table 2 4, we also examine the sensitivity of the relationship between CDS momentum and ratings changes to the rating level. We divide firms into investment grade and junk grade and redo the analysis. Our first observation is that removing firms with credit rating changes eliminates momentum returns in both investment grade andjunk grade groups. The long/short strategy produces returns of 0.0784% and 0.148% for investment grade an d junk grade firms, respectively, neither of which is statistically significant. Absent firms with future ratings changes, there is no significant difference in momentum returns between investment grade and junk grade firms (0.0696% with a t statistic of 0 .19). The better performance of the momentum strategy among junk grade firms clearly comes from the difference in future rating changes in anticipated directions as documented, with a monthly return difference of 2.607% and a t statistic of 2.12 at the bot tom of Table 2 4.
40 What is the role of CDS depth in the relationship between CDS momentum and ratings changes? In Table 2 4, we divide firms into high depth and low depth and repeat the previous analysis. It is clear that firms with high CDS depth produce g reater CDS momentum profits than those with low CDS depth because the long/short momentum portfolios in the high depth group more effectively predict future ratings changes in the anticipated directions. For example, the excess frequency of upgrades in P5 over P1 in high depth group (0.0494) is greater than that in the low depth group (0.0487), and such a tendency is much stronger in its magnitude for the rating downgrades. The long/short portfolio in the high depth group ( 0.120) twice as accurately predic ts the credit rating downgrades as the portfolio in low depth group ( 0.0640). This superior predictability of high depth CDS returns on future rating changes in the anticipated directions yields a 2.430% monthly long/short portfolio return, whereas the mo mentum return generated through the same mechanisms in the low depth group (1.084%) is just half of the return. The difference in these returns is substantial at 1.346% (=2.430% 1.084%) and statistically significant at the 5% level. Our findings regarding these CDS depth effects are consistent with Qiu and Yu (2012) who recently document that CDS depth is a proxy for informed trading in the CDS market. The at market spreads of CDS contracts with high endogenous liquidity are also shown to reveal information earlier than stock returns, particularly when the credit quality of reference entities deteriorates. The ability of past CDS returns with high depth contracts to predict rating downgrades better than rating upgrades is also in line with their findings. Th e analysis in Table 2 4 allows for the possibility that past CDS returns have an informational advantage over stock returns in predicting future ratings changes. If so, it is more
41 likely to be for the entities with lower credit ratings and high contract de pth, as suggested in our analysis in Table 2 4. CDS Momentum and Stock Returns In the previous section, we demonstrated a close relation between CDS momentum and the ability of past CDS returns to anticipate future ratings changes. However, it may be that past CDS returns are just noisy signals of past stock returns. In this section, we address this concern and show that there are indeed incremental information benefits in CDS returns above and beyond the information contained in stock returns, and that suc h informational benefits in the CDS returns generate a momentum spillover from the CDS to the stock market. Gebhardt, Hvidkjaer, and Swaminathan (2005) do not find a momentum spillover from the corporate bond market to the stock market, which suggests impo rtant information differences between corporate bond and CDS returns. Lastly, we provide evidence that past CDS returns, when combined w ith past stock returns, dramat ically improve the performance of the traditional stock to stock momentum strategy of Jega deesh and Titman (1993) by sharpening the signals that generate stock momentum profits. Credit Rating Changes In Table 2 5, we first show that CDS momentum returns are highly correlated with other mar . However, we also emphasize that the CDS mo mentum returns lead those of bond momentum , and the CDS returns have incremental information not fully cap tured by stock returns . Table 2 5 displays time series correlation coefficients between CDS momen tum and momentum in the stock and bond mark ets. We use the bond momentum series provided by Jostova, Nikolova, Philipov, and Stahel (2013): all firms, investment grade firms, and junk grade bonds. There is a slight mismatch in timing between our CDS momentum series and their bond
42 momentum series an d the firms are not an exact match, so the reported correlation might misstate the true correlation. Over the full matched period, the correlation between CDS momentum and bond momentum is 0.255. Across sample periods, the crisis period shows the highest c orrelation (0.281). The correlation is much higher for investment grade firms (0.304) than junk grade firms ( 0.053). Among investment grade firms, the correlation peaks during the crisis period (0.359). The stock momentum factor, UMD, exactly matches the rebalancing timing of our CDS momentum. The correlation between CDS momentum and the UMD factor is 0.388, and also peaks during the crisis period (0.560). We construct the other stock momentum return series (All Firms, Investment and Junk Grade, and High a nd Low Depth) to exactly match both our CDS return rebalancing timing as well as our exact firm sample. Among all firms and over the entire sample period (see All Firms), the correlation between CDS momentum and stock momentum is 0.533. The correlatio ns ar e very low during the pre crisis period and highest during the crisis and post crisis periods. Overall, stock and CDS momentum correlation is higher for investment grade firms than junk grade firms (0.443 versus 0.267) and for high depth contracts (0.536 v ersus 0.407). Despite these positive correlations between CDS momentum and momentum in the other reported in Table 2 5, there is no significant lead/lag relatio n between CDS momentum and stock momentum; however, both lead bond momentum. The results are quite striking because the contemporaneous bond momentum returns are in fact measured in advance to both the CDS momentum and the stock momentum returns. The bond momentum returns we obtained from Jostova, Nikolova, Philipov, and Stahel (2013) are measured at the end of each month, while the other two momentum returns CDS and stock are measured on the 19th of the month. Despite
43 this timing informational advantage th at is readily available in the bond momentum returns, both the CDS and stock markets precede the momentum in corporate bond markets, which re ensures the superior information capacity we expect a priori in these two markets over the corporate bond market. In Table 2 5, we examine whether CDS returns have an incremental informa tion advantage above and beyond that embedded in stock returns. We isolate the CDS return signal from that of the stock market (and vice versa) by independently sorting the CDS entitie s by past stock returns (sP1 to sP5) and past CDS returns (P1 to P5). Then, we examine whether past CDS returns generate significant CDS momentum profits, while we control for the signals in the past stock returns (Relative Value of CDS). Similarly, to inv estigate whether there is valuable incremental information in past stock returns, we form the CDS long/short portfolios based on past stock returns while controlling for past CDS returns (Relative Value of Stock). Overall, we find that the marginal informa tion in past CDS returns generates signifi cant momentum profits for both entities in sP1 (1.006% monthly return) and sP5 (0.372% monthly return). However, we only find marginal informa tion benefits of past stock re turns for the entities in P1, the CDS lose rs (0.588% monthly). For the entities in P5, the CDS winners, we in fact find return reversal ( 0.0462%) based on past stock returns as incremental sorting signals. These incremental information benefits of CDS returns over stock returns are evident partic ularly for the entities with lower credit ratings and high CDS depth. The CDS based long/short return of junk grade firms is 2.707% or 0.570% conditional on the firm falling into sP1 or sP5, respectively, and both are statistically significant. In the oppo site case, the relative advantage of the stock signal is positive and statistically significant for P1 (= 1.347%) but
44 that greater CDS depth is also associated w ith a stronger relative advantage of the CDS signal, supporting the argument that high depth CDS contracts contain greater information flow and suggesting potential segmentation in the information content of CDS and stock markets. Conditioning on the firm falling into sP1 (sP5), the CDS based long/short strategy for high depth firms produces statistically significant returns of 0.325% (0.230%). In contrast, for low depth firms, the CDS signal is weak after considering the stock signal i.e., the CDS based lo ng/short returns are insignificant conditioning on falling into one of the extreme stock momentum portfolios. The findings in Table 2 5 suggest that the CDS market has incremental information above and beyond the stock market information and that the liqui dity of a CDS contract measured by its contract depth is consistent with the notion of the endogenous liquidity in the CDS market (Qiu and Yu, 2012). That is, higher CDS liquidity is associated with greater information flow originating in the CDS market. M omentum Spillover from CDS to Stock An important question that remains is whether or not momentum spills over from the CDS market to the stock market? An affirmative response would reconfirm the incremental informational advantage in using CDS signals to g enerate stock return momentum. Table 6 presents future stock returns of CDS based momentum portfolios. In this strategy, we sort firms based on past three month (i.e., J = 3m, K = 1m). We purchase stocks of firms in the CDS based winner portfolio (P5) and sell short the stock of firms in the CDS based loser portfolio (P1). Among all firms in our sample, the long/short strategy generates a return of 0.517%, with a t statistic of 1.69 and Sharpe ratio of 0.1 42. Although not statistically significant, the magnitude of the return is largest during the crisis period. In addition, the return is independent of common stock risk factors: Mkt,
45 (Loc4 Alpha) constructed using our sample of firms (alpha coefficients of 0.638% and 0.635%, respectively). To better understand the source of the CDS momentum spillover, we further di vide firms by investment grade/junk grade and recompute the cross mar ket momen tum returns. The momentum spillover is concentrated in junk grade firms, producing a long/short return of 1.365%, with a t statistic of 2.29 and Sharpe ratio of 0.230. The CDS to stock cross market momentum strategy produces no significant retur ns among investment grade firms. These results as consistent with the finding in Acharya and Johnson (2007) who document that the CDS market reveals information earlier than the stock market for poorly rated firms. The CDS to stock cross market momentum ef fect is also concentrated in the most liquid CDS contracts. The long/short return is 0.806% for firms w ith high CDS depth compared to 0.0473% for firms with low depth. It should be noted that the long/short strategy of the high depth group has the largest Sharpe ratio (0.247), which indicates the investment efficiency of this subgroup among all the subgroups we consider in this table . Again, the returns generated by the high depth firms are not explained by common stock based risk factors or a stock momen tum risk factor constructed for each specific subgroup. The alpha coefficients are 0.954% and 0.956% for the Carhart four factor model and our localized four factor model, respectively with both being significant at the 1% level. Our results so far suggest that the CDS and stoc k markets are somewhat informa tionally segmented and that for groups of firms there is information flow from the CDS market to the stock market. That is, the results in Table 2 6 together with those in Table 2 5 show that potential in formation flow from the CDS market to the stock market is more likely for firms in financial distress and/or those with high CDS contract depth. In the following analysis, we show
46 how an investor in the stock market could utilize the superior information c apacity of the CDS market to enhance his/her stock investment performance. Table 6 also presents the stock returns of two momentum strategies: a single market stock based strategy and a joint market strategy that only trades firms at the intersection of t he winner and loser stock and CDS momentum portfolios. The joint market momentum strategy is constructed in two steps. First, stock and CDS momentum portfolios are created independently based on their own market signals. Then, we purchase the stock of fir ms that fall into both past stock signal winners (sP5) and past CDS signal winners (P5) and sell short the stock of those in past stock signal losers (sP1) and past CDS signal losers (P1). In this analysis, we take advantage of the positive and negative si gnals coming from both the stock and CDS markets. Given that we trade stocks instead of CDS contracts, transaction cost concerns in this joint market momentum strategy are also minimal (Frazzini, Israel, and Moskowitz, 2012). Across all firms, the traditio nal stock momentum stra tegy does not produce statisti cally significant returns during our time period (0.124% with a t statistics of 0.16). This result is not specific to our stock momentum ret urns. As illustrated in Figure 5 , insignificant stock momentum profits are also shown using the Carhart momentum factor, UMD, which is based on a broader cross section of stocks. Both our stock momentum and UMD returns suffer from abrupt losses during the crisis period (Daniel, 2011; Daniel, Jagannathan, and Kim, 2012 ). In contrast, the joint market strategy produces a large long/short strategy return of 116 bps per month (=1.160%) that is statistically significa nt at the 1% level, an improve ment of net 104 bps monthly return (=1.037%). The crisis period appears to be driving the magnitude of the joint market momentum returns (= 2.647%), which is the period when traditional momentum strategies are we ll known to break down. Figure 5 shows that strategies incorporating CDS
47 returns as additional sorting signals are able to avoid dra matic losses during the crisis period. Both the cross market (CDS to stock) momentum strategy and the joint market strategy significantly outperform traditional stock based strategies during our sample period. We posit that anticipated future cre dit rating changes are the mechanism through which CDS spills over to stock in either the pure cros s market momentum or the joint market momentum strategy. CDS returns assist stock returns in correctly anticipating credit rating changes and then benefit fr om the realized price divergence between up graded and downgraded firms. This result can be seen in the relative fraction of upgrades (downgrades) in the winner (loser) portfolio that we document in Table 2 6. For instance, 32.4% of firms in the joint mar ket loser portfolio (P1, sP1) experience down grades as opposed to 18.9% in the single market stock momentum loser portfolio (sP1). Likewise, 18.1% of firms in the joint market winner portfolio (P5, sP5) are upgraded, which is much more than the 9.01% in the single mar ket stock momentum winner portfo lio (sP5). In all subgroups, the join t market momentum portfolios have the greater fraction of realized anticipated rating changes. The resulting improvement in performance is evident, particularly for the enti ties with low credit ratings and/or high CDS depth. When we further compare the return difference between joint and single market strategies in the last four columns of Table 2 6, it is evident that high CDS contract depth plays an important information r ole in explaining CDS information flows to stocks, consistent with endogenous CDS liquidity notions (Qiu and Yu, 2012). Overall, the results in Table 2 6 confirm the greater information capacity of the CDS market for certain types of firms and its associat directions as a way for stock investors to improve their momentum investment performance.
48 Final Thoughts We examine the extent to which momentum exists in the CDS market and whether it spills over to another closely related asset market that is linked through common firm fundamentals the stock market. Despite the CDS mark ciency, we document both economically and statisticall y significant CDS return momen tum. We further show that t he underlying mechanisms driving CDS return momentum are distinct from those associated with corporate bond return momentum (Jostova, Nikolova, Philipov, and Stahel, 2013) and highlight the role of CDS return information in predicting future credit rating changes with past CDS winners (losers) undergoing more frequent rating upgrades (downgrades) in subsequent periods. We also document the ability of a future rating change channel in driving the within CDS market momentum, partic ularly for firms with poor credit ratings (Acharya and Johnson, 2007) and/or high CDS contract depth (Qiu and Yu, 2012). Importantly, we also document that there is useful incremental information in CDS returns beyond the information in stock returns. We s how significant cross market and joint market CDS to stock momentum returns, both of w hich are closely related to fu ture rating with poor credit ratings and/o r greater CDS market liquidity drive these interesting cross market CDS to stock information dynamics. In particular, our novel joint market momentum strategy that uses both stock and CDS market signals in the portfolio formation process helps traditional stock momentum strategies avoid well documented break downs in momentum returns during the recent crisis period (Daniel, 2011; Daniel, Jagannathan, and Kim, 2012). Our ne w strategy improves monthly mo mentum returns by net 104 bps during this period. Overa ll, our results provide some explanatory evidence regarding a number of impor tant empirical regularities documented in the recent momentum literature: momentum exists
49 across several asset classes and exhibits a strong comovement across these assets and m arkets (Asness, Moskowitz, and Pedersen, 2013). We uncover some important sources that underlie this cross market comovement in momentum returns; common cor porate credit rating change generates a CDS return momentum anomaly associated with timely CDS cre dit risk information and sluggish ratings responses. We provide evidence consistent with the existence of such a mechanism by focusing on two closely related asset classes stocks and CDS that are linked through underlying firm fundamentals. At the same tim e, our study also suggests that stock investors can substantially improve their investment returns by paying attention to the incremental information originating in highly liquid CDS contracts. Our results suggest that further investigating the efficacy of CDS market information on investment strategies or in the design of other financial claim contracts could be a useful avenue for future research.
50 Table 2 1. Sample s tatistics . Mean St. Dev . Min Median Max N CDS based CDS s pread (bps) 202.0 6 0 400.654 1 .000 85 .000 9714 .000 75522 CDS r eturn (%) 0.018 3.947 62.255 0.003 302.635 75522 CDS d epth 6.396 4.764 2 .000 5 .000 33 .000 75322 Stock based Stock r eturn (%) 0.396 10.743 89.038 0.563 337.294 75378 Size ($Bil) 20.428 39.993 0.0 04 7.282 519.894 74922 Credit rating and firm size d ecile S&P r ating (1=AAA) 9.116 3.296 1 .000 9 .000 22 .000 71717 Investment grade d ummy 0.597 0.491 0 .000 0 .000 1 .000 71717 NYSE/AMEX size d ecile 9.229 1.326 1 .000 10 .000 10 .000 74922 Correlation m atrix 1. 2. 3. 4. 5. 6. 7. 1. CDS s pread (bps) 1.000 2. CDS r eturn ( % ) 0.151 1.000 3. CDS d epth 0.114 0.016 1.000 4. Stock r eturn ( %) 0.049 0.347 0.013 1.000 5. Size 0.158 0.008 0.107 0.021 1.000 6. S&P r ating 0.477 0.023 0.181 0.012 0.514 1.000 7. Size d ecile 0.386 0.013 0.226 0.029 0.275 0.508 1.000 Note: Table 2 1 presents the summary statistics and correlation matrix of variables used in this study. Data are monthly from January 2003 to December 2011. CDS data are provided by Markit, stock returns are from CRSP, and S&P Ratings are acquired from Compustat. Investment m onth observations.
51 Table 2 2 . CDS Momentum . Momentum Portfolios P K=1m K=3m K=6m P1 P2 P3 P4 P5 P5 P1 P5 P1 P5 P1 J=1m 0.086 0.061 0.013 0.016 0.309 0.396 0.333 0.238 ( 0.51 ) ( 0.90 ) ( 0.29 ) (0.27 ) (1.63 ) (3.55 ) (2.31 ) (1.70 ) Size d ecile 9.004 9.458 9.540 9.426 9.018 0.0 13 S&P r ating 9.891 8.088 7.779 8.421 10.67 0 0.779 Depth 6.376 6.404 6.598 6.615 6.018 0.358 J=3m 0.193 0.158 0.033 0.143 0.327 0.52 0 0.467 0.31 0 ( 0.86 ) ( 1.79 ) ( 0.60 ) (2.54 ) (2.52 ) (3.87 ) (3. 10 ) (2.39 ) Size d ecile 8.923 9.465 9.558 9.450 9.069 0.146 S&P r ating 10.01 0 8.071 7.625 8.357 10.77 0 0.763 Depth 6.560 6.420 6.715 6.726 5.943 0.617 J=6m 0.160 0.054 0.002 0.003 0.283 0.443 0.342 0.261 ( 0.73 ) ( 0.68 ) ( 0.05 ) (0.05 ) (1.85 ) (3.19 ) (2.41 ) (1.85 ) Size d ecile 8.890 9.514 9.574 9.462 9.077 0.187 S&P r ating 9.946 7.857 7.610 8.419 10.88 0 0.935 Depth 6.713 6.644 6.925 6.808 5.872 0.841 J=9m 0.169 0.039 0.020 0.014 0.243 0.412 0.292 0.219 ( 0.67 ) ( 0.3 9 ) ( 0.35 ) (0.27 ) (1.94 ) (2.16 ) (1.91 ) (1.24 ) J=12m 0.153 0.034 0.023 0.004 0.198 0.351 0.249 0.182 ( 0.56 ) ( 0.32 ) ( 0.46 ) (0.10 ) (1.91 ) (1.62 ) (1.32 ) (0.94 ) Pre crisis p eriod (January 2003 June 2007) J=3m 0.081 0.039 0.032 0.053 0. 309 0.228 0.266 0.24 0 (0.57 ) (1.4 3) (1.30 ) (1.38 ) (2.99 ) (2.15 ) (2.12 ) (1.90 ) Crisis p eriod (July 2007 April 2010) J=3m 0.639 0.356 0.136 0.150 0.333 0.971 0.815 0.484 ( 0.99 ) ( 1.44 ) ( 0.86 ) (0.98 ) (0.98 ) (2.56 ) (2.11 ) (1.95 ) Pos t crisis p eriod (May 2010 December 2011) J=3m 0.370 0.197 0.015 * 0.129 0.271 0.64 0 0.386 0.186 ( 0.95 ) ( 1.54 ) ( 0.24 ) (1.65 ) (1.19 ) (2.85 ) (1.96 ) (1.54 ) Investment g rade (BBB and above) J=3m 0.125 0.091 0. 029 0.0541 0.143 0.269 0.271 0 .209 ( 0.92 ) ( 1.35) ( 0.67) (1.26) (2.04) (2.88) (2.19) (1.78) Junk g rade (BB+ and below) J=3m 0.216 0.026 0.133 0.2 0.617 0.833 0.669 0.443 ( 0.54 ) ( 0.15) (0.76) (1.70) (2.36) (3.81) (3.05) (2.19) Note: Table 2 2 presents our basic CDS moment um results. Firms are sorted on past CDS return into five equally sized portfolios in which portfolio 1 (5) is the group with the lowest (highest) past CDS return using a formation period J. The long/short strategy is constructed by purchasing CDS contract s of firms in portfolio 5 (High) and selling CDS contracts in portfolio 1 (Low) and holding for period K. The Newey West t statistic is provided in parenthesis.
52 Table 2 3 . CDS momentum returns: size, rating and d epth . Model Definitions ( 1) brDEFAULT,b rTERM ( 2) Mkt,SMB,HML,UMD ( 3) brDEFAULT,brTERM,Mkt,SMB,HML,UMD ( 4) brDEFAULT,brTERM,Mkt,SMB,HML,sMOM CDS momentum alphas Full Pre Crisis Post Model P1 P2 P3 P4 P5 P5 P1 P5 P1 P5 P1 P5 P1 ( 1) 0.298 0.131 0.057 0.024 0.313 0.611 0.363 0.894 0.648 ( 1.46 ) ( 1.62 ) ( 1.36 ) ( 0.55 ) ( 2.68 ) ( 4.35 ) ( 4.54 ) ( 3.25 ) ( 4.07 ) ( 2) 0.313 0.113 0.061 0.011 0.270 0.583 0.416 0.570 0.809 ( 2.75 ) ( 1.96 ) ( 1.71 ) ( 0.37 ) ( 3.05 ) ( 6.04 ) ( 4.52 ) ( 2.91 ) ( 3.58 ) ( 3) 0.362 0.144 0.06 3 0.017 0.290 0.652 0.416 0.718 0.829 ( 3.08 ) ( 2.59 ) ( 2.20 ) ( 0.54 ) ( 3.19 ) ( 5.96 ) ( 4.14 ) ( 3.11 ) ( 3.66 ) ( 4) 0.173 0.122 0.031 0.101 0.333 0.506 0.348 0.692 0.604 ( 0.81 ) ( 1.52 ) ( 0.58 ) ( 1.91 ) ( 2.56 ) ( 4.08 ) ( 4.56 ) ( 2.07 ) ( 2.91 ) Characteristic adjusted returns Size 0.184 0.117 0.065 0.001 0.224 0.408 0.286 0.550 0.488 ( 1.64 ) ( 3.09 ) ( 2.03 ) ( 0.01 ) ( 4.22 ) ( 3.58 ) ( 4.04 ) ( 1.76 ) ( 2.74 ) Rating 0.173 0.168 0.030 0.113 0.171 0.344 0.220 0.5 15 0.377 ( 1.64 ) ( 2.03 ) ( 1.25 ) ( 3.12 ) ( 3.82 ) ( 2.93 ) ( 2.91 ) ( 1.59 ) ( 2.41 ) Depth 0.153 0.110 0.081 0.033 0.210 0.363 0.244 0.471 0.496 ( 1.58 ) ( 2.24 ) ( 1.29 ) ( 0.57 ) ( 4.29 ) ( 3.34 ) ( 3.46 ) ( 1.61 ) ( 2.60 ) All 3 0 .170 0.059 0.030 0.001 0.144 0.314 0.234 0.444 0.293 ( 1.74 ) ( 2.07 ) ( 1.56 ) ( 0.07 ) ( 3.64 ) ( 3.33 ) ( 2.98 ) ( 1.99 ) ( 3.42 ) CDS momentum returns by quintiles of size, rating and depth Q1 Q2 Q3 Q4 Q5 Q5 Q1 Q5 Q1 Q5 Q1 Q5 Q1 Size 0.811 0.357 0.506 0.435 0.129 0.682 0.416 0.758 1.043 ( 2.42 ) ( 2.00 ) ( 3.03 ) ( 3.11 ) ( 1.56 ) ( 2.07 ) ( 1.22 ) ( 2.33 ) ( 1.23 ) Rating 0.082 0.106 0.145 0.359 1.113 1.031 0.767 1.034 1.429 ( 0.77 ) ( 0.98 ) ( 1.83 ) ( 2.44 ) ( 4.10 ) ( 4.56 ) ( 3.29 ) ( 3.99 ) ( 2.47 ) Depth 0.294 0.326 0.733 0.687 0.798 1.092 0.416 1.250 2.030 ( 1.78 ) ( 1.76 ) ( 3.20 ) ( 3.64 ) ( 4.56 ) ( 5.07 ) ( 2.13 ) ( 3.33 ) ( 3.93 ) Size ( O) 0.418 0.334 0.242 0.334 0.107 0.311 0.143 0.227 0.858 ( 2.23 ) ( 2 .20 ) ( 2.16 ) ( 2.20 ) ( 0.91 ) ( 1.63 ) ( 0.71 ) ( 0.97 ) ( 1.82 ) Rating ( O) 0.246 0.309 0.368 0.441 0.939 0.693 0.408 0.908 1.007 ( 1.40 ) ( 1.73 ) ( 3.00 ) ( 3.14 ) ( 3.59 ) ( 3.20 ) ( 1.66 ) ( 2.15 ) ( 2.02 ) Depth ( O) 0.220 0.125 0.095 0.65 3 0.705 0.925 0.588 0.942 1.427 ( 1.84 ) ( 0.96 ) ( 0.48 ) ( 3.27 ) ( 3.96 ) ( 4.98 ) ( 3.55 ) ( 3.49 ) ( 2.96 ) Note: Table 2 3 shows the robustness of CDS momentum to common risk factors and firm characteristics. It reports the alpha coefficients o f four factor model regressions using common stock and bond risk factors. It also reports the characteristic adjusted returns of CDS momentum portfolios and the long/short strategy. Returns are adjusted for size, rating and depth. The size (depth) adjustme nt is made by subtracting the mean return at each size (depth) decile from the actual return each month. The rating adjustment is made by subtracting the mean return at each rating. It also reports momentum profits by quintiles of raw size, rating and dept h as well as orthogonalized size, rating and depth (indicated by O). The difference in average returns between Q5 and Q1 is also given and a t test is given for equivalence to zero. Each CDS momentum strategy utilizes a formation period J = 3m and a holdin g period K = 1m. The Newey West t statistic is provided in parenthesis.
53 Table 2 4 . CDS momentum returns and credit rating c hanges . Cumulative CDS returns prior to ratings changes t 6 t 5 t 4 t 3 t 2 t 1 t Upgrade event 0.422 0.691 0.870 1.372 1.633 2. 206 0.44 0 (3.06 ) (3.31 ) (3.25 ) (3.39 ) (3.27 ) (4.00 ) (1.98 ) Downgrade event 0.871 1.123 1.954 2.246 2.528 3.242 1.327 ( 6.68 ) ( 5.34 ) ( 7.88 ) ( 7.57 ) ( 7.42 ) ( 7.97 ) ( 3.89 ) CDS momentum and rating changes Full Pre Crisis Post P1 P5 P5 P1 P5 P1 P5 P1 P5 P1 No rating change 0.019 0.121 0.102 0.027 0.064 0.379 (0.12 ) (1.04 ) (1.04 ) (0.31 ) (0.26 ) (2.54 ) Upgrade firms 0.282 0.840 0.558 0.317 0.925 0.551 (1.59 ) (2.52 ) (2.16 ) (1.52 ) (1.40 ) (1.38 ) Downgrade firms 1.204 0.305 0.899 0.424 1.217 1.638 ( 2.13 ) ( 1.84 ) (2.01 ) (1.36 ) (1.21 ) (1.15 ) Upgrade frac. 0.050 0.102 0.052 0.032 0.067 0.085 Downgrade frac. 0.206 0.110 0.096 0.084 0.143 0.037 Anticipated 1.204 0.840 2.044 0.904 3.395 2.734 D irection ( 2.13 ) (2.52 ) (4.70 ) (3.57 ) (4.57 ) (1.84 ) Unanticipated 0.282 0.305 0.587 0.128 1.282 0.441 Direction (1.59 ) ( 1.84 ) (2.14 ) ( 0.70 ) ( 3.30 ) ( 1.08 ) Investment grade firms only No rating change 0.0195 0.097 0.078 0.035 0.12 4 0.112 (0.18 ) (1.30 ) (1.13 ) (0.70 ) (0.65 ) (1.23 ) Upgrade frac. 0.037 0.059 0.021 0.013 0.019 0.053 Downgrade frac. 0.184 0.093 0.090 0.081 0.127 0.041 Anticipated 0.521 0.232 0.753 0.535 1.038 0.838 Direction ( 1.46 ) (1.89 ) (2.55 ) (1.86 ) (1.39 ) (1.76 ) Junk grade firms only No rating change 0.090 0.238 0.148 0.029 0.009 0.760 (0.35 ) (0.98 ) (0.89 ) (0.14 ) ( 0.03 ) (3.81 ) Upgrade frac. 0.092 0.167 0.075 0.029 0.126 0.157 Downgrade frac. 0.273 0.122 0.151 0.147 0.2 05 0.059 Anticipated 2.289 1.071 3.360 2.091 4.972 3.930 Direction ( 2.99 ) (1.96 ) (5.79 ) (3.53 ) (6.12 ) (1.95 ) High depth firms only No rating change 0.011 0.224 0.213 0.105 0.248 0.445 (0.06 ) (1.69 ) (2.05 ) (1.25 ) (0.94 ) (2.62 ) Upg rade frac. 0.047 0.097 0.049 0.027 0.073 0.107 Downgrade frac. 0.235 0.115 0.120 0.102 0.186 0.065 Anticipated 1.393 1.037 2.430 0.994 4.154 3.260 Direction ( 2.14 ) (2.27 ) (4.45 ) (2.95 ) (4.08 ) (1.98 ) Low depth firms only No rating chang e 0.025 0.017 0.008 0.055 0.109 0.307 (0.19 ) (0.16 ) ( 0.08 ) ( 0.54 ) ( 0.49 ) (2.45 ) Upgrade frac. 0.059 0.108 0.048 0.033 0.055 0.085 Downgrade frac. 0.175 0.111 0.064 0.056 0.094 0.010 Anticipated 0.642 0.442 1.084 0.78 1.958 0.322 Direction ( 1.48 ) (1.80 ) (3.15 ) (2.12 ) (2.45 ) (0.70 ) Note: Table 2 4 provides results of (a) CDS momentum and future S&P credit rating changes , (b) cumulative CDS returns leading up to a rating change in month t , and (c) CDS momentum returns and c redit rating changes. The Newey West t statistic is provided in parenthesis.
54 Table 2 5 . CDS momentum and momentum e lsewhere . Correlation matrix of momentum returns Full Pre Crisis Post Bond momentum All firms 0.255 0.103 0.281 0.043 Investment grade 0.304 0.163 0.359 0.261 Junk grade 0.053 0.071 0.082 0.102 Stock momentum UMD factor 0.388 0.024 0.560 0.599 All firms 0.533 0.258 0.574 0.657 Investment grade 0.443 0.077 0.484 0.599 Junk grade 0.267 0.015 0.440 0.271 High depth 0.536 0.196 0.598 0.585 Low depth 0.407 0.275 0.432 0.483 Lead/lag effects in momentum returns CDS MOM CDS MOM Bond MOM Bond MOM Bond MOM Stock MOM CDS MOM 0.432 0.475 2.472 (3.26 ) (3.10 ) (6.5 4 ) Lagged 0.104 0.017 0.629 0.367 0.122 CDS MOM ( 0.95 ) (0.17 ) (4.58 ) (2.30 ) ( 0.27 ) Bond MOM 0.228 (3.26 ) Lagged Bond MOM 0.104 0.380 0.492 0.411 ( 1.64 ) (4.77 ) (6.11 ) (5.15 ) Stock MOM 0.120 0.044 0 .016 (6.54 ) (1.47 ) ( 0.48 ) Lagged Stock MOM 0.010 0.145 0.096 0.081 (0.46 ) (4.98 ) (2.90 ) ( 0.82 ) Observations 101 106 101 101 101 106 Adj. R Squared 0.101 0.296 0.419 0.385 0.470 0.302 CDS returns of joint market momentu m strategies P1,sP1 P1,sP5 P5,sP1 P5,sP5 P5 P1 in sP1 P5 P1 in sP5 sP5 sP1 in P1 sP5 sP1 in P5 All firms 0.631 0.043 0.375 0.329 1.006 0.372 0.588 0.046 ( 2.55 ) ( 0.27 ) (2.17 ) (3.25 ) (6.23 ) (2.72 ) (2.05 ) ( 0.37 ) Inv. grade firms 0 .218 0.047 0.031 0.117 0.249 0.0702 0.265 0.085 ( 1.28 ) (0.40 ) (0.30 ) (1.82 ) (1.79 ) (0.61 ) (1.64 ) (0.97 ) Junk grade firms 1.387 0.040 1.320 0.53 0 2.707 0.57 0 1.347 0.79 0 ( 3.60 ) ( 0.18 ) (2.79 ) (2.99 ) (6.12 ) (2.53 ) (3.86 ) ( 1.83 ) Hi gh depth firms 0.243 0.013 0.082 0.243 0.325 0.23 0 0.257 0.161 ( 0.99 ) (0.09 ) (0.40 ) (2.24 ) (1.88 ) (1.80 ) (1.57 ) (1.00 ) Low depth firms 0.200 0.963 0.181 0.423 0.019 0.540 0.763 0.242 (0.75 ) (2.65 ) (1.10 ) (1.95 ) ( 0.03 ) ( 1.40 ) (1.83 ) (0.93 ) Note: Table 2 5 explores the relation between CDS momentum and return momentum in other markets. It shows the time series correlation between of CDS momentum and stock and bond momentum. It also reports lead/lag effects in the momentum time seri es of CDS, bonds and stock. It also presents joint market momentum results. Reported are CDS returns at intersections of CDS momentum portfolios P and stock momentum portfolios sP. The relative value of the CDS (stock) is defined as the CDS based (stock ba sed) long/short return conditional on the firm measures the marginal ability of the CDS signal (past CDS return) to predict future CDS returns conditional on the negat ive stock signal. The Newey West t statistic is provided in parenthesis.
55 Table 2 6 . Momentum spillover from CDS to s tocks . Cross market momentum returns: CDS to s tocks P1 P2 P3 P4 P5 Full Pre Crisis Post Car4 Alpha All f irms 0.057 0.392 0.530 0.401 0 .575 0.517 0.128 1.263 0.230 0.638 (0.07 ) (0.65 ) (0.99 ) (0.73 ) (0.81 ) (1.69 ) (0.58 ) (1.58 ) (0.47 ) (1.91 ) 0.142 0.074 0.246 0.108 Inv est . g rade 0.350 0.403 0.490 0.333 0.313 0.036 0.037 0.148 0.374 0.074 (0.52 ) (0.77 ) (0.99 ) (0. 67 ) (0.56 ) ( 0.16 ) ( 0.31 ) (0.23 ) ( 1.20 ) (0.39 ) 0.055 0.029 0.037 0.295 Junk g rade 0.363 0.516 0.890 0.723 1.002 1.365 1.163 2.168 0.449 1.492 ( 0.34 ) (0.52 ) (0.87 ) (0.77 ) (0.99 ) (2.29 ) (1.76 ) (1.66 ) (0.36 ) (2.31 ) 0. 23 0 0.238 0.300 0.082 High d epth 0.130 0.091 0.434 0.447 0.676 0.806 0.395 1.488 0.696 0.954 ( 0.17 ) (0.19 ) (0.96 ) (0.98 ) (1.11 ) (2.53 ) (1.88 ) (1.83 ) (1.18 ) (2.94 ) 0.247 0.226 0.242 0.313 Low d epth 0.213 0.335 0.529 0.466 0.166 0.0 47 0.027 0.119 0.407 0.014 (0.35 ) (0.69 ) (1.06 ) (0.96 ) (0.28 ) ( 0.23 ) ( 0.12 ) (0.24 ) ( 1.70 ) (0.06 ) 0.044 0.014 0.036 0.278 Stock Returns of Joint Market Momentum Strategies sP1 sP5 sP5 sP1 P1,sP1 P5,sP5 P5,sP5 P1,sP1 Full Pre Crisis Post All Firms 0.140 0.263 0.124 0.340 0.82 0 1.16 0 1.037 0.142 2.647 0.565 (0.13 ) (0.41 ) (0.16 ) ( 0.50 ) (1.72 ) (2.95 ) (1.72 ) (0.29 ) (1.63 ) (0.81 ) Upgrade f rac. 0.067 0.090 0.022 0.0525 0.181 0.129 0.107 Down f rac. 0.189 0.103 0. 086 0.324 0.108 0.216 0.13 0 Ant. f rac. 0.189 0.090 0.324 0.181 Inv est . g rade 0.127 0.157 0.284 0.713 0.463 0.250 0.534 0.745 2.699 0.8 00 (0.15 ) ( 0.29 ) ( 0.43 ) ( 0.73 ) ( 1.09 ) (0.47 ) (1.17 ) ( 1.19 ) (2.11 ) (0.93 ) Ant. f rac. 0. 158 0.055 0.290 0.075 Junk g rade 0.284 1.170 1.453 0.742 1.171 1.913 0.46 0 0.004 1.437 1.03 0 ( 0.19 ) (1.22 ) (1.40 ) ( 0.74 ) (1.61 ) (2.97 ) (0.79 ) (0.01 ) (0.67 ) (1.23 ) Ant. f rac. 0.258 0.172 0.417 0.284 High d epth 0.208 0.181 0.0 27 0.823 1.12 0 1.944 1.971 0.733 4.029 1.635 (0.20 ) (0.29 ) ( 0.04 ) ( 0.84 ) (1.72 ) (2.45 ) (3.22 ) (1.80 ) (2.38 ) (2.45 ) Ant. f rac. 0.206 0.089 0.341 0.182 Low d epth 0.379 0.456 0.077 0.210 0.548 0.758 0.681 0.185 0.630 1.599 (0.34 ) (0 .68 ) (0.09 ) ( 0.27 ) (1.29 ) (1.11 ) (1.27 ) (0.44 ) (0.54 ) (2.03 ) Ant. f rac. 0.176 0.100 0.278 0.180 Note: Table 2 6 presents evidence of momentum spillover from the CDS market to the stock market. It presents cross market momentum results. Re ported are stock returns and Sharpe ratios (directly below the returns) of momentum portfolios based on the past 3 month CDS return. Also reported are the alpha coefficients of a Carhart four factor model (Car4) that includes the factors Mkt, SMB, HML. It also presents joint market momentum results. Reported are the stock returns of stock based momentum portfolios and the intersection of CDS and stock portfolios. For example, (P1, sP1) includes firms that fall into both the CDS based momentum portfolio P1 a nd the stock based momentum portfolio sP1. The fraction of firms that have a rating change is also reported. The Newey West t statistic is provided in parenthesis.
56 Figure 2 1 . CDS spreads and credit rating changes . We plo t the cumulative percentage change in CDS spread over the interval [ 90,90] days centered around credit rating upgrade and downgrade events. Depicted are 1,093 downgrade events averaging a 1.32 notch deterioration in credit quality at the plus (+) and minu s ( ) level and 739 upgrade events averaging a 1.21 notch improvement in quality. CDS data is provided by Markit and covers the period January 2003 to December 2011.
57 A B Figure 2 2 . CDS momentum r eturns . A ) C umulative wealth of a portfolio invested in the J=3m K=1m CDS momentum strategy at the beginni ng of the sample period and individual monthly returns of the momentum strategy. The initial portfolio value is $1,000. The momentum strat egy is constructed by purchasing the CDS contracts in the highest quintile of 3 month performance and selling short the CDS contracts in the lowest quintile. The left y axis tracks the portfolio wealth and the right y axis measures the monthly return. B ) 1 2 month moving average of monthly CDS momentum returns. CDS data are provided by Markit. Time period is January 2003 to December 2011.
58 A B C D Figure 2 3 . Sample c omposition. Figure 3 describes the composition of the sample used in this study. A ) H istogram that shows the percentage of the sample in each decile of all NYSE/AMEX listed firms. B ) H is togram that shows the percentage of the sample in each S&P credit rating category. C ) I ndustry histogram. D ) H istogram of CDS depth (i.e., number of contributors to the quoted spread). CDS data are provided by Markit. Time period is January 2003 to Decembe r 2011.
59 Figure 2 4 . Time series of CDS spreads, CDS returns and stock r eturns. Figure 4 displays the time series of CDS spreads (top graph), CDS returns (bottom graph, solid line) and stock returns (bottom graph, dashed line) for all 1,297 firms in our sample from January 2003 to December 2011. CDS data are provided by Markit and stock data are acquired from CRSP.
60 Figure 2 5 . CDS to s t ock m omentum. Figure 5 portrays the profitability of the CDS to stock momentum strategy, the stock momentum strategy using the UMD factor, the stock momentum strategy using our sample of firms, and the joint momentum strategy sorting on both the past CDS return and past stock return. The CDS to stock momentu m strategy is constructed by purchasing the equity of firms in the highest quintile of 3 month CDS return (winners) and selling short the equity of firms in the lowest quintile (losers). The y axis tracks the dollar value of a portfolio that invests in eac h momentum strategy. The initial portfolio value is $1000. CDS data are provided by Markit. Time period is January 2003 to December 2011.
61 CHAPTER 3 THE EXODUS FROM SOVEREIGN RISK To what extent can private sector firms delink themselves from sovere ign risk? It is well known that while governments play an important role in financial markets, they can also generate externalities arising from their actions and their ability to compel and proscribe. For example, governments suffering from large budget d eficits and slow economic growth may limit the ability of private sector firms to service their external debt obligations. Budget deficits may also presage increases in corporate taxes or other corporate infringements, increasing potential default risks. In the aftermath of the recent global credit risk crisis where sovereign credit default swap (CDS) spreads rose dramatically from an average of 30 bps to above 250 bps in 2008 and remained above 100 bps thereafter (see Figure 1), there is an increasing con cern about the transferring of sovereign risks to private sector firms. Anecdotal evidence further suggests that these corporate and investor concerns are warranted, as governments such as Argentina have turned towards heterodox, interventionist economic p olicies whereby they have imposed increasingly invasive corporate restrictive measures such as expropriations and foreign exchange, trade, and capital controls. While both resident and non resident claimants are exposed to potential sovereign risk effects, distant non resident claimants are often particularly concerned about the violation and enforcement of their property rights by foreign governments. Major credit rating agencies term the risk of exchange ability to convert local currency into foreign currency and make transfers to non resident wealth, with T&C risk, private sector debt claims cannot usually be offered better terms and
62 applied broadly to p rivate sector credit ratings by the three major rating agencies for many years. countries with ratings exceeding that of their sovereign. In this paper, we ex amine sovereign ceiling violations (SCVs) in international CDS markets situations in which a corporate CDS spread is lower than its sovereign counterpart with equal contractual terms such as tier, maturity, restructuring clause, and settlement currency. Using credit spread information on firm level 5 year CDS contracts for 2,364 companies in 54 countries during 2004 2011, we provide a comprehensive picture on SCVs and investigate the channels through which private sector firms, in part, delink themselves from sovereign risk. Controlling for firm and country level fundamentals, we find firms exposed to better property rights institutions through their foreign asset positions and stock listings on exchanges with stricter disclosure requirements tend to viol ate the sovereign ceiling rule. These channels capture distinct effects beyond those associated with exposures to foreign country fundamentals. We further find that SCVs from the CDS market unidirectionally predict SCVs in S&P credit ratings. Put together, our results suggest that firm level global asset and information network connections are important in firms delinking themselves from sovereign risk. Our study contributes to three important research streams. First, we contribute to the sovereign ceiling literature (Durbin and Ng, 2005) by using the international CDS market to examine the patterns and determinants of SCVs. Prior studies are limited to a small number of firms with international bond market data, while we provide a comprehensive picture usin g 2,364 firms in 54 countries over an 8 year period. The international CDS market provides an ideal setting to analyze the patterns and determinants of SCVs relative to bond markets. For
63 example, CDS contracts are standardized in their maturities and rest ructuring clauses, regardless of whether the underlying credit entity is corporate or sovereign. This standardization of the counterpart with equal contractual terms. CDS contracts are also more liquid than the bonds issued by the same entities (Longstaff, Mithal, and Neis, 2005). Finally, CDS investors are often perceived as more sophisticated and informed than investors in other financial markets (Acharya and Johnson , 2007; Berndt and Ostrovnaya, 2008), which suggests potentially greater market information content than credit ratings. Second, we also contribute to the international corporate linkages and exposure literature equity cross informational environments across countries. Several studies in this literature highlight how the foreign aff capital structure through the effects of varying tax rates and creditor rights in the countries of their foreign affiliates (Noe, 2000; Desai, Foley, and Hines, 2004). The literature also emphasizes that stricter disclosure of information through equity cross listings in more local external capital markets (Pagano, Roell, and Zechner, 2002; Lins, Strickland, and Zenner, 2005; Bailey, Karolyi, and Salva, 2006). Our global asset distribution and equity cross listing channels extend these international corporate linkage notions. Third legal and institutional characteristics on the structure and pricing of financial claims (La Porta,
64 Lopez de Silanes, Shleifer, and Vishny, 2002; Shleifer and Wolfenzon, 20 02; Qian and Strahan, institutions that the firm is connected to throug h global asset and information connections. We provide evidence on the importance of these two channels as firm level liberalization mechanisms. As shown in Part A of Figure 2, the frequency and magnitude of SCVs in the international CDS market have incre ased, particularly during 2008 2011. The magnitude of the SCVs reached over 100 bps per annum, on average, which is substantial given that the average CDS spreads for companies at the AAA and BBB ratings levels during the same time period were 79 and 189 b ps, respectively. Looking at the cross sectional patterns in Part B of Figure 2, we further see that the SCVs are more prevalent in countries with weak institutional characteristics in terms of property rights/creditor rights protections and low informatio n disclosure requirements. geographic location and its stock cross listing status in foreign markets serve as mechanisms to s T&C risk. We hypothesize that firms whose foreign assets are located in countries with better property rights institutions than their home that govern the fir stock is cross listed alleviates T&C risk concerns through improved transparency a nd reductions in firm informational frictions (Information channel, hereafter). We expect that the influence of
65 both channels is greater during the recent sovereign credit risk crisis given the increases in ign T&C risk concerns) and deterioration of firm operational information. An underlying assumption of our foregoing arguments is that both institutional and informational qualities influence the degree of sovereign T&C risk. We first validate this assumpt ion at the sovereign contract level and document that countries with better property rights, better creditor rights, and more transparent informational environments have lower CDS spreads. More importantly, these effects exist above and beyond those associ ated with both local and regional fundamental economic factors (Ang and Longstaff, 2011; Longstaff, Pan, Pedersen, and Singleton, 2011). We use various proxies for institutional factors, including measures of property rights protection compiled by the Inte rnational Country Risk Guide (ICRG) and Heritage Foundation databases as well as creditor rights proxies constructed by La Porta, Lopez de Silanes, Shleifer, and Vishny (1998) and Djankov, McLeish, and Shleifer (2007). We use the required number of disclos ed items and disclosure frequency mandated by local exchanges as proxies for informational factors (Bushman, Piotroski, and Smith, 2004). We find that for a one standard deviation increase in the degree of property rights, there is a statistically signific ant 64% to 76% drop in the annual average sovereign CDS spread relative to its sample average, depending on the proxy we use for the property rights measures. We also find both economically and statistically significant effects of creditor rights and discl osure requirements on sovereign CDS spreads, even though the magnitudes of their effects are less than half that associated with property rights protection. While we show that both institutional and informational factors influence sovereign level CDS sprea listing status
66 explain SCVs in corporate CDS contracts? As an initial step to answer this question, we analyze the determinants of the difference in CDS spreads between firms and t heir local sovereign market. We find that firms whose foreign assets are located in countries with stronger property rights and creditor rights institutions than their country of domicile have substantially lower relative CDS spreads to that of their home sovereign. Further analyses reveal that stricter listings also reduces the informational c hannel to induce SCVs in the CDS market. Both the institutional and informational factors are economically significant, with an average 26 bps combined impact on the corporate CDS spreads over their sovereigns. We document stronger effects from these two channels during the recent sovereign credit risk crisis. CDS spreads be tween a firm and its home sovereign. This identification highlights the institutional exposure that a multinational company has. Our results are also robust to control ling connected foreign countries through their foreign asset positions. Since we pair firm and sovereign CDS contracts with equal contractual terms, any differenc e in restructuring clause that may affect their recovery rates also does not explain our findings. Do credit ratings reflect these factors in their sovereign ceiling rules? For this test, we use the S&P credit rating difference between a firm and its sover eign. We find that S&P does not
67 However, they do reflect some firm level aggregate foreign sales information, but not the information on locational exposures. Taken together, our credit rating results suggest that major rating agencies often underestimate the effect of the institutional channel in their corporate credit rating assessment, particularly during the recent crisis period. As a further robustness tes t, we confirm that both the institutional and informational channels continue to explain SCVs using various probit regressions with a set of SCV indicator variables that are both transaction cost adjusted and signed. The results are also robust to alternat ive modeling approaches that include ordered probit models using bucketed violation count variables as dependent variables. Coming full circle, we compare the information content of SCVs in the CDS market to those defined with S&P credit ratings. We find t hat the SCVs from the CDS market have strong predictive power on credit ratings based SVCs in subsequent years. In sharp contrast, we do not find evidence of SCVs from credit ratings predicting SCVs in CDS markets. These results suggest that information fr om the CDS market flows to rating agencies, supporting the view that market driven CDS spreads are useful in assessing the credit quality of underlying private sector companies in a more timely and informative manner (Flannery, Houston, and Partnoy, 2010). The remainder of this paper is organized as follows: In Section 2, we provide institutional and pricing background on the CDS market, while in the following Section 3 we develop our main testing hypotheses. In Section 4, we describe our sample and main va riables. We provide our main findings in Section 5. Section 6 provides some concluding comments. Institutional Background on the CDS Market A CDS is a bilateral insurance contract over a fixed term whose payment is contingent on the default of an underlyin
68 obligation acceleration, obligation default, and restructuring. A CDS contract is associated with bonds or any credit assets issued by a reference entity, and it trades on the over the counter market. The party that sells the default insurance (protection seller) pays the full face value of the underlying bond issued by a reference enti purchases the insurance claim (protection buyer) periodically pays fees over the term of the swap to the protection seller until the reference entity defaults and upon the default, delivers the defau lted underlying bond to the protection seller. With this CDS contract, the buyer transfers the default risk of the reference entity to the seller, which bears the full loss given default (LGD). The seller collects fees from protection buyers that are propo rtional to the expected discounted value of LGD. The fees are called CDS spreads, which are quoted in basis points per annum of recovery rate of the underlying bond and the hazard rate that the reference entity could default in an infinitesimal time interval conditioning that the entity does not default by that time. A fair ic discounted value of LGD. Under the flat hazard rate and constant recovery rate assumptions, the fair spread can be simplified to the product of the hazard rate and the constant LGD per $1 notional in a continuous time setup. The reference entity of the swap could be a private sector firm or a government. In the former case, we call the CDS contract a corporate CDS, and in the latter case a sovereign CDS. Ea ch CDS contract comes with its own maturity of the insurance period, a restructuring clause that defines trigger events and further denotes the types of bond that a protection buyer
69 could deliver to a protection seller upon the trigger events, and the base currency of its settlement. Maturities span from a few months to 10 years or more, but 5 year maturities are the clause could have a material impact on a CDS spre ad since it determines the types of trigger events and affects the recovery rate of an underlying obligation by limiting the types of deliverable obligations upon such events. Therefore, it is possible that two CDS contracts with equal contractual terms bu t different restructuring clauses could come up with different spread values, even if they are written on the same reference entity (Packer and Zhu, 2005; Berndt et al., 2007). There are four distinct restructuring types that are mostly different in the de livery options of underlying obligations: 1) Full restructuring (FR), which allows the delivery of any bond of maturity up to 30 years issued by the underlying credit entity; 2) Modified restructuring (MR), which limits the deliverable obligations to those with a maturity of 30 months or less after the termination date of the CDS contract; 3) Modified modified restructuring (MMR), which limits the deliverable assets to be the bonds with a maturity shorter than 60 months for restructured obligations and 30 m onths for all other obligations; and 4) No restructuring (XR), which excludes all restructuring events from trigger events. To make direct comparisons on the credit risks of two different reference entities, all contractual terms of the two CDS contracts m ust be matched to filter out other potential contract risk profiles that are unrelated to the underlying credit risks of the two reference entities. In our analysis, we carefully match and compare the spread of a 5 year private sector CDS to that of its so vereign counterpart with equal maturity, tier, restructuring clause, and settlement currency to
70 eliminate the non credit risk profiles that may cause the two spreads to diverge from each other, even if the two reference entities are perceived to have simil ar credit qualities. Hypothesis Development Suppose that a country is in financial distress and unable to pay back its debt obligations. Consider a hypothetical firm that has all of its assets located within that country, and thus, the assets are largely g overned by the rules and conditions imposed by the distressed government in which the firm and its assets reside. One can envision two scenarios for this purely domestic firm, default (D state) and non default (ND state). In a D may not fully default restructuring process. The government can affect the restructuring process either directly by changing the rules and conditions on ho w the creditors can pull out the residual value of the in an ND proba bility that is induced by potential increases in taxes or other forms of government property rights in either state, lenders are not willing to pay favorabl obligations. However, if the firm had a portion of its assets in other countries with better property expect more transparent and credible enforcement of the rules se
71 (recovery rate effect) and could also lead to a lower perceived def ault probability of the underlying credit entity (default probability effect). This could result in a lower CDS contract spread is lower, ceteris paribus. Better creditor rights enhance the efficiency and speed of post default restructuring and lowers within creditor conflict of interest. Creditor rights are particularly relevant when firms are in a D state, with better creditor righ ts increasing the recovery value that creditors can potentially pull out of debtors in the restructuring process (recovery rate effect) and thus leading to a higher ex ante corporate bond price. This implies a lower CDS contract spread on that bond. When spread is lower, ceteris paribus. error in default risk forecast) and the potential price of the bond issued by a firm whose balance sheet information is limitedly disclosed to its creditors c ould be discounted due to these information opacity reasons. Following the notions used in the home bias literature (Coval and Moskowitz, 2001; Mian, 2006), informational opacity could be a more serious concern to foreign investors, and such investors woul d not be willing to pay favorable bond prices unless the issuer disclosed its accounting information in a more frequent and comprehensive manner. More stringent equity cross listing requirements help alleviate foreign investor concerns since they often req uire additional degrees of disclosure on the frequency and number of items regarding corporate balance sheet information. The aforementioned discussion giv es rise to our last hypothesis:
72 closure requirements, From the above hypotheses, we expect more CDS sovereign ceiling violations for firms whose assets are located in countries with better property rights and creditor rights than those of listed on exchanges with stricter disclosure requirements than its domestic exchange. The effects of these institutional and informational channels on corporate CDS spreads are expecte d to be stronger as the probability Data and Variables We gather daily CDS contract data for corporations and governments over the years 2004 2011 from Markit Group, and we use the 5 year CDS contract since it is the most liquidly trading contract among the CDS maturities (Longstaff et al., 2005; Acharya and Johnson, 2007; level in formation (ticker, country, region, industry, and credit rating composite) and contract specific information (tier, restructuring type, currency, and depth). Depth refers to the number of distinct contributors to each daily CDS spread collected by Markit. We use depth as a measure of CDS liquidity (e.g., high depth indicates high liquidity). the same contract characteristics, including tier, restructuring type, and curren cy. This matching to Sovereign Level Variables We collect country level economic, in stitutional, and other control data from various
73 them into a numerical score that ranges from 0 21 (21 is equal to a AAA rating, 20 is equal to a AA+ rating, etc.) t o form a variable called Sovereign S&P Credit Rating. GDP and Government Debt (% of GDP) are downloaded from the International Monetary Fund (IMF). External government debt is retrieved from the World Bank and scaled by GDP to form the variable External De bt (% of GDP). We collect stock market index returns for each country from Datastream and compute an annual volatility (standard deviation) of weekly log returns to form local Stock Market Volatility. al environments, we collect data from several sources. W e ( La Porta et al. , 1998 and Djankov et al. , 2007 ) : English, French, Germanic, Scandinavian, and Socialist legal origins. We also utilize four va Property Rights measures the extent to which government creates and enforces laws that protect private property and the extent to which government expropriates private property (sour ce: Heritage Foundation). Rule of Law is an assessment of the law and order tradition of the country (source: International Country Risk Guide, ICRG). due to budgetary is sues, political pressure, or a change in government (source: ICRG). nationalizes private property or enterprise (source: ICRG). Next, we use two variables that represent protection to creditors: Creditor Rights measures the legal protection allocated to creditors to have influence over decisions that affect the value of their position. Specifically, this variable takes into acc priority given to creditors in bankruptcy, and (4) the degree to which creditors have 1998; Djankov et al., 2007).
74 Ln(Contract Enforcement Days) records the average number of days it takes to resolve a creditor payment dispute in the courts. We use this variable as a measure of the efficiency and enforcement of the legal system as it perta ins to creditor rights (source: Djankov et al., 2007). requirement variables that capture distinct dimensions of disclosure. Disclosure: Number of Items records a countr wide range of topics, including general information, financial condition, and corporate governance (source: Bushman et al., 2004). rting through the frequency of interim reports (source: Bushman et al., 2004). All these institutional and informational variables are standardized to have mean zero and the standard deviation of one. The country level variable summary statistics are prov ided in Table 3 1. There are 54 countries, with an average annualized 5 year sovereign CDS daily spread of 122.33 bps. The average depth of the 5 year sovereign contract is 7.428. The average sovereign S&P credit rating score is 17.173, which approximately corresponds to A+ rating. The average Ln(GDP) is 6.013 and is equivalent to $408.70 billion (=exp(6.013)). The average government debt issuance Firm Level Variables We gather firm level financial data from Datastream, Worldscope, and Thomson One. While we can readily match firms across these three databases, no reliable firm identifier exists to link the Markit CDS database to these other firm level financial databases. To link the Markit CDS data with the firm level financial data, we manually match the Markit and Datastream databases by firm name, country, and industry. We successfully match 2,364 firms in 54 countries.
75 From the Thomson One database, we collect an annual S&P credit rating for ea ch firm and convert it into a numerical score that ranges from 0 21 (21 is equal to a AAA rating, 20 is rating to form S&P Credit Rating Difference. A difference o f 1 signifies a one sub notch (+/ ) difference. We collect market capitalization, total assets, total debt, long term debt, short term debt, and net cash flow at the end of each calendar year from Worldscope. The variable Ln(Market Capitalization) is defi ned as the natural log of market capitalization. Leverage is defined as total debt over total assets. Short term Debt Fraction is equal to short term debt over the sum of short term debt and long term debt. Cash Flow to Assets is defined as net cash flow o ver total assets. given year. We compute Number of Stock Exchanges by coun ting in Datastream the number of exchanges on which the firm has listed its equity. Scaled Exposure and Extra Disclosure Variables There are a few ways of measuring the percentage of total foreign assets out of total assets for each firm in a given year. W orldscope provides a variable that explicitly states the percentage of total foreign assets, and it also provides the amounts of total assets and total international assets so that foreign assets percentage can be computed by dividing total international a ssets by total assets. We use the former to measure the total foreign asset percentage, Foreign Assets (fraction of Total Assets). However, in instances where no foreign assets percentage is explicitly provided, we use the computed value from the latter. W e similarly construct our total foreign sales percentage variable, Foreign Sales (fraction of Total Sales).
76 In addition to the aggregate view of foreign assets, Worldscope also provides geographically segmented financial information. Up to ten geographic s egments are reported, and companies can input the description of the geographic segment as they choose. Thus, a firm has flexibility to assign a single country or multiple countries to a single geographic segment. We count the number of geographic segments reported by the firm to construct the variable Number of Geographic Segments. level has a net the institutional characteristics of the foreign countries in which the firm has assets with the c environment variables that segments, we define for firm i in year y: where For eign Asset % iy is Foreign Assets (fraction of Total Assets), Segment Institutional Value iys is the institutional value of segment s, Segment Assets iys is the amount of assets in segment s, F oreign Assets iy is the total amount of foreign assets of the firm, and Home Institutional Value iy
77 We create a firm level macroecono mic and institutional variables: GDP, local Stock Market Volatility, Property Rights, Rule of Law, Repudiation Risk, Expropriation Risk, Creditor Rights, and Ln(Contract Enforcement Days). A firm based in a country with low disclosure requirements may ha ve to increase its informational transparency when it lists its equity on an exchange in a country with higher requirements of the countries in which the firm has listed its equity on an exchange. We create i in year y, which are equal to the maximum disclosure ome country: where Home Exchang ec ome country and Exchange c1 , Exchange c2 , etc. refers to the disclosure requirement index values of each country in which the firm lists its equity. Using this method, we construct Extra Disclosure: Number of Items Reported and Extra Disclosure: Frequency an d Count. have mean zero and the standard deviation of one. In Table 3 1, we provide summary statistics on our firm level variables. There are 2,364 firms from 54 c ountries. The average value of the annualized 5 year firm CDS daily spread is 195.511 bps, and the firm CDS contracts have an average market depth of 5.628. The average firm size in our sample is $18.37 billion (=exp(23.634)), with an average firm leverag e of 29.9%.
78 On average, firms in our sample cross list their stock on 2.348 different exchanges (Number of Stock Exchanges) and operate in the 3.693 different segments (Number of Geographic Segments). Main Results Local Institutional and Informational Fac tors in Sovereign CDS Spreads In this section, we first examine the extent to which local institutional characteristics and the degree of disclosure requirement mandated by local stock exchanges are priced in sovereign CDS spreads. A central assumption of our main hypotheses developed in Section 3 is that could be exacerbated if their claims reside in countries with weak property rights institutions and poor informat ion quality. These T&C risk concerns could also apply to the government securities since both the enforceability of the sovereign financial contracts and the availability of the information associated with those sovereign issued claims differ across countr ies, depending on their local institutional and informational qualities. Hence, we could view this sovereign CDS test as a thought experiment such that we treat governm government, we can then examine how such factors affect their CDS spreads in a purely multinational setting. Confirming the influence of those factors on sovereign CDS spreads rights protection (Property Rights, Rule of Law, Repudiation Risk, and Risk of Expropriation), and two proxies for creditor rights protection (Creditor Rights and Ln(Contract Enforcement
79 Days)). For informational factors, we use Disclosure: Number of Items Reported and Disclosure Requirement: Frequency and Count. In Figure 3, we provide scatter average value of 5 year sovereign CDS daily spreads against each of our institutional and informational fact ors using our full 54 country sample. In Parts A, B, C, and D of Figure 3, one can see that sovereign CDS spreads decrease in each of our four local property rights protection measures (Property Rights, Rule of Law, Repudiation Risk, and Risk of Expropriat ion). A similar tendency is observed with our two local creditor rights protection proxies. This tendency is clearly depicted in the results where we use Ln(Contract Enforcement Days) as a proxy for the local creditor rights. The plot shows that the longer it takes for creditors to seize their recovery parts mandates stri cter disclosure requirement in both the number of items to be reported and the reporting frequency. While the figures are suggestive of a relation between sovereign CDS spreads and their institutional and informational environments, they do not control for other factors that are likely to affect these spreads. In Table 3 2 we report annual panel regressions to confirm these findings in multivariate settings. Our main left hand side (LHS) variable is the natural logarithm of the annual average value of 5 yea r sovereign CDS daily spreads. The main right hand side (RHS) variables of this regression are French Legal Origin dummy (La Porta et al., 1998; Djankov et al., 2007) and our institutional and informational variables. Following the sovereign CDS and bond p ricing literature (Claessens et al., 2007; Ejsing and Lemke, 2009; Longstaff et al., 2011; Dieckmann, and Plank, 2011), we control for local
80 fundamental variables, including Stock Market Volatility, Ln(GDP), Government Debt (% of GDP), External Debt (% of GDP). R egional and systemic factors a re important in explaining sovereign CDS spreads ( Longstaff et al. , 20 11 and Ang and Longstaff , 2011 ) . To control for such effects, we additionally control for Ln(Region CDS Spread) -the natural logarith m of average CDS spread for other countries in the same reg ion. W e group corresponding countries into North America, Latin America, Asia, Europe, and the Middle East/Other and compute Ln(Region CDS Spread) for each regio n ( see Longstaff et al. , 2011 ) . We control for year fixed effects to control for a potential global factor effect (Longstaff et al., 2011), which will capture average sovereign CDS spreads effects in each year. Sovereign CDS in our sample could have different restructuring clauses and denominated currencies. We control for the fixed effects associated with these contractual term differences (restructuring type and currency dummies), and we also control for liquidity effects by including Sovereign CDS Depth in the RHS. Standard errors a re clustered at each country level to adjust for any within country autocorrelations of the regression residuals. In column (1) of Table 3 2, we primarily focus on the influence of legal origins on sovereign CDS spreads. L egal rules originated from French civil law are associated with poor protection for outside investors/creditors and less developed financial markets than those from civil law origin ( La Porta et al. , 2008 ) . In column (1), we find significantly higher CDS spreads for the countries from French legal origin. The point estimate of 0.393 for French Legal Origin is statistically significant the 5% level, c onsistent with prior research (La Porta et al., 2008) . Columns (2) to (5) provide the regression results with local property rights m easures. We find a significantly negative point estimates for all four property rights variables (Property Rights, Rule of Law, Repudiation Risk, and Expropriation Risk). The results indicate that
81 countries with poor property rights protection indeed pay h igher costs to raise their government debt. The effects are both economically and statistically significant at the 1% level. In column (2), the point estimate of Property Rights, 0.654, implies 65.4% decrease in sovereign CDS spread for a one standard dev iation increase of that property rights measure. This effect corresponds to a substantial 80 bps decrease in sovereign CDS spreads given that our sample average of sovereign CDS spread is 122.33 bps. We repeat the same exercise for Creditor Rights (colum n (6)) and Ln(Contract Enforcement Days) (column (7)), finding a similar negative effect of stronger creditor rights protection on the sovereign spreads. Both effects are statistically significant at least at the 10% level, and among the two, the effect as sociated with Ln(Contract Enforcement Days) is stronger in both its economic and statistical significance. In columns (8) and (9), we examine the influence of our local informational factors on sovereign CDS spreads. We find a significantly negative poin t estimate, particularly for Disclosure: Number of Items. Overall, these results, together with our previous results with local property rights and creditor rights protection measures, indicate that local institutional and informational qualities affect so vereign credit spreads. These findings validate our key We conduct a horse race regression in the remaining columns (10) to (13) to disentangle the relative effects of our variables. Using the complete specifications, we find that local property rights measures have the most significant effect on sovereign CDS spreads in terms of their economic and statistical significance. Notably, the statistical significance of French Le gal Origin dummy disappears in these four columns except column (11). This suggests that the effect
82 previously captured by the French legal origin dummy in column (1) could be driven by the In each model specification in Table 3 2, our fundamental control variables typically have correct expected signs and statistical significance (positive sign for Ln(Regional CDS Spread), Stock Market Volatility, Government Debt (% of GDP), External Debt (% of GDP ); and negative sign for Ln(GDP)). However, the liquidity measure, Sovereign CDS Depth, has the opposite sign to our prior. In the next section where we analyze the determinants of CDS spread difference between a firm and its local government, we also cont rol for their liquidity effects and find that the sovereign liquidity measure recovers its expected negative sign. Determinants of Sovereign Ceiling Violations: Firm minus Sovereign CDS Spread In this section, we examine the determinants of SCVs in CDS mar ket and find that the difference in spreads tends to be smaller and becomes negative as the firm is exposed to better institutional and informational environments through its foreign asset holdings and its stock cross listing status. Major rating agencies provide guidelines on their sovereign ceiling rule. For example, sovereign are possible where there is strong implicit or explicit support from a highly rated parent in another jurisdiction, and/or there is significant cash flow diversity derived from With this guideline in mind, we first conduct a univariate analysis to examine the extent results in Table 3 status: 1) No. of Stock Exchanges, 2) No. of Geog. Segments, 3) Foreign Assets (fraction of Total), an d 4) Foreign Sales (fraction of Total). Using each variable, we sort our sample firms
83 zero international exposure. As discussed earlier in the introduction, SCVs in CDS market are more pervasive during the recent crisis period, 2008 2011 (see Part A of Figure 2). To incorporate this observed time trend in our analysis, we further decompose our sample into two sub periods: pre crisis period (2004 2007) and crisis perio d (2008 2011). Then, we compute the average firm CDS spread in each bucket for the two sub periods and report the results in Table 3 3. In Table 3 3, one can see that average corporate CDS spreads in both sub periods are lower for the firms whose stocks a re cross listed in multiple foreign exchanges (No. of Stock Exchanges) and whose assets are located in multiple countries (No. of Geog. Segments). These CDS s preads between purely domestic firms (Zero) and multinational firms (High) become larger during the crisis period. Assets (fraction of Total)), the CDS spreads of multina tional firms (High) are lower than purely domestic companies (Zero). These results are also consistent with the sovereign ceiling guidelines reported by major rating agencies. The differences again increase in their magnitudes during the crisis period (pre crisis, 22.55 bps vs. crisis, 113.56 bps). We find similar results when we sort firms using Foreign Sales (fraction of Total). Overall, Table 3 3 shows that the CDS spreads of multinational firms are lower than those of purely domestic firms in terms of their asset and equity cross listing information exposures. In Table 3 exposure variables (Foreign Assets (fraction of Total) and Foreign Sales (fraction of Total)). We
84 incorp institutional qualities that the multinational firm is exposed to across its various markets as discussed in Section 4.3. In Table 3 3, we provide evidence on the exten multinational status affects its own CDS spread and use our scaled institutional exposure measures as sorting variables. We consider our four property rights proxies and two creditor rights proxies to show the relevance of these scaled spread. CDS spreads of firms whose foreign assets are exposed to better property rights institutions (Net High) have lower average CDS spreads than those exposed to poor property rights institutions (Net Low). We also find a similar tendency with creditor rights protection measures. The difference in average corporate CDS spreads between Net High and Net Low firms are indeed widened during the crisis period, suggesting that T&C risk concerns became more severe d uring the crisis period. Overall, our univariate results in Table 3 3 provide some support for our main hypotheses, though they do not control for other factors that are likely to influence the CDS spread. In Table 3 4, we extend our findings to examine th e potential determinants of SCVs in the CDS market by running annual panel regressions in a multivariate setting. In these regressions, we use the CDS spread difference between a firm and its local government as our main LHS variable (i.e., Firm Sovereign CDS Difference). As explanatory variables, we consider our scaled institutional and informational factors that fully incorporate the locational listing on foreign exchanges. The scaled variables a re: 1) for the property rights protection, Scaled Exposure: Property Rights, Rule of Law, Repudiation Risk, and Expropriation Risk; 2) for the creditor rights protection, Scaled
85 Exposure: Creditor Rights Protection, Ln(Contract Enforcement Days); and 3) fo r disclosure requirement, Extra Disclosure: Number of Items Reported and Frequency and Count. We motivate our use of the Firm Sovereign CDS Difference as our LHS variable from the definition of a SCV. A SCV is defined as an event where Firm Sovereign CDS Difference becomes strictly negative. Hence, a lower value of Firm Sovereign CDS Difference in a given firm year does not necessarily imply that the firm year is indeed a SCV event. However, the analysis on the determinants of this CDS spread difference be tween a firm and its sovereign counterpart could be informative on the potential determinants of the actual SCV events. As a first step to explore the potential determinants of SCVs, we use a regression framework to examine the relationship between the Fir m Sovereign CDS Difference and both of our institutional and informational factors. The regression is conducted at each firm year level, and we include the following firm level, time varying controls: Ln(Market Capitalization), Leverage, Short term Debt Fr action, Cash Flow to Assets, and Stock Return Volatility. The use of all these control variables is well motivated in the corporate credit risk literature (see Collin Dufresne et al., 2001, among others). We control for country and the year fixed effects. The inclusion of both fixed effects alleviates potential concerns of any omitted regional and global factors that may affect both the firm and sovereign CDS spreads (Longstaff et al., 2011). Any regional fixed effects are implicitly explained by our count ry level fixed effects. Different industries may have different degrees of exposures to the domestic business cycles (Durbin and Ng, 2005). To capture this industry heterogeneity, we also control for the industry fixed effects in the regression. To account for the liquidity effects of both firm and sovereign CDS contracts, we include both Firm CDS Depth and
86 base currencies, and thus, each matched firm sovereign CDS pair used in this analysis could vary in these two contractual terms. We control for such contractual term variations by including restructuring type and settlement currency dummies, and cluster t he standard errors at the firm level to adjust for within firm persistence of the regression residuals. Table 4 reports our main results. In the first four columns (1) to (4), one can see significant negative effects of scaled property rights variables on the firm sovereign CDS spread difference. In column (1), the point estimate on Scaled Exposure: Property Rights is 12.09, indicating that for a one standard deviation increase in that scaled property rights variable, there is approximately a 12.09 bps de crease in excess corporate CDS spread over its local three columns, we see both qualitatively and quantitatively similar results for the other three scaled property rights variables. The effects with one standard deviation increase in their values range from 9 to 13 bps. In the next two columns (5) and (6), we use the scaled creditor rights measures. Both measures have the correct expected signs (negative fo r Scaled Exposure: Creditors Rights Protection and positive for Scaled Exposure: Ln(Contract Enforcement Days)), and their point estimates are statistically significant. Their economic impact with a one standard deviation increase suggests a 2.41 to 9.22 b ps lower corporate CDS spread relative to its local In columns (7) and (8), we repeat the same exercise with our scaled informational factors, Extra Disclosure: Number of Items Reported and Frequency and Count. Both variables have significantly negative point estimates. For example, the economic magnitude of the effect of
87 Extra Disclosure: Number of Items Reported is 13.36 bps for a one standard deviation increase in that variable. In the last four columns (9) to (12), we run a hor se race regression with all of our scaled institutional and informational variables. We use each of two scaled creditor rights and disclosure related variables in those specifications, while we use each scaled property rights measures one at a time. We fin d that most of the scaled exposures based on property rights, creditor rights, and extra disclosure related variables significantly explain the deviation of corporate CDS spreads from their sovereign counterparts. Among them, scaled property rights and ext ra disclosure related variables are most significant in explaining SCVs in the CDS market in terms of both their economic and statistical significance. It could be argued that our scaled institutional variables and extra disclosure requirement measures sim ply capture the aggregate exposure of the firm to foreign institutions, not the listings. To show the robustness of our results to such concerns, we repeat the same regressions with add itional controls, Foreign Assets (fraction of Total Assets) and Foreign Sales (fraction of Total Sales). Table 3 5 provides these results. The point estimates of these two additional controls are not statistically significant in most of our specifications, except columns (6) and (11). At the same time, the point estimates of both our scaled property rights and extra disclosure requirement variables are little changed in their economic magnitudes and statistical significance, even after controlling for these additional aggregate foreign assets/sales exposures. This confirms that the unterpart. It should be noted that the effect of Scaled Exposure: Creditor Rights Protection is not generally
88 robust, showing the opposite sign of its effect on the Firm Sovereign CDS Difference, which is particularly observed in column (5). However, the v ariable recovers its expected sign and statistical significance in the last four columns of Table 3 5 where we conduct the horse race regressions using all of our institutional and informational variables. Another potential concern is that our scaled insti tutional and informational variables are informational variables have expl anatory power for the Firm Sovereign CDS spread difference, we control for scaled foreign country fundamental exposures: Scaled Exposure: Ln(GDP) and Stock Market Volatility. The results in Table 3 5 show that our findings are not driven by these foreign f undamental exposure concerns and confirm that indeed our scaled institutional factors and extra disclosure requirement measures capture their own incremental effects on the firm sovereign CDS spread difference. T&C risk concerns are expected to be more se vere to foreign investors when the local government is in distress, with a high default probability. From Figure 1, we show that there is a dramatic increase in global 5 year sovereign CDS spreads, and thus, we expect that the effects of scaled property ri ghts, creditor rights, and informational factors are stronger during this crisis period. To test this hypothesis, we interact each of these variables with a crisis indicator variable that takes a value of one for 2008 2011 and zero otherwise. The results a r e reported in Table 3 5 . We find that scaled property rights measures and extra disclosure requirement have significant incremental effects in explaining the SCVs, especially during the crisis. The coefficient estimate on the interaction term between Scal ed Exposure: Creditor Rights and the crisis dummy has the
89 opposite sign to our expectation. However, the effect is not statistically significant, with a t stat on the interaction term of approximately 0.12 0.88. While we control for industry fixed effects, there may be some additional potentially interesting cross industry T&C risk effects. In particular, banks arguably face potentially higher T&C risk since they are often perceived to be more likely to be nationalized than retailers, and governments in tro uble may find banks as the most readily accessible source for foreign exchange (Durbin a nd Ng, 2005). A recent stud y theoretically demonstrates how a financial edit risk upon the announcement ( Acharya et al. , 2011 ) . In untabulated results, we examine such interactive effects using a financial sector dummy with our four scaled property rights, two scaled creditor rights, and two extra di sclosure variables. We find some support consistent with higher banking T&C risk with our four scaled property rights variables and Extra Disclosure: Number of Items Reported, particularly during the recent crisis period. However, the financial sector inte ractive effect is generally weak for the other measures such as the two scaled creditor rights variables and another extra disclosure variable related to the frequency and count. In additional untabulated robustness checks, we further conducted various co nditional analyses using alternative regression specifications such as quantile regression approaches that are less sensitive to potential outlier effects. We again obtained similar quantitative and qualitative results to our reported findings. All these untabulated results are available on request. Determinants of Sovereign Ceiling Violations: S&P Ratings To what extent do ratings agencies account for geographical foreign asset segmentation information or foreign equity cross listing disclosure requiremen t information in their sovereign operations? More importantly, how do their ratings perform compared to CDS market results,
90 particularly during the crisis pe riod? In this section we address these questions by using an experimental design similar to our earlier firm sovereign CDS spread difference analysis. difference between a firm and its local government. Table 6 provides some preliminary evidence on the information content of sovereign ceiling credit rating differences. We start our analysis with aggregate foreign exposure measures, together with the simple counts of the number of stock exchanges and the number of geographic number of stock exchanges. Number of Geographic Segments and Foreign Sales (fraction of Total Sales) have a significantly positive influence on the relative corporate credit ratings to their sovereign counterparts, particularly during the crisis. However, the rating spreads do not appear to utilize the aggregate foreign asset information in either sample perio d. In Table 3 6 , we also directly examine whether our scaled property rights, creditor rights, and extra disclosure variables are incorporated into the corporate credit ratings relative to their sovereign counterparts. Looking at columns (1) to (4), none o f the scaled property rights variables affects the credit rating difference between a firm and its local government, indicating credit rating. In columns (5) and (8), there is some evidence that the Scaled Exposure: Creditor Rights and Extra Disclosure: Frequency and Count are reflected in the relative corporate credit ratings. However, in column (6), the point estimate sign on Scaled Exposure: Ln(Contract Enfo rcement Days) is opposite to our expectation (i.e., negative expected sign for this variable). Overall, our results suggest that S&P sovereign ceiling credit rating differences do not However, we provide some
91 evidence consistent with S&P sovereign ceiling credit rating differences taking into account a listing locations, particularly during the crisis period. In comparison with our cross listing evidence from the CDS marke t, the overall cross listing impact on the sovereign ceiling credit rating difference is statistically weaker. asset geographical information as well as equity cross li sting information into their corporate and sovereign CDS pricing, we re run our earlier firm sovereign CDS spread difference regressions, while also controlling for S&P credit rating differences. Looking at t he results in Table 3 6 , we find that the effect ive property rights and extra disclosure requirement in the number of reported items still well explain the CDS spread difference between a firm and its local government, especially during the crisis period. These results suggest some superior informationa l efficiency in the CDS market relative to credit ratings, and also suggest that there could be potential informational spillovers from CDS markets to credit ratings. We address these informational spillovers in Section 5.3. Determinants of Sovereign Ceili ng Violations: Probit and Ordered Probit Analysis So far, our results suggest that cross sectional differences in firm relative CDS spreads to their local government spreads are well exposures and inf listing status. A more direct test on the determinants of SCVs in CDS market requires that we define a SCV of its local government. We define an SCV event at each firm year level in the following two ways: 1) a dummy counterpart in at least one trading day during a year (F irm Violator Dummy (Simple method))
92 and 2) a dummy variable that takes a value of one if the annual average value of daily CDS spreads of a firm is lower than that of its sovereign counterpart (Firm Violator Dummy (Mean method)). The former dummy captures both transient and permanent violations, whereas the latter captures relatively more permanent violations that last approximately for a year. To account for violation frequency, we also use a third method Firm Violator Buckets (Bucket Method). In this me thod, we classify firms into four violation buckets (0, 1, 2, 3) based on the number of days a firm experiences a SCV using the Simple method. The bucket number of 0 indicates a non violator, 1 indicates an infrequent violator, 2 indicates a medium violato r, and 3 indicates a frequent violator. These SCV classification approaches, however, are subject to measurement errors that mainly arise from transaction costs involved in both corporate and sovereign CDS contracts. Instead of using ad hoc cutoffs to defi ne meaningfully lower corporate CDS spreads relative to their sovereign counterparts, we utilize the information on average bid ask spreads available for both firm and sovereign CDS contracts to create transaction cost cutoff bounds. We compile the bid and ask quotes for both firm and sovereign CDS contracts from an alternative data source, Credit Market Analysis (CMA), provided through Datastream. The CMA data is less accurate and complete than the Markit data, so merging these datasets at the firm level r esults in a sample reduction and introduces potential errors. To avoid these problems, we filter out distinctively inconsistent spread observations between the two databases, take a country average by year, and apply the transactions costs across all firms in the respective country. This approach enables us to minimize both the sample reduction and the introduction of data errors. Using the transaction cost adjusted dummy variables and bucketed frequency variables, we run both probit and ordered probit reg ressions. As shown previously in Figure 2, SCVs are
93 more pervasive during the crisis period, so we focus on the crisis period in this analysis. We use the s ame specification from Table 3 4 and cluster the standard errors at the firm level. In t he first fou r columns of Table 3 7 , we use Firm Violator Dummy (Simple Method). The results reported in columns (1) to (4) confirm that both scaled institutional and informational factors well explain the intensity of a firm to violate the sovereign ceiling rule durin g the crisis period. Among them, our four scaled property rights protection proxies and the extra disclosure variable in number of items reported appear to more significantly explain the intensity of a firm to violate the sovereign ceiling rule. When we re peat the same analysis using Firm Violator Dummy (Mean Method) in columns (5) to (8), we obtain largely similar results to the Simple Method results. In those columns, all of our institutional and informational factors significantly explain the intensity o f a firm to violate the sovereign ceiling rule in the CDS market at least at the 10% level. Looking at the ordered probit results in columns (9) to (12), we again obtain results consistent with our earlier findings. Overall, we find that the better proper ty and creditor rights that a firm is exposed to through its foreign asset positions and the stricter disclosure requirement mandated by foreign stock exchanges where its stocks are cross listed, the more likely it is that the firm is regarded as a safer e ntity than its local government. Taken together, our results in Table 3 7 confirm our earlier findings reported in Table 3 4 where we use the firm sovereign CDS spread difference regressions. Information Spillovers from CDS Markets to Credit Ratings We pre viously show that our scaled institutional and informational factors explain CDS spread differences between a firm and its local government, above and beyond what is explained by their credit rating di fference (Table 6 ). In this section, we revisit these r esults and more
94 directly test whether there is any information spillover on SCVs from CDS markets to credit ratings. As a motivating analysis, in Figure 4, we show the percentage of firms that are already violators of the sovereign ceiling rule in the CD S market 1 or 2 year prior to each event in year t that a firm first time becomes a violator of the sovereign ceiling rule based on the S&P credit ratings. There, we find that more than 80% of the firms that become S&P sovereign ceiling violators in year t are also contemporaneously classified as CDS SCVs. In addition, among these becoming S&P violators in year t, more than 60% of the them (=50%/83.30%) were already correctly classified as SCVs in the CDS markets 1 year prior to the event year t. We also f ind that even 2 years prior to the event year t, the CDS markets correctly classified approximately a half of the firms (=40.27%/83.30%) that will eventually become sovereign ceiling rule violators in the S&P credit ratings in two year time. A sharp contra st is that we do not find any significant contemporaneous nor leading associations of the S&P SCVs with future violations of the firms in the sovereign ceiling rule with CDS spreads. This result clearly illustrates how much richer information the CDS marke t has on the fundamental credit risk difference between a private sector firm and its government of domicile, relative to the information available in the S&P credit ratings. Next, we confirm these findings in Figure 4 in a multivariate regression sett ing a nd report the results in Table 3 8 . W e measure the extent to which the CDS market has information content in S&P credit ratings. In this specification, we use a dummy that denotes a firm becomes Violation variable has a binary value of either one or zero, if it is used in the LHS of a
95 n the previous year. In the RHS of the regression in column (1), we use the 1 year and 2 year lagged values of change in the CDS violation dummy (based on Firm Violator Dummy (Mean Method)) where the change in the CDS violation it is used in the RHS of a regression, could have a value of one (or negative one) if the firm becomes (or stops being) a CDS violator in a given year and zero otherwise. We further control for Ln(Market Capitalization), Leverage, Short term Debt Fraction , Cash Flow to Assets, and Stock Return Volatility in the RHS of the regression. We also control for industry, region, and crisis period, restructuring type fixed effects. In the next model (2), we additionally control for the 1 year and 2 year lagged valu es of change in the S&P in the RHS of a regression, could have a value of one (or negative one) if a firm becomes (or stops being) an S&P violator in a given year a nd zero otherwise. These regressions are all probit regressions, and we cluster the standard errors at the firm level. From column (1), it appears that both 1 year and 2 year lagged values of change in CDS violation dummy significantly predict the event t hat a firm becomes an S&P violator in the subsequent year at the 1% level. These results are consistent with our earlier findings in Figure 4 that there is leading information in the CDS markets on the potential violations of the sovereign ceiling rule in th e S&P credit ratings. In Table 3 8 where we further control for the 1 and 2 year iolation is hardly changed. The results in the first two columns are not fully convincing to show one way information spillovers from CDS markets to credit ratings since the opposite prediction could
96 also be as likely. To show that this reverse causality i independent variables in model (3). In the next model (4), we further include the 1 and 2 year in the RHS of the regression. In both columns, we do not find any evidence that SCV information flows from credit ratings to the CDS market, which is again consistent with our earlier results reported in Figure 4. We conduct several robustness tests in the remaining models (5) to (8). In model (5), we focus on not only the becoming an S&P violator event but also the stopping an S&P violation event and jointly estimate the likelihood of the occurrence of each event using a multinomial logit regression. We re peat the same multinomial logit regression in model (6), while we instead use the CDS market events as our main dependent variables. There we find that the change in violation status of a firm in the CDS market predicts both becoming an S&P violator and st opping an S&P violation in the subsequent years ( model (5)), whereas we do not find any significant predictability of the variables based on the S&P credit ratings on the future events of SCVs in the CDS market ( model (6)). Further analysis using an ordere d probit regression confirms the robustness of our results to alternative regression specifications. Overall, our results in Table 3 8 , together with those in Figure 4, confirm one way information spillover on SCVs from the CDS market to credit ratings. T hese results support the view recently proposed by several researchers that market CDS spreads are useful in assessing the credit quality of underlying private sector companies in a more timely and informative manner (Flannery et al., 2010). Concluding Rem arks risk literature. We fill this gap by investigating the underlying nature and determinants of
97 sovereign ceiling violations (SCVs) in both the international CDS market and the credit ratings assigned by one of the three major rating agencies, S&P. Using 5 year CDS spreads on 2,364 private sector companies in 54 countries, together with their sovereign CDS counterparts with equal contractual terms, we document that SCVs in the CDS market become more pervasive during the recent sovereign credit risk crisis, and they are more frequently observed in countries with weak institutional and informational qualities. To explain these temporal and cross sectional patterns of SCVs in the international CDS market, we first hypothesize that both the institutional and informational qualities of a local transfer and convertibility risk, t (Institutional and Informational) through which private sector companies could delink placing some firm assets in foreign countries with better property and creditor rights protection, and 2) cross listing their stocks on foreign exchanges where stricter information disclosure is mandated. We provide evidence showing that both our institutional and informational cha nnels help explain the temporal and cross sectional patterns of SCVs. These channels also capture distinct effects beyond those associated with firm and country level fundamentals as well as their exposures to foreign country fundamentals. Indeed, firms t hat are effectively exposed to better property/creditor rights institutions and firms whose stocks are trading on foreign exchanges with stricter information revelation requirements exhibit meaningfully lower credit spreads than those of their home governm ents. However, we find little support that S&P ratings fully reflect these important global networks that multinational firms have to improve their foreign institutional and
98 information exposures. In line with this finding, we provide further evidence show ing that SCVs from the CDS market unidirectionally predict SCVs in S&P credit ratings, suggesting that CDS spreads measure underlying corporate credit risks in a more timely and informative manner than credit ratings assigned by S&P. Taken together, our r esults suggest that firms can, in part, avoid their sovereign level global asset connections in countries with better institutional and informational environments.
9 9 Table 3 1. Summary s ta tistics . Note: T able 3 1 presents the sovereign and firm level summary statistics for 54 countries and 2,364 firms. Credit Default Swap (CDS) spreads are provided in basis points (bps). N represents country year observations at the sovereign level and firm year observati ons at the firm level. Variable Name Mean Stan. Dev. Median Min Max N Sovereign l evel Sovereign CDS s pread (bps) 122.33 246.63 56.91 1.00 3148.80 361 Ln(s overeign CDS, bps) 0.89 1.67 0.56 4.60 3.45 361 Sovereign CDS d epth 7.42 3.37 7.35 2.00 16.10 361 Sove reign Credit r ating 17.17 3.95 18.00 1.00 21.00 329 Ln(region s overeign CDS, bps) 4.01 1.58 4.57 0.12 7.49 351 Stock market v olatility 0.04 0.01 0.03 0.01 0.11 344 Ln(GDP, $ Billions) 6.01 1.35 5.79 2.51 9.60 357 Govt d ebt (% of GDP) 55.10 34.16 49.01 4.40 222. 53 355 External d ebt (% of GDP) 0.71 1.28 0.38 5.48 10.37 344 Property r ights 0.00 1.00 0.14 2.09 1.26 360 Rule of l aw 0.00 1.00 0.16 2.43 1.14 341 Repudiation r isk 0.00 1.00 0.46 1.86 1.35 341 Expropriation r isk 0.00 1.00 0.54 2.57 1.0 6 289 Creditor r ights 0.00 1.00 0.05 1.93 1.83 342 Ln(contract enforcement d ays) 0.00 1.00 0.25 2.06 2.29 336 Disclosure (no. of i tems) 0.00 1.00 0.27 1.99 1.65 274 Disclosure (f req uency and c ount) 0.00 1.00 0.23 2.71 1.24 274 Firm l evel Firm CDS s pread (bps) 195.51 468.41 85.82 4.06 8972.10 13204 Firm Sov CDS d iff (bps) 155.26 463.90 48.78 3106.18 8929.78 13204 Firm CDS d epth 5.62 4.12 4.13 2.00 27.36 13204 S&P r ating d ifference 6.7 6 3.55 7.00 20.00 12.00 9370 Ln( m arket c ap, $) 23 .63 2.36 23.36 10.10 32.62 13204 Leverage ( d ebt to a ssets) 0.29 0.16 0.28 0.00 0.72 13204 Short term d ebt f raction 0.49 0.25 0.49 0.00 1.00 13204 Cash f low to a ssets 0.07 0.05 0.0 7 0.01 0.20 13204 Stock r eturn v olatility 0.04 0.02 0.04 0.00 0.17 11850 Number of s tock e xchanges 2.34 1.08 2.00 1.00 8.00 13148 Geographic s egments 3.69 2.33 3.00 1.00 10.00 11966 Foreign a ssets 0.18 0.24 0.06 0.00 1.00 13204 Foreign s ales 0.29 0.29 0.21 0.00 1.00 13167 Scaled exposure and extra disclosure Ln(GDP) 0.00 1.00 0.00 5.11 5.76 12223 Stock m arket v olatility 0.00 1.00 0.12 11.26 12.97 12363 Property r ights 0.00 1.00 0.32 4.37 5.35 12240 Rule of l aw 0.00 1.00 0.34 5.14 5.81 11663 Repudiation r isk 0.00 1.00 0.43 3.86 5.68 11569 Expropriation r isk 0 .00 1.00 0.03 7.79 8.35 11653 Creditor r ights 0.00 1.00 0.03 4.30 6.07 11618 Ln(contract e nforce ment d ays) 0.00 1.00 0.24 13.62 9.10 11141 No. of i tems 0.00 1.00 0.32 0.32 8.16 12861 Freq uency and c ount 0.00 1.00 0.28 0.28 10.56 12867
100 Table 3 2 . Determinants of sovereign credit r isk . (1 ) (9 ) (10) (11 ) (12 ) (13 ) Control v ariables Ln(region CDS s pread ) 0.418 ( 9.72 ) 0.369 ( 8.75 ) 0.331 ( 7.38 ) 0.319 ( 7.59 ) 0.332 ( 8.03 ) Stock market v olatility 1.61 6 ( 4.16 ) 2.981 ( 5.67 ) 3.196 ( 5.94 ) 2.742 ( 5.38 ) 2.859 ( 5.58 ) Ln(GDP ) 0.409 ( 6.35 ) 0.243 ( 4.54 ) 0.227 ( 4.08 ) 0.176 ( 3.18 ) 0.111 ( 2.06 ) Government d ebt (% of GDP ) 0.011 ( 5.06 ) 0.006 ( 3.82 ) 0.008 ( 4.75 ) 0.009 ( 5.23 ) 0 .0083 ( 4.95 ) External d ebt (% of GDP ) 0. 046 ( 1.50 ) 0.113 ( 3.73 ) 0.098 ( 3.21 ) 0.111 ( 3.84 ) 0.130 ( 4.33 ) Sovereign CDS d epth 0.111 ( 6.25 ) 0.077 ( 4.17 ) 0.087 ( 4.63 ) 0.098 ( 5.68 ) 0.076 ( 4.30 ) French legal o rigin 0.393 ( 2.07 ) 0.158 ( 1.01 ) 0.291 ( 1.88 ) 0.071 ( 0.44 ) 0.103 ( 0.73 ) Property r ights 0.654 ( 9.21 ) 0.560 ( 6.03 ) Rule of l aw 0.639 ( 8.77 ) 0.426 ( 5.00 ) Repudiation r isk 0.701 ( 10.47 ) 0.749 ( 7.37 ) Expropriation r isk 0.7 63 ( 11.03 ) 0.629 ( 7.14 ) Credito r r ights 0.154 ( 1.78 ) 0.048 ( 0.80 ) 0.018 ( 0.30 ) 0.047 ( 0.80 ) 0.028 ( 0.53 ) Ln(contract enforcement d ays ) 0.370 ( 3.98 ) Disclosure: number of i tems 0.251 ( 2.96 ) 0.037 ( 0.53 ) 0.050 ( 0.69 ) 0.044 ( 0.63 ) 0.056 ( 0.89 ) Disc losure: f requency 0.037 ( 0.46 ) O bservations 252 252 252 252 Adjusted R squared 0.897 0.880 0.906 0.898 Note: Table 3 2 presents panel regressions on the natural log of the annual mean sov ereign Credit Default Swap (CDS) spread. CDS data is provided by Markit, and spreads are expressed in basis points (bps). The analysis covers 54 countries over the years 2004 2011. The panel regressions are at yearly intervals and utilize standard errors t hat are clustered at the country level. T statistics are provided in parenthesis .
101 Table 3 3 . CDS spreads and multinational f irms . Pre crisis p eriod (2004 2007) Crisis p eriod (2008 2011) International p resence Zero Low Med High High Zero Zero Low Med High High Zero Stock e xchanges 106.05 78.14 45.42 27.16 78.89 *** 260.71 206.72 116.89 143.58 117.13 *** 140.65 112.89 73.46 25.00 0.0000 289.88 239.63 90.28 121.32 0.0000 Geographic s egments 96.18 84.02 69.30 66.56 29.62 *** 275.13 223.9 1 167.68 124.00 151.14 *** 122.23 114.14 113.92 67.99 0.0000 311.89 256.57 173.35 64.49 0.0000 Foreign a ssets 87.50 81.40 68.61 64.95 22.55 *** 250.96 202.16 176.83 137.40 113.56 *** 132.73 101.96 113.35 80.01 0.0000 293.49 224.32 203.52 96.70 0.0000 Foreign s ales 88.74 78.24 68.94 74.35 14.39 *** 260.24 241.15 200.51 136.86 123.38 *** 125.37 101.26 134.90 89.55 0.0001 305.88 259.02 238.92 97.45 0.0000 Net Low Net High High Low Net Low Net High High Low Property r i ghts 84.52 80.95 3.560 0 238.59 215.05 23.54 *** 131.50 119.11 0.3263 284.88 238.81 0.0005 Rule of l aw 84.05 82.44 1.610 0 235.79 222.85 12.95 ** 131.46 121.67 0.6404 280.25 254.68 0.0491 Repudiation r isk 91.64 69.82 21.82 ** * 240.77 216.18 24.59 *** 138.70 106.28 0.0000 281.52 253.25 0.0002 Expropriation r isk 95.38 83.46 11.92 *** 236.90 220.03 16.87 ** 154.24 127.94 0.0000 276.96 258.89 0.0114 Creditor r ights 91.01 65.74 25.28 *** 238.45 212.56 2 5.88 *** 136.01 104.58 0.0000 278.98 248.84 0.0000 Ln(contract enf. d ays) 60.00 90.68 30.67 *** 191.91 241.83 49.92 *** 99.84 134.61 0.0000 229.83 280.19 0.0002 Note: Table 3 3 presents the average firm CDS spread (in basis points) of 2,3 64 firms in 54 countries at varying degrees of international presence (for a total of 13,204 firm years). It examines the variables Number of Stock Exchanges, Number of Geo graphic Segments, Foreign Assets (fraction of Total Assets), and Foreign Sales (frac tion of Total Sales). Each variable is divided into four nt cludes firms that have a net negative exposure. A t test is performed between High and Zero or High and Low bucket groups. T statistics are provided in italics , and statistical significance is indicated by *, **, *** for 10%, 5% and 1%, respectively.
102 Table 3 4 . Firm sovereign CDS spread d ifference . (1) (8) (9) (10) (11) (12) Firm fundamentals & liquidity e ffects Ln(market c apitalization) 96.74 *** ( 3.82) 85.72 *** ( 3.28) 83.99 *** ( 3.25) 81.85 *** ( 3.21) 83.94 *** ( 3.25) Leverage 307.9 0 * ** (7.84) 302.2 0 ** (7.33) 301.3 0 *** (7.31) 302.8 0 *** (7.32) 303.1 0 *** (7.36) Short term debt f raction 32.77 * (1.80) 39.00 * (1.84) 38.53 * (1.85) 39.89 * (1.88) 38.42 * (1.83) Cash flow to a ssets 50.44 ( 0.40) 25.69 ( 0.18) 35.19 ( 0.25) 31.38 ( 0.22) 32.17 ( 0.24) Stock return v olatility 749.45 *** (10.36) 768.93 *** (9.95) 766.63 *** (9.94) 767.75 *** (9.92) 765.89 *** (9.96) Firm CDS d epth 1.92 *** ( 2.95) 1.50 ** ( 2.18) 1.51 ** ( 2.20) 1.64 ** ( 2.37) 1.70 ** ( 2.47) Sovereign CDS d epth 3.39 * (1.74) 4.47 ** (2.12) 4.54 ** (2.17) 4.60 ** (2.19) 4.59 ** (2.19) Scaled exposure & extra d isclosure Property r ights 12.09 ** ( 2.34) 12.26 ** ( 2.21) Rule of l aw 9.01 *** ( 2.72) 9.22 ** ( 2.43) Repudiation r isk 13.09 *** ( 2.92) 12.67 *** ( 2.69) Expropriation r isk 12.21 *** ( 3.22) 11.59 *** ( 2.91) Creditor rights p rotection 2.41 * ( 1.69) 2.29 * ( 1.68) 2 .48 * ( 1.65) 1.04 ( 1.30) 0.78 (1.22) Ln(contract enforcement d ays) 9.22 ** (1.97) Number of items r eported 13.36 *** ( 5.30) 12.18 *** ( 4.58) 12.17 *** ( 4.53) 11.18 *** ( 4.09) 11.19 *** ( 4.36) Frequency and c ount 12.77 ** ( 2.46) Observations 10079 10140 10068 10158 Adjusted r s quared 0.249 0.248 0.249 0.248 Note: Table 3 4 present s the corresponding annual m ean sovereign CDS spread. CDS spreads are provided by Markit and are expressed in basis points. The analysis includes 2,364 firms in 54 cou ntries over the years 2004 2011 . The panel regressions are at yearly intervals and utilize standard errors that are c lustered at the firm level. Fixed effects include year, country, industry, CDS restructuring type, and CDS Currency. T statistics are provided in parenthesis, and statistical significance is indicated by *, **, *** for 10%, 5% and 1%, respectively.
103 Table 3 5 . Firm sovereign CDS spread difference: robustness c hecks . (1) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) International presence Variable Variable x Crisis Foreign assets 0.555 ( 0.86 ) 0.682 ( 1.06 ) 0.754 ( 1.13 ) 0.998 ( 1.71 ) 0.606 ( 0.89 ) Foreign sales 0.227 ( 0.89 ) 0.324 ( 1.14 ) 0.357 ( 1.21 ) 0.470 ( 1.81 ) 0.289 ( 0.95 ) Scaled exp: Ln(GDP ) 3.140 ( 2.39 ) 5.721 ( 1.60 ) Scaled exp: market vol 3.141 ( 1.99 ) 1.195 ( 1.29 ) Scaled exposure and extra disclosure Property rights 10.840 ( 2.11 ) 10.770 ( 2.46 ) 13.880 ( 2.14 ) 12.510 ( 2.14 ) 9.429 ( 2.11 ) 4.521 ( 1.76 ) Rule of law 7.295 ( 2.30 ) 7.412 ( 2.04 ) Repudiat ion risk 10.800 ( 2.57 ) 10.480 ( 2.37 ) Expropriation risk 9.900 ( 2.80 ) 9.348 ( 2.51 ) Creditor rights 3.118 ( 1.90 ) 0.213 ( 1.76 ) 1.588 ( 1.72 ) 1.131 ( 1.63 ) 1.286 ( 1.56 ) 3.396 ( 1.60 ) 1.11 7 ( 1.31 ) 3.153 ( 1.76 ) 4.347 (0.88 ) Ln(contract enf. days ) 7.728 ( 1.72 ) No. of items reported 12.670 ( 5.04 ) 11.470 ( 4.37 ) 11.440 ( 4.35 ) 10.510 ( 3.90 ) 10.590 ( 4.16 ) 12.680 ( 4.66 ) 12.690 ( 4.68 ) 2.501 ( 1.43 ) 28.610 ( 5.80 ) Frequency and count 12.080 ( 2.39 ) Observations 10079 10140 10068 10158 10964 9843 11088 9915 10079 Adjusted r squared 0.250 0.249 0.251 0.249 0.251 0.261 0.249 0.259 0.251 Note: Table 3 5 presents pane l regre ssions on CDS Spread Difference (firm mean CDS spread minus the mean sovereign CDS spread (in bps) ) . The analysis includes 2,364 firms in 54 countries over the years 2004 2011. The crisis is years 2008 2011. The panel regressions are yearly using st andard errors clustered by firm. Controls include Ln(Market Cap), Leverage, Short term Debt Fraction, Cash Flow to Assets, Stock Return Vol atility , Firm CDS Depth, Sovereign CDS Depth. Fixed effects include year, country, industry, CDS restructuring type, and CDS Currency. T statistics are provided in parenthesis.
104 Table 3 6 . Firm sovereign credit rating d ifference . Credit rating: international p resence Var Var x Crisis Obs. Adj. R Sq. (1) Number of stock e xchanges 0.609 ( 7.23 ) 0.030 ( 0.63 ) 6915 0.6 20 (2) Number of geographic s egments 0.041 ( 1.26 ) 0.057 ( 2.46 ) 6545 0.606 (3) Foreign assets (fraction of total a ssets) 0.243 ( 0.88 ) 0.087 ( 0.39 ) 6953 0.611 (4) Foreign sales (fraction of total s ales) 0.070 ( 0.29 ) 0.481 ( 2.61 ) 6953 0.611 Credit rating: f undamentals (5) Scaled e xposure: Ln(GDP) 0.076 ( 0.97 ) 0.007 ( 0.11 ) 6434 0.574 (6) Scaled exposure: stock market v olatility 0.034 ( 0.48 ) 0.001 ( 0.01 ) 6446 0.575 Credit Rating: Scaled Exposure and Extra Disclosure (7) Prop erty r ights 0.028 ( 0.92 ) 0.015 ( 0.44 ) 6266 0.576 (8) Rule of l aw 0.002 ( 0.05 ) 0.022 ( 0.54 ) 5499 0.594 (9) Repudiation r isk 0.007 ( 0.18 ) 0.039 ( 0.90 ) 5499 0.594 (10) Expropriation r isk 0.035 ( 0.53 ) 0.068 ( 1.09 ) 5502 0.611 (11) Creditor rights 0.010 ( 1.17 ) 0.150 ( 2.12 ) 5523 0.623 (12) Ln(contract enforcement d ays) 0.097 ( 1.77 ) 0.025 ( 0.44 ) 5520 0.620 (13) Number of items r eported 0.026 ( 0.08 ) 0.003 ( 0.59 ) 6552 0.615 (14) Frequency and c ount 0.260 ( 0.33 ) 0.008 ( 2.06 ) 6552 0.589 CDS spread with rating control: scaled exposure and extra disclosure (15) Property r ights 0.0286 ( 0.92 ) 0.015 ( 0.44 ) 6266 0.576 (16) Creditor rights 0.010 ( 1.17 ) 0.150 ( 2.12 ) 5523 0.623 (17) Number of items r epo rted 0.026 ( 0.08 ) 0.003 ( 0.59 ) 6552 0.615 Note: Table 3 6 presents regressions on S&P Credit Rating Difference, which is defined as the number of the rating co rresponds to a higher credit quality. Table 3 6 includes 1,374 companies in 43 countries over the years 2004 2011. The crisis is the years 2008 2011. The panel regressions use standard errors clustered at the firm level. T statistics are provided in parent hesis .
105 Table 3 7 . C DS sovereign ceiling violations and multinational f irms . Scaled exposure variables: Variable indicated below Creditor r ights Disclosure: no. items Obs R sq. Firm violator dummy ( simple m ethod ) Scaled e x posure: ( 1 ) Property r ights 0.051 ( 1.99 ) 0.033 ( 1.68 ) 0.030 ( 2.12 ) 5173 0.303 ( 2 ) Rule of l aw 0.023 ( 2.24 ) 0.044 ( 1.79 ) 0.033 ( 2.20 ) 5197 0.305 ( 3 ) Repudiation r isk 0.023 ( 1.99 ) 0.043 ( 1.63 ) 0.030 ( 2.08 ) 5152 0.306 ( 4 ) Expropr iation r isk 0.043 ( 1.81 ) 0.042 ( 1.64 ) 0.037 ( 2.34 ) 5207 0.302 Firm violator dummy ( mean m ethod ) Scaled e xposure: ( 5 ) Property r ights 0.055 ( 2.34 ) 0.065 ( 1.74 ) 0.014 ( 1.94 ) 5002 0.386 ( 6 ) Rule of l aw 0.036 ( 2.11 ) 0.062 ( 1.72 ) 0.024 ( 1.86 ) 5007 0.384 ( 7 ) Repudiation r isk 0.064 ( 2.14 ) 0.077 ( 1.75 ) 0.010 ( 2.03 ) 4969 0.386 ( 8 ) Expropriation r isk 0.018 ( 1.76 ) 0.070 ( 1.73 ) 0.022 ( 1.92 ) 5025 0.384 Firm violator b uckets ( 0,1,2,3 ) Scaled e xpo sure: ( 9 ) Property r ights 0.051 ( 1.97 ) 0.042 ( 1.62 ) 0.030 ( 2.17 ) 5173 0.231 ( 10 ) Rule of l aw 0.024 ( 2.01 ) 0.052 ( 1.61 ) 0.040 ( 2.22 ) 5197 0.232 ( 11 ) Repudiation r isk 0.013 ( 1.88 ) 0.052 ( 1.73 ) 0.033 ( 2.25 ) 5152 0.233 ( 12 ) Exprop ri ation r isk 0.011 ( 1.84 ) 0.037 ( 1.75 ) 0.025 ( 2.03 ) 5207 0.240 Note: Table 3 7 presents probit regressions on three variations of the firm violator dummy variable, which is equal to one if a firm was a sovereign ceiling violator in a given year. The first , which spread fell below its sovereign CDS spread in at least one day during the year. The second, iven year if the annual mean of a regressions, shown in models (9) variable, which measures the sever ity of the sovereign ceiling violation. All Firm Violator Dummy variables are transaction cost adjusted. In this table, we exclusively focus on the crisis period, 2008 2011. T statistics are provided in parenthesis.
106 Table 3 8 . CDS versus S&P rating: lead /lag predictions in sovereign ceiling v iolations (SCVs) . Becomes an S&P violator Becomes a CDS violator Become S&P viol Stop S&P viol 1 ) Become CDS viol Stop S&P viol 1 ) Change in S&P violation Change in CDS violation (1) Probit (2) Probit (3) Probi t (4) Probit (5) Multinomial l ogit (6) Multinomial l ogit (7) Ordered p robit (8) Ordered p robit 1 Year l ag: CDS viol 1.249 ( 3.03 ) 2.109 ( 3.48 ) 1.702 ( 5.94 ) 3.011 ( 3.51 ) 2.984 ( 3.07 ) 2.489 ( 4.93 ) 8.537 ( 0.83 ) 1.462 ( 6.26 ) 3 .123 ( 6.33 ) 2 Year l ag: iol 1.990 ( 3.53 ) 3.226 ( 4.29 ) 0.370 ( 2.08 ) 4.133 ( 4.26 ) 2.424 ( 1.78 ) 0.719 ( 1.58 ) 8.721 ( 0.85 ) 1.186 ( 1.57 ) 2.297 ( 5.91 ) 1 Year l ag: iol 4.202 ( 4.26 ) 0.264 ( 0 .56 ) 0.273 ( 0.57 ) 4.368 ( 3.18 ) 21.64 0 ( 0.02 ) 0.883 ( 0.74 ) 12.97 0 ( 0.03 ) 0.286 ( 1.16 ) 0.771 ( 1.25 ) 2 Year l ag: iol 0.262 ( 0.48 ) 0.664 ( 1.40 ) 0.762 ( 1.57 ) 0.099 ( 0.03 ) 0.450 ( 0.10 ) 0.547 ( 0.40 ) 4.227 ( 0.85 ) 0.148 ( 0.43 ) 0.027 ( 0.05 ) Observations 4301 4201 4375 4375 4386 4228 4386 4228 R squared 0.537 0.652 0.150 0.185 0.378 0.266 0.331 0.236 Note: Table 3 8 provides results on whether the CDS SCV has power to predict SCVs in S&P credit ratings (and vice versa). Columns (1) (2) are probit regressions in which the binary dependent variable is equal to one when the firm becomes an SCV in its S&P rat ing. Columns (3) (4) are probit regressions in which the binary dependent variable is equal to one when th e firm becomes a CDS SCV ). Column (5) is a multinomial logit regression in which the dependent variable is equal to one when the firm becomes an S&P SCV, negative one when the firm stops being an S&P SCV, and zero otherwise (base c ase). Column (6) is a multinomial logit regression in which the dependent variable is equal to one when the firm becomes a CDS SCV, negative one wh en the firms stops being a CDS SCV, and zero otherwise (base case). The multinomial logit models provide a co efficient estimate for each value of the dependent variable. Column (7) and (8) are ordered probit regressions in which the dependent variable is eq ual the ndependent variables are 1 year and 2 2011. Observations are at the fi rm year level. CDS data are provided by Markit. S&P ratings are acquired from Thomson One and Worldscope. All regressions are clustered at the firm level and contain Ln(Market Cap), Leverage, Short term Debt Fraction, Cash Flow to Assets, and Stock Return Volatility as controls. Fixed effects include industry, region, crisis period, and CDS restructuring type. T statistics are provided in parenthesis.
107 1. July 31, 2007. Bear Stearns liquidates hedge funds. 2. September 15, 2008. Lehman Bros. files for Ch. 11 bankruptcy. 3. October 3, 2008. US congress passes Troubled Asset Relief Fund (TARP). 4. 5. April 27 28, 2010. S&P cuts credit rating of Greece, Portugal and Spain. 6. May 2 9, 2010. Greece acc epts bailout; European Financial Stability Facility (EFSF) created. 7. November 28, 2010. Spain accepts bailout. 8. May 17, 2011. Portugal accepts bailout. 9. September 18 October 13, 2011. S&P cuts credit rating of Spain, Italy, and 24 Italian banks. Figure 3 1 . G lobal sovereign CDS s preads . Figure 1 presents the time series of the GDP weighted and equally weighted mean of global 5 year sovereign CDS daily spreads (in basis points) of 54 countries over the years 2004 2011. CDS data are provided by Markit. Numbers are placed on the graph to mark important events and outline the initial rising credit risk of the financial sector, government bailouts that followed, and the concurrent increase in sovereign credit risk across the globe.
108 A B C D Fig ure 3 2 . Global sovereign ceiling violations (SCVs) in the CDS market . A) D aily frequency of SCVs in the CDS market than sovereign CDS ). B ) M agnitude of SCVs in the CDS market during 2008 2011 . C ) Q uantity of SCVs in the CDS market in countries with strong and weak institutions during the pre crisis period . D ) Quantity of SCVs in the CDS market during the crisis period . CDS spread data are retrieved from Markit and include 2,364 firms in 54 countries over 2004 2011.
109 A B C D E F G H Figure 3 3 . Sovereign CDS s preads and country institutional c haracteristics . For each part of Figure 3 , the y axis is the sovereign CDS spread and the x axis is as follows: A) Property Rights, B) Rul e of Law, C) Repudiation Risk, D) Expropriation Risk, E) Creditor Rights, F) Ln(Contract Enforcement Days), G) Disclosure: Number of Items, and H) Disclosure: Frequency.
110 Figure 3 4 . Sovereign ceiling violations (SCVs): CDS precedes S&P ratings . Th e bar graph in Figure 4 shows the percentage of firms that are already sovereign ceiling violators in the CDS market prior to becoming a violator in the S&P rating (and vice versa). The x axis marks the years preceding the beginning of the SCV event at year t. T he beginning of the SCV is defined as the first year in our sample in which the firm starts to violate the sovereign ceiling rule. The y axis marks the % of firms that are violators at year t 2, t 1 and t. For example, 40.27% of firms that experience an S& P SCV at year t have already seen a CDS SCV two years prior. On the other hand, only 0.95% of firms that experience a CDS SCV have already seen an S&P SCV two years prior. CDS data are provided by Markit. S&P ratings are available through Thomson One and D firm as a violator if the average difference between the firm CDS spread and the sovereign CDS spread throughout the year is negative.
111 CHAPTER 4 CONCLUSION My dissertation touches on two important credit risk issues in corporate finance and investments. First, I provide insight into determinants of CDS contract returns, the impact of information and liquidity on CDS returns, and the relationship between returns in the CDS market and the stock market. Second, I show that firms are able to reduce their sovereign risk exposure by allocating assets to environments with strong legal protections and issuing equity on foreign stock exchanges that require a high level of disclosure. CDS Mome ntum explores the asset pricing characteristics of the CDS market. I test for return momentum, one of the most persistent and pe rvasive asset pricing anomalies . Despite the role of more sophisticated institutional market participants in the CDS market, I f ind that recent performance positively predicts future performance . Performance is best among lower grade entities and higher depth CDS contracts . CDS momentum is driven by correct anticipation of future credit rating changes. T he winner (loser) portfolio is driven by firms undergoing rating upgrades (downgrades). There is significant return run up (drop off) leading up to a rating upgrade (downgrade), and then positive (negative) returns in the month of the announcement. This mechanism works particularly well among high depth contracts. C ross market tests show that this information channel is relevant to stock momentum. By incorporating CDS return information into the portfolio formation process, the traditional stock momentum strategy avoids abrupt losse s during the financial crisis period and dramatically improves its performance. In this joint market momentum strategy, CDS returns, particularly those of high depth contracts, assist stocks in more accurately predicting important future rating change even ts.
112 T he Exodus from Sovereign Risk extends the literature that relates the strength of country institutions and the pricing of corporate credit. G lobal sovereign risk increased dramatically during the financial crisis , increasing worries about how sovereig n risk would be transferred to the private sector. I examine the extent to which firms reduce their linkage to sovereign risk . Using CDS data for a large sample of firms, I highlight two important channels through which sovereign risk and corporate credit risk are related. First, I find that a firm with foreign assets in a country with relatively stronger property rights than its home country has a lower CDS spread relative to the sovereign spread. Second , I find that if a firm has equity listed on a forei gn stock exchange with relatively stricter disclosure requirements, then the firm has a lower CDS spread relative to the sovereign spread . Further, these firm level asset and information network connections lead to corporate credit risk improving beyond th e level of the sovereign,
113 LIST OF REFERENCES Acharya, V iral V., and Timothy C. Johnson, 2007 , Inside r trading in credit derivatives. Journal of Financial Economics 84, 110 141. Acharya, Viral V., Itama r Drechsler, Philipp Schnabl, 2011, A Pyrrhic Victory? Bank Bailouts and Sovereign Credit Risk, NYU Working Paper. Ang, Andrew and Francis Longstaff, 2011, Systematic Sovereign Credit Risk: Lessons from the U.S. and Europe, NBER Working Paper No. 16982. As ness, Clifford S., Tobias J. Moskowitz, and Lasse Heje Pedersen, 2013, Value and Momentum Everywhere, The Journal of Finance 68, 929 985. Avramov, Doron, Tarun Chordia, Gergana Jostova, an d Alexander Philipov, 2007, Mo mentum and Credit Rating, The Journal of Finance 62, 2503 2520. Avramov, Doron, Tarun Chordia, Gergana Jostova, and Alexander Philipov, 2012, Anomalies and financial distress, Journal of Financial Economics 108, 139 159. Bae, Kee Hong and Vidhan K. Goyal, 2009, Creditor Rights, Enforcement, an d Bank Loans, Journal of Finance 64, 823 860. Bai, Jennie, and Pierre Collin Dufresne, 2013, The CDS Bond Basis, Working Paper, Ecole Polytechnique Federale de Lausanne and Georgetown University. Bai, Jennie and Shang Jin Wei, 2012, When Is There a Strong Transfer Risk from the Sovereigns to the Corporates? Property Rights Gaps and CDS Spreads, NBER Working Paper No. 18600. Bailey, W., G. Karolyi, and C. Salva, 2006, The Economic Consequences of Increased Disclosure: Evidence from International Cross listin gs, Journal of Financial Economics 81, 175 213. Bekaert, Geert and Campbell Harvey, 2000, Foreign Speculators and Emerging Equity Markets, Journal of Finance 55, 565 613. Bekaert, Geert, Campbell Harvey, and C. Lundblad, 2005, Does Financial Liberalization Spur Growth, Journal of Financial Economics 77, 3 55. Berndt, Antje and Iulian Obreja, 2010, Decomposing European CDS Returns, Review of Finance 14, 189 233. Berndt, Antje and Anastasiya Ostrovnaya, 2008, Do Equity Markets Favor Credit Market News Over Op tions Market News?, Working Paper, Carnegie Mellon University.
114 Berndt, Antje, Robert A. Jarrow, and ChoongOh Kang, 2007, Restructuring Risk in Credit Default Swaps: An Empirical Analysis, Stochastic Processes and their Applications 117, 1724 1749. Blanco, Roberto, Simon Brennan, and Ian W. Marsh, 2005, An Empirical Analysis of the Dynamic Relation between Investment Grade Bonds and Credit Default Swaps, The Journal of Finance 60, 2255 2281. Borensztein, Eduardo, Kevin Cowan, and P atricio Valenzuela, The Impact of Sovereign Ratings on Corporate Ratings in Emerging Market Economies, IMF Working Paper. Bukspa n, Neri, Emmanuel Dubois Pelerin, and Solomon B. Samson, 2008, 2008 Corporate Criteria: Analytical Methodology, St Bushman, Robert M., Joseph D. Piotroski, and Abbie J. Smith, 2004, What Determines Corporate Transparency?, Journal of Accounting Research 42, 207 252. Cailleteau, Pierre and Vincent J. Truglia, 2006, A G Campbell, John Y., Jens Hilscher, and Jan Szilagyi, 2008, In Search of Distress Risk, The Journal of Finance 63, 2899 2939. Caprio, Lorenzo, Mara F accio, and J. McConnell, 2011, Sheltering Corporate Assets from Political Extraction, Journal of Law, Economics, & Organization, forthcoming. Chava, Sudheer, Rohan Ganduri, and Chayawat Ornthanalai, 2012, Are Credit Ratings Still Relevant?, Working Paper, Georgia Institute of Technology and University of Toronto. Choi, Stephen J., Mitu Gulati, and Eric Posner, 2011, Pricing Terms in Sovereign Debt Contracts: A Greek Case Study with Implications for the European Crisis Resolution Mechanism, John M. Olin Law & Economics Working Paper No. 541. Claessens, Stijn , Daniela Klingebie l, and Sergio L. Schmuckler, 2007, Government Bonds in Domestic and Foreign Currency: The Role of Institutional and Macroeconomic Factors, Review of International Economics 15, 370 413. Collin Dufresne, Pierre, Robert S. Goldstein, and J. Spencer Martin, 2001, The Determinants of Credit Spread Changes, Journal of Finance 56, 2177 2207. Coulton, Brian, Richard Fox, David Riley, and Roger Scher, 2004, Country Ceiling Ratings and Rating Abo ve the Sovereign, Fitch Ratings Criteria Report: Sovereigns, June 2004. Coval, Joshua D. and Tobias J. Moskowitz, 2001, The Geography of Investment: Informed Trading and Asset Prices, Journal of Political Economy 109, 811 841. Daniel, Kent, 2011, Momentum Crashes, Working Paper, Columbia University.
115 Daniel, Kent, Ravi Jagannathan, and Soohun Kim, 2012, Tail Risk in Momentum Strategy Returns, Working Paper, Columbia University and Northwestern University. Das, Sanjiv, Madhu Kalimipalli, and Subhankar Nayak, 2014, Did CDS trading improve the market for corporate bonds?, Journal of Financial Economics 111, 495 525. Desai, Mihir A., C. Fritz Foley, James R. Hines Jr., 2004, A Multinational Perspective on Capital Structure Choice and Internal Capital Markets, Journal of Finance 59, 2451 2487. Dichev, Ilia D., and Joseph D. Piotroski, 2001, The Long Run Stock Returns Following Bond Ratings Changes, The Journal of Finance 56, 173 203. Dieckmann, Stephan and Thomas Plank, 2011, Default Risk of Advanced Economies: An Empirical Analysis of Credit Default Swaps during the Financial Crisis, Review of Finance 0, 1 32. Dittmar, Robert F. and Kathy Yuan, 2008, Do Sovereign Bonds Benefit Corporate Bonds in Emerging Markets?, Review of Financial Studies 21, 1983 2014. Djank ov, Simeon, Caralee McLeish, Andrei Shleifer, 2007, Private Credit in 129 Countries, Journal of Financial Economics 84, 299 329. Djankov, Simoen, Rafael La Porta, Florencio Lopez de Silanes, Andrei Shleifer, 2003, Courts, Quarterly Journal of Economics 118 , 453 517. Durbin, Erik and David T. Ng, 2005, The Sovereign Ceiling and Emerging Market Corporate Bond Spreads, Journal of International Money and Finance 24, 631 649. Ejs ing, Jacob W. and Wolfgang Lemke, 2009, Sovereign and Bank Credit Risk Premia during 2008 09, Working Paper No 1127, The Janus Headed Salvation. Ericcson, Jan, Kris Jacobs, Rodolfo Oviedo, 2009, The Determinants of Credit Default Swap Premia, Journal of Financial and Quantitative Analysis 44, 109 132. Flannery, Mark, Joel Houston, and Fra nk Partnoy, 2010, Credit Default Swap Spreads as Viable Substitutes for Credit Ratings, University of Pennsylvania Law Review 158, 10 031. Frazzini, Andrea, Ronen Israel, and Tobias J. Moskowitz, 2012, Trading Costs of As set Pricing Anomalies, Working Pa per, AQR Capital Management and University of Chicago. Friewald, Nils, Christian Wagner, and Josef Zechner, 2012, The Cross Section of Credit Risk Premia and Equity Returns, The Journal of Finance (forthcoming). Gebhardt, William R., Soeren Hvidkjaer, and Bhaskaran Swaminathan, 2005, Stock and bond market interaction: Does momentum spill over?, Journal of Financial Economics 75, 651 690.
116 Han, Bing, and Yi Zhou, 2011, Term Structure of Credit Default Swap Spreads and Cross Section of Stock Returns, Working Paper, Florida State University and University of Texas at Austin. Hilscher, Jens, Joshua M. Pollet, and Mungo Wilson, 2014, Are Credit Default Swaps a Sideshow? Evidence that Information Flows from Equity to CDS Markets, Journal of Financial and Quantitat ive Analysis (forthcoming). Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad News Travels Slowly: Size, Analyst Coverage, and the Profitability of Momentum Strategies, The Journal of Finance 55, 265 295. Hong, Harrison, and Jeremy C. Stein, 1999 , A Unified Theory of Underreaction, Mo mentum Trading, and Overreaction in Asset Markets, The Journal of Finance 54, 2143 2184. Huang, Xing , 2012, Gradual Information Diffusion in the Stock Market: Evidence from U.S. Multinational Firms, Working Paper, U niversity of California Berkeley. Hull, John, Mirela Predescu, and Alan White, 2004, The relationship between credit default swap spreads, bond yields, and credit rating announcements, Journal of Banking and Finance 28, 2789 2811. Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency, The Journal of Finance 48, 65 91. Jegadeesh, Narasimhan, and Sheridan Titman, 2001, Profitability of Momentum Strategies: An Evaluation of Al ternative Explanations, The Journal of Finance 56, 699 720. Jegadeesh, Narasimhan, and Sheridan Titman, 2011, Momentum, Annual Review of Financial Economics 3, 493 509. Jostova, Gergana, Stanislava Nikolova, Alexander Philipov, and Christof W. Stahel, 2013 , Momentum in Corporate Bond Returns, Review of Financial Studies 26, 1649 1693. Kim, Gi Hyun, Haitao Li, and Weina Zhang, 2009, The CDS Bond Basis and the Cross Section of Corporate Bond Returns, Working Paper, National University of Singapore and Univers ity of Michigan. Kisgen, Darren J, 2006, Credit Ratings and Capital Structure, The Journal of Finance 61, 1035 1072. Kisgen, Darren J, 2007, The Influence of Credit Ratings on Corporate Capital Structure Decisions, Journal of Applied Corporate Finance 19, 65 73. Kliger, Doron, and Oded Sarig, 2000, The Information Value of Bond Ratings, The Journal of Finance 55, 2879 2902.
117 La Porta, Rafael, Florencio Lopez de Silanes, Andrei Shleifer, and Robert W. Vishny, 1998, Law and Finance, Journal of Political Econom y 106, 1113 1155. La Porta, Rafael, Florencio Lopez de Silanes, Andrei Shleifer, and Robert W. Vishny, 2002, Investor Protection and Corporate Valuation, Journal of Finance 57, 1147 1170. La Porta, Rafael, Florencio Lopez de Silanes, and Andrei Shleifer, 2008, The Economic Consequences of Legal Origins, Journal of Economic Literature 46:2, 285 332. Lee, Charles, and Bhaskaran Swaminathan, 2000, Price Momentum and Trading Volume, The Journal of Finance 55, 2017 2069. Li, Ningzhong, Scott Richardson, and Ire m Tuna, 2012, Macro to Micro: Country Exposures, Firm Fundamentals and Stock Returns, Working Paper, London Business School. Lins, Karl, Deon Strickland, and Marc Zenner, 2005. Do Non U.S. Firms Issue Equity on U.S. Exchanges to Relax Capital Constraints?, Journal of Financial and Quantitative Analysis 40, 109 133. Lon gstaff, Francis A., Sanjay Mithal, and Eric Neis, 2005, Corporate Yield Spreads: Default Risk or Liquidity? New Evidence from the Credit Default Swap Market, Journal of Finance 60, 2213 2253. Longstaff, Francis A., Jun Pan, Lasse H. Pedersen, and Kenneth J. Singleton, 2011, How Sovereign is Sovereign Credit Risk?, American Economic Journal: Macroeconomics 3, 75 103. Manso, Gustavo, 2013, Feedback effects of credit ratings, Journal of Financial Economics 109, 535 548. Menkhoff, Lukas, Lucio Sarno, Maik Schmeling, and Andreas Schrimpf, 2012, Currency momentum strategies, Journal of Financial Economics 106, 660 684. Mian, Atif , 2006, Distance Constraints: The Limits of Foreign Lending in Poor Econo mies, Journal of Finance 61, 1465 1505. Miller, Darius P. and Natalia Reisel, 2011, Do Country level Investor Protections Impact Security level Contract Design? Evidence from Foreign Bond Covenants, Review of Financial Studies 25, 408 438. Mitchell, Mark, and Todd Pulvino, 2012, Arbitrage crashes and the speed of capital, Journal of Financial Economics 104, 469 490. Nashikkar, Amrut, Marti G Subrahmanyam, and Sriketan Mahanti, 2011, Liquidity and Arbitrage in the Market for Credit Risk, Journal of Financial and Quantitative Analysis 46, 627 656.
118 Ni, Sophie, and Jun Pan, 2010, Trading Puts and CDS on Stocks with Short Sale Ban, Working Paper, Hong Kong University and MIT. Nguyen, Quoc H., 2012, Geographic Momentum, Working Paper, University of Illinois at Urb ana Champagne. Noe, Thomas H., 2000, Creditor Rights and Multinational Capital Structure, Working Paper, Tulane University. Norden, Lars, and Martin Weber, 2004, Informational efficiency of credit default swap and stock markets: The impact of credit rating announcements, Journal of Banking and Finance 28, 2813 2843. name and Multi name Credit Derivatives (John Wiley & Sons). Okunev, John, and Derek White, 2003, Do Momentum Based Strategies Still Work in Foreign Curren cy Markets?, Journal of Financial and Quantitative Analysis 38, 425 448. Packer, Frank and Haibin Zhu, 2005, Contractual Terms and CDS Pricing, BIS Quarterly Review March 2005, 89 100. Pagano, Marco, Ailsa Roell, Josef Zechner, 2002, The Geography of Equi ty Listing: Why Do Companies List Abroad?, Journal of Finance 57, 2651 2694. Pirrong, Craig, 2005, Momentum in Futures Markets, Working Paper, University of Houston. Qian, Jun and Philip E. Strahan, 2007, How Laws and Institutions Shape Financial Contracts : The Case of Bank Loans, Journal of Finance 62, 2803 2834. Qiu, Jiaping, and Fan Yu, 2012, Endogenous liquidity in credit derivatives, Journal of Financial Economics 103, 611 631. Shleifer, Andrei and Daniel Wolfenzon, 2002, Investor Protection and Equity Markets, Journal of Financial Economics 58, 3 27. Tang, Dragon, and Hong Yan, 2007, Liquidity and Credit Default Swap Spreads, Working paper, University of Hong Kong and University of South Carolina. Zhu, Haibin, 2006, An Empirical Comparison of Credit Sp reads between the Bond Market and the Credit Default Swap Market, Journal of Financial Services Research 29, 211 235.
119 BIOGRAPHICAL SKETCH Stace Sirmans is an Assistant Professor of Finance in the Sam Walton College of Business at the University of Arka nsas. He received his Ph.D. in f inance in August of 2014 from the University of Florida where he studied the role of credit risk in financial markets. research has been featured in a number of conferences, such as the SFS Cavalcade and the WU Gutma nn Symposium. In 2013, he was awarded the WRDS Best Paper in Empirical Finance award for his research on credit default swaps. Stace completed his undergraduate studies at Florida State University in the spring of 2009.
THEJOURNALOFFINANCE VOL.LXII,NO.5 OCTOBER2007MomentumandCreditRatingDORONAVRAMOV,TARUNCHORDIA,GERGANAJOSTOVA, andALEXANDERPHILIPOVABSTRACTThispaperestablishesarobustlinkbetweenmomentumandcreditrating.Momentum profitabilityislargeandsignificantamonglow-gradefirms,butitisnonexistent amonghigh-gradefirms.Themomentumpayoffsdocumentedintheliteratureare generatedbylow-gradefirmsthataccountforlessthan4%oftheoverallmarket capitalizationofratedfirms.Themomentumpayoffdifferentialacrosscreditrating groupsisunexplainedbyfirmsize,firmage,analystforecastdispersion,leverage, returnvolatility,andcashflowvolatility.JEGADEESHANDTITMAN(1993)DOCUMENTthatthemomentum-basedtradingstrategyofbuyingpastwinnersandsellingpastlosersprovidesstatisticallysignificantandeconomicallylargepayoffs.Theempiricalevidenceonstockreturn momentumisintriguingbecauseitpointstoaviolationofweak-formmarketefficiency.Inparticular,FamaandFrench(1996)showthatmomentum profitabilityistheonlyCAPM-relatedanomalyunexplainedbytheFamaand French(1993)three-factormodel.Moreover,Schwert(2003)demonstratesthat marketanomaliesrelatedtoprofitopportunities,includingthesizeandvalue effectsinthecross-sectionofaveragereturnsaswellastime-seriespredictabilityofreturnsbythedividendyield,typicallydisappear,reverse,orattenuate followingtheirdiscovery.Incontrast,JegadeeshandTitman(2001,2002)documenttheprofitabilityofmomentumstrategiesaftertheirinitialdiscovery.The robustnessofmomentumprofitabilityhasgeneratedavarietyofexplanations, bothbehavioralandriskbased.1 DoronAvramovisattheRobertH.SmithSchoolofBusiness,UniversityofMaryland;Tarun ChordiaisattheGoizuetaBusinessSchool,EmoryUniversity;GerganaJostovaisattheSchool ofBusiness,GeorgeWashingtonUniversity;andAlexanderPhilipovisattheSchoolofManagement,GeorgeMasonUniversity.TheauthorsthankYakovAmihud,MichaelCooper,KoreshGalil, NarasimhanJegadeesh,SheridanTitman, Lubo sP Â« astor,seminarparticipantsatAmericanUniversity,BankofCanada,EmoryUniversity,UniversityofMaryland,McGillUniversity,University ofSouthernCalifornia,the10thAnnualFinanceandAccountingConferenceatTelAvivUniversity,YaleUniversity,theChicagoQuantitativeAlliance,theWashingtonAreaFinanceAssociation conference,andespeciallyananonymousrefereeandanassociateeditorforhelpfulcomments.All errorsareourown.1See,forexample,Barberis,Shleifer,andVishny(1998),Daniel,Hirshleifer,andSubrahmanyam(1998),Hong,Lim,andStein(2000),ChordiaandShivakumar(2002),Grinblattand Han(2005),andAvramovandChordia(2006a),amongothers.2503
2504 TheJournalofFinance Ithasalsobeenshownthatmomentumprofitabilityisrelatedtobusiness conditions.Specifically,ChordiaandShivakumar(2002)documentthatmomentumpayoffsarelargeduringexpansionsandnonexistentduringrecessions. AvramovandChordia(2006a)demonstratethattheimpactofpastreturnson futurereturnscannotbecapturedbyconditionalandunconditionalrisk-based assetpricingmodels.However,theyshowthatthemomentumpayoffsarerelatedtothecomponentofmodelmispricingthatvarieswithbusinesscyclevariablessuchastheTreasurybillyield,thetermspread,andthedefaultspread. Moreover,AvramovandChordia(2006b)showthatanoptimizinginvestorwho usesthesebusinesscyclevariablesisabletosuccessfullyloadonthemomentumstrategyduringdifferentphasesoftheeconomy.Sincecreditriskvaries overthebusinesscycle,itisnaturaltoaskwhetherthemomentumpayoffsare relatedtothecreditriskoffirms.Inthispaper,weprovideanewandunexploreddimensioninunderstandingtheprofitabilityofmomentumstrategies. Weshowthatmomentumprofitsarerestrictedtohighcreditriskfirmsand arenonexistentforfirmsofhighcreditquality. Specifically,basedonasampleof3578NYSE,AMEX,andNASDAQfirms ratedbyS&PovertheJuly1985toDecember2003period,2weshowthat overformationperiodsofthree,6,9,and12months,theextremeloserand winnerportfoliosofJegadeeshandTitman(1993)consistofstockswiththe lowestandthenext-lowestcreditrating,respectively.Theaverageratingofthe entiresampleofratedfirmsisBBB.Theextremeloser(winner)portfoliohas anaverageratingofBB Ã (BB ).Theextremelosersandwinnersaretheonly noninvestmentgradeportfoliosinthesampleofratedfirms. Tradingstrategiesthatconditiononthreecreditratingand10prior6-month returngroupsyieldmomentumpayoffsthatincreasemonotonicallywithcredit risk Ã‘ theyincreasefromaninsignificant0.27%permonthforthebestquality debtterciletoasignificant2.35%fortheworst.Similarly,basedon10credit ratingand3pastreturnportfolios,momentumpayoffsincreasefromaninsignificant0.07%permonthforthehighestcreditqualitydeciletoasignificant 2.04%fortheworst.Amongthelow-ratedfirms,loserstocksarethedominant sourceofreturncontinuationandtheprofitabilityofmomentumstrategies. Basedon10creditriskand3pastreturngroups,thereturndifferentialbetweenthelowestandhighestcreditriskloserfirmsaverages1.60%permonth, whereasthereturndifferentialforthewinnerfirmsis,onaverage,only0.37%. Wealsoimplementmomentumstrategiesbasedontheprior6-monthreturn fordifferentsamplesofratedfirms,aswesequentiallyexcludethelowest-rated firms.Strikingly,thesignificantprofitstomomentumstrategiesarederived fromasampleoffirmsthataccountsforlessthan4%ofthemarketcapitalizationofallratedfirmsandforabout22%ofthetotalnumberofratedfirms.When weexcludefirmswithanoverallS&PratingofD,C,CC,CCC Ã ,CCC,CCC ,B Ã , B,B ,andBB Ã ,themomentumstrategypayoffsfromtheremainingfirms,2WeusetheS&PLong-TermDomesticIssuerCreditRating.Dataonthisvariableareavailable onCompustatonaquarterlybasisstartingfromthesecondquarterof1985.
MomentumandCreditRating 2505 whichaccountfor96.6%oftheoverallmarketcapitalizationofratedfirms, becomestatisticallyinsignificant. Recentworkdemonstratesthesignificanceofmomentumforcertainsubsamplesofstocks.Forinstance,Jiang,Lee,andZhang(2005)andZhang(2006)find evidenceofhighermomentumpayoffsamongfirmswithhigherinformation uncertainty.Informationuncertaintyisproxiedbyfirmsize,firmage,return volatility,cashflowvolatility,andanalystforecastdispersion.However,our findingssuggestthatthecreditratingeffectonmomentumisbothindependentofandmuchstrongerthantheeffectofalltheseinformationuncertainty variables.Inparticular,theinformationuncertaintyvariablesdonotcapture themomentumprofitsacrosscreditratinggroupswhereascreditratingdoes capturethemomentumprofitsacrosstheuncertaintyvariables.Specifically, momentumpayoffsexistamonglarge-capitalizationfirmsthatarelowrated, butareabsentinhighlyratedsmall-capitalizationfirms.Thus,whilemomentumprofitabilitydoesnotariseexclusivelyinsmallstocks,itisfoundexclusivelyamonglow-ratedstocks. Therestofthepaperisorganizedasfollows.SectionIpresentsthedata. SectionIIpresentstheresultsandSectionIIIpresentsrobustnesschecks.SectionIVconcludes. I.Data WeextractmonthlyreturnsonallNYSE,AMEX,andNASDAQstockslisted intheCRSPdatabase,subjecttoseveralselectioncriteria.First,stocksmust haveatleastsixconsecutivemonthlyreturnobservations.Inaddition,asin JegadeeshandTitman(2001),weexcludestocksthat,atthebeginningofthe holdingperiod,arepricedbelow$5orhavemarketcapitalizationthatwould placetheminthebottomNYSEdecile.Thisisdonetoensurethattheempirical findingsarenotdrivenbylow-pricedandextremelyilliquidstocks.However, wefindthatourresultsarerobusttotheinclusionofstocksbelow$5andthose thatbelongtothesmallestdecile.Thefilteringproceduredeliversauniverse of13,018stocks.Fromthisuniverse,wechoosethosestocksthatarerated byStandard&Poor ' s,leavinguswith3,578ratedstocksovertheJuly1985 throughDecember2003period.Thebeginningofoursampleisdeterminedby thefirstdateforwhichfirmratingsbyStandard&Poor ' sareavailableonthe COMPUSTATtapes. TheS&Pissuerratingusedinthispaperisanessentialcomponentofour analysis.Standard&Poor ' sassignsthisratingtoafirm,notanindividual bond.AsdefinedbyS&P,priorto1998,thisissuerratingisbasedonthefirm ' s mostseniorpubliclytradeddebt.After1998,thisratingisbasedontheoverall qualityofthefirm ' soutstandingdebt,eitherpublicorprivate.Thepre-1998 issuerratingthereforerepresentsaselectsubsampleofcompanybonds,while after1998itrepresentsallcompanydebt.WetransformtheS&Pratingsinto conventionalnumericalscores,where1representsaAAAratingand22reflects
2506 TheJournalofFinance aDrating.3Thus,ahighernumericalscorecorrespondstoalowercreditrating orhighercreditrisk.Numericalratingsof10orbelow(BBB Ã orbetter)are consideredinvestmentgrade,andratingsof11orhigher(BB orworse)are labeledhighyieldornoninvestmentgrade.Theequallyweightedaveragerating ofthe3,578firmsinoursampleis8.83(approximatelyBBB,theinvestmentgradethreshold)andthemedianis9(BBB). Toensurethatoursampleofstocksisrepresentative,inTableIwecompare ratedandunratedfirms.Itisimportanttonotethatalthoughthetotalnumber ofratedfirmsismuchsmallerthanthatofunratedfirms(3,578ratedfirms and9,440unratedfirms,foraratioof1to2.6),theaveragepermonthnumber ofratedandunratedfirmsisconsiderablycloser(1,639ratedfirmsand2,246 unratedfirms,foramoreappealingratioof1to1.4). PanelAofTableIpresentsmonthlyreturnsfortheloserportfolio(P1),the winnerportfolio(P10),andthemomentumstrategyofbuyingthewinnerportfolioandsellingtheloserportfolio(P10 P1).MomentumportfoliosareconstructedasinJegadeeshandTitman(1993).Atthebeginningofeachmonth t , werankalleligiblestocksonthebasisoftheircumulativereturnovertheformationperiod(months t Ã 6to t Ã 1)andassignthemto1of10portfoliosbased ontheirprior6-monthreturn.Theseportfoliosarethenheldfor K months. Weskipamonthbetweentheformationandholdingperiods(months t 1to t K ).Eachportfolioreturniscalculatedastheequallyweightedaveragereturnofthecorrespondingstocks.Themonthlymomentumstrategyreturnfora K -monthholdingperiodisbasedonanequallyweightedaverageoftheportfolioreturnsfromstrategiesimplementedinthecurrentmonthandtheprevious K Ã 1months.4TheevidenceinPanelAsuggestssimilarmomentumprofitabilityamong ratedandunratedstocks.Inparticular,themomentumstrategy(P10 Ã P1)averagesaprofitof1.29%( t -stat 3.15)permonthforratedfirmsand1.43% ( t -stat 3.41)forunratedfirms.Forbothratedandunratedfirms,momentum profitsareprominentoverexpansionaryperiods,aswellasinnon-January months;consistentwithJegadeeshandTitman(1993),momentumprofitsare negativeinJanuary.Wealsoexaminetheindustrydistributionofoursampleof3,578S&Pratedfirmsrelativetotheoverallsampleof13,018NYSE, AMEX,andNASDAQfirmslistedonCRSP.The20industriesconsideredare thoseanalyzedbyMoskowitzandGrinblatt(1999).Theevidenceshows(resultsareavailableuponrequest)thattheindustrydistributionsofratedand unratedfirmsaresimilar,rulingoutconcernsthatratedfirmsareconcentrated inparticularindustries.3Theentirespectrumofratingsisasfollows:AAA 1,AA 2,AA 3,AA 4,A 5, A 6,A 7,BBB 8,BBB 9,BBB 10,BB 11,BB 12,BB 13,B 14,B 15,B 16,CCC 17,CCC 18,CCC 19,CC 20,C 21,andD 22.4Anumberofstocksdelistfromoursampleovertheholdingperiod.Loserstocksarelikelyto delistduetolowpricesorbankruptcywhilewinnerstocksmaydelistduetoanacquisition.This couldpotentiallyleadtobiasedresults.Toensurethattherearenodelistingbiases,throughout thepaperweusethedelistingreturnwheneverastockdisappearsfromoursample.
MomentumandCreditRating 2507TableIDescriptiveStatisticsPanelApresentsrawmomentuminratedandunratedfirms.Foreachmonth t ,allNYSE,AMEX,and NASDAQstocksonthemonthlyCRSPtapewithreturnsformonths t Ã 6through t Ã 1arerankedinto decileportfoliosaccordingtotheircumulativereturnduringthatperiod.Weexcludestocksthat,atthe endofmonth t Ã 1,arepricedbelow$5oraresmallerthanthesmallestNYSEsizedecile.Decileportfolios areformedmonthlyandtheirreturnsarecomputedbyweightingequallyallfirmsinthatdecileranking. Themomentumstrategyinvolvesbuyingthewinnerportfolio(P10)andsellingtheloserportfolio(P1). Thepositionsareheldforthefollowing6months( t 1through t 6).Thereisa1-monthlagbetweenthe formationandtheholdingperiods.Monthlyreturnsrepresenttheequallyweightedaveragereturnfrom thismonth ' smomentumstrategyandallstrategiesfromupto5monthsago.Thetableshowstheaverage rawmonthlyprofitsduringtheholdingperiodofthewinnerandloserportfoliosaswellasthemomentum strategyreturns. t -statisticsareinparentheses,withdenotingsignificanceatthe5%level.Thesample periodisJuly1985toDecember2003.PanelBpresentsdescriptivestatisticsofmonthlyreturnsfor stocksratedbyStandard&Poor ' sandforallstockslistedonCRSP.Returnsarecomputedasthetimeseriesmeanofthecross-sectionalaveragereturnforeachmonth(in%permonth).Standarddeviation, skewness,kurtosis,alpha(%permonth),andbetaarecomputedforeachstockandthenaveragedacross allstocks.Alphasandbetasarebasedonstockswithatleast25returnobservationsduringthesample period.Sizeiscomputedasthetime-seriesmeanofthecross-sectionalmeanofallmarketcapitalizations ineachmonth(in$billions). PanelA:RawMomentuminRatedandUnratedFirms AllRatedUnrated FirmsFirmsFirms #ofFirms13,0183,5789,440 OverallP10 P11 491 291 43 (3 48)(3 15)(3 41)P10 170 25 Ã 0 05 (0 29)(0 45)( Ã 0 07) P101 661 541 39 (3 15)(3 74)(2 46)Non-JanuaryP10 P11 821 541 81 (4 55)(3 96)(4 70)P1 Ã 0 32 Ã 0 07 Ã 0 60 ( Ã 0 55)( Ã 0 13)( Ã 0 99) P101 511 471 21 (2 69)(3 37)(2 02)JanuaryP10 P1 2 36 1 58 2 86 ( Ã 0 92)( Ã 0 65)( 1 08) P15 723 976 21 (1 90)(1 53)(1 91) P103 372 393 34 (2 59)(1 91)(2 49)ExpansionP10 P11 491 271 43 (3 39)(3 03)(3 31)P10 120 30 Ã 0 14 (0 20)(0 55)( Ã 0 23) P101 611 571 29 (2 95)(3 72)(2 21)RecessionP10 P11 421 481 46 (0 80)(0 83)(0 81) P10 83 Ã 0 291 14 (0 24)( Ã 0 09)(0 32) P102 251 182 60 (1 09)(0 65)(1 18) ( continued )
2508 TheJournalofFinanceTableI Ã‘Continued PanelB:ReturnandSizeCharacteristicsofSampleFirms Firms RatedbyAll S&PFirms Return Ã equallyweightedmean1 351 24 Return Ã valueweightedmean1 111 09 Return Ã standarddeviation12 3913 50 Return Ã skewness0 250 34 Return Ã kurtosis5 005 13 CAPMalpha Ã mean0 160 05 CAPMbeta Ã mean1 041 06 FFalpha Ã mean Ã 0 010 02 FFmktbeta Ã mean1 141 04 Size Ã mean3 060 98 PanelBofTableIprovidesdescriptivestatisticsforthedistributionofraw monthlyreturnsinthesampleofratedandunratedfirms.Themomentsof thestockreturndistribution,aswellastheaveragealphasandmarketbetas, aresimilaracrossthetwocategories.Forinstance,themeanmonthlystock returnis1.35%amongratedfirmsand1.24%amongallfirmsduringtheperiod July1985toDecember2003.ThemeanCAPMalpha(beta)ofratedfirmsis 0.16%(1.04),and0.05%(1.06)amongallfirms.ThemeanFama-Frenchalpha is 0.01%(0.02%)permonthforrated(all)firms.ItisalsoevidentfromPanelB thatratedfirmshavesubstantiallylargermarketcapitalizationthanunrated firms. Overall,TableIconfirmsthatoursampleofratedfirmsisrepresentative. Bothratedandunratedfirmsproducesimilarmomentumprofits,theyshare similarindustrydistributions,andtheyhavesimilarstockreturndistributions. II.Results A.MomentumandFirmCreditRatingovertheFormationPeriod Toestablishthefirstlinkbetweenmomentumtradingstrategiesandcredit risk,weexaminetheaveragenumericalcreditratingforeachofthe10momentumportfoliosoverformationperiodsof3,6,9,and12months.Theresultsare presentedinTableII.Theextremeloserportfolio(P1)isheavilytiltedtowards firmswiththelowestqualitydebt.Forexample,focusingona6-monthformationperiod,theaveragenumericalratingoftheloserportfoliois13.06(BB Ã ), whichismuchhigherthantheaverageratingof8.83(BBB).Theextremewinnerportfolio(P10)alsoconsistsofhighcreditriskstocks,recordinganaverage creditratingof11.19(BB ).Themiddleportfolio(P6)hasthebestcreditrating of7.64(BBB ).Indeed,theaveragecreditratingformsaU-shapeacrossthe variousmomentumportfolios.Thissuggeststhatthemomentumstrategyof
MomentumandCreditRating 2509TableIICreditRatingProfileofMomentumPortfoliosoverFormationPeriodForeachmonth t ,allstocksratedbyStandard&Poor ' swithreturnsformonths t Ã J through t Ã 1(formationperiod)availableonCRSParerankedintodecileportfoliosaccordingtotheirreturn duringtheformationperiod.Weexcludestocksthat,attheendofmonth t Ã 1,arepricedbelow $5oraresmallerthanthesmallestNYSEsizedecile.Thetableshowsforeachdecileportfoliothe mediannumericS&Pratingduringformationperiodsof J 3,6,9,and12months.ThisS&P ratingisassignedbyStandard&Poor ' stoafirm(notabond)basedontheoverallqualityofthe firm ' soutstandingdebt,eitherpublicorprivate.TheratingisavailablefromCOMPUSTATona quarterlybasisstartingin1985.WetransformtheS&Pratingsintoconventionalnumericscores. Thenumericratingcorrespondsto:AAA 1,AA 2,AA 3,AA 4,A 5,A 6,A 7,BBB 8,BBB 9,BBB Ã 10,BB 11,BB 12,BB Ã 13,B 14,B 15,B Ã 16, CCC 17,CCC 18,CCC Ã 19,CC 20,C 21,andD 22.ThesampleperiodisJuly1985to December2003. J 3 J 6 J 9 J 12 P112.8513.0613.1813.22 P29.8410.1210.2910.30 P38.668.648.698.62 P48.068.078.007.93 P57.777.757.647.58 P67.727.647.617.49 P77.817.697.607.53 P88.087.897.707.66 P98.918.598.348.22 P1011.4411.1911.0110.91 buyingpreviouslosersandsellingpreviouswinnersessentiallytakeslongand shortpositionsinfirmswiththehighestcreditrisk. TableIIIpresentsthecompositionofunrated,investmentgrade,andnoninvestmentgradefirmsindecileportfoliossortedonpast6-monthreturns.There aremoreunratedfirmsintheextremewinnerandloserportfolios.Also,there aresignificantlyfewerfirmswithinvestmentgraderatingandmorefirms withnoninvestmentgraderatingintheextremeportfolios.Finally,thereturn differentialbetweenthewinnerandloserportfoliosisastatisticallyinsignificant(significant)0.77%(2.12%)permonthfortheinvestment(noninvestment) gradefirmsandis1.48%fortheunratedfirms.Overall,theevidencesupports ourclaimthatfirmswithalowcreditrating,orfirmsthatwouldbelowrated iftheyhadarating,drivethemomentumphenomenon. B.MomentumPro Ãž tabilityandCreditRating Weimplementmomentumstrategiesbyconditioningonbothcreditrating andcumulative6-monthformationperiodreturns.Wefirstconsider3credit ratinggroupsand10formationperiodreturnportfolios.Wethenstudy10credit ratinggroupsand3past6-monthreturnportfolios.Creditrisk-pastreturn groupsareformedonasequentialbasis,sortingfirstoncreditratingandthen
2510 TheJournalofFinanceTableIIICompositionofMomentumPortfoliosForeachmonth t ,allstocksratedbyStandard&Poor ' swithreturnsformonths t Ã 6through t Ã 1(formationperiod)availableonCRSParerankedintodecileportfoliosaccordingtotheirreturn duringtheformationperiod.Weexcludestocksthat,attheendofmonth t Ã 1,arepricedbelow $5oraresmallerthanthesmallestNYSEsizedecile.Thefirstthreecolumnsinthetableshowfor eachdecileportfoliothepercentageofstockswithnoratingandthepercentageofstocksthatare investmentgrade(IG)andnoninvestmentgrade(NIG).Thelastthreecolumnsshowtheequally weightedaveragereturnofthethreegroupsineachportfolio.IGrepresentsanS&PratingofBBB Ã orbetterandNIGrepresentsanS&PratingofBB orworse.ThesampleperiodisJuly1985to December2003.Thedenotessignificanceatthe5%level. Composition(%ofStocks)Returns(%permonth) NoNo PortfolioRatingIGNIGRatingIGNIG P175 249 3215 440 170 97 Ã 0 32 P270 9217 6311 440 571 080 22 P369 8021 209 000 811 090 52 P470 5222 017 470 931 110 81 P570 6822 636 690 991 120 81 P669 7023 706 601 031 130 77 P769 3123 557 151 141 120 78 P870 0421 478 501 191 171 01 P973 0816 8110 111 351 301 11 P1081 587 1311 291 651 741 80 P10 P11 480 772 12 (3 70)(1 77)(4 29) onpastreturns.5Foreachmonth t ,thelow(high)creditriskgroup(group1 (group3))containsthe30%best(worst)-ratedstocksbasedontheirS&Prating forthisparticularmonth.Thestocksineachgrouparethendividedinto10 momentumportfoliosbasedontheirreturnovermonths t Ã 6to t Ã 1.The 10creditriskgroupsareformedeachmonthbydividingthesampleoffirms inthatmonthintodecilesbasedonthecreditratings.Eachoftheresulting creditratinggroupsisthendividedintothreemomentumportfolios(P1,P2, andP3)containingtheworst30%,middle40%,andtop30%performersbased ontheirpast6-monthreturns.Thetwosequentialrankingsgenerate30credit risk-momentumportfolios. PanelAofTableIVpresentsthemomentumprofitscorrespondingtothe3 creditriskand10momentumgroups.Payoffstomomentumstrategiesstrongly dependuponcreditrating.Focusingonthelow(stockswithanaverageratingof4.97 A )andmedium(averageratingof8.5 BBB )creditrisk groups,theaveragepayofftotheP10 P1strategyis0.27%( t -stat 0.88)and 0.75%( t -stat 2.12)permonth,respectively.Thepayoffismuchlargeras5Weverifythatourresultsholdforindependentsortsaswell.
MomentumandCreditRating 2511TableIVMomentumbyCreditRiskGroupForeachmonth t ,allstocksratedbyStandard&Poor ' swithavailablereturndataformonths t Ã 6through t Ã 1aredividedintothreegroups,top30%,middle40%,andbottom30%(PanelA),aswellasdeciles(PanelB), basedontheircreditrating.Weexcludestocksthat,attheendofmonth t Ã 1,arepricedbelow$5oraresmaller thanthesmallestNYSEsizedecile.Foreachcreditratinggroup,wecomputethereturnoftheloserportfolio P1astheequallyweightedaveragereturnovertheholdingperiodoftheworstperforming10%(PanelA,30% PanelB)andthewinnerportfolioP10(P3inPanelB)ofthebest-performing10%(PanelA,30%PanelB)of thestocksbasedontheirreturnsovertheformationperiod.Thereisa1-monthlagbetweentheformation andtheholdingperiods.Themomentumstrategyinvolvesbuyingthewinnerportfolioandsellingtheloser portfolioandholdingthepositionfor6months.Sincethemomentumstrategyisimplementedeachmonth, themonthlyreturnsrepresenttheequallyweightedaveragereturnfromthismonth ' smomentumstrategy andallstrategiesfromupto5monthsago.Foreachcreditratinggroup,thetableshowstheaveragereturns ofthemomentumstrategy,aswellastheaveragereturnoftheloserandwinnerportfolios.Thesampleperiod isJuly1985toDecember2003.ThenumericS&Pratingispresentedinascendingorderbycreditrisk,that is,1 AAA,2 AA ,3 AA, ,21 C,and22 D.Thedenotessignificanceatthe5%level. PanelA:10Momentumand3CreditRatingGroups RatingGroup (1 LowestRisk,3 HighestRisk) 123 AverageA BBB BB Rating4 978 5013 02 OverallP10 P10 270 752 35 (0 88)(2 12)(4 21)P11 120 81 Ã 0 43 (2 81)(1 68)( Ã 0 59) P101 401 561 92 (4 13)(4 26)(3 77)Non-JanuaryP10 P10 430 952 70 (1 38)(2 69)(5 21)P10 980 61 Ã 0 92 (2 36)(1 21)( Ã 1 28) P101 411 561 78 (4 03)(4 05)(3 31)JanuaryP10 P1 Ã 1 54 Ã 1 55 Ã 1 59 ( Ã 1 10)( Ã 0 93)( Ã 0 45) P12 763 105 08 (1 87)(1 80)(1 29) P101 221 553 48 (0 94)(1 31)(2 36)ExpansionP10 P10 300 782 30 (0 94)(2 12)(4 02)P11 140 85 Ã 0 39 (2 95)(1 78)( Ã 0 55) P101 441 631 91 (4 11)(4 32)(3 69)RecessionP10 P1 Ã 0 060 383 01 ( Ã 0 04)(0 29)(1 23) P10 890 34 Ã 0 94 (0 38)(0 13)( Ã 0 21) P100 840 722 07 (0 65)(0 47)(0 86) ( continued )
2512 TheJournalofFinanceTableIV Ã‘ Continued PanelB:3Momentumand10CreditRatingGroups RatingDecile(1 LowestRisk,10 HighestRisk) AverageAAA AA BBB BBBBBB Ã BBBB Ã B Rating3 174 986 137 098 049 0310 1311 8213 1914 52 OverallP3 P10 070 070 150 200 210 320 550 731 122 04 (0 32)(0 36)(0 73)(0 94)(1 01)(1 52)(2 18)(2 46)(3 46)(4 63)P11 131 111 141 080 950 950 850 560 17 Ã 0 47 (3 67)(3 36)(3 20)(2 98)(2 63)(2 55)(1 98)(1 14)(0 31)( Ã 0 69) P31 191 191 281 281 171 271 401 291 291 56 (3 96)(3 94)(4 17)(4 11)(3 74)(3 93)(4 00)(3 14)(2 88)(3 04)Non-JanuaryP3 P10 220 170 250 300 300 420 700 961 342 28 (1 05)(0 84)(1 25)(1 43)(1 46)(2 03)(2 81)(3 34)(4 35)(5 52)P11 031 041 030 980 860 820 670 27 Ã 0 20 Ã 0 93 (3 25)(3 02)(2 81)(2 61)(2 30)(2 13)(1 51)(0 53)( Ã 0 36)( Ã 1 40) P31 261 211 281 271 171 241 381 231 151 36 (4 06)(3 89)(4 01)(3 91)(3 57)(3 67)(3 75)(2 85)(2 42)(2 52)JanuaryP3 P1 Ã 1 68 Ã 1 07 Ã 1 05 Ã 0 94 Ã 0 84 Ã 0 87 Ã 1 23 Ã 1 98 Ã 1 43 Ã 0 77 ( Ã 1 80)( Ã 1 24)( Ã 1 10)( Ã 0 88)( Ã 0 80)( Ã 0 86)( Ã 1 06)( Ã 1 33)( Ã 0 77)( Ã 0 29) P12 161 952 312 272 002 462 833 884 404 68 (1 86)(1 58)(1 72)(1 59)(1 40)(1 72)(1 92)(2 26)(1 68)(1 28) P30 480 871 261 331 161 591 601 902 963 91 (0 38)(0 75)(1 11)(1 23)(1 10)(1 41)(1 43)(1 50)(2 20)(2 39)ExpansionP3 P10 100 110 170 210 250 370 600 781 061 92 (0 43)(0 51)(0 82)(0 96)(1 17)(1 71)(2 34)(2 55)(3 21)(4 30)P11 121 111 141 110 960 940 850 560 19 Ã 0 40 (3 70)(3 40)(3 29)(3 12)(2 68)(2 54)(2 01)(1 16)(0 35)( Ã 0 61) P31 221 221 311 321 211 301 451 331 251 51 (3 90)(3 91)(4 13)(4 13)(3 76)(3 92)(4 01)(3 18)(2 79)(2 92)RecessionP3 P1 Ã 0 26 Ã 0 35 Ã 0 100 07 Ã 0 26 Ã 0 28 Ã 0 100 121 803 49 ( Ã 0 37)( Ã 0 42)( Ã 0 11)(0 08)( Ã 0 28)( Ã 0 32)( Ã 0 09)(0 10)(1 28)(1 68) P11 141 131 040 720 921 140 850 59 Ã 0 00 Ã 1 33 (0 68)(0 61)(0 51)(0 35)(0 45)(0 56)(0 36)(0 20)( Ã 0 00)( Ã 0 33) P30 870 780 940 790 660 860 750 711 802 16 (0 76)(0 66)(0 74)(0 59)(0 51)(0 64)(0 54)(0 39)(0 77)(0 87) wellasstatisticallyandeconomicallysignificantat2.35%( t -stat 4.21)for thehighestcreditriskgroup(ratingof13.02 BB Ã ).Momentumprofitsare highestinfirmswiththepoorestqualityofoutstandingdebt,asratedbyS&P. Thisisanewfindingthatshedslightonthesourceofprofitabilityofmomentum strategies. Momentumstrategypayoffsinthenon-Januarymonthsarealsoinsignificantforthelowestrisktercile.Forthemediumriskstocksthemomentum payoffsareasignificant0.95%permonthandforthehighriskstocksthepayoffsare2.70%permonth.ThepayoffsinJanuaryarenegativealbeitstatisticallyinsignificant.Duringrecessions,themomentumstrategypayoffsincrease monotonicallywithcreditriskbutarestatisticallyinsignificant.6Ontheother hand,duringexpansions,notonlydothepayoffsincreasemonotonicallywith6RecessionaryandexpansionarymonthsareidentifiedbyNBER.
MomentumandCreditRating 2513 creditrisk,buttheyareastatisticallyandeconomicallysignificant2.30%per monthforthepoorestcreditqualityfirms. PanelBofTableIVpresentstheresultsfor10creditriskand3momentumportfolios.Again,theevidenceshowsthatmomentumprofitsstrongly dependoncreditrisk.Focusingonthelowestriskgroup(anaverageratingof3.17 AA),themonthlymomentumprofit(P3 P1)isaninsignificant0.07%.Payoffstomomentumstrategiesincreasemonotonicallyacrossthe creditratinggroups.Thehighestmomentumpayoffof2.04%( t -stat 4.63) permonthisrecordedforthehighestcreditriskgroup(anaverageratingof 14.52 B).ConsistentwiththeresultsinTableIII,momentumprofitsbecomestatisticallysignificantonlywhencreditqualitydeterioratestoarating ofBBB Ã orbelow(BBBorbelowforthenon-Januarymonths).Further,duringeconomicexpansions,itisonceagainonlystocksratedBBB Ã orlower thatexhibitsignificantmomentumprofits.PanelBofTableIVdocuments thatthedifferenceinmomentumprofitsacrosscreditriskgroupsisdriven primarilybyloserstocks.Thereturndifferentialbetweentheloserportfolios (P1)forthelowestandhighestcreditriskfirmsaverages1.60%permonth [1.13 ( 0.47)],whereasthewinnerportfolio(P3)forthehighestcreditrisk firmsearns,onaverage,only0.37%morethanitslowestcreditcounterpart [1.56 Ã 1.19]. Thusfar,wehaveexaminedtherelationbetweenmomentumprofitability andcreditriskusingportfoliostrategiesbasedondoublesorting,firstbycredit riskthenbyprior6-monthreturn.Wenowturntoimplementingthetraditional momentumstrategies,thatis,thosebasedonlyontheprior6-monthreturn, butweconsiderdifferentinvestmentsubsamples.Inparticular,westartwith theentiresampleofratedfirmsandthensequentiallyexcludefirmswiththe highestcreditrisk(worstcreditrating).Thisanalysisrevealsthesubsampleof firmsthatdrivemomentumprofits. TableVreportstheaveragepayoffsfrommomentumstrategiesineachsubsampleasweprogressivelydroptheworst-ratedfirms.Italsoprovidesthe percentageofmarketcapitalizationrepresentedbyeachsubsample,aswellas thepercentageofthetotalnumberoffirmsincludedineachsubsample.These twomeasuresarecomputedeachmonth,andwereportthetime-seriesaverage.Thepayoffstomomentumstrategiesareinsignificantatthe5%levelwhen theinvestmentsamplecontainsstocksintheratingrangeAAAthroughBB. Remarkably,thissampleaccountsfor96.62%ofthemarketcapitalizationof theratedfirmsanditcontains78.84%ofthetotalnumberoftheratedfirms. Inotherwords,themomentumprofitsarederivedfromasampleoffirmsthat accountsforlessthan4%ofthetotalmarketcapitalizationofallratedfirms orlessthan22%ofallratedfirms. Asweprogressivelydropthebest-ratedfirms(resultsavailableuponrequest),themomentumprofitsincreasemonotonicallyasonlytheworst-rated firmsremaininthesample.ForasampleofstocksratedBorlower,themomentumprofitamountsto3.74%permonth.Moreremarkably,thereareonly about70firmsonaveragepermonththatareratedBorlower.These70firms compriseonly0.77%ofthesamplebymarketcapitalizationand4.22%ofthe
2514 TheJournalofFinanceTableVUnconditionalMomentumoverDifferentRatingSubsamplesForeachmonth t ,allNYSE,AMEX,andNASDAQstocksratedbyS&PandavailableonCRSPwithreturns formonths t Ã 6through t Ã 1arerankedintodecileportfoliosbasedontheirreturnduringthatperiod.We excludestocksthat,attheendofmonth t Ã 1,arepricedbelow$5oraresmallerthanthesmallestNYSE sizedecile.Portfolioreturnsarecomputedmonthlybyweightingequallyallfirmsinthatdecileranking.The momentumstrategyinvolvesbuyingthewinnerandsellingtheloserportfolioandholdingthepositionfor6 months(from t 1to t 6).Themonthlyreturnsrepresenttheequallyweightedaveragereturnfromthis month ' smomentumstrategyandallstrategiesfromupto5monthsago.Eachsubsequentrowinthetable representsamonotonicallydecreasingsampleofstocksobtainedbysequentiallyexcludingfirmswiththe lowestcreditrating.Thefirstcolumnshowstherawmonthlyprofitsfromthemomentumstrategyforeach subsampleoffirms. t -statisticsareinparentheses.Thesecondcolumnshowsthemarketcapitalizationofthe givensubsampleasapercentageoftheoverallsampleofS&Pratedfirms.Thethird(forth)columnprovides theaveragenumber(percentage)offirmspermonthineachsubsample.Sample:July1985toDecember2003. Thedenotessignificanceatthe5%level. Percent StockMomentumofTotalNumberofPercentage SampleProfitsMarketCapFirmsofFirms Allfirms1 29100 001,639 00100 00 (3 15)AAA-D1 28100 001,638 7999 99 (3 13)AAA-C1 2399 981,637 6999 92 (2 98)AAA-CC1 2399 981,637 6999 92 (2 98)AAA-CCC1 2199 971,636 9199 87 (2 96)AAA-CCC1 1899 971,635 8399 81 (2 89)AAA-CCC 1 1399 951,632 7099 62 (2 79)AAA-B1 1299 901,625 3599 17 (2 81)AAA-B1 0099 651,603 3397 82 (2 62)AAA-B 0 8499 121,559 4895 15 (2 33)AAA-BB0 6898 121,426 1087 01 (2 02)AAA-BB0 5696 621,292 1478 84 (1 73) AAA-BB 0 4395 031,181 4372 08 (1 38) AAA-BBB0 3992 961,085 8066 25 (1 26) AAA-BBB0 3189 06943 7357 58 (1 02) AAA-BBB 0 2682 96762 5646 53 (0 84) AAA-A0 2375 65612 6737 38 (0 75) AAA-A0 2168 02467 6428 53 (0 69) AAA-A 0 1351 97287 0017 51 (0 42) AAA-AA0 3338 94176 4410 76 (1 12)
MomentumandCreditRating 2515 totalnumberoffirms.Inotherwords,themomentumphenomenonoccursina smallfractionoftheworst-ratedstocks. III.RobustnessChecks Inthissectionweconductnumerouscheckstoensurethattheimpactof creditratingonmomentumisrobusttovariousalternativeexplanations. A.CouldCreditRatingsProxyforSystematicRisk? Thusfar,wehaveexaminedrawmomentumstrategypayoffs.Anaturalexercisewouldbetorisk-adjusttherawpayoffstoensurethattheprofitabilityof momentumstrategiesamonghighcreditriskfirmsdoesnotmerelycompensate forexposuretocommonsourcesofrisk.Weregressthemomentumpayoffsfor thethreecreditriskgroupsonthethreeFamaandFrench(1993)factorsaswell ontheexcessmarketreturn.FocusingontheFama-Frenchfactors(available uponrequest),wefindthatthemonthlyalphasare0.41%( t -stat 1.28),1.02% ( t -stat 2.85),and2.53%( t -stat 4.47)forthelow,middle,andhighcreditrisk groups,respectively.Ifanything,thealphasarehigherthantherawmomentumpayoffsreportedinTableIV,suggestingthatloserstocksareriskierthan winnerstocksandthatthemomentumstrategydoesnothavepositiveexposuretosystematicriskfactors.Theevidencestronglysuggeststhatmomentum profitabilityacrosshighcreditriskfirmsdoesnotrepresentcompensationfor systematicrisk,atleastbasedontheCAPMandtheFama-Frenchthree-factor model. B.MomentumPro Ãž tsinVariousSubsamples Recentworkarguesthatmomentumisstrongerinstocksthathavehigh informationuncertainty.Informationuncertaintyisthedegreeofambiguity aboutfirmfundamentals.Highinformationuncertaintyfirmscanbeassociated withhigherinformationacquisitioncostsandlessreliableestimatesoftheir value.Specifically,Jiang,Lee,andZhang(2005)andZhang(2006)arguethat pricedriftislargerinstockswithgreaterinformationuncertainty,whichis proxiedbyfirmsize,firmage,analystcoverage,dispersioninanalystforecasts, returnvolatility,andcashflowvolatility.7Anessentialquestionthatarisesiswhethertheimpactofcreditratingson momentumprofitabilityissubsumedbyinformationuncertainty.Toaddress thisquestion,weassesstherobustnessofmomentumprofitabilityacrossthe creditratingdimensionbasedon3 3portfoliossortedindependentlyoncredit ratingandvariablesthatproxyforinformationuncertainty. PanelAofTableVIpresentsresultsforsortsbycreditratingandfirmsize. Momentumreturnsincreasewithcreditriskacrossallsizegroups.Forinstance,forthesmall(large)firms,momentumreturnsincreasemonotonically7Jiang,Lee,andZhang(2005)alsoshowthathighinformationuncertaintystockshavelower futurereturns.
2516 TheJournalofFinanceTableVIIndependentSortsbyCreditRisk andAlternativeFirmCharacteristicsForeachmonth t ,allstocksratedbyStandard&Poor ' swithavailablereturndataformonths t Ã 6through t Ã 1(formationperiod)aredividedintoninegroupsbased,alternatively,ontheirsize, volatility,leverage,cashflowvolatility,age,analystfollowing,dispersion(bottom30%,average 40%,andtop30%),andS&Prating(best30%,average40%,andworst30%).Weexcludestocks that,attheendofmonth t Ã 1,arepricedbelow$5oraresmallerthanthesmallestNYSEsize decile.Foreachgroup,thetableshowstheaveragereturnsofthemomentumstrategy,which involvesbuyingthewinnerportfolioP10ofthebestperforming10%ofthestocksbasedontheir returnsovertheformationperiodandsellingtheloserportfolioP1andholdingthepositionfor6 months( t 1through t 6).CashFlowVolatility(CVOL)iscomputedasinZhang(2006)asthe standarddeviationofcashflowfromoperationsinthepast5years(withaminimumof3years). TheAgevariablerepresentsthenumberofmonthssincethefirm ' sIPO.IftheIPOdateisnot availableinCompustat,thentheAgevariablesrepresentsthenumberofmonthssinceCRSPfirst reportedreturndataforthisfirm.Analystcoverageiscomputedastheaveragenumberofanalysts followingafirm.AnalystdispersionismeasuredasthestandarddeviationinanalystEPSforecasts forthenextquarter,extractedfromI/B/E/S.Thedenotessignificanceatthe5%level. RatingTercile LowAverageHigh RiskRiskRisk PanelA:IndependentSortbyCreditRiskandSize Small0 310 752 66 (0 64)(1 70)(4 94)Average0 340 582 04 (1 05)(1 67)(3 23)Big0 280 941 79 (0 84)(2 10)(2 25) PanelB:IndependentSortbyCreditRiskandVolatility Lowvolatility0 110 431 18 (0 53)(1 95)(3 36)Averagevolatility0 400 791 69 (1 29)(2 49)(4 59)Highvolatility Ã 0 071 302 68 ( Ã 0 14)(2 59)(4 17) PanelC:IndependentSortbyCreditRiskandLeverage(BV(Debt)/MV(Equity)) LowLeverage Ã 0 080 962 76 ( Ã 0 21)(2 24)(2 87)AverageLeverage Ã 0 200 141 29 ( Ã 0 56)(0 40)(2 57)HighLeverage0 790 512 80 (1 50)(1 12)(4 10) PanelD:IndependentSortbyCreditRiskandAgeofFirm Young0 451 022 75 (0 99)(2 31)(4 26)Average0 230 651 82 (0 80)(1 96)(3 63)Old0 170 511 50 (0 59)(1 59)(3 14) ( continued )
MomentumandCreditRating 2517TableVI Ã‘ Continued RatingTercile LowAverageHigh RiskRiskRisk PanelE:IndependentSortbyCreditRiskandCashFlowVolatility LowCVOL0 200 381 21 (0 69)(1 03)(1 96)AverageCVOL0 510 772 51 (1 21)(1 79)(3 59)HighCVOL0 270 682 49 (0 46)(1 10)(3 25) PanelF:IndependentSortbyCreditRiskandDispersioninAnalystForecasts Lowdispersion0 280 761 57 (0 90)(2 01)(2 84)Averagedispersion0 070 372 04 (0 24)(0 98)(3 77)Highdispersion0 280 592 11 (0 74)(1 40)(2 81) from0.31%(0.28%)to2.66%(1.79%)permonthmovingfromlowrisktohigh riskfirms.Whilethemomentumprofitsdecreasewithsizeforthehighrisk firms,thereisnosizeimpactinthelowriskfirms.Thereissomeinteraction betweenfirmsizeandcreditriskasthehighestmomentumreturnexistsin thesmall,highriskfirms(2.66%)andthelowestexistsinthelarge,lowrisk firms(0.28%).Overall,theevidencesuggeststhatitiscreditriskandnotfirm sizethatprovidesthedivergentmomentumreturns. PanelsBandCshowsimilarresultsforfirmvolatilityandleverage.8For instance,whensortingindependentlyoncreditriskandvolatility,themonthly momentumreturnstolowcreditrisk,high(low)volatilitystocksareastatisticallyinsignificant Ã 0.07%(0.11%).Inotherwords,thereisnodifferential momentumreturnacrossvolatilityforthelowcreditriskstocks.Whensorting oncreditriskandleverage,themonthlymomentumpayoffstothehighrisk, low(high)leveragestocksare2.76%(2.80%).Onceagain,thereisnodifferential momentumreturnacrossleverageamongstthehigh-riskstocks. PanelDpresentstheresultsforsortsoncreditratingandage.9Momentumreturnsincreasemonotonicallywithcreditriskacrossallagegroups. Also,themomentumstrategyprofitsdecreasewithfirmagebuttheeffect isabsentamongstthelowriskfirms.Importantly,thedifferentialimpactof firmageonmomentumprofitsisfarsmallerthatthatofcreditrisk.Similarresultsobtainforsortsoncreditratingandcashflowvolatility(CVOL)108Monthlyvolatilityforastockisthesumofthesquareofthedailyreturnswithinthemonth andleverageisdefinedastheratioofbookvalueofdebttothemarketvalueofequity.9Firmageismeasuredasthenumberofmonthssincethefirm ' sIPO.10CashflowvolatilityiscomputedasinZhang(2006).
2518 TheJournalofFinance inPanelEandforsortsoncreditratingandanalystforecastdispersionin PanelF.Whilemomentumreturnsincreasemonotonicallywithcreditrisk acrossallCVOLandanalystforecastdispersiongroups,thereverseisnot true.Moreimportantly,thedifferentialimpactofcreditriskonmomentum profitsisfarlargerthantheimpactofCVOLorthatofanalystforecast dispersion. Insum,sortingoncreditratingprovidesapayoffdifferentialinmomentumstrategies,butthesamedoesnotnecessarilyholdforsortsonsize,return volatility,leverage,cashflowvolatility,firmage,andanalystforecastdispersion.11Theseproxiesforinformationuncertaintyseemtoprovidedifferential momentumpayoffsonlyinthecaseofhighcreditriskstocks,whereascredit riskprovidesdifferentialmomentumpayoffsacrossdifferentvaluesoftheinformationuncertaintyvariables.Theevidencestronglysuggeststhatcreditrisk hasanindependenteffectnotcapturedbyvariablesthatproxyforinformation uncertainty. C.TheImpactofDistress TableIIshowsthat,overtheformationperiod,theextremeloserandwinner portfolioscontainadisproportionatelylargenumberofhighcreditriskfirms. Moreover,theaveragecreditratingoftheloserstocks(BB Ã )islowerthan thatofthewinnerstocks(BB ).Wealsofindthatthedifferenceinreturns betweenthehighestandthelowestdecileratinggroupis0.61%permonth andthedifferencebetweenthehighestandthesecond-to-lastratingdecile portfoliois0.18%permonth,suggestingthatdistressedstocksexperiencelower averagereturns.Tosummarize,wehavethefollowingthreeobservations:(i) Themomentumstrategygoeslong(short)thewinner(loser)stocks,(ii)loser stockshave,onaverage,lowerratingsthanwinnerstocks,and(iii)lower-rated stocksearnlowerreturns.Thesethreeobservationstogethersuggestthatthe impactofdistressshouldresultinhigherreturnsformomentumportfoliosthat arelongwinnersandshortlosers. Thus,anessentialquestionthatarisesiswhethertheimpactofcreditratings onmomentumprofitabilityisentirelyexplainedbydistressedstocksthatrealizelowerreturns.Werulethispossibilityoutforseveralreasons.First,themaximumreturndifferentialacrossdecileratingportfoliosisonly61basispoints permonth(resultsnotreported),whereas,asnotedearlier,thereturndifferentialacrosswinnerandloserlow-ratedstocksisover2%permonth.Moreover, weimplementmomentumstrategiesoncreditrating Ã adjustedreturnsbysubtractingthematched-decilecreditratingportfolioholdingperiodreturnfrom theindividualstockholdingperiodreturn.Therating-momentumrelationis robusttosuchanadjustment(resultsavailableuponrequest).Finally,observe fromPanelCofTableVIthattheimpactofleverageonmomentumstrategy profitsisfarsmallerthanthatofcreditrating.Sinceleveragecanbethought11Wedonotpresentresultsforanalystfollowingbecausefirmsizeandanalystfollowingare highlycorrelated.
MomentumandCreditRating 2519 ofasaproxyfordistress,thissuggeststhatitisnotdistressbutcreditratings thatdriveourresults. D.OtherRobustnessChecks MoskowitzandGrinblatt(1999)documentthatindustrymomentumaccounts formuchoftheindividualstockreturnmomentum.Hence,strongermomentum inlower-ratedstockscouldbeattributedtosuchstocksbeingconcentratedin oneparticularindustrythatconsistentlyexhibitshighermomentum.However, weconfirmthatourfindingsarenotdrivenbyindustrymomentum.Inparticular,followingMoskowitzandGrinblatt(1999),wecomputeindustry-adjusted stockreturnsbysubtractingfromeachstockreturnovertheholdingperiodthe returnofthecorrespondingindustryoverthesameperiod.Thecreditriskeffect onmomentumprofitabilityisrobusttosuchanindustryadjustment(results areunreportedbutavailableuponrequest). Inasimilarmanner,weimplementfurtherrobustnesschecks,controlling forsize,volatility,tradingvolume,illiquidity,analystcoverage,andanalyst forecastdispersion.Indeed,low-ratedstocksaresmaller,havehighervolatility, lowerliquidity,loweranalystcoverage,andhigherforecastdispersionthan high-ratedstocks.Wesubtractthedecileportfolioreturncorrespondingtothe abovecharacteristicsfromtheholdingperiodreturnsoftheindividualstocks inthewinnerandtheloserportfolios.Theresults(availableuponrequest) showthatthelinkbetweenmomentumandcreditriskremainsstrongand significantevenaftercontrollingfortheabovepotentiallyrelevantmomentum determinants. IV.Conclusion Thispaperestablishesastronglinkbetweenmomentumprofitabilityand firmcreditrating.Theempiricalfindingsarebasedonasampleof3,578NYSE, AMEX,andNASDAQfirmsratedbyS&PovertheJuly1985toDecember2003 period.Theselectedsampleisrepresentative,asratedandnonratedfirms sharesimilarcharacteristicsintermsof(i)theirstockreturndistribution,(ii) themomentumprofitstheygenerate,and(iii)theirindustrydistributionamong the20industriesstudiedbyMoskowitzandGrinblatt(1999). Theextremewinnerandloserportfoliosarecomprisedmainlyofhighcredit riskstocks.Momentumprofitabilityisstatisticallysignificantandeconomically largeamonglow-ratedfirms,butitisnonexistentamonghigh-gradefirms.The resultsarerobustandcannotbeexplainedbyinformationuncertaintyasproxiedbyfirmsize,firmage,analystforecastdispersion,leverage,returnvolatility,andcashflowvolatility.Excludingfromtheanalysisthehighestcreditrisk firms,whichaltogetheraccountforlessthan4%ofthemarketcapitalization ofratedfirms,rendersthemomentumprofitabilitystatisticallyinsignificant. Indeed,ourcross-sectionalanalysisexplicitlyshowsthatmomentumtradingstrategiesareprofitableonlyamongthehighestcreditriskfirms.This maysuggestthataggregatemomentumpayoffsarehigherduringrecessionary
2520 TheJournalofFinance periodswhencreditriskisamajorconcern.However,asnotedearlier,thetimeseriesanalysisdemonstratesthatmomentumprofitabilitydoesvarywiththe businesscycle,butapparentlyinthewrongdirection,thatis,momentumpayoffsareeconomicallyandstatisticallysignificantonlyduringexpansionswhen therearefewerdefaults.Thisdisagreementbetweenthecross-sectionaland time-seriesfindingsisapuzzlethatfutureworkshouldaddress. REFERENCESAvramov,Doron,andTarunChordia,2006a,Assetpricingmodelsandfinancialmarketanomalies, ReviewofFinancialStudies 19,1001 Ã 1040. Avramov,Doron,andTarunChordia,2006b,Predictingstockreturns, JournalofFinancialEconomics 82,387 Ã 415. Barberis,Nicholas,AndreiShleifer,andRobertVishny,1998,Amodelofinvestorsentiment, JournalofFinancialEconomics 49,307 Ã 343. Chordia,Tarun,andLakshmananShivakumar,2002,Momentum,businesscycle,andtime-varying expectedreturns, JournalofFinance 57,985 Ã 1019. Daniel,Kent,DavidHirshleifer,andAvanidharSubrahmanyam,1998,Investorpsychologyand securitymarketunder-andoverreactions, JournalofFinance 53,1839 Ã 1885. Fama,EugeneF.,andKennethR.French,1993,Commonriskfactorsinthereturnsonstocksand bonds, JournalofFinancialEconomics 33,3 Ã 56. Fama,EugeneF.,andKennethR.French,1996,Multifactorexplanationsofassetpricinganomalies, JournalofFinance 51,55 Ã 84. Grinblatt,Mark,andBingHan,2005,Prospecttheory,mentalaccounting,andmomentum, Journal ofFinancialEconomics 78,311 Ã 339. Hong,Harrison,TerenceLim,andJeremyC.Stein,2000,Badnewstravelsslowly:Size,analyst coverage,andtheprofitabilityofmomentumstrategies, JournalofFinance 55,265 Ã 295. Jegadeesh,Narasimhan,andSheridanTitman,1993,Returnstobuyingwinnersandsellinglosers: Implicationsforstockmarketefficiency, JournalofFinance 48,35 Ã 91. Jegadeesh,Narasimhan,andSheridanTitman,2001,Profitabilityofmomentumstrategies:An evaluationofalternativeexplanations, JournalofFinance 56,699 Ã 720. Jegadeesh,Narasimhan,andSheridanTitman,2002,Cross-sectionalandtime-seriesdeterminants ofmomentumreturns, ReviewofFinancialStudies 15,143 Ã 157. Jiang,Guohua,CharlesM.C.Lee,andGraceYiZhang,2005,Informationuncertaintyandexpected stockreturns, ReviewofAccountingStudies 10,185 Ã 221. Moskowitz,TobiasJ.,andMarkGrinblatt,1999,Doindustriesexplainmomentum? Journalof Finance 54,1249 Ã 1289. Schwert,G.William,2003,Anomaliesandmarketefficiency,inGeorgeConstantinides,MiltonHarris,andReneStulz,eds.: HandbookoftheEconomicsofFinance (North-Holland,Amsterdam). Zhang,X.Frank,2006,Informationuncertaintyandstockreturns, JournalofFinance 61,105 Ã 136.