The Bond Market Effects of Reputational Shocks to Credit Rating Agencies Kirti Sinha* Kellogg School of Management, Northwestern University Linda Vincent Kellogg School of Management, Northwestern University March 2018 Abstract Credit rating agencies (CRAs) play a unique role in capital markets, subject to neither the discipline of a competitive market nor the incentives of an unregulated market. As a result, the expected effects of a reputational shock to CRAs’ credibility are difficult to predict. We examine whether investors decrease their reliance on credit ratings after two reputational shocks, the Enron and WorldCom bankruptcies in 2001-2 and the 2008 financial crisis. For new bond issues, we find that the association between ratings and the bond spread decreases after each reputational shock. For bond rating changes, we find statistically lower bond market reactions to downgrades and upgrades after each reputational shock compared to before the shock. Overall, our findings suggest that investors place less reliance on ratings after a CRA has been hit by a reputational shock. JEL classification: G12, G18, G24 Keywords: Credit rating agencies, bond spread, reputation * Corresponding author. Kellogg School of Management, Northwestern University, 2001 Sheridan Road Acknowledgements: We thank David Dranove, Michael Fishman, Benjamin Iverson, Nayab Khan, Nicola Persico, James Schummer, and Rajkamal Vasu for their helpful comments and suggestions. Additional thanks to all the faculty and PhD students at Kellogg School of Management, Northwestern University.
43
Embed
Abstract - Fox School of Business | Temple University...Kirti Sinha* Kellogg School of Management, Northwestern University Linda Vincent Kellogg School of Management, Northwestern
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Bond Market Effects of Reputational Shocks to Credit Rating Agencies
Kirti Sinha*
Kellogg School of Management, Northwestern University
Linda Vincent
Kellogg School of Management, Northwestern University
March 2018
Abstract
Credit rating agencies (CRAs) play a unique role in capital markets, subject to neither the discipline of a
competitive market nor the incentives of an unregulated market. As a result, the expected effects of a
reputational shock to CRAs’ credibility are difficult to predict. We examine whether investors decrease
their reliance on credit ratings after two reputational shocks, the Enron and WorldCom bankruptcies in
2001-2 and the 2008 financial crisis. For new bond issues, we find that the association between ratings
and the bond spread decreases after each reputational shock. For bond rating changes, we find statistically
lower bond market reactions to downgrades and upgrades after each reputational shock compared to
before the shock. Overall, our findings suggest that investors place less reliance on ratings after a CRA
has been hit by a reputational shock.
JEL classification: G12, G18, G24
Keywords: Credit rating agencies, bond spread, reputation
* Corresponding author. Kellogg School of Management, Northwestern University, 2001 Sheridan Road
Acknowledgements: We thank David Dranove, Michael Fishman, Benjamin Iverson, Nayab Khan, Nicola
Persico, James Schummer, and Rajkamal Vasu for their helpful comments and suggestions. Additional
thanks to all the faculty and PhD students at Kellogg School of Management, Northwestern University.
1 INTRODUCTION
Credit rating agencies (CRAs) play a unique role in capital markets. If other information
intermediaries such as sell side analysts, investment advisors, mutual funds, or asset managers provide
low quality or inaccurate recommendations, or make inappropriate investment decisions, investors may
terminate the relationship with the person or institution and substitute another. Loss of business thus
serves as a disciplining mechanism for these information intermediaries and the threat of the loss of
income provides incentives for credible performance. Unlike other information intermediaries, if a CRA
provides inappropriate ratings, investors do not have the option to terminate the relationship with the
CRA or to stop paying for its services. Furthermore, the CRAs are virtually assured of future business
and corresponding income by the governmental requirements for debt ratings under many circumstances
and the entrenched market expectation for debt ratings for virtually all debt issues. Under these
circumstances, what disciplines CRAs to provide quality, accurate bond ratings and what ramifications, if
any, are there in the face of inappropriate ratings?
The important role of credit rating agencies (CRAs) in the capital markets is well-established in
both academic (e.g., Holthausen and Leftwich (1986), Hand et al. (1992), and Dichev and Piotroski
(2001)) and practitioner literature ( e.g., Partnoy (2009)). The incentives faced by the CRAs and the
related reliability of credit ratings have been the subjects of significant debate for many reasons. One of
the main concerns is that the issuer generally pays the CRAs for the rating resulting in the potential for
conflict of interest. After the establishment of the NRSRO requirement in 1975, the debt ratings market
became a government sanctioned oligopoly, potentially precluding incentive for improving performance.
Two recent events increased scrutiny of the CRAs: the Enron/WorldCom bankruptcies in
2001/2002 and the 2008 financial crisis. Both Enron and WorldCom had investment grade debt ratings
until just prior to their respective bankruptcy filings. The structured finance instruments, including
mortgage backed securities and collateralized debt obligations, that arguably contributed to the financial
crisis of 2008-2009, were frequently rated not only investment grade, but given the highest AAA rating.
2
These two events drew attention not just to the conflict of interest but also to the potentially “excessive”
reliance of investors on ratings.1
Academic research in the aftermath of these two events has focused primarily on how the
conflicts of interest inherent in the CRA’s issuer-pay model and the fallacies of the efficacy of existing
regulatory mechanisms contributed to these events (e.g., Benmelech and Dlugosz (2010), Efing and Hau
(2015), Bolton et al. (2012)). Other studies examine how regulatory changes after these events have
affected the quality of credit ratings (e.g., Jorion et al. (2005), Cheng and Neamtiu (2009), Dimitrov et al.
(2015)). There is little evidence on whether CRA’s loss of reputation as a result of these events has
affected investors’ reliance on credit ratings. This is the research question that we explore in this paper.
Specifically, we study whether reputational shocks to CRAs alter the association between
ratings and returns in the bond markets. Because there are arguments on both sides of this question, it
becomes an empirical issue to resolve. One hypothesis is that investors decrease their reliance on credit
ratings due to their perception that CRAs failed to exercise sufficient care and professional judgment in
developing ratings for the securities at issue and/or inflated their ratings in order to gain business under
the issuer-pay model. The then existing regulatory framework did not hold CRAs accountable for their
actions; that is, there were no regulatory costs imposed for ratings subsequently found to be inappropriate
and/or inaccurate. At the same time, regulations effectively required all bond market securities to be
rated by NRSRO-designated CRAs so there were no obvious reputational costs given there were only
three NRSROs during much of that period.2 This lack of reputational and regulatory costs raises the
question of whether CRAs have any incentives to modify their behavior after what would likely be
perceived as a reputational shock. If investors believe that CRAs will not improve on their rating
1 Bolton et al. (2012) discuss that the combination of CRA reliance on fees from issuers, investors who were too
trusting and issuers looking to benefit from mispricing of their issues could have led to substantial rating inflation
that contributed to the 2008 financial crisis. 2 As Hunt (2009) notes: “The dominant view of rating quality in the legal literature and among policymakers comes
from the “reputational capital” model of rating agencies, which holds that a well-functioning reputation mechanism
will give rating agencies optimum incentives for producing high quality ratings. The underlying idea is that if
investors determine that CRAs’ ratings are of low quality, they will stop crediting the ratings and the agency’s
business will lose value. At the same time, it has been recognized that real world characteristics of the rating market
may cause the reputation mechanism not to function well.” (p 113)
3
practices after such a shock, their only recourse might be to place less weight on the ratings relative to the
weight placed prior to the shock and relative to other publicly available information on the credit
instruments. In support of this hypothesis, Sethuraman (2016) finds that after a reputational shock bond
issuers increase their voluntary disclosures as investors start relying on information sources other than
credit ratings. On the other side of the argument, if investors believe that CRAs conducted unbiased
investigations and exercised appropriate due diligence and professional judgement in developing their
ratings prior to the reputational shock, then investors might continue to view the ratings with the same
confidence as before the shock. Relatedly, investors might believe in the disciplining force of the
reputational shock resulting in higher quality ratings and continue to view the ratings with the same or
increased confidence. Our null hypothesis is that the reputational shock does not affect investors’ reliance
on credit ratings.
To explore this hypothesis, we focus on the two reputational shocks noted above.3 Enron filed for
bankruptcy in November 2001 and WorldCom filed for bankruptcy in July 2002. Rating agencies were
widely criticized following these events because Enron’s bonds were rated investment grade four days
prior to the bankruptcy filing and WorldCom’s bonds were investment grade three months before
WorldCom filed for bankruptcy. These two bankruptcies constitute our “Reputational Shock 1” and the
period of the shock is November 2001 to July 2002. “Reputational Shock 2” is the period of the financial
crisis between September 2008 and August 2009.
We analyze individual issue ratings (bond-level approach) instead of issuer ratings (firm-level
approach) for corporate bonds. For the two reputational shocks, we analyze both the ratings at-issue and
subsequent rating changes to assess investor reliance on credit ratings. In addition to analyzing the total
sample, we also partition the bonds by investment grade (IG) and non-investment grade (NIG) or junk
3 One might question why there would be a second reputational shock if investors had already decreased their
reliance on ratings after the first shock. This brings us to an important issue. The goals of the regulatory reforms
such as the Credit Rating Agency Reform Act of 2006 (CRARA 2006) after the Enron/WorldCom scandals, were to
increase the quality of the ratings and investor confidence in them. Sethuraman (2016) uses the introduction of
CRARA 2006 as an event that would restore CRAs reputation in the financial market. But the 2008 financial crisis
provided evidence on shortcomings of CRARA 2006. Even though it increased investors’ confidence in ratings, it
could not enhance rating quality or discipline rating agencies through enhanced competition.
4
based on evidence markets’ perceptions differ between the two categories. A common criticism of the
Enron and WorldCom ratings was not just that they were too high but rather that they were investment
grade. Benmelech and Dlugosz (2010) find evidence that the 2008 financial crisis was partly caused by
IG designated securities. Based on conversations with fixed income traders at Nomura Securities in New
York, financial market analysts consider ratings more important for IG bonds than for NIG bonds.
Therefore, we expect to find that investors’ reliance on ratings for IG bonds will decrease more than their
reliance on ratings for NIG bonds.
For ratings at-issue, we examine whether the association between credit ratings and the bond
spread over the appropriate treasury (matched on maturity) differs for bonds issued after each shock
compared to bonds issued prior to each shock. Any deviation in the association between actual spread
and ratings is consistent with a change in investor reliance on the ratings.
We find that, for both reputational shocks, the association between credit ratings and bond interest
rate spread at issue decreases significantly after the respective reputational shock compared to the
association before the shock. As hypothesized, this change is more significant for IG bonds than for NIG
bonds. These results are consistent with investors decreasing their reliance on ratings after CRAs suffer a
reputational shock. While this result is consistent with our hypothesis, we recognize that there are other
influences on bond spreads, including levels and changes in macroeconomic conditions. During
economic expansions (recessions), investors become more (less) trusting of CRAs and the bond spreads
tend to be lower (higher) than in “normal” periods. In such periods, the correlation between bond ratings
and default rates, changes, suggesting that other factors such as recovery rates and risk premia also affect
the movement in spreads (Chen (2010)). While our specifications of the reputation shock windows are
chosen to minimize the effects of such economic periods (particularly for the 2008 financial crisis, for
which we remove the entire period of September 2008 to August 2009 because macroeconomic
conditions might have had an impact on the bond spreads), the results could still be affected by risk
premia. To confirm that the results are not driven by macroeconomic factors, we also examine market
reactions to credit rating changes.
5
Our alternate hypothesis is that investors rely less on credit ratings during the period after each
shock compared to the period before. To measure market reaction to credit rating changes, we employ
standard event study methodology. We compute abnormal bond returns around rating change
announcements in periods before and after each shock. Based on our predictions of a lower level of
reliance on credit ratings, we expect rating changes, both downgrades and upgrades, to result in lower
magnitude abnormal returns, on average, after a CRA has experienced a reputational loss. We find results
generally consistent with our hypotheses for both downgrades and upgrades.
We contribute to the accounting and finance literature in at least two ways. First, with respect to the
literature on the information content of credit ratings, we provide evidence that investors’ reliance on
credit ratings changes as a result of a reputational shock to the CRAs. Second, we add to the literature on
the effectiveness of the posited reputational mechanism in the context of CRAs. Decreasing investor
reliance on ratings is one of the biggest motives of regulatory authorities after the 2008 financial crisis
(Partnoy (2009)).
Our work is related to four other papers. deHaan (2016) shows that rating performance improves
after the financial crisis. He argues that this result is consistent with the rating agencies positively
responding to public criticism and regulatory pressures. Using loan-level data, he also shows that debt
participants reduce their reliance on credit ratings after the financial crisis. Jaballah (2015) also studies the
impact of the 2008 financial crisis on the reputation of CRAs by measuring the stock market reactions to
changes in credit ratings before and during the crisis. He documents significantly negative stock market
reactions for downgrades before the crisis and less significant reactions after the crisis. Bedendo et al.
(2013) analyze the credit default swaps (CDS) market immediately following the 2008 financial crisis and
conclude that corporate credit ratings are viewed as less credible during the crisis compared to the period
immediately preceding the crisis. However, the same authors, in a recent working paper, Bedendo et al.
(2016), find contrasting results when they look at the stock market reactions to issuer rating
announcements after a reputational shock to CRAs. They argue that the latter results are consistent with
6
the scenario where investors believe that rating agencies will self-discipline and rebuild their reputation
by increasing rating quality.
Our paper differs from these three papers in several respects. First, we focus on bond markets
rather than equity or private debt markets, and specifically on individual bond ratings instead of issuer
ratings in order to increase the power of our results. Second, we provide evidence that investors discount
ratings after a reputation shock, which contrasts with the result in Bedendo et al. (2016). Third, in
addition to studying the impact of ratings on bond spreads at the time of new bond issues, we find
corroborating results by analyzing investor reactions to rating change announcements for bonds. Fourth,
we provide consistent results for two reputational shocks, questioning the regulatory frameworks ability
to discipline CRAs and their goal of decreased investor reliance through regulatory changes.
This paper proceeds as follows. Section 2 describes the institutional background, discusses related
research, and develops the hypotheses. Section 3 describes the methodology and data and Section 4
provides the main results and robustness checks. Section 5 summarizes and concludes.
2 BACKGROUND, RELATED LITERATURE, AND HYPOTHESES DEVELOPMENT
2.1 Institutional Background
Credit ratings gained importance in capital markets after the great depression of the 1930s. With
the establishment of the Securities and Exchange Commission (SEC) in 1934, certain regulated industries
were permitted to invest only in bonds having satisfactory credit ratings. At that time, credit rating
agencies (CRAs) followed an investor-pay model; that is, investors paid fees for ratings provided by the
CRAs.
Two important changes occurred in the 1970s. In 1975, the SEC created the Nationally
Recognized Statistical Rating Organization (NRSRO) designation for CRAs. Pension funds and money
market mutual funds, for example, can invest only in NRSRO rated investment grade (IG) bonds.
NRSRO credit ratings became part of the regulator’s determination of the reserves required to be held by
7
banks and insurance companies. An IG rating is required for an SEC Rule 415 shelf registration. Credit
ratings became widely used for commercial purposes such as long term leases and requirements for letters
of credit. In other words, NRSRO credit ratings are well-entrenched in the operation of capital markets.
Secondly, most CRAs changed in the 1970s from an investor-pay model to an issuer-pay model
following the invention of the photocopy machine, citing loss of revenue due to photocopying of rating
reports by investors as one of the main reasons for the change. This latter change created a potential
conflict of interest because the largest source of income for CRAs is rating fees. CRAs needed to win
business from other CRAs in order to gain revenue, potentially leading to biased ratings. An issuer can
approach multiple rating agencies to get a specific issue rated prior to issue and choose one or more CRAs
based on a comparison of the ratings, an action known as “rating shopping.” Since the rating agencies
follow an issuer-pay model and issuers can engage in “rating shopping”, the rating agencies have
incentives to inflate ratings at the cost of investors.4
At the time of the Enron and WorldCom bankruptcies in 2001/2002, the SEC had granted NRSRO
status to only S&P, Moody’s and Fitch. There was a lack of transparency in the NRSRO certification
process by the SEC and the process served as an effective barrier to entry. The “Sarbanes-Oxley Act
(SOX)”, passed by the U S Congress in July 2002, requires the SEC to provide details on the
determination procedure for designating NRSROs. The SEC’s January 2003 report, as required by SOX,
resulted in extensive Congressional hearings on the NRSRO process culminating in the Credit Rating
Agency Reform Act of 2006 (CRARA)” (Cheng and Neamtiu (2009)). CRARA established the criteria for
NRSRO certification and imposed a strict timetable on the SEC for granting NRSRO recognition with the
goal of increasing competition among the CRAs so they would have incentives to put more effort into the
ratings generating process and thus decrease the conflicts of interest inherent in the issuer-pay model.5
Prior to CRARA, there were seven NRSROs and three more CRAs were subsequently designated.6
4 He et al. (2015) study rating shopping on the MBS market. They show that single rated tranches have been
“shopped”, and pessimistic ratings never reach the market. 5 See Hunt (2009). Applicants for registration must provide credit ratings performance measurement statistics over
short-term, mid-term, and long-term periods, 15 U.S.C. 78o-7(a)(1)(B)(i), describe the procedures and
8
In addition to passage of SOX after the Enron bankruptcy, the SEC adopted Regulation Fair
Disclosure (Reg FD) in 2002 to prevent selective disclosure of information by publicly traded firms to
analysts and institutional investors. However, Rule 100(b)(2)(iii) of Reg FD provided an exemption for
disclosures made to CRAs which meant that issuers could continue to disclose private information to
CRAs.
The claimed goal of the regulatory changes (SOX and CRARA) was to improve investor
confidence in ratings provided by the CRAs (Sethuraman, 2016). But the 2008 financial crisis suggests
that these new regulatory provisions were inadequate and did not solve quality and conflict of interest
problems associated with CRAs.
In response to the 2008 financial crisis, Congress passed the “Dodd-Frank Act” in 2010, which
included two important provisions relating to CRAs. First, Dodd-Frank Act removed the exemption of
CRAs from the provisions of Reg FD, effective on October 4, 2010. However, Reg FD applies to
“covered persons” and at the time of the introduction of Reg FD, CRAs were registered as investment
advisors and thus qualified as covered persons but were specifically exempted. 7 But the CRARA of 2006
amended Section 2(a)(11)(F) of the Investment Advisors Act of 1940 so that NRSROs were specifically
excluded from the definition of “investment advisor”. Thus, none of the NRSROs are registered as
investment advisors, implying that the Reg FD amendment by the Dodd-Frank Act had no effect on
NRSROs. Interestingly, if a credit rating agency is not an NRSRO it can remain exempt as before because
Rule 100(b)(2)(ii) allows companies to enter a confidentiality agreement with the CRA.
methodologies that the applicant uses in determining credit ratings, 15 U.S.C. 78o-7(a)(1)(B)(ii), and provide
certifications from at least 10 unaffiliated qualified institutional buyers, with each certification indicating that the
buyer has used the credit ratings of the applicant for at least the 3 years immediately preceding the date of the
certification. Applications that contain the prescribed information are to be granted unless the Commission
determines that (a) the applicant does not have adequate and managerial resources to consistently produce credit
ratings with integrity and to materially comply with the rating procedures it says it follows or (b) the applicant or
person controlling the applicant has been convicted of a felony or has been punished for committing certain
securities violations. 15 U.S.C. 78o-7(d) 6 The list of NRSROs is available at http://www.sec.gov/ocr/ocr-current-nrsros.html 7 Under Rule 100(b)(1) of Reg FD, the list of covered persons includes:
• Broker-dealers and their associated persons,
• Investment companies, hedge funds and their affiliated persons, or
• Any security holder or person for whom it is reasonably foreseeable that such person would buy or sell
Second, the Dodd-Frank Act repealed Rule 436(g) of the Securities Act of 1933. Rule 436(g)
exempted rating agencies from “expert” liability when they issue credit ratings. However, The Asset-
Backed Market Stabilization Act of 2011 (H.R. 1539), passed six months after Dodd-Frank, restored Rule
436(g). H.R. 1539 came in six months after the passage of Dodd-Frank Act, which implies that the
provisions of Dodd-Frank, related to lawsuits against CRAs, were no longer valid after 6 months.
In summary, the Dodd-Frank Act 2010 did not have any provisions that would arguably result in
greater reliability of credit ratings after the financial crisis.
2.2 Prior Research
Much of the earlier academic research focuses on the information content of credit ratings.
Holthausen and Leftwich (1986), Hand et al. (1992), and Dichev and Piotroski (2001) all find that equity
investors react to bond rating announcements and the reaction is greater in magnitude for rating
downgrades than for upgrades. Kao and Wu (1990) and May (2010) provide evidence that both ratings
levels and changes, respectively, have predictive ability for subsequent firm performance and credit risk.
Some recent research examines whether credit ratings have decreased in relevance to capital
markets over time. Chava et al. (2012) show that CDS spreads have greater explanatory power for the
cross sectional variation in bond yields in both the primary and secondary bond markets than do credit
ratings. On the other hand, Benmelech and Dlugosz (2010) find that rating inflation, due to rating
shopping, was one of the major drivers of the 2008 financial crisis. Relatedly, Efing and Hau (2015)
provide evidence that CRAs provide higher ratings for issuers that provide the CRAs with more bilateral
securitization business. The authors document that the size of the rating favors is positively related to the
complexity of the structured debt deals and to the activity level in the credit market in terms of bond
issuances.
Within the extensive conflict of interest literature, we focus on two streams of research most
relevant for our study. The first analyzes whether increased regulation has succeeded in improving the
quality of bond ratings by CRAs. Jorion et al. (2005) find that the information content of both credit
rating downgrades and upgrades is greater following the passage of Reg FD (2000). Similarly, Cheng and
10
Neamtiu (2009) find that CRAs issue more timely downgrades and more accurate ratings following the
passage of SOX in 2002. These two papers suggest that the regulatory mechanisms have been successful
but Dimitrov et al. (2015) finds no evidence that the Dodd-Frank Act (2010) disciplines CRAs to provide
more accurate ratings. Instead, they find that CRAs issue lower ratings, give more false warnings, and
issue more but less informative downgrades after Dodd-Frank. They suggest that these results are more
consistent with the reputation model of Morris (2001), according to which CRAs became more concerned
about their reputation following Dodd-Frank because of the new provision that made the rating agencies
liable for the ratings they provided and subject to lawsuits for damages.8
The second strand of literature relates to conflicts of interest and the reputational capital of CRAs.
Theoretical literature in this area focuses on whether reputational concerns, in equilibrium, result in truth-
telling by the rating agencies. Mathis et al. (2009) finds that reputation cycles may exist during which a
CRA first builds up its reputation by relaying information accurately but subsequently exploits this
reputation by collecting fees for inflated ratings. The authors demonstrate that truth telling incentives are
weaker when CRAs are rating complex debt products such as structured securities (e.g., asset backed
securities (ABS) and mortgage backed securities (MBS)). Bolton et al. (2012) model an equilibrium in
which CRAs inflate credit ratings with both endogenous and exogenous reputation costs, suggesting that
reputational concerns are not strong enough or sufficient to discipline CRAs. In the microeconomics
literature, an information intermediary is modeled as engaging in acquiring and certifying information and
committing to it through disclosures. Also, the reputational costs are usually modeled as no business in
the later periods if the intermediary lies in the first period. But in the case of CRAs, these models do not
apply, as documented by Mathis et al. (2009), and Bolton et al. (2012). The results of these two papers tie
well with the institutional setting in which CRAs work. The requirement of being an NRSRO to be able to
8 Dimitrov et al. (2015) refers to the repeal of 436(g) by the Dodd-Frank Act, which made credit rating agencies
liable for increased lawsuit exposure. But, this result seems rather puzzling particularly because, as mentioned
earlier, the Asset-Backed Market Stabilization Act of 2011 (H.R. 1539), restored rule 436(g) of The Securities Act
of 1933, which exempted rating agencies from “expert” liability when they issue credit ratings. H.R. 1539 came in
six months after the passage of Dodd-Frank Act, implying that the provisions of Dodd-Frank, related to lawsuits
against CRAs, were no longer valid after 6 months
11
rate securities and the mandatory requirement of getting a bond rated by a CRA, make it difficult for
reputational concerns to work as a disciplining device, unlike traditional microeconomics reputation
models.
This brings us to our main research question: in the absence of disclosure rules, legal liability, and
rights to terminate (due to regulatory requirements to get the bonds rated by an NRSRO), do investors
decrease reliance on ratings after a CRA has been hit by a reputational shock?
We identify three related papers that explore empirically the effectiveness of the reputational
effects and regulatory mechanisms discussed above; however, the results are mixed. Jaballah (2015)
studies the impact of the Subprime crisis on the reputation of CRAs by measuring the stock market
reactions to changes in credit ratings before and during the crisis. Using data for European and American
stock markets for the period of 2005-2009, he documents negative and statistically significant stock
market reaction to rating downgrades before the crisis. However, during the crisis, he only finds negative
and significant reaction for the European stock markets. He argues that these results suggest that U.S.
market participants ignored rating changes during the crisis, suggesting that they found the ratings
unreliable. Bedendo et al. (2013) analyze the credit default swaps (CDS) of 205 issuers immediately
following the 2008 financial crisis and find that corporate credit ratings became less credible following
the 2008 crisis as reflected in the diminished price impact of ratings changes compared to before the
crisis. However, the same authors, in a more recent working paper, Bedendo et al. (2016), find
contrasting results with respect to the stock market reactions to issuer rating announcements following
three blows to CRAs’ reputation: the Enron/WorldCom bankruptcies; the mass structured product
downgrade by Moody’s in 2007; and the federal government’s lawsuit against S&P in 2013. They find a
stronger response by equity investors to ratings downgrades following these three reputational shocks
compared to before. The authors explain these results as consistent with investors’ beliefs that the CRAs
choose to rebuild their reputation by increasing rating quality.
Our paper differs from these three papers in several respects. First, we focus on corporate bond
markets and specifically, on individual corporate bond ratings. In the United States, the corporate bond
12
market comprises 21% of the $41 trillion total bonds outstanding as of September, 2016.9 CRAs have
played an important role in bond markets since at least 1909 with the founding of Moody’s. Although the
importance of ratings to equity investors is well-documented, the ratings effect is indirectly manifested in
the firm’s cost of debt, a component of the cost of capital used in equity valuation. Thus we focus on the
direct information content of credit ratings for individual bonds. Second, we provide evidence that the
investors discount the credibility of ratings after a reputation shock in contrast to Bedendo et al. (2016).
Our results from new bond issues as well as for rating changes are consistent with investors discounting
ratings due to the reputational loss suffered by CRAs. Third, in addition to bond rating change
announcements, we also look at how investors perceive ratings at the time of new bond issues.
2.3 Hypothesis Development
Under the assumptions that CRAs incorporate private information in determining their ratings and
that investors rely on these ratings, a reputation loss for the CRAs would adversely affect investors’
beliefs about the quality of the credit ratings, and in turn, about the credit quality of the bond issue.10 We
thus predict that investors reduce their reliance on a credit rating after a reputation shock to CRAs. We
analyze new bond issues as well as bond rating changes to develop specific hypotheses:
2.3.1 Bond Ratings and Bond Spreads at Issue
Before we discuss our hypotheses, we first note the relationship between bond spreads and bond
ratings. The bond spread is the difference between two bonds with the same maturity but different credit
ratings; often the difference between a corporate bond and a risk-free treasury bond. Bond spreads
depend on several risk factors, including credit risk, prepayment risk, liquidity risk, legal risk, maturity
risk, and complexity risk.11 Credit risk refers to a bond’s inability to repay all its principal and interest on
time as promised. Credit ratings typically address this loss risk. Loss risk incorporates the probability of
9 http://www.sifma.org/research/statistics.aspx US Bond Market Issuance and Outstanding (xls) - annual, quarterly,
or monthly issuance to December 2016 (issuance) and from 1980 to 2016 Q3 (outstanding) 10 Zeibar and Reiter (1992) shows that bond ratings affect bond yields (i.e., investors consider bond ratings when
Fixed income managers primarily use two metrics to evaluate bonds – bond yield and bond spread.
Bond spread is the yield of a corporate bond adjusted for the yield of a treasury bond with the same
maturity. We focus on bond spread in studying the effect of ratings on the default risk of bonds,
consistent with prior literature.
Ratings can be either for the issuer or for the specific bond issue. An issuer rating is based on the
assessed creditworthiness of the borrower’s overall financial condition whereas an issue rating is based on
the assessed probability of default and expected loss given default for the given instrument, as discussed
above. We examine issue ratings for corporate bonds, not for the issuer, because the greater number and
greater variance across issues increase the power of the tests. In addition, although bond issue ratings are
“sticky”, issuer ratings are even stickier.16 In addition, the sample has almost as many distinct issuers as
distinct issues; that is, most of the bonds in our sample are being issued by different issuers.17
Ratings are broadly divided into “Investment Grade” (IG), (i.e., bonds rated BBB- and above by
S&P and Fitch and Baa3 and above by Moody’s), and “Speculative Grade”/“Non-Investment
Grade”/”junk” (NIG) for all other bonds. Institutional investors such as pension funds are prohibited from
investing in NIG grade bonds and a downgrade from IG to NIG requires them to sell the affected bonds.
Therefore, we analyze the entire sample, the IG sample, and the NIG sample separately.18
16 Moody’s Rating Symbols and Definitions (December 2016, p42): “KRAs (Key Rating Assumptions) are, by their
nature, relatively stable inputs to the analytical process, and because they seek to bring a degree of stability,
consistency and transparency to something that may in practice be uncertain, they are intended to be reasonably
resilient to change. They may change over time in response to long-term structural changes or as more is learned
about long-run relationships between risk factors, but they would be very unlikely to change as a result of a short-
run change in economic or financial market conditions.” 17 Maul and Schiereck (2016) discuss the advantages and disadvantages of using bond-level and firm-level data.
Bond-level ratings provide a larger number of observations, thus increasing the power of the tests. The choice of
issues rather than issuer could result in clustering of observations by issuer in large samples, but because we study
the 12-month period before and after each shock, firm clustering is not an issue as our sample has almost as many
distinct issuers as distinct issues. As a precaution against the potential effects of firm-level clustering, we rerun our
tests with clustering standard errors by firm; our results remain unchanged. 18 It is important to distinguish between an issuer-paid rating agency (e.g. S&P, Moody’s, Fitch) and investor-paid
or a subscriber-paid rating agency (e.g. Egan-Jones Rating Company). Companies provide material non-public
information to rating agencies that are issuer paid. In other words, issuer-paid rating agencies claim to have an
advantage over subscriber-paid rating agencies because their ratings are based on private information rather than
only on publicly available information. Therefore, in our study, we focus on issuer-paid ratings, specifically S&P
Ratings. Another important fact in this regard, as discussed above, is that issuer-paid rating NRSROs still have an
17
To test our first hypothesis, we estimate the following regression model:
Log(Bond Spreadi) = β
0+ β
1*Rating
i+ β
2*Shock + β
3*Rating
i×Shock + Controlsi+ εi (3)
Where:
Log(Bond Spread) = the natural logarithm of the bond spread of the new bond issue
Rating = the initial rating of the bond at the time of issue
Shock = an indicator variable taking a value of 1 if the observation belongs to
the Post-Period and 0 otherwise
Controls = include issue specific risks, other than credit risk, that determine the
spread such as maturity, size of the issue, callable, put-able, and
sinking fund.
If investors change their reliance on ratings after a reputational shock, then we expect the coefficient 𝛽3 to
be significant, indicating a deviation between ratings-based spreads and actual spreads.
For our next two hypotheses, we use event study methodology to estimate market reaction to
bond rating changes. Consistent with prior literature, we measure investors’ reliance on credit ratings as
the bond price reaction to a rating change announcement. To study whether investors discount
information in bond rating changes, we compare bond rating reactions in the Pre-Period to those in the
Post-Period for both reputational shocks.
To calculate abnormal bond returns for the event studies, we first estimate normal bond returns.
There are several alternatives for specifying bond event studies.19 We use daily bond returns and form
matching portfolios with treasury securities, consistent with Bessembinder et al. (2009).20 Treasury
securities yield curves are available for seven major maturity categories (1, 2, 3, 5, 7, 10, 20, and 30
access to all non-material public information even after Dodd-Frank Act 2010 (Section 939B) as they do not belong
to the category of “covered persons”.
19 Bessembinder et al. (2009) and Maul and Schiereck (2016) review bond event studies used in the literature.
Bessembinder et al. (2009) analyze various event study methods and provide recommendations on which ones are
more appropriate in terms of power of the test. They find that the use of the daily bond returns significantly
increases the power of the test. The authors provide evidence that calculating a bonds’ excess return against a
matched firm’s or matched portfolio’s bond return is superior to using mean adjusted returns. 20 https://fred.stlouisfed.org/tags/series?t=bofaml
years) on the St. Louis Fed website. We interpolate the yield curve whenever a bond maturity does not
match the maturity in the category.
The corporate bond return for the event study is calculated as follows:
RETt = Pt - Pt-1
Pt-1
(4)
The subscript t denotes the Post-Period and t-1 denotes the Pre-Period. The price relative is thus
measured as the first trading price subsequent to the ratings change less the last trading price prior to the
ratings change announcement (i.e., the event window). These prices exclude accrued interest and are
referred to “clean prices.”
The abnormal return (ABRET) for bond i is calculated as the difference between the observed
return (RET) and the expected return (E(RET)). The expected return is computed as the matched (on
maturity) treasury bond return over the event window.
ABRETi = RETi - E(RETi ) (5)
We also estimate the multivariate regression of ABRET on our indicators of shock and on the set of
controls identified in prior literature as significant determinants of abnormal returns (e.g. Holthausen and
Leftwich (1986), Hand et al. (1992), Dichev and Piotroski (2001), and May (2010)):
ABRETi = β0 + β
1Shock + Controlsi + εi (6)
Shock is an indicator variable with a value of 1 if the observation belongs to the Post-Period and 0
otherwise. Controls here include the change in level of ratings (notches) and an indicator variable that
denotes whether the downgrade or upgrade was from IG to NIG or from NIG to IG respectively.
3.2 Data
During the period of our study, there are three major NRSRO designated CRAs – S&P, Moody’s,
and Fitch with either S&P and Moody’s, and often both, rating most of the corporate bond issues.
Although we execute our analyses using both S&P and Moody’s ratings separately, we report only S&P
results in the interests of economy unless the results for Moody’s differ from those for S&P.
19
We use data from several sources: the National Association of Insurance Commissioners (NAIC)
database; the Trade Reporting and Compliance Engine (TRACE) database; and Mergent’s Fixed
Investment Securities Database (FISD).
3.2.1 Bond Spreads and Initial Ratings
Data on bond ratings was gathered from the Mergent FISD database. The bond ratings dataset
consists of issue details for over 150,000 issuers, US Agencies and US treasury debt securities. The raw
dataset includes 2,614,166 rating changes and initial ratings issued for bonds between 1950 and 2015. We
delete observations whenever the rating is not between AAA and D.21 Table 1 provides the mapping from
the CRAs ratings to the cardinal scaled used in our analyses. For our analysis, we report results using
S&P ratings because most of the issues are rated by both the rating agencies and there is no significant
variation. There are 45,163 observations of new issues of corporate fixed coupon bonds rated by S&P for
the sample period between 1950 to 2015.
We further partition this sample for our two reputational shocks. Table 3 provides summary
statistics for these two samples. Reputational Shock 1 contains 869 bond issues in the Pre-Period and
671 bond issues in the Post-Period. Reputational Shock 2 contains 257 bond issues in the Pre-Period and
750 bond issues in the Post-Period.
3.2.2 Bond Market Reaction and Rating Changes
For the event studies, we require bond transaction data for both the Pre-Period and Post-Period for
both reputational shocks. The preferred database for bond transactions data is the Trade Reporting and
Compliance Engine (TRACE) but it is only available from 2002, whereas Reputational Shock 1 requires
data for the 12-month period before November 2001 and 12-month period after July 2002. 22 To execute
21 We follow prior literature and map the rating codes to the cardinal scale (Table 1). Moody’s uses code from Aaa
down to C to rate bonds whereas Fitch and S&P rate bonds from AAA down to D. We transformed the credit ratings
for S&P, Moody’s and Fitch into a cardinal scale starting with 1 as AAA/Aaa, 2 as AA+/Aa1, and so on until 23
(DDD/DD/D) as the default category. Following Jorion et al. (2005), we chose 23 instead of 22 for the default
category because Fitch provides three ratings (DDD/DD/D) for default, so 23 represents the average of the default
DD rating. 22 TRACE database contains price, time and size of transactions for all publicly traded over the counter (OTC)
corporate bonds. Even though TRACE was introduced in July 2002, it started covering all publicly traded bonds
20
the event study for Reputational Shock 1, we combine the TRACE dataset with National Association of
Insurance Commissioners (NAIC) to create a panel dataset of bond trades from 1994 to 2015. NAIC
contains bond transactions for insurance companies starting in 1994. 23
The NAIC and TRACE databases contain bond transaction data for corporate bonds. The TRACE
dataset is more comprehensive and covers transactions of all publicly traded corporate bonds, beginning
in July 2002. For bonds traded multiple times on a given day, TRACE covers all transactions with
individual time stamps. We calculate closing bond price for each bond on a specific transaction date using
the last trade price approach. That is, we extract the last traded price of the day for each bond on a
transaction date. To create unique bond-day transactions data, we create a trade weighted price using
volume for the trades whenever there are multiple trades on the same time stamp.
Consistent with prior literature, we consolidate the two databases to create a long span transaction
data sample (Lin et al. (2011)). We keep transaction records reported by TRACE only if transactions of
same bond are included in both NAIC and TRACE databases after July 2002. Our final sample includes
corporate bond transactions from January 1994 to December 2015. We combine CUSIP and trade data
from NAIC and TRACE datasets to get unique bond-date combinations. We then add price data from
NAIC and TRACE to this dataset.
While we report results only for S&P ratings, Table 2 provides ratings upgrades and downgrades
by all rating agencies for corporate bonds. We have a total of 117,631 rating changes for S&P, 117,002
rating changes for Moody’s, and 55,967 rating changes for Fitch. As mentioned earlier, Moody’s and
S&P are the two primary players in the corporate bond markets, with more than twice the rating changes
as Fitch. Focusing on S&P ratings, we see that number of downgrades increase substantially after both
reputational shocks (highlighted in grey).
only from October 2004 onwards. Currently, TRACE covers 100 percent of OTC activity representing 99 percent of
total US corporate bonds market activity in over 30,000 securities. 23 NAIC consists of all transactions of publicly traded corporate bonds beginning January 1994 by life insurance
companies, property and casualty insurance companies, and health maintenance organizations (HMOs).
21
In order to ensure that the event window captures the rating event of S&P without any
contamination from other concurrent rating changes, we identify the closest rating changes by Moody’s
and Fitch for any S&P rating change. If the S&P rating change event had either Moody’s or Fitch rating
change event for the same bond within 10 days, then we expand our event window from the day of rating
change by S&P to the event start date as the first rating change by either of the CRAs and determine the
event end date as the last rating change by either of the CRAs. We only keep rating events for which we
have rating consensus among all rating agencies. This leaves us with 117,091 rating change events by
S&P with 48,751 distinct issues and 6,617 distinct issuers.
Corporate bonds are generally illiquid and our sample includes bonds for which the first transaction
after a rating change event occurs as long as several weeks after the rating change. To capture the effect
of the rating change event, we exclude observations for which the difference between the last transaction
date before (Last Trans Date) the rating change and the first transaction date after (Next Trans Date) the
rating change is more than 20 days.24 The Reputational Shock 1 sample contains 3,833 downgrades and