Is the credit default swap market more timely in predicting a credit rating event and more accurate in measuring credit risk than the corporate bond market?

(1)

MSc Finance

“Is the credit default swap market more timely in predicting a credit rating event and more accurate in measuring credit risk than the corporate bond market?”

Student: Philip Latour Student number: 11420553

Supervisor: Dr. R. (Rafael) Almeida da Matta

July 2018

Abstract

This thesis performs a comparative study between the performances of CDS spreads and corporate bond spreads. The study in this thesis is two-fold. First, the timeliness of both spreads in predicting a credit rating event is investigated by conducting an event study. This event study examines the impact of credit rating announcements on both spreads in various event windows. I find no evidence that one spread precedes the other in their reaction following a downgrade or review for downgrade. Upgrades are faster incorporated in the CDS market but this finding is not very persistent if other methodology is applied. The second part of the study analyzes and compares the accuracy of both spreads in measuring credit risk. This accuracy analysis is performed by comparing the actual historical probability of default with the risk-neutral probability of default implied by the credit spread or the CDS spread. I find that CDS spreads outperform credit spreads in measuring the credit risk of investment-grade debt. However, for

(2)

Statement of Originality

This document is written by Philip Latour who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this

document is original and that no sources other than those mentioned in the text and its references have been used creating it. The faculty of Economics and Business is

(3)

Introduction

The 2015 Oscar-nominated film The Big Short describes a slightly romanticized version of the financial crisis of 2007-2008. The movie tells the story of various investors who recognized the crisis well in advance and bet against the housing market before the housing bubble bursts in mid-2007. To make this bet, one hedge fund manager, Dr. Michael Burry (portrayed by Christian Bale), buys credit default swaps against subprime mortgage deals from investment banks like Goldman Sachs, the Deutsche Bank and Bear Stearns which allowed him to short the housing market. In January 2007, the mortgages defaults started to rise dramatically. However, the credit ratings of the mortgage-backed securities remained stable and very high (AAA) and the prices of these bonds even increased. One scene in the movie shows clearly that credit rating agencies (i.e. Standard & Poor’s) continued to provide unauthenticated high ratings, despite if they already knew that these ratings were inflated and incorrect. One explanation for this behavior was the competition between the rating agencies. If Standard & Poor’s would not provide the AAA-rating to a bond issuer, Moody’s or another AAA-rating agency would. As a result of this competition, the credit ratings of investment banks remained very high until Lehmann Brothers defaulted in September 2008, which led to massive credit downgrading.

Of course, the movie illustrates a lightly romanticized true-based story. Nevertheless, most of the events described in this movie actually happened. According to Flannery et al. (2010), the credit ratings of five large investments banks (Bear Stearns, Goldman Sachs, Lehman Brothers, Merrill Lynch and Morgan Stanley) barely changed during the financial crisis. Lehman Brothers still had an A-rating when it declared bankruptcy in September 2008. According to Opp et al. (2013), these misleading high ratings, followed by massive downgrading and defaults in 2008, have led regulators, politicians and the popular media to conclude that the business-model of the rating agency industry is essentially flawed. As a consequence of this subprime debacle during the financial crisis, rating agencies are monitored more strictly, because the market started to doubt the validity of the credit ratings they were releasing (Jacobs, 2010).

Due to their poor predictive power during the financial crisis, credit rating agencies (CRAs) have come under rising criticism. Surveys revealed that most investors in the US believe that CRAs are too slow in adjusting their ratings to new available information

(5)

about the creditworthiness of the assessed firms. An explanation for this lack of timeliness of rating changes is the through-the-cycle methodology which is applied by most CRAs. In this methodology, the credit rating is focused on the persisting credit risk component and prudent “migration” policy (Altman & Rijken, 2006). Rating agencies use this through-the-cycle approach to decrease rating volatility and to avoid a “rating bounce”, the reversion of the rating within a relatively short period of time (Löffler, 2002). Furthermore, CRAs have been highly criticized because of their changing role over time in the financial market. Where the rating agencies began as information intermediaries, selling assessments of corporate obligations to investors who were considering to buy those obligations, they eventually shifted the focus of their business models from investor to issuer, in which the rating agencies were hired by the corporations selling the obligations to provide that obligation a seal of approval. With this issuer-paid business model, CRAs face a huge conflict of interest, as they have to impartially rate securities of the corporations from who the CRAs gather their revenues (Krugman, 2010; Flannery et al., 2010).

The recently increased criticism and skepticism about the accuracy and timeliness of CRAs in predicting a credit event makes it interesting to investigate whether there are valuable alternatives to measure the credit risk of a debt-issuer. Daniels and Jensen (2005) find that credit ratings are a significant determinant of both credit spreads and CDS spreads. Furthermore, their results indicate that credit rating changes are anticipated in both the corporate bond market and the CDS market. So, these two markets could be seen as potential valuable substitutes for credit ratings in measuring credit risk and predicting credit events.

As mentioned above, alternative sources of information about the creditworthiness of a financial entity are the credit default swaps (CDSs) spread and the corporate bond spread. In general, CDSs are deeply and frequently over-the-counter traded derivatives which help to reflect valuable market information about the credit risk of the underlying financial obligations (Flannery et al., 2010). A CDS contract can be seen as an insurance that is purchased to protect the CDS buyer against the reference entity’s credit risk. In the case of a default, the CDS buyer receives the difference between the face value and the

(6)

seller, the CDS buyer pays a periodic fee to the seller. This periodic fee is called the CDS spread and is a fair representation of the market perception of the reference entity’s creditworthiness (Duffie, 1999). The corporate bond spread (or credit spread) is the difference in yield between a risky corporate bond and a risk-free (treasury) bond with the same maturity. This difference in yields compensates an investor for bearing the default risk of a corporate bond.

In this thesis, I will conduct a comparative study in which I compare the CDS market and the corporate bond market regarding the timeliness of predicting a credit rating event and the accuracy of measuring the credit risk by answering the following research question:

“Is the credit default swap market more timely in predicting a credit rating event and more accurate in measuring credit risk than the corporate bond market?”

The study in this thesis is two-fold. The first part of the study investigates and compares the timeliness of credit spreads and CDS spreads in their reaction to credit rating announcements from Moody’s. To analyze the timeliness of these spreads, an event study is conducted to determine the abnormal (excess) spreads, as well as abnormal returns following a credit rating announcement (downgrade, upgrade, review for downgrade, or review for upgrade). Furthermore, a cross-sectional multivariate regression is performed to identify which determinants have a significant influence on these abnormal spreads. In the second part, the accuracy of the credit spread and CDS spread is analyzed by comparing the actual historical probability of default with the implied risk-neutral probability of default which can be deducted from the credit spreads and the CDS spreads

This thesis is structured as followed. In Chapter 1, the existing literature about the credit rating agencies, CDS market and the corporate bond market is reviewed. Chapter 2 describes the applied methodology regarding the performed event studies, the regression analysis and the accuracy analysis. Chapter 3 represents the data description. In Chapter 4, the empirical results of the event studies, regression analysis and accuracy analysis are presented and discussed. In Chapter 5, the concluding remarks of this thesis are presented.

(7)

1. Literature Review

1.1. Credit rating agencies

Since John Moody started a small rating company in the early 1900s, which essentially synthesized the complex data in miscellaneous reports on the railroad industry into a single letter rating, the credit rating business has developed into a worldwide multi-billion dollar industry (Haan & Amtenbrink, 2011). Within a few years after Moody’s started to distribute its first ratings, three other main competitors entered the credit rating market. Poor’s Publishing Company and Standard Statistics Company (who were merged into Standard & Poor’s (S&P) in 1941) and Fitch Publishing Company (Fitch) started to rate stocks and bonds in the mid-1920s (Partnoy, 1999). These three main credit rating agencies are known as “The Big 3”.

Credit rating agencies essentially provide two services. First, they independently assess the ability of a debt issuer to meet its debt obligations. This assessed creditworthiness is expressed in a related credit rating. These credit ratings provide the public with information about the creditworthiness of governments, corporates, or financial instruments individually. Second, CRAs offer “monitoring services” by controlling debt issuers to act correctly and thereby prevent downgrades via ‘watch’ procedures (Haan & Amtenbrink, 2011).

In the current financial market, most debt issues in the US are rated by at least one of “The Big 3” CRAs. The rating is divided into various rating classes which are all assigned by a rating letter ranging from AAA to D (depending on the CRA). Appendix 1 illustrates all corresponding ratings for all three CRAs. In the case of Moody’s, AAA is the best rating. Firms or bonds that are provided by this rating are considered to have a probability of default (inability of repayment) of almost zero. Firms or bonds that are assigned with a CCC or D rating are close to or in default. A main distinction is made between investment grade rated debt and non-investment grade (or speculative grade) rated debt, where investment grade rated debt ranged between AAA and BBB, and non-investment grade rated debt are BB or below. The term “investment grade” was basically used to indicate financial securities which are eligible for investments by institutions such as insurance companies and banks (Standard & Poor’s, 2002).

(8)

1.1.1. Credit rating determination

CRAs provide ratings to debt issuers only when there is sufficient information available to construct a trustworthy assessment and only after pertinent qualitative, quantitative and legal analyses are conducted. This rating process is not restricted to the examination of certain financial measures. An appropriate rating assessment of a company’s credit quality includes a profound and complete review of the company’s business fundamentals, including the prospects for the business industry’s growth, the exposure to technological changes, labor agitation, or regulatory actions. CRAs determine the rating based on their assessment on the firm’s comprehensive creditworthiness and capacity to meet its financial obligations (Standard & Poor’s, 2002).

In the scenario of a rating change, CRAs typically signal these upcoming change using ‘outlooks’ and rating reviews (so-called ‘watchlists’). Whereas outlooks are more an evaluation of the development of a credit rating over the medium-term (approx. 18 months), reviews are more short-term focused (approx. 90 days). These watchlist and review procedures are providing a generally strong indication of the rating action (Hill et al., 2010; Haan & Amtenbrink, 2011).

1.1.2. Criticism on credit rating agencies

In the last decades, but especially since the credit crisis of 2007, the CRAs are subject to rising criticism. The credit crisis painfully exposed the inaccuracy of the credit ratings, as CRAs were not able to identify and signal the upcoming events. The ratings of the five largest investment banks and five largest commercial banks of the US remained remarkably stable in the build-up to the crisis and the ratings of Lehman Brothers and Bear Stearns remained within the A-category when they both failed in September and March 2008, respectively (Flannery et al., 2010). The following massive downgrading and defaults during the crisis have resulted in rising critique from politicians, regulators, the media and, the investors (Opp et al., 2013).

One of the main points of critique is the conflict of interest that the CRAs face. Where CRAs essentially were investor-paid information intermediaries who were selling credit assessments of financial obligations to investors who were considering to buy these obligations, the approach of CRAs to generate revenues changed in the mid-1970s

(9)

(Flannery et al., 2010; Partnoy, 1999). During that time the CRAs adjusted their business model from the originally investor-paid model into an paid model. In this issuer-paid model, debt-issuing companies pay fees to the rating agencies to rate the issuer’s debt obligations. This issuer-paid model results in a conflict of interest for the CRAs, since their debt-issuing clients prefer more favorable ratings as it directly reduces their cost of capital, and they do not necessarily prefer authentic and correct ones (Becker & Milbourn, 2011). Thereby, because the CRAs generate their revenues from these debt-issuers, this issuer-paid model could potentially influence the CRAs’ objectivity and independency (Krugman, 2010).

Bolton, Freixas and Shapiro (2012) describe three main sources of conflicts for CRAs. First, they claim that CRAs face a conflict of interest in understating the credit risk of a debt-issuer in order to attract new business or to build a steady relationship with an issuer. Faltin-Traeger (2009) finds that repeating debt-issuing firms are more loyal to a certain CRA if they were provided by a more favorable rating by the same CRA in the past. Furthermore, He, Qian & Strahan (2012) state that debt-issuing companies of large structured products are often provided with more favorable ratings by CRAs, confirming the presence of a conflict of interest for the CRAs, since these larger debt-issuing firms potentially bring or take away substantial business and revenues

Second, the issuers’ ability to purchase only the most favorable rating at the CRA that is providing this, also brings an incentive for CRAs to provide higher or inflated ratings. Since the credit rating industry has three main rating companies, a debt-issuer has multiple opportunities to shop for the most favorable ratings (also-called “rating

shopping”) and to take advantage of credulous investors by only buying the best ratings.

Becker & Milbourn (2011) also show that increased competition in the rating industry concurs with more inflated rating and more deterioration in rating quality.

A third source of conflict is the trusting nature of some investor clienteles. Especially during boom times, when the investors are more trusting and the probability of getting exposed is relatively small, ratings are being more often inflated (Bolton, Freixas and Shapiro, 2012). These various conflicts of interest have led to a rising critique about the

(10)

Another point of critique on the CRAs is regarding the lack of timeliness of credit rating changes. Moody’s takes a rating action only “when it is unlikely to be reversed within a relatively short period of time” (Cantor, 2001). CRAs apply this through-the-cycle approach to maintain the ratings stable, since a change in rating has a large impact on the creditworthiness and thereby the cost of capital of a debt-issuing firm. CRAs only take rating actions when they believe a debt-issuer experiences a presumable enduring change in its fundamental creditworthiness (Altman & Rijken, 2006). This avoidance of a rating bounce is the result of the conflicting objectives of CRAs to provide investors with stable but also timely and accurate ratings (Löffler, 2005).

1.1.3. Informational Content of Credit Ratings

According to the information content hypothesis, CRAs have access to information that is not available for the market, which implies that any rating announcement provides new and valuable information to the public market. This new credit information about a financial entity will be immediately incorporated by the market after the announcement (Steiner & Heinke, 2001). However, since the CRAs apply the through-the-cycle

methodology, credit ratings are relatively stable with the result that rating changes are potentially lagging the market’s credit perception (Löffler, 2005). In that scenario, rating changes do not provide any new information to the market and price reactions will be already observable before the rating announcement.

Several prior studies have investigated whether credit rating announcement provide new information to the corporate bond market or CDS market. Katz (1974) was one of the first who investigated the impact of rating changes on bond prices. His results indicate no market anticipation prior to a credit rating announcement and suggest that the bond market is inefficient and slow to assimilate the new credit information after the announcement. Contrary findings by Hite & Warga (1997) report a significant announcement effect in the month prior to negative credit announcements (downgrades). This effect is especially high for firms that are moving from investment-grade to speculative-grade debt as a consequence of the downgrade. The reaction following upgrades is much weaker and insignificant. This asymmetric effect is in line with the findings of Hand, Holthausen & Leftwich (1992). They suggest that the potential explanation for this asymmetric effect is that CRAs assign more resources on disclosing

(11)

potential downgrades, as providing inflated rating to their clients could affect their reputation more severely. Consistent with this asymmetric effect, Steiner & Heinke (2001) report a significant abnormal return in bond prices following a downgrade or a negative watchlist, but do not find any significant abnormal return in the event of a positive rating announcement.

Since the CDS market is relatively new compared to the traditional bond market, the number of studies about the effect of rating announcements on the CDS market is substantially lower. Norden & Weber (2004) find a significant anticipation in the CDS spread prior to a downgrade and a review for downgrade. However, they find no significant abnormal returns for positive rating announcements. In line with these findings, Hull, Predescu & White (2004) also report significant anticipation in case of a negative rating announcement, but no significant effect for positive announcements. These asymmetric results for downgrades and upgrades suggest that negative rating announcement from CRAs do convey new and valuable information to the CDS market and that information coming from positive rating announcements is already incorporated by the CDS market.

1.2. The Credit Default Swap

A credit default swap (CDS) is an insurance or protection that the holder of a defaultable bond can purchase as protection against a given credit event, such as a default, of a particular company or sovereign entity. So, a CDS contract transfers the credit exposure of a reference entity (the company that has issued the underlying bond) from the CDS buyer to the CDS seller. In the case of a credit event, the CDS buyer receives a contingent amount from the CDS seller. This contingent amount is often the difference between the face value of the underlying bond and its market value (e.g. recoverable amount). In return, the buyer of a CDS contract makes periodic payments to the CDS seller until the maturity date of the CDS contract or until a credit event occurs. Credit events that typically trigger a CDS include bankruptcy, failure to make a principal or interest payment, obligation acceleration, a repudiation or moratorium (refusal of the reference entity to acknowledge or pay a debt), or a restructuring (Longstaff, Mithal & Neis, 2004). A credit event must be officially documented with a notice and supported with a public

(12)

announcement of the firm (Duffie, 1999). Figure 1 illustrates a schematic illustration of a CDS transaction and the consequent transfer of credit risk.

Figure 1: Illustration of CDS transaction and transfer of credit risk

The periodic premium paid by the protection buyer to the protection seller is called the CDS spread and is typically quoted in basis points (1 bps = 0.01%) of the notional amount of the underlying bond. So, for example, a protection buyer who buys a CDS with a CDS spread of 60 bps and a protected notional amount of $10 million should pay a periodic fee of $60.000 to the CDS seller. The CDS spread has become a popular measure of a firm’s credit risk over the years and is a fair representation of the market perception of a firm’s credit risk. If the market expectation of a certain reference entity’s creditworthiness decreases, the price of protection against a credit event of this reference entity is expected to increase, and accordingly the CDS spread is also expected to increase (Duffie, 1999).

1.2.1. Relationship between Credit Rating changes, Credit spreads and CDS Spreads Credit ratings, credit spreads, and CDS spreads are all measures of credit risk in their own respect. Credit ratings can be considered as mathematically derived rankings of creditworthiness by analysts, with the objective to provide accurate and stable rankings to the investors (Altman & Rijken, 2006). Where these credit ratings are assigned by professionals, the credit spread and CDS spread are reflections of the market’s perception of the creditworthiness of a financial entity, which are updated constantly. The credit spread reflects the additional return that an investor demands over the risk-free rate, and the CDS spread reflects the premium that an investor has to pay to protect itself against a default (Duffie, 1999). As one would expect, the credit spread and CDS spread for a

(13)

reference entity are both negatively related to its credit rating: the worse the credit rating, the higher the credit spread and CDS spread (Hull, Predescu & White, 2004).

1.2.2. CDS – Bond Basis

The corporate bond spread or also-called credit spread of a defaultable corporate bond is the difference in yield between this corporate defaultable bond and a risk-free non-defaultable bond (e.g. treasury bond) with the same maturity.

Theoretically, the CDS spread of a financial entity should be approximately equal to the credit spread. Under the assumption of no-arbitrage, the cash flows from a portfolio containing a n-year par yield corporate bond and a n-year CDS spread on this corporate bond should approximately be equal to the cash flows from a n-year par yield risk-free bond (Hull, Predescu & White, 2004). This relationship could be summarized into the following theoretical equation:

Credit spread  Corporate Bond Yield – Risk-free Rate  CDS spread

In the case that the CDS spread is greater than the difference between the corporate bond yield and the risk-free rate, an arbitrage opportunity will exist where it is

profitable to take a short position in the CDS contract and the corporate bond and buy the risk-free bond, under the assumption that investors are allowed to take a short position in corporate bonds and to borrow at the risk-free rate (Hull, Predescu & White, 2004).

Various prior studies have examined the existence of this theoretical equality of the CDS spread and the credit spread. The studies of Blanco et al. (2004) and Norden & Weber (2004) suggest that the approximate theoretical equivalence of the CDS spread and the credit spread broadly holds. However, these studies also find considerable short-term deviations between the two spreads which are due to liquidity effects. These findings are consistent with Longstaff, Mithal & Neis (2004) who also suggest that the differences in spreads could be explained by differences in liquidity of the CDS market and the corporate bond market.

(14)

The difference between the CDS spread and the credit spread is also known as the CDS-Bond basis:

CDS-Bond Basis = CDS Spread – Corporate Bond Spread (or Credit Spread)

The CDS-Bond basis could potentially also be explained by the additional counterparty risk on the CDS market. In the case of a default, the CDS buyer is then owed a payment from its counterparty, which is equal to the difference between the bond’s face value and recovering value. However, if the default was unexpected by the counterparty, the CDS seller could suddenly face large losses, which potentially could force the CDS seller into financial distress. Thus, the CDS buyer might not receive the agreed protection payment from the CDS seller. To compensate for this additional risk, CDS buyers might demand a lower CDS spread (De Wit, 2006; Arora et al., 2011).

(15)

2. Methodology

In this section, the applied methodology and approach in this study are described. The comparative study in this thesis is two-fold. The first part of the study focuses on the timeliness of credit spreads and CDS spreads in their reaction to credit rating announcements and compares them to each other. To analyze the timeliness of these spreads, an event study is conducted to determine the abnormal (excess) spreads, as well as the abnormal returns following a credit rating announcement (downgrade, upgrade, review for downgrade, or review for upgrade). Furthermore, a cross-sectional multivariate regression is performed to identify which determinants have a significant influence on these abnormal spreads. In the second part, the accuracy of the credit spread and CDS spread is analyzed by comparing the actual historical probability of default with the implied risk-neutral probability of default which can be derived from the credit spreads and the CDS spreads.

2.1. Timeliness of Credit Spreads and CDS Spreads

The first part of my study investigates the effect of a credit rating event (e.g. downgrade, review for downgrade, upgrade or review for upgrade) on the credit spread and the CDS spread. An event study is conducted to study whether the credit spread and the CDS spread have a significant reaction before or after a credit rating event. Furthermore, I will test whether significant differences exist in the timeliness of the reactions of the two markets to see whether one market incorporates the information of the credit event faster than the other.

2.1.1. Hypotheses regarding Timeliness

Prior academic studies examining the effect of rating events on CDS spreads and credit spreads have found mixed results regarding the timeliness of these spreads. According to the information content hypothesis (Steiner & Heinke, 2001; Micu et al., 2006), negative credit rating announcement do convey new information to the market which should result in an increase in both the credit spread and CDS spread. Since actual rating changes are often preceded by reviews for changes, I distinguish these two types of rating events.

(16)

In accordance with prior research (Norden & Weber, 2004; Micu et al. 2006), I expect that if various rating events reflect the identical credit information, only the earliest event incorporates the pricing-relevant information. Thus, if a company experiences a credit rating review which is followed by an actual rating change, I expect the review to have a more significant impact on both the spreads, as the market will instantly adjust their opinion about the creditworthiness of the company, implying that all new revealed information is incorporated in both spreads at the moment of the actual rating change announcement. Consequently, the impact of the subsequent rating change is relatively small, except if the amount of notches of the rating change is lower or higher than expected by the market. This effect holds only under the assumptions that all reviews are reliable sources of information and the time between the review announcement and the actual rating change is not too long (Norden & Weber, 2004). Furthermore, prior research (e.g. Bannier & Hirsch, 2010) expects that unexpected rating changes which are not preceded by a rating review have a more severe impact on both spreads.

Hence, in my first two hypotheses I will test the effect of negative rating events on both the credit spreads (H1) and CDS spreads (H2).

H1a: A credit rating downgrade has a positive effect on the credit spread H1b: A review for downgrade has a positive effect on the credit spread H2a: A credit rating downgrade has a positive effect on the CDS spread

H2b: A review for downgrade has a positive effect on the CDS spread

Because of the comprehensive credit monitoring by debt investors and credit analysts, prior academic research suggests that negative rating events (e.g. downgrades and review for downgrades) are better anticipated by market participants than positive rating events. In contrast to negative rating events, most previous studies finds either an insignificant or at most a weak market reaction to positive rating events (Finnerty et al., 2013). Consequently, I do not expect to find a significant effect of positive rating events on both spreads.

H3a: A credit rating upgrade has no significant effect on the credit spread H3b: A review for upgrade has no significant effect on the credit spread

(17)

H4a: A credit rating upgrade has no significant effect on the CDS spread H4b: A review for upgrade has no significant effect on the CDS spread

Furthermore, I will investigate whether significant differences exist in the timeliness of the reactions of the bond market and CDS market on credit rating events to see whether one market incorporates the new information more quickly than the other. This part of the study is most closely related to the research of Daniels & Jensen (2005). Their findings suggest that credit rating changes are anticipated in both the corporate bond market and the CDS market, but the CDS market reacts faster and more significantly to rating changes than the bond market. In line with these findings, I expect the CDS spread to have a faster reaction to credit rating events than the credit spread.

H5: The CDS market responds earlier to rating events than the corporate bond market

2.1.2. Event Study Methodology

The event study approach is applied to determine the cumulative abnormal spreads for both the credit spreads and the CDS spreads following a rating event. An event study attempts to measure the valuation effects of a specific corporate event by examining the response of the prices, spreads or returns around the announcement of the event. The underlying idea is to compare the actual spread around the announcement of the event with a theoretical spread that would be expected if the event would not have taken place. The abnormal spread (return) is the difference between the actual spread (return) and the expected spread (return) (MacKinlay, 1997).

The event study assumes that the Efficient Market Hypothesis (EMH) holds. Building on the Efficient Market Hypothesis, originally expounded by Fama (1970), event studies imply semi-strong efficient markets, namely that all publicly available information is reflected in the firm’s credit spread or CDS spread. When investors receive new information (e.g. the announcement of a credit rating event), they react immediately and the spreads react accordingly (Fama, 1970).

(18)

as the difference between the risky bond yield and the risk-free bond yield with the same maturity. The sample period covers 5 years beginning in January 2012 and ending in December 2016. In this event study, an event is any credit rating announcement (downgrade, review for downgrade, upgrade or review for upgrade) coming from Moody’s within the sample period. Further information about the data sample is provided in the data description.

2.1.3. Estimation Period and Event Windows

The study of MacKinlay (1997) shows that there are a few steps involved in performing an event study. The first step is to set a timeline wherein the event study is conducted. Before the abnormal spreads are measured, the event has to be split into two periods: the estimation period and the event window.

Figure 2: Event Study Model

First, the considered event and the corresponding event date (T0) have to be identified and the period in which the abnormal spreads (returns) are measured should be defined. This period is designated as the event window (EW). The considered event is the announcement of a credit rating event (review or rating change) by Moody’s and the corresponding event date is the date of this announcement. The event window covers the period 90 days (T-90) prior to the event until 10 days (T10) after the event, which is in line with the event window applied in the research of Hull, Predescu & White (2004). Figure 2 provides a schematic illustration of the event window and the estimation period.

This event window is subdivided into five different event windows which makes it possible to classify the abnormal spreads into certain periods of time. The first event window covers the period 90 days (T-90) to 61 days (T-61) prior to the event (EW1 = [-90/-61]). The second event window spans the time 60 days (T-60) until 31 days (T-31) prior to the event (EW2 = [-60/-31]), the third event window spans the time 30 days (T-30) until 3 days (T-3) prior the event (EW3 = [-30/-3]), the fourth event window spans 2 days (T-2)

Estimation Period ( Event Window (

(19)

before the event until 2 days (T+2) after the event (EW4 = [-2/+2]), and the fifth event window covers 3 days (T+3) after until 10 days (T+10) after the event (EW5 = [+3/+10]).

As mentioned before, the underlying idea of an event study is to compare the actual observed spreads around the event date with the expected spreads. The expected (normal) spreads are the theoretical spreads that would have expected if the event would not have taken place. These expected spreads are calculated in the period prior to the event window and this period is called the estimation period. In this study the estimation period begins 180 days (T-180) before the event and ends 91 days (T-91) before the event and contains 90 daily observations.

2.1.4. Computation of Excess Spreads

To compute the excess spreads for the bond market and the CDS market I adopt the methodology and assumptions of Daniels & Jensen (2005). They assume that for a given sample of N events, the spreads stm = [s1m, … , sNm] are independently and multivariate normally distributed for all t and m = Credit spread, CDS spread. The estimated spread over the estimation window is calculated using a constant mean model in which the expected spread is the mean spread over the 90 days within the estimation period. This estimated spread is calculated using the following equation:

𝑠_𝑖𝑡𝑚 = _im+ _itm, 𝑡  _,𝑚 = 𝐶𝑟𝑒𝑑𝑖𝑡 𝑠𝑝𝑟𝑒𝑎𝑑, 𝐶𝐷𝑆 𝑠𝑝𝑟𝑒𝑎𝑑 (

for each event i, where _im_{is the mean spread taken over the estimation period (L1) and} _itm_{is a normally distributed error term with an expected mean equal to zero.}

The abnormal (excess) spread (AS) for each day within the event window (L2) is calculated by subtracting the estimated mean spread over the estimation period from the observed spread:

𝐴𝑆_𝑖𝑡𝑚 = s_itm− µ̂_im, 𝑡  ,𝑚= 𝐶𝑟𝑒𝑑𝑖𝑡 𝑠𝑝𝑟𝑒𝑎𝑑, 𝐶𝐷𝑆 𝑠𝑝𝑟𝑒𝑎𝑑 (2

(20)

The cumulative abnormal (excess) spread (CAS) is measured by summing up the abnormal spreads of all the days in the event window:

𝐶𝐴𝑆_𝑖𝑚 = ∑ 𝐴𝑆_𝑖𝑡𝑚 𝑡𝐿2,

m = Credit spread, CDS spread ₍₃ with a sample variance (̂_𝑖𝑚 ₂_{= ∑}

𝑉̂_{𝑖,𝑡𝑠}𝑚

𝑡,𝑠𝐿2 where 𝑉̂𝑖

𝑚_{is the covariance matrix of the} abnormal spreads which is assumed to be equal to zero (no correlation between the abnormal spreads across time and events). This absence of correlation implies that we can aggregate the cumulative abnormal spreads over a subsample (T) of n events.

𝐶𝐴𝑆𝑚( = 𝑛∑ 𝐶𝐴𝑆𝑖 𝑚 𝑖T 𝑎𝑛𝑑 (̂𝑚 _{( =} 𝑛 ∑(̂𝑖 𝑚 𝑖T (4 and (5

The null hypothesis H0 that the given credit rating announcements have no significant impact on the spreads is tested using the following test-statistic:

𝐽 = 𝐶𝐴𝑆 𝑚

̂𝑚 (6

2.1.5. Cumulative Abnormal Return using the Market Model

Next to the described abnormal excess spread model adopted from Daniels & Jensen (2005), a second model is constructed to investigate the impact of rating events on both credit spreads and CDS spreads. This alternative model is applied to check the robustness of the results achieved by the abnormal excess spread model. This second model is consistent with Micu et al. (2006), and computes the cumulative abnormal return of the credit spread and the CDS spread by applying the market model. The usage of abnormal returns differs from abnormal spreads since abnormal spreads focus on absolute changes where abnormal returns focus on relative changes. The credit spread returns and CDS spread return are calculated by:

𝑅_𝑖,𝑡𝑚 = 𝑠𝑖,𝑡 𝑚_{− 𝑠}

𝑖,𝑡 𝑚

(21)

The abnormal return is calculated by subtracting the expected return estimated over the estimation period from the actual observed return.

The abnormal returns (AR) can be estimated by:

𝐴𝑅_𝑖,𝑡𝑚 = 𝑅_𝑖,𝑡𝑚 − (_𝑖𝑚+ _𝑖𝑚𝑅_𝑘,𝑡𝑚 + _𝑖𝑡m), 𝑚 = 𝐶𝑟𝑒𝑑𝑖𝑡 𝑠𝑝𝑟𝑒𝑎𝑑, 𝐶𝐷𝑆 𝑠𝑝𝑟𝑒𝑎𝑑 (8

where (_𝑖𝑚₊_ 𝑖 𝑚_𝑅

𝑘,𝑡𝑚 + 𝑖𝑡𝑚 ) is the expected return estimated over the estimation period by the market model. The parameters _𝑖𝑚_and_

𝑖

𝑚_{are the intercept and the slope} estimated over the 90 days prior to the event window and _𝑖𝑡𝑚_{represents the mean zero} error term. 𝑅_𝑘,𝑡𝑚_{represents the average return of the firms within the same rating class as} the firm who experiences the rating event.

The cumulative abnormal return is measured by summing up the abnormal returns of the days in each event window:

𝐶𝐴𝑅_𝑖𝑚 = ∑ 𝐴𝑅_𝑖𝑡𝑚 𝑡𝐿2,

m = Credit spread, CDS spread ₍₉

where the covariance matrix of the abnormal returns is assumed to be zero. Hence, the aggregated cumulative abnormal return for each subsample (T) with n events is given by:

𝐶𝐴𝑅𝑚( = 𝑛∑ 𝐶𝐴𝑅𝑖 𝑚 𝑖T 𝑎𝑛𝑑 (̂𝑚 _{( =} 𝑛 ∑(̂𝑖 𝑚 𝑖T ( 0 and (

2.1.6. Cross-sectional Multivariate Regression

In this section I will identify the determining factors of the abnormal spreads for rating changes by performing a cross-sectional multivariate OLS regression. In this regression analysis the dependent variable is the cumulative abnormal spread (credit spread or CDS spread) for each event window. Due to the small number of observations for credit rating upgrades, only the abnormal spreads following downgrades are analysed in this regression.

(22)

In this regression analysis the used explanatory variables are Initial Rating, More, Review and IGSG. The initial rating of the firm (A or higher, Baa, Ba, B, and Caa or lower) represents the firm’s credit rating before the rating change. Firms that are listed with an A-rating or higher are considered the base group (with a value of 0 for all rating dummies). These dummy variables are in line with the regression model of Daniels & Jensen (2005). More is a dummy variable that is equal to 0 if the rating change is by one notch, and equal to 1 if the rating change is by two or more notches. Including this dummy variable is consistent with Norden & Weber (2004) and Daniels & Jensen (2005). Review represents a dummy variable which is equal to 0 if the rating change was unexpected and not preceded by an associated review, and equal to 1 if the rating change was preceded by a review. This dummy variable is consistent with the research of Bannier & Hirsch (2010). IGSG is a dummy variable which is equal to 1 if the rating change leads to a transition from investment grade to speculative grade, and equal to 0 if not. The results of Hite & Warga (1997) indicate that the corporate bond spread has a significant positive reaction for downgrades that lead to a transition from investment-grade to speculative-grade. Furthermore, a control variable is included which reflects the various types of industries represented in the sample. A list with all variables and the associated descriptions and expected directions are provided in table 1.

The cross-sectional multivariate regression analysis of the credit spread (CS) and the CDS spread (CDS) is estimated using the following equations:

𝐶𝐴𝑆𝐶𝑆 =  +  𝐵𝑎𝑎𝑖+ 𝐵𝑎𝑖+ 𝐵𝑖+₄𝐶𝑎𝑎𝑖+₅𝑀𝑜𝑟𝑒𝑖+ 𝑅𝑒𝑣𝑖𝑒𝑤𝑖 +₇𝐼𝐺𝑆𝐺𝑖 + 𝐶𝑜𝑛𝑠𝑢𝑚𝑒𝑟𝐺𝑜𝑜𝑑𝑠_𝑖 + 𝐸𝑛𝑒𝑟𝑔𝑦_𝑖+𝑀𝑎𝑛𝑢𝑓𝑎𝑐𝑡𝑢𝑟𝑖𝑛𝑔_𝑖 +𝑆𝑒𝑟𝑣𝑖𝑐𝑒𝑠𝑖 +𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙𝑠𝑖 + 𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖 + 𝑖 𝐶𝐴𝑆_𝐶𝐷𝑆=  +  𝐵𝑎𝑎_𝑖 + 𝐵𝑎_𝑖 + 𝐵_𝑖 +₄𝐶𝑎𝑎_𝑖 +₅𝑀𝑜𝑟𝑒_𝑖 + 𝑅𝑒𝑣𝑖𝑒𝑤_𝑖+₇𝐼𝐺𝑆𝐺_𝑖 + 𝐶𝑜𝑛𝑠𝑢𝑚𝑒𝑟𝐺𝑜𝑜𝑑𝑠_𝑖 + 𝐸𝑛𝑒𝑟𝑔𝑦_𝑖 +𝑀𝑎𝑛𝑢𝑓𝑎𝑐𝑡𝑢𝑟𝑖𝑛𝑔_𝑖 +𝑆𝑒𝑟𝑣𝑖𝑐𝑒𝑠_𝑖+𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙𝑠_𝑖+ 𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛_𝑖+ _𝑖

(23)

Variable Description Expected Direction Initial Rating Dummy variable representing the initial

rating of the firm prior to the credit rating event (e.g. dummy variable for Baa, Ba, B, and Caa. Companies with an A-rating or higher are considered the base group (value of 0 for all dummies), to prevent collinearity. There are no other rating grades within the sample.

(+) for downgrades

A lower initial rating is expected to have a higher impact on the abnormal spread.

More Dummy variable for the size of the rating

change (equal to 0 if rating change is by one notch, equal to 1 if rating change is by two or more notches).

(+) for downgrades

The size of the rating change is expected to have a higher impact on the spread.

Review Dummy variable indicating if the rating

change was preceded by an associated review within 60 days prior to the actual rating change (equal to 0 if there was no preceding review, equal to 1 if there was a review).

(-) for downgrades

An expected rating change is expected to have a smaller impact on the spread than an unexpected rating change.

IGSG Dummy variable representing whether the

firm is making the transition from investment grade to speculative grade due to the rating change or vice versa (0 if there is no transition, 1 if there is a transition).

(+) for downgrades

The transition from IG to SG (SG to IG) is expected to have a larger impact on the spreads for downgrades (upgrades).

Industry Dummy control variable representing the type

of industry.

CAS Dependent variable in the regression analysis representing the cumulative abnormal spread (credit spread or CDS spread).

Table 1: List of variables included in the regression analysis with associated description and expected direction

2.2. Accuracy of Credit Spreads and CDS Spreads

In the second part of this study, the accuracy of the corporate bond market and CDS market in measuring credit risk is investigated. To perform this accuracy analysis the implied risk-neutral probability of default (RNPD) of a firm with a certain credit rating is compared to the actual historical probability of default (APD) of firms within that same

(24)

Almeida & Philippon, 2007; Murthy, 2011), however these studies only compare the APD with the RNPD implied by the bond market or the CDS market alone, and not include a comparison between the accuracy performances of the two markets.

2.2.1. Credit Risk Premium

Prior research suggests that the spreads on corporate bonds tend to be much higher than the spreads that would be implied by the expected losses in the scenario of a default alone. Even if the impact of a default is taken into account, an investor could expect significantly higher returns from investing in risky corporate bonds than from investing in non-defaultable risk-free bonds. This discrepancy between the historical implied spread and the risk-neutral implied spread is referred to as the credit spread puzzle (Altman, 1989; Remolona, Allen, Borio, et al., 2003; Hull, Predescu & White, 2005). An explanation for the existence of the credit spread puzzle is that bond traders do not exactly base their bond prices only on the actual probability of a default. They also include an additional premium to compensate for the risks they are bearing. This credit risk premium has various determinants and differs over time and across the quality of the underlying bond (Hull, Predescu & White, 2005; Heynderickx et al., 2016).

The credit risk premium implies that the corporate credit spread is not entirely accounted by the default risk component. There are other non-default factors which also have an impact on the credit spread. Longstaff, Mithal & Neis (2004) test whether the non-default component in the credit spread is related to the illiquidity of corporate bonds. They find evidence that the nondefault component is strongly related to the illiquidity measures of individual corporate bonds, such as the bid/ask spread. This result is supported by Hull, Predescu & White (2005) suggesting that liquidity is an important component of the risk premium, especially for high-quality bonds.

Another part of the nondefault component of the credit spread could be attributed to the tax effect. This tax effect is caused by different tax treatment of risky corporate bonds and risk-free treasury bonds. In the United States, corporate bonds are subject to taxes at the state level whereas treasury bonds are not. Longstaff, Mithal & Neis (2004) only finds a weak evidence for this tax effect. However, Driessen (2004) finds that this tax effect accounts for more than 30% of the credit spreads.

(25)

2.2.2. Hypothesis regarding Accuracy

According to Longstaff, Mithal & Neis (2004) the usage of CDS spreads instead of credit spreads eliminates the tax effect and reduces the liquidity effect. Therefore, the credit risk premium is lower and the implied risk-neutral probability of default from CDSs are expected to be closer to the actual historical probability of default.

H6: CDS spreads are a more accurate measure of credit risk than credit spreads

2.2.3. Risk-Neutral Probability of Default and Actual Probability of Default

To derive the risk-neutral probability of default (RNPD) from the credit spread or the CDS spread I follow the methodology applied by Hull, Predescu & White (2005). They state that an appropriate estimation of the risk-neutral default intensity per year () for a bond is given by:

_{𝑅𝑖𝑠𝑘 𝑛𝑒𝑢𝑡𝑟𝑎𝑙,𝑖} = 𝑦𝑖− 𝑟 ( − 𝑅𝑖

( 2

where 𝑦_𝑖 is the corporate bond’s yield, 𝑟 is the yield on a risk-free bond with the same maturity as the corporate bond, and 𝑅_𝑖 is the recovery rate of corporate bond i in the scenario of a default. The recovery rate 𝑅𝑖 is assumed to be 40% for the corporate bonds which is in line with the common assumption of the market participants (Hull, Predescu & White, 2005; Heynderickx, 2016).

This formula can be converted to estimate the risk-neutral default intensity per year derived from the CDS spread. In order to construct this equation, the theoretical equality between the CDS spread and the corporate bonds spread must be assumed (CDS = y – r). If we insert this equality equation, the following formula is obtained:

_{𝑅𝑖𝑠𝑘 𝑛𝑒𝑢𝑡𝑟𝑎𝑙,𝑖} = 𝐶𝐷𝑆 𝑆𝑝𝑟𝑒𝑎𝑑𝑖 ( − 𝑅𝑖

( 3

The actual historical probabilities of default (APD) are obtained from Moody’s (2016), where they report the cumulative default rate for bonds with maturity of T years. In order

(26)

to obtain the average default intensity per year over the T years, the formula of Hull, Predescu & White (2005) is applied:

𝑒 ℎ𝑇 = − 𝑑 ( 4

where 𝑑 is the cumulative default rate for T years and h is the average default intensity over the T years. This equation can be rewritten as:

ℎ = − ln( − 𝑑 = _{𝐴𝑐𝑡𝑢𝑎𝑙} ( 5

Heynderickx et al. (2016) defines the ratio between the two default intensities as the relative credit risk premium or the coverage ratio ():

 = 𝜆𝑅𝑖𝑠𝑘 𝑁𝑒𝑢𝑡𝑟𝑎𝑙,𝑖 𝜆_{𝐴𝑐𝑡𝑢𝑎𝑙} 

𝑅𝑁𝑃𝐷

𝐴𝑃𝐷 ( 6

A coverage ratio close to 1 indicates that the implied RNPD is close to the APD and thereby an accurate measure of the credit risk.

3. Data Description

This section describes the used data to investigate the timeliness of credit spreads and CDS spreads in their reaction on credit rating events as well as the accuracy of these spreads in measuring credit risk.

3.1. Credit Default Swap data

The 5-year CDS spreads of US firms are obtained from Thomson Reuters’s DataStream. The data covers the period from 1 January 2012 until 31 December 2017. The CDS contract with a maturity of 5 years is the most commonly traded CDS contract. I obtained daily quotes on 5-year CDS contracts on senior unsecured debt and exclude all contract that are written on subordinate or junior debt obligations.

(27)

3.2. Corporate bond data

The corporate bond spreads are calculated by subtracting a certain risk-free rate from the corporate bond yield. In order to match the 5-year maturity of the CDS contracts I also require the yields of 5-year bonds. However, in practice, a corporate bond with a maturity of exactly 5 years is seldom available. To tackle this problem I use linear interpolation to obtain the maturity-matched 5-year bond yield. This method of linear interpolation is in line with prior research of Blanco et al. (2004) and Houweling & Vorst (2005). To conduct this linear interpolation of a 5-year bond yield, I need the availability of a bond with a maturity shorter than 5 years and longer than 5 years over the entire sample period. By linearly interpolating the yields of these bonds with a shorter and longer maturity, it is possible to estimate the 5-year bond yield for the entire sample.

For each reference entity with suitable CDS data, the corporate bond yield for all available bonds are obtained using TRACE. I exclude non-plain vanilla bonds, such as floating-rate bonds, convertible bonds, and all securities that have imbedded step-up coupons, options, or other special features that could potentially cause different pricing. Furthermore, only senior bonds are included in the data sample.

The applied default-free interest rate for a 5-year risk free bond is the 5-year US Treasury rate, which is obtained from Bloomberg. This risk-free rate is subtracted from the estimated bond yield to obtain the reference entity’s credit spread.

3.3. Credit rating changes data

The corporate credit rating events (reviews and rating changes) are collected from the defaults and recovery database from Moody’s. From this database, the reference entity, the type of event, the rating date, the number of notches, and the initial rating class are obtained. Outlooks are excluded from the data sample as they represent a more medium-time period. This exclusion is consistent with Norden (2008) who suggests that outlooks only reflect the belief of individual rating analysts.

Since rating changes from the different rating agencies are typically correlated with each other, only the rating events from Moody’s are included. Furthermore, it is important to

(28)

findings. To control for contamination, all events that were preceded by another event within 10 days were excluded from the data sample, which is consistent with Micu et al. (2006).

The final data sample included 110 rating announcement (39 downgrades, 7 upgrades, 48 review for downgrades, and 16 review for upgrades).

(29)

4. Empirical Results

In this section, the results of the event studies, the multivariate cross-sectional regression and the accuracy analysis are reported and discussed. The first part regards the descriptive statistics of the event study, the second part presents the findings of the timeliness study (event study and multivariate regression) and the third part concerns the results of the accuracy analysis.

4.1. Descriptive Statistics

In accordance with the data description above, the final data sample in the event study consists of 110 rating events for the period between 1 January 2012 and 31 December 2016. The final data sample contains 39 downgrades, 48 reviews for downgrade, 7 upgrades, and 16 reviews for upgrade. An overview of the data sample is provided in table 2.

Downgrades Reviews for

Downgrade Upgrades Reviews for Upgrade Total Final Sample 39 48 7 16 110 Bond Rating AAA/Aa 2 1 0 0 3 A 15 17 0 4 36 Baa 11 20 3 10 44 Ba 7 4 3 1 15 B 2 4 1 1 8 Caa/lower 2 2 0 0 4 Size of Rating Change

One notch 27 n/a 6 n/a 33

Two or more 12 n/a 1 n/a 13

Expected Rating Changes

Expected by

review 16 n/a 1 n/a 17

Unexpected

by review 23 n/a 6 n/a 29

Table 2: Data sample characteristics

Table 3 illustrates the mean, median, minimum value, maximum value, range, standard deviation, skewness and the kurtosis for the excess spread of both the credit spread and the CDS spread within the event window around the announcement date [-2/+2].

The high skewness and kurtosis of downgrades and reviews for downgrade indicate that the distribution of the excess spreads are highly skewed and highly peaked around the

(30)

The large differences between the mean and median for these subsamples also indicate the existence of extreme values. The subsamples upgrades and reviews for upgrades are less skewed and lower peaked around the center. However, this could be due to the small sample sizes of these both subsamples.

The mean indicates whether the excess spread becomes larger or smaller on average for a specific rating event. A negative rating event (e.g. downgrade or review for downgrade) is expected to widen both the credit spread and the CDS spread. Consequently, a positive rating event (e.g. upgrade or review for upgrade) is expected to tighten both the credit spread and the CDS spread. The mean seems to comprehend these expected directions reasonably well. For both downgrades and reviews for downgrade the mean (and median) is positive, which indicates that both the credit spread and CDS spread are expanding. For upgrades and reviews for upgrade the mean is negative as expected, with an exception of the credit spreads for reviews for upgrade. However, this unexpected positive effect is very close to zero.

Event window

[-2/+2] Downgrades Reviews for Downgrade Upgrades Reviews for Upgrade Number of

observations 39 48 7 16

Credit

spreads spreads CDS spreads Credit spreads CDS spreads Credit spreads CDS spreads Credit spreads CDS

Mean 15.996 16.022 2.027 1.096 -1.784 -1.106 0.066 -0.447 Median 1.380 0.634 0.982 0.005 -0.458 -1.117 0.138 -0.260 Min -3.212 -5.694 -9.488 -6.023 -9.915 -2.909 -2.800 -3.303 Max 216.707 267.26 41.343 8.627 1.835 0.292 1.595 1.297 Range 219.919 273.320 50.831 14.650 11.750 3.202 4.395 4.600 Std. Dev 41.368 48.435 6.414 2.740 3.892 1.329 1.127 1.102 Skewness 3.738 4.407 4.983 1.072 -1.861 -0.455 -1.049 -1.036 Kurtosis 15.402 20.809 31.226 1.866 3.920 -1.499 1.580 2.115

(31)

4.2. Timeliness Results 4.2.1. Event Study Results

In this part the findings of the event studies are discussed. To investigate whether credit spreads and CDS spreads have a significant reaction prior to or after a credit rating event, and to test whether one market incorporates this credit information faster than the other, I have subdivided the event window in five smaller event windows. Three of these event windows are prior to the event ([-90/-61],[-60/-31] and [-30/-3]), one event window covers the days around the announcement date ([-2/+2]), and one event window covers the post-event period ([-3/-10]). These different event windows make it possible to classify the abnormal spreads into certain periods of time and to test whether one spread reacts earlier than the other.

Two different models are applied to estimate the abnormal spreads/returns within the different event windows. The first model is consistent with the methods and assumptions of Daniels & Jensen (2005). In this model, the abnormal excess spreads are calculated by using the constant mean model, where the mean spread is calculated over the estimation period prior to the event window. In this study, the estimation period contains 90 daily observations. For each event window, the cumulative abnormal (excess) spread is calculated by subtracting the mean spread from the actual observed spread. To test the robustness of the excess spread model’s results a second model is applied. This second model is consistent with Micu et al. (2006), and uses the cumulative abnormal returns to test the effect on the bond market and the CDS market. These abnormal returns are estimated by applying the market model.

4.2.1.1. Excess Spread Method Results

Table 4 reports the test results for all types of credit rating events using the cumulative abnormal excess spread model consistent with Daniels & Jensen (2005). Panel A represents the excess credit spread for all event windows and Panel B represents the excess CDS spreads.

(32)

Table 4: Event Study Analysis – Cumulative abnormal (excess) spreads using the constant mean model

This table reports the test-statistics based on the cumulative excess spread method for all downgrades, reviews for downgrade, upgrades and reviews for upgrade in the data sample and for all five different event windows. The events column states the number of events in the sample for each type of credit event. Cumulative excess spreads for each event window is calculated by subtracting the mean spread, which is computed over the estimation window covering 90 days prior the first event window, from the actual observed daily spread. Panel A shows the t-statistics for cumulative excess credit spreads and Panel B shows the t-statistics for the cumulative excess CDS spreads. T-statistics marked with (*) are significant at a 10%-level, (**) at a 5%-level, and (***) at a 1% significance level.

Downgrades

To analyze whether credit spreads and CDS spreads have a significant reaction to the announcement of a credit rating downgrade the test-statistics of the cumulative abnormal (excess) spreads for both markets should be investigated for all event windows. The test-statistics for the credit spreads indicate that the cumulative abnormal spread is positively significant at a 5% significance level for all five event windows. Furthermore, the cumulative abnormal CDS spread is also positively significant at a 5% level for the event windows [-90/-61],[-2/+2] and [+3/+10]. In the event windows [-60/-31] and [-30/-3] the increase in CDS spread is only significant at a 10% level.

These results are consistent with the findings of Daniels & Jensen (2005), who also report that credit rating downgrades have a significant impact on both the credit spread and the CDS spread in all event windows. The positive reaction (widening of the spreads) in the post-event event window [+3/+10) is in contrast with the results of Steiner & Heinke (2001), who find evidence that downgrades cause market overreaction. Due to this overreaction, the positive effect on the spreads (widening) prior to the event is followed by a negative effect (tightening) on the spreads directly after the downgrade. Norden & Weber (2004) support this finding and also observe a market overreaction on the CDS market which results in a decline in the cumulative abnormal spreads directly after a downgrade. Hull, Predescu & White (2004) find a significant positive effect on the

Event window [-90/-61] [-60/-31] [-30/-3] [-2/+2] [+3/+10] Events Panel A: Excess Credit Spreads

Downgrade 39 2.35** 1.98** 2.25** 2.41** 2.32** Review Downgrade 48 1.57 1.54 1.54 2.19** 2.02** Upgrade 7 -1.22 1.00 -1.60 -1.21 -1.53 Review Upgrade 16 -0.16 -0.68 -0.31 0.24 -0.91

Events Panel B: Excess CDS Spreads

Downgrade 39 2.04** 1.69* 1.86* 2.07** 2.17** Review Downgrade 48 1.19 1.61 1.13 2.77*** 2.82*** Upgrade 7 -1.04 -1.45 -2.12** -2.20** -1.99** Review Upgrade 16 -1.09 -1.07 -1.46 -1.62 -1.44

(33)

cumulative abnormal CDS spread in the event windows prior to the event, but do not report any evidence for market overreaction. On the other hand, Micu et al. (2006) reports a positive effect on the CDS market only in the event window after the downgrade.

The results of my study provide enough evidence to confirm both hypotheses H1a and H2a that a credit rating downgrade has a positive effect on both the credit spread and the CDS spread. My results indicate that this positive effect is significant in all event windows.

Reviews for Downgrade

The results for the subsample reviews for downgrade illustrate that the credit spread and the CDS spread reacts less early on reviews for downgrade than for actual rating downgrades. The test-statistics show that there is only a significant increase in both the credit spread and the CDS spread in the event windows [-2/+2] and [+3/+10]. This indicates that reviews for downgrade are not anticipated on both markets in the event windows prior to the event.

These findings are in line with Micu et al.(2006) who also reports significant increase in CDS spreads in the event windows around and after the announcement. They suggest that reviews for downgrade convey more new relevant information to the market than actual downgrades, and are thereby less anticipated by the market.

The results of my study confirm both hypotheses H1b and H2b. A review for downgrade has a positive effect on both the credit spread and the CDS spread. However, this effect is less anticipated than actual downgrades, since the positive effect is only significant in the event windows [-2/+2] and [+3/+10] for both spreads.

Upgrades

The results for the subsample upgrades indicate that there is no significant effect on the credit spread in any event window. This finding is consistent with prior research (Steiner & Heinke, 2001; Bannier & Hirsch, 2010; Finnerty et al., 2013). This absence of a significant market reaction to upgrades could be explained by the fact that upgrades are less monitored by credit investors than downgrades, and thereby less anticipated and

(34)

incorporated by the market. Daniels & Jensen (2005) also finds no significant decrease in credit spreads in any event window in the case of an upgrade.

The test-statistics of the excess credit spreads in the reaction of upgrades provide evidence for H3a. However, since the size of this subsample is very small, making inferences could be potentially inaccurate and unjustified.

The CDS spread shows a significant decrease during the event windows [-30/-3], [-2/+2] and [+3/+10] in the scenario of an upgrade. This result is in line with Micu et al. (2006) whose results also indicate that upgrades have a significant impact on CDS spreads. Daniels & Jensen (2005) also reports a significant decrease in CDS spreads in the event window direct after the rating announcement. The significant impact of upgrades on CDS spreads do not confirm hypothesis H4a.

Reviews for Upgrade

The test-statistics for the subsample reviews for upgrade suggest there is no significant impact on both the credit spreads and CDS spreads in the event of a review for upgrade. These results are in line with prior research (Steiner & Heinke, 2001; Bannier & Hirsch, 2010; Finnerty et al., 2013) and confirm hypotheses H3b and H4b.

Credit Spreads vs. CDS spreads

The results of the event study provide not enough evidence to conform hypothesis H5 that the CDS market responds earlier to rating events than the bond market. In the event of downgrades both markets show a significant increase in credit spread or CDS spread in all five different event windows. It should be noted that the significance for the credit spread is higher than for the CDS spread in the event windows [-60/-31] and [-30/-3], but these differences are too small to state that the corporate bond market reacts earlier than the CDS market.

In the event of reviews for downgrade and reviews for upgrade both markets present the same t-statistics for all the event windows. Only in the case of an upgrade there are significant differences in reactions observable between the two markets. Where the corporate bond market does not show a significant decrease in the credit spread in any

Is the credit default swap market more timely in predicting a credit rating event and more accurate in measuring credit risk than the corporate bond market?

Statement of Originality

Table of Contents

Introduction

1. Literature Review

2. Methodology

3. Data Description

4. Empirical Results