University of Groningen, Faculty of Economic and Business Master Business Administration; Organizational and Management Control
Downgrading the
reputational capital theory?
An event study testing the reaction of stock prices on 41 downgraded banks before- and after the financial crisis of 2008.
Author: L.M.Wiggers
Student number: s1615629
Supervisor: Drs. D.J.J. Heslinga
Second Supervisor: Dr. B.A. Boonstra
Groningen, 24
thof April, 2012
1
Abstract
The financial crisis of 2008 put credit rating agencies into the spotlight. Critisism about the functioning of agencies triggered doubtions about their main underlying theory, the reputational capital. This paper examines the efficient market hypothesis in relation to credit rating downgrades of banks before- and after the crisis. Moreover, it investigates which implications this result has for the reputational capital theory. Both theories are tested with the event study methodology; where 41 downgraded banks in the period November 2005 till October 2007 represent the dataset ‘before the crisis’ and 41 downgraded banks in the period January 2009 till December 2011 the dataset
‘after the crisis’. Significant results ‘after the crisis’ give evidence to reject the efficient market hypothesis and doubt about the reputational capital theory. Nevertheless, justified results are not found to reject the reputational capital theory.
Furthermore, this study acknowledge the importance of using crude adjustment when the event dates are clustered and gives an indication of the presence of information leaking of downgrade announcements.
Keywords: event study, credit rating agencies, rating downgrades, efficient market hypothesis, reputational capital theory, financial crisis 2008
JEL codes: G01, G14, G24, L14
2
Preface
This thesis is the last step of completing my master Organizational and Management Control at the University of Groningen. The research process took me more than 7 months and every day I enjoyed the study and learned something more. Credit rating agencies were at the start of my research a hot topic in the news, especially because Standard & Poor’s downgraded the United States for the first time in 70 years. The movie ‘Inside Job’ triggered me to look further at the criticism on rating agencies caused by the financial crisis of 2008. Finally, when the news showed on 29 November that 15 banks were downgraded by Standard & Poor’s, the last missing piece for the research design was completed.
Looking back at my research period, I can see a great difference between my knowledge of the rating agency industry and event studies before the research and after. This is especially due to the useful comments and advice of my supervisors’ Drs .D.J.J. Heslinga and Dr. B.A. Boonstra. Mr.
Heslinga kept me on the right track discussing the subject and theory and Mr. Boonstra was able to teach me all the skills for an event study. I would like to thank them for their help and keeping me motivated all the time with new suggestions and insights.
Furthermore, I would like to thank my brother Jeroen for his comments and improvements of my thesis. In addition, I would like to thank my study friend Laurens helping me with the right start in conducting the research design.
L.M. Wiggers
Nieuwe Amstelstraat 27 1011 PL Amsterdam
lisanne_wiggers@hotmail.com
Student number: 1615629
3
Table of Contents
1. Introduction ... 4
2. Concepts and literature review ... 6
2.1. The credit rating agency ... 6
2.2. What are credit ratings? ... 6
2.3. Why credit ratings are useful? ... 7
2.4. The credit rating process ... 8
2.5. How agencies are paid for their services ... 9
2.6. Efficient market hypothesis ... 9
2.7. Reputational capital theory ... 10
2.8. Critiques on the reputational capital model ... 11
2.9. The principal- agency theory ... 13
2.10.Tools to mitigate the agency-problem... 16
2.11. Hypothesis formulation ... 17
3. Methodology and data ... 19
3.1. Methodology ... 19
3.2. Data ... 23
3.3. Tests ... 25
4. Results and discussions ... 27
4.1. Daily abnormal returns ... 28
4.2. Results with crude adjustment ... 29
4.3. Cumulative abnormal returns ... 30
4.4. Robustness check ... 31
4.5. Comparison of the means of the two sample groups ... 32
5. Conclusion, limitations and recommendations for future research ... 33
5.1. General conclusions and discussion ... 33
5.2. Limitations of this research ... 34
5.3. Recommendations for future research ... 34
6. Literature ... 35
Appendix I: Descriptive statistics ... 39
Appendix II: Description of the ANOVA F-test ... 40
Appendix III: Daily abnormal returns ... 41
Appendix IV: Differences between tests with crude adjustment and without ... 43
Appendix V: Cumulative abnormal returns ... 45
4
1. Introduction
The bankruptcy of Lehman Brothers in September 2008 and the following credit crisis were reasons to hesitate about the function of several audit agencies in the financial market. Credit rating agencies were one of them. They still had an ‘investment grade’ rating on Lehman’s commercial paper on the morning that Lehman declared bankruptcy. This is just one example of the mistakes credit rating agencies made during the crisis (White, 2010). Therefore, people critiqued the rating agencies and doubted about their relevance as information supplier (Hunt, 2009).
After the credit crisis, the news was full of critiques on the credit rating agencies. On the 23th of April, 2010 the CNN headline state: ‘Credit rating agencies in the hot seat’, announcing the testifying of a number of credit rating agency leaders in a trial about the causes of the financial crisis. More than a year later, on 30 September 2011, the BBC made known: ‘SEC find ‘apparent failures’ at credit rating agencies’, describing the concerns of the Security and Exchange Commission about the rating methodologies of the rating agencies. Even the Chairman Guan Jianzhong from the Chinese rating agency Dagong Global Credit mentioned his critique in an interview with Financial Times on the 21
stof July, 2010: ‘The financial crisis was caused because rating agencies did not properly disclose risk and this brought the entire US financial system to the verge of collapse, causing huge damage to the US and its strategic interests’. These criticisms were especially mentioned towards the three biggest rating agencies; Moody’s, Standard & Poor's and Fitch.
Perceived problems like conflicts of interest, dependent regulation of credit ratings, the lack of
‘transparency’ and the limitation of competition are topics which were mentioned in the critiques towards the rating agencies (Hunt, 2009; Partnoy, 1999; Frost, 2007 and Kerwer, 2005).
Policymakers have undertaken several initiatives to address these problems with ratings (Hunt, 2009). These subjects put rating agencies into the spotlight, stirring up questions about the working of the credit rating market.
The efficient market hypothesis (EMH) is in this context an important theory which forecasts that stock prices will not adjust in response to the rating change event if rating agencies base their ratings on publicly available information (Brooks, 2004). Nevertheless, a lot of researchers found evidence associated with significant negative returns in equity and bond markets as response on a rating downgrade (Barron et al., 1997; Brooks, 2004; Ederington and Goh, 1998; Glascock et al., 1987; Goh and Ederington, 1993; Griffin and Sanvicente, 1982; Holthausen and Leftwich, 1986; Liu et al., 1999).
This implies that or the efficient market hypothesis does not hold or private information is available
to the rating agencies (Brooks, 2004).
5
Moreover, the reputational capital theory is the leading theory describing the functioning of credit rating agencies in society. This theory explains that a rating agencies purpose is to make high-quality assessments of issuer creditworthiness available to market participants. The value of the rating agency’s business is derived from the agency’s reputation of conducting high-quality ratings. This model implies that under the right circumstances a well-functioning reputation mechanism will avoid low-quality ratings (Hunt, 2009).
Especially after the credit crisis there was doubt about the reputational capital theory. By example;
Hunt (2009) mentions that the reputational-capital theory could be fundamentally inadequate in defining the rating market. He suggests that credit rating agencies no incentive have to guarantee their rating quality, when their high quality ratings not have been rewarded with a good reputation.
Hunt argues that the fear of losing reputation is not likely to constrain rating agencies from issuing low-quality ratings. Therefore, the dominant academic view of rating agencies, the reputational capital theory, is likely to be wrong. Partnoy already argued in 1999 that the rating-dependent regulation undermines the validity of the reputational capital model.
The major objective of this research is to examine if there is a different relation between credit rating downgrades and stock price movements before and after the crisis. More specific, this paper tests if the efficient market hypothesis can be rejected, which may be in favour of the reputational capital theory. According to the reputational capital model, the market should react more tremendous before the crisis than after because of the damaged image of the credit rating agencies after the crisis.
A lot of research has done before analysing the efficient market hypothesis, but linking the results to the reputational capital theory in an empirical manner has not been done before. This research is based on the credit rating downgrades of banks. The banking industry is chosen because the financial crisis impacted this industry in Europe and the United States massively and the whole society is involved hereby.
The remainder of this research is organized as followed. The following chapter will describe the
concepts of credit ratings and their agencies, this segment is important to understand the critics in
the next part. This consists of a description of the efficient market hypothesis, a theoretical
discussion about the reputational capital model and will finish with the critiques on the rating
agencies. Because the agency theory is seen as one of the major theories explaining the critiques, a
special part is devoted to this model. Thereafter, describes chapter 3 the methodology, data and
lastly the tests which are used. The results are presented in chapter 4 and finally in chapter 5 the
research will be concluded.
6
2. Concepts and literature review
2.1. The credit rating agency
Since 1841, as reaction on the financial credit crisis of 1837, the first form of a credit rating agency was developed by Louis Tappan in New York. Together with a similar rating agency formed by John Brandstreet in 1849, they formed in 1933 Dun and Bradstreet, which became the owner of Moody’s Investor Services in 1962.
Poor’s Publishing Company issued its first ratings in 1916 followed by Standard Statistics Company in 1922, which merged to form Standard & Poor’s in 1941. In 1924 Fitch started to publish their first ratings (Cantor & Packer, 1994).
These companies are till today the most well-known and largest credit rating agencies in the world and are therefore called the ‘big three’. When these rating agencies started to rate a wider variety of industries and capital markets, they started to grow and their credit ratings became more popular among investors in determining creditworthiness. Therefore, a lot of new rating agencies were developed the years after, as reaction on the increasing reliance of investors and regulators on the opinions of credit rating agencies, since they saved them the time and effort performing the analysis themselves (Božović, 2011).
2.2. What are credit ratings?
The task of credit rating agency is to provide evaluations of the likelihood that obligations will be repaid, or in other words estimate the credit quality. Standard & Poor’s (2010) describe a credit rating as ‘the rating agency’s opinion about the ability and willingness of an issuer, such as a corporation or state or city government, to meet its financial obligations in full and on time’. On the other hand, a credit rating can give a value on the credit quality of an individual debt issue, and the possibility that the issue may default.
Every rating agency applies its own methodology in measuring creditworthiness and their ratings are
based on a letter scale, which differences per credit rating agency. Table 1 shows the ratings of
Moody’s, Fitch and Standard & Poor’s. In the description that Standard & Poor’s gives for a credit
rating they clearly state that credit ratings are not a guarantee that an investment will pay out or
that it will not default. The ratings are clearly not buy, sell, or hold recommendations or a measure
of asset value, but an opinion about the creditworthiness of an issuer or credit quality of an
individual debt (Standard & Poor’s, 2010).
7
Fitch S&P Moody’s Interpretation
AAA AAA Aaa Highest quality
AA+ AA+ Aa1 High quality
AA AA Aa2
AA- AA- Aa3
A+ A+ A1 Strong payment capacity
A A A2
A- A- A3
BBB+ BBB+ Baa1 Adequate payment capacity
BBB BBB Baa2
BBB- BBB- Baa3
BB+ BB+ Ba1 Likely to fulfil obligations,
BB BB Ba2 on-going uncertainty
BB- BB- Ba1
B+ B+ B1 High risk obligations
B B B2
B- B- B3
CCC+ Current vulnerability to
CCC CCC Caa Default, or in default
CCC-
C Ca
DDD C In bankruptcy or default,
DD D or other marked
D shortcomings
Table 1: long term debt credit rating symbols (Source: S&P, 2012; Fitch, 2012; Moody’s; 2012)
2.3. Why credit ratings are useful?
The role of credit ratings is an informational one, which is crucial for the functioning of modern financial markets. The rating can play a useful role in two perceptions; first in enabling corporations and governments to raise money in the capital markets and second in purchasing these debt
securities by investors to receive interest plus the return of their principal. Following, two basic types of credit ratings can be distinguished: the bond rating and the issuer credit rating. The bond rating measures the likelihood of default or delayed payments of a bond issue; therefore, they support the borrowers because they are able to improve conditions for raising capital and give a picture of the overall perception of quality of the bond on the market. On the other side, issuer credit rating takes the issuer’s creditworthiness into consideration, which is crucial for the pricing of securities. Overall, credit rating agencies provide signals to market participants on the credit quality of financial
securities (Božović, 2011).
8 2.4. The credit rating process
The rating process includes several aspects before publishing the credit rating to the public. The process distinguishes several aspects like quantitative, qualitative and legal analysis. Standard &
Poor’s (2010) makes a difference between model driven ratings and analyst driven ratings.
The model driven ratings are focused on quantitative data, which are mainly based on the financial figures which are found in the financial reports of the company, these data is incorporated into a mathematical model. Only a few small credit rating agencies use only this approach, most of the time it’s combined with the following approach (Standard & Poor’s, 2010).
The analyst drive rating is most of the time used by assigning an analyst, often in combination with a team of specialists, to take the lead in evaluating the entity’s creditworthiness. The analysts look at quantitative and legal aspects like the companies’ quality of management, a review of the companies’ competitiveness and predicted growth figures within its industry. This information is obtained by published reports, interviews and discussions with the issuer’s management. Together with the legal analysis, which measures the companies’ sensitivity for technological changes, regulatory changes and labour relations they use the information to assess the entity’s financial situation, operating performance, policies and risk management strategies (Standard & Poor’s, 2010). Figure 1 shows an analyst driven rating process developed by Standard & Poor’s.
Figure 1: Standard & Poor's (2010) analyst driven rating process
9 2.5. How agencies are paid for their services
The provided ratings are most of the time requested and paid for by the issuer or originator of the financial instrument in question, which then will be published to the public for free (Hunt, 2009).
There a two paying models: the issuer-pay model and the subscription model.
Under the issuer-pay model, rating agencies charge issuers a payment for the requested rating opinion. Their analysis is valuable because they make use of information obtained from the issuers, what might not accessible for the public and use this information in their process of forming a rating.
The rating agencies can publish the ratings to the public for free because they are not dependent on subscribers for fees.
The other model, the subscription model, works the other way around. The ratings are not available for free to the public, but charge investors and other market participants a fee for access to the agency’s ratings (Standard & Poor’s, 2010).
There is a conflict of interest between the rating entity and agency. Critics point out that, these models are sensitive for agency problem since the entities are paying for the ratings, and may be influenced by the investors. To discuss this topic further, the efficient market hypothesis and the role of reputation will be discussed first.
2.6. Efficient market hypothesis
According to Fama (1970), the most ideal economic world is a market in which prices provide perfect signals for resource allocation: a market where firms make investment decisions based on security prices that ‘fully reflect’’ all available information at any time and investors can choose among those securities that represent ownership of firm’s activities. Fama (1970) mentioned a market where prices always ‘fully reflect’ available information ‘efficient’. The article distinguished three subsets of market efficiency: the weak, semi-strong and strong form. The weak form suggests that all historical prices are reflected in the current price. The semi-strong form considers all publicly available information in the current price and the strong form reflects all information in their stock prices, so publicly available information as well as private information (Fama, 1970). If rating agencies base their rating changes on publicly available information, the efficient market hypothesis predicts that stock prices will not adjust in response to the ratings change event. Nevertheless, in the literature can be found that rating downgrades are consistently associated with significant negative returns in equity and bond markets (Barron et al., 1997; Brooks, 2004; Ederington and Goh, 1998;
Glascock et al., 1987; Goh and Ederington, 1993; Griffin and Sanvicente, 1982; Holthausen and
10
Leftwich, 1986; Liu et. al., 1999). This suggests that either proof against the semi-strong form EMH exist, or, the availability of some private information only to rating agencies, that has come into the public field (Brooks, 2004). In this research it is assumed that the efficient market hypothesis holds, which means that the stock prices fully reflect available information efficient and does not adjust in response to the rating event. The following part which describes the reputational capital theory can give some more insight about the working of credit ratings and their agencies.
2.7. Reputational capital theory
This theory describes the function of rating agencies to make high-quality assessments of issuer creditworthiness available to market participants and that the agency’s reputation for developing high quality ratings creates the value of a credit rating agency. This reputation can be seen as the product that produces returns (Hunt, 2009) and where the power of the market reaction is dependent on. Reputational capital leads to another decision-making factor for parties selecting to hire a credit rating agency or not, this factor is ‘trust’. Rating agencies who invest in the quality of their rating process acquire reputational capital; individuals and institutions will look at the reputational capital of the credit rating agency in deciding whether to rely on the agency, or, instead, to make the analysis themselves. This model implies that under the right circumstances a well- functioning reputation mechanism will deter low-quality ratings. This is based on the following assumption; when individuals and institutions perceive the ratings as inaccurate or not trustful, the rating agency will suffer a loss of reputation and there is more chance of development for their competition (Partnoy, 2001). Nevertheless, the question remains whether the reputational story is the primary explanation of the credit rating industry, or whether there is another explanation for the functioning of the credit rating industry. The efficient market hypothesis argues that the stock prices do not adjust to a rating change. However, the reputational capital theory assumes that there is an adjustment and the power of the reaction is dependent on the reputation of the credit rating agency. In theory this means that when the reputation of a credit rating agency suffers, the market would react less extreme as before. It will be assumed that when the efficient market theory not holds, the market is based on the reputational capital theory.
Several authors were critical about the reputational capital view. Hunt (2009) argues that the fear of
losing reputational capital is not decisive to limit credit rating agencies from issuing low-quality
ratings, mainly because they don’t have any reputation to lose in this type of product rating. Also
Kerwer (2005) states that financial market participants cannot hold rating agencies accountable in a
routine way, and Partnoy (1999) gives arguments that the reputational capital view is not supported
by history or economic analysis. He states that the credit rating agencies have not maintained good
11
reputations based on the informational content, but became powerful thankfully to the selling of regulatory licenses, the right to be in compliance with regulation.
2.8. Critiques on the reputational capital model
Critiques doubt about the independency of credit rating agencies and expose several critical points.
Main points mentioned are limitation of competition, the lack of ‘transparency’, the dependent relation between ratings and regulation and the conflict of interest, following from the way credit rating agencies are paid for their ratings by the issuer-pay model and the subscription model (Hunt, 2009, Frost, 2007, Partnoy, 1999 and Smith, 2002). Nevertheless, opinions on the importance of every point vary per author. Partnoy (1999) argues that the reputational capital model is undermined by the rating-dependent regulation and proposes a new model instead. Smith (2002) discusses mainly the key dimensions of the credit rating business in the context of the potential for exploitation of conflict of interest and Frost (2007) researched if the criticisms are supported by empirical evidence. The main points will be discussed shortly hereafter.
2.8.1. Limited competition
Moody’s and Standard & Poor’s are the two major rating agencies for decades, Fitch played a good third role since the 90’s but is by far not so big as the other two mentioned. In 2009, the three major agencies had a market share between 85% and 95% (Hunt, 2009). Hunt (2009) mentions that these figures can suggest that the market is non-competitive and even based on the two-rating norm, the practice of receiving ratings from two different agencies on each issue, an oligopoly market.
Following by the argument that the agencies Moody’s and Standard & Poor’s don’t have to compete at all. The concentration is also blamed by the result of natural barriers to entry the market and economies of scale. The S.E.C. (Securities and Exchange Commission) issued in 1973 a rule which incorporated ratings of Nationally Recognized Statistical Rating Organization (NRSRO), and only three rating agencies were given this status, Standard & Poor’s, Moody’s and Fitch, at the time.
Dominion Bond Rating Service Ltd and AM Best Co received later also the status. Nevertheless, this
title ensured the existing credit rating agencies minimal competition by limiting new entrants (Pinto,
2006). Also the S.E.C. (2003) made a note to further investigate to which extent allegations of
anticompetitive or unfair practices by large credit rating agencies exist. But Frost (2007) makes a
useful comment about the little evidence available on credit rating agency’s possible anticompetitive
or unfair practices and besides the suspicion may not be based on publicly available information.
12 2.8.2. Lack of ‘transparency’
Also the promotion of transparency about credit rating agencies became much more important after the credit crisis. Hunt (2009) mentioned two types of transparency; the methodology- and the performance transparency. The first mentioned is the transparency in which outsiders are informed by the way credit rating agencies develop their ratings, the performance transparency describes the ability to recognize how well the ratings perform.
Concerns about the methodological transparency are based on the lack of transparency in the non- public information provided by the credit rating agency. Because rating agencies have access to information that is not available to the public, it may provide confusing signals to the markets because investors lack that information and are not able to control the ratings (Pinto, 2006). Frost (2007) mentions besides the failing of adequately disclose information about their procedures by credit rating agencies also the selectively disclose information of credit rating agencies to their subscribers. This discussion is about whether information concerning a rating action is made available to subscribers prior to public issuance of the rating and the extent to which information concerning a rating is made available to subscribers but is not available to the public at all.
Nevertheless, the S.E.C. proposed some rules in 2008 to enhance methodology transparency.
However, Hunt (2009) mentions that the most important aspect of this discussion is in which extent the methodological transparency undermines the reputational model. Agencies are prepared to show their models which suggest that the reputational model doesn’t function as it should be. It suggests that the ratings receive their value from something else than high quality derived from their rating techniques.
Also the performance transparency is essential in the functioning of the reputation model. Without information about the performance of the credit ratings determined by agency, it is not possible to create an objective view about the quality of those ratings. In 2008 the S.E.C. already proposed some rules to improve performance transparency (Hunt, 2009).
2.8.3. Rating – dependent regulation
Partnoy (1999) concludes that the reputational capital view is not supported by history of economic analysis. This is based on the fact that credit rating agencies haven’t survived six decades yet based on their production of credible and accurate information. Partnoy (1999) argues that good reputations of credit rating agencies are not based on the informational content of their credit ratings, but on their selling of regulatory licenses, i.e., the right to be in compliance with regulation.
Hunt (2009) also explains that agency ratings are intertwined into financial regulations and this
13
could form a demand for ratings based on other grounds than quality but a demand based on the fulfilling of regulatory requirements by investors.
At last Pinto (2006) argues that the role of credit rating agencies is not required by legislation like other gatekeepers but their position is created due to their regulatory license and reliance of the market for their services.
2.8.4. Conflicts of interest
Rating agencies are confronted with several potential conflicts of interest, mostly formed by the fact that credit rating agencies are paid for their services by issuers or originators of the products.
This problem should be solved by the reputational capital, the content that credit rating agencies have more to lose by risking their reputation for objectivity than to please the issuers or originators of the products. Nevertheless, the arguments mentioned above show the weak side of this concept (Hunt, 2009). The main issue is therefore whether exploitation of this essential conflict is in fact problematic, and whether it is one that might bear negatively on the functioning of financial markets. Therefore, this topic will be further and extensively discussed in the following part in the context of the principal-agency theory.
2.9. The principal- agency theory
This part will describe the principal- agency theory in relation with the credit rating agencies. The principal agency theory in relation with credit rating agencies can be divided in three problems; the moral hazard, adverse selection and accountability. Finally, some suggestions mentioned in literature to mitigate the agency problem will be described.
2.9.1. The agency theory
Jensen and Meckling (1976) define an agency relationship as: ‘.. a contract under which one or more persons (the principal(s)) engage another person (the agent) to perform some service on their behalves which involves delegating some decisions making authority to the agent’. According to Eisenhardt (1989) the agency theory is based on the key idea that the principal- agent relationship should reflect efficient organization of information and risk-bearing costs. The problem arises when the principal and agent have different goals and risk preferences.
The agency-problem knows two parties, the agent and the principal where the agent is supposed to
act in the interest of the principal. Looking at credit rating agencies in this context, the credit rating
14
agency can be seen as the agent and the principal can be described as all user groups of the credit ratings. Looking better at this relationship, it can maybe better be described as the principal- trustee relationship because the user groups make use of the trust function of the credit agency. The basis of this trust function is grounded on the reputational capital view which is explained in the previous chapter. This theory describes that parties include ‘trust’ as a factor in their decision-making process and that the success of the credit rating agencies are based on their reputation (Partnoy, 2001).
Based on the critique delivered on the credit rating agencies it can be concluded that there is a persistent mismatch between demand and supply between the credit rating agencies and their user group. The way that the agency problem establishes itself, can be in various ways. First the problem of moral hazard arises because the agent may simply not put forth the agreed- upon effort (Eisenhardt, 1989). In the case of credit rating agencies it is caused due to the circumstance that the evaluation standards set by rating agencies are unobservable to outsiders (Mukhopadhyay, 2004).
The second problem arises from the misrepresentation of ability by the agent. This problem is called adverse selection and arises when the principal is not able to verify the skills or abilities of the agent where it was hired for at the time of selecting the agent or while the agent is working (Eisenhardt, 1989). Ashcraft and Schuermann (2006) explain that adverse selection is created in this context by the situation that the rating agencies logically still know less than the issuer. Therefore it is hard to select a suitable rating agency because the issuer has always more knowledge about the product than the rating agency and cannot compare their knowledge.
The final topic is risk sharing or also called accountability. This problem comes from the fact that the principal and the agent have a different attitude toward risk and therefore prefer different actions.
As Kerwer (2005) concludes that credit rating agencies are often criticized for holding illegitimate power and the broad demand for accountability had no great effect on how they operate. Therefore, there is a persistent mismatch between demand and supply of accountability.
2.9.2. Moral hazard
Božović (2011) discusses broadly the moral hazard problem identified in the credit rating industry.
Firstly it is important to make a distinction between solicited, the case where the credit rating
agency is paid for the rating by the issuer of the security and unsolicited ratings, which is a rating
where a credit rating agency did not get paid for. Because this distinction can be made, a conflict of
interest arises. This conflict of interest comes from the fact that a profit-maximizing credit rating
agency will issue the rating in such a way that it has a positive effect for the issuer, and stimulate
15
repeated business with them. This problem is only protected by the reputational capital view, the fear of losing reputation by the credit rating agency (Božović, 2011).
The second point comes from the fact that the credit rating industry is a regulated oligopolistic industry. It is logical that an issuer would shop for better ratings in the market and make their choice to hire a credit rating agency on this point. Nevertheless, the duopolistic market in the beginning made it very difficult to shop for better ratings. First there were only Standard & Poor’s and Moody’s known as recognised credit rating agencies, Fitch later became the third player in the field and made it an oligopolistic market. Contrary to common intuition, this deteriorated the quality of corporate bond ratings because the issuer now has the possibility to choose the best available rating. It also reduces the expected long-term gains of the rating companies, and therefore it weakens their efforts to provide a rating of good quality (Božović, 2011). A research of Becker and Milbourn (2011) investigated the impact of the third player in the market and found that it was related to a reduction of predictive power of corporate bond ratings by credit rating agencies.
2.9.3. Adverse selection
Ashcraft & Schuermann (2006) explain the working of credit rating agencies in relation to adverse selection. First point mentioned is the fact that rating agencies are assigned on trust to make an analysis to attain a given credit rating. The opinion of the credit rating agency is sensitive for the fact that the issuer likely still knows more than themselves.
This ‘lemons model’ explains that if a profit-maximizing agent once is hired, it will always deliver the lowest quality of service no matter what level of service the agent has contracted to provide. This occurs because the level of service with the lowest cost will maximize the agent’s profit on a contract (Dejong, 1985). So in the case of a credit rating agency, they will put once they are hired to attain a credit rating, less effort for the quality of the rating and will deliver ‘lemons’.
2.9.4. Accountability
Despite the critique on credit rating agencies and the demand for accountability, their way of operating did not change. So there can be concluded that there is a persistent mismatch between demand and supply of accountability – ‘an accountability gap’ as Kerwer called it (2005).
The problem of accountability flows from the information asymmetry between the principal and the
agent. The biggest problem that arises in this case originates from the situation that the relationship
between the principal and the agent is external and therefore is the principal hierarchically not
16
superior to the agent. As mentioned before, the relationship can better be seen as a principal- trustee relationship instead of the most generalized principal-agency relationship. Credit rating agencies should be the solution according to the standard principal-agency theory instead of the problem. Nevertheless, the credit rating agencies point out that their ratings should not be misunderstood as advice to buy or sell a security and dispose themselves from their accountability.
Despite the fact that borrowers should have a good chance to influence the credit rating agencies according to the principal-agency theory, they have very little influence. The situation that borrowers pay for being rated should give them power as a principal. Nevertheless is this situation an abnormality, as the real consumers of this information product are the investors instead of the issuers (Smith & Walter, 2002)
2.10.Tools to mitigate the agency-problem
2.10.1. Moral Hazard
To overcome the moral hazard problem Mukhopadyay (2004) suggests that designing an appropriate contract is the solution. The contract should specify the payment to the rating agency which is based on the evaluation standard the rating agency claims it has set. Their performance should be linked to the performance of the rated debt.
Bolton, Frexas and Shapiro (2009) investigated the role of a monopoly and duopoly in their paper and concluded that a duopoly is not better because issuers are able to shop around for better ratings. They suggest several solutions like upfront payments to eliminate conflict of interests, disclosure of all ratings and an investors-pay solution. Nevertheless they give theirselves feedback that these interventions do not stimulate credit rating agencies to gather information and perform their jobs better.
2.10.2. Adverse selection
Dejong et al. (1985) describe in their research ‘the lemon problem’ and investigated some of the
remedies to solve this problem. They investigated the power of a negligence liability rule where
agents could, for a cost, examine and perfectly conclude an agent’s actions if a loss occurred. When
the agent delivers a ‘lemon’, the agent will be held responsible for the loss. Dejong et al. (1985)
found that this liability rule would eliminate almost the whole problem when this would be done
with publication of the outcome of all investigations. In this case, the credit rating agencies will also
be hold responsible for their actions but as mentioned before they hold themselves irresponsible
because their ratings are according them their ‘opinion’ and not an advice.
17 2.10.3. Accountability
It is difficult to hold the credit rating agencies accountable for their ratings because of their vision that their ratings are ‘opinions’. A possible solution is to change the regulation by the US government to hold rating agencies accountable; nevertheless, this is difficult because the government is also dependent on the judgment of the ratings agencies because they borrow through the financial market. Another solution mentioned is to diminish the impact of the credit rating agencies wherever imaginable. Eliminating the original proposal of relying the measurement of risk exclusively on credit rating agencies would lead to a reduction of the negative impact. The third solution would be the improvement of the quality of the ratings made by the agencies.
However, this solution will be in conflict with the moral hazard problem mentioned before (Kerwer, 2005).
Finally, Kerwer (2005) mentions that another solution would be the introduction of more financing options to the public and private borrowers in order to make them less dependent on credit rating agencies.
2.11. Hypothesis formulation
As explained in the literature the efficient market hypothesis assumes that stock prices do not react on a changed rating. Nevertheless, research shows that stock prices react negatively to a downgrade. This could lead to a rejection of the efficient market hypothesis or the assumption that private information is available to the rating agencies. The power of the reaction is dependent on the reputation of the credit rating agency according to the reputational capital theory. However, critics point out that the reputational capital does not function as it is explained in the theory.
Because the rating agencies suffered on critics after the crisis, a damaged image is expected.
According to the reputation capital theory the market would react less extreme after the crisis then before.
This research will focus on these theories and the research objective is formulated as follow:
Does the efficient market hypothesis hold comparing stock price reactions of credit rating
downgrades before- and after the financial crisis of 2008 and which implications does this have for
the reputational capital theory?
18 The following research hypothesis can be established:
H
0: The efficient market hypothesis holds
H
1: The efficient market hypothesis does not hold
First, the methodology and downgrades of the credit rating agencies will be defined, selected and further explained in the following chapter. This hypothesis will be tested with an event study, comparing the reaction of stock prices on 41 downgraded banks before- and after the financial crisis of 2008. The efficient market hypothesis holds if abnormal returns show no reversal pattern. In other words, investors could not profit from arbitrage opportunities with efficient markets. This would have negative implications for the reputational capital theory. When there is a significant effect on the stock prices, the alternative hypothesis is assumed to be true and the efficient market hypothesis is rejected. The following hypothesis can then be tested:
H
0: The reputational capital theory does hold H
1: The reputational capital theory does not hold
H
0is true when the abnormal returns have a higher significant result before the crisis than after the
crisis. The alternative hypothesis will be accepted when there is a more significant effect after the
crisis compared to before or if there is no significant evidence that the results differ before and after
the crisis.
19
3. Methodology and data
This chapter is divided in two parts; the first part will explain the methodology which is used and the second part describes the data used to conduct the analyses.
3.1. Methodology
This study is performed with an event study which is based on the methodology from Brown &
Warner (1980, 1985), MacKinlay (1997), Serra (2004), Jarcque & Bera (1980), Corrado (1989) and Anderson et al. (2010). To perform the event study it is important to select the event date. In this research the event date can be defined as the announcement date of a credit rating downgrade by a credit rating agency. Following it is important to select the estimation window and the event window. The event date is selected as t = 0 and the estimation window starts with 120 days till 6 days prior the event day [-120 ; -6]. Despite of the fact that Brown and Warner (1985) and MacKinlay (1997) both recommend an estimation period around 250 days, there is chosen for a smaller estimation period because many banks of the dataset have a downgrade earlier in this period which would pollute the data. The 114 days is based on a period that all firms from the dataset do not have another downgrade in this period next to the event date. The event period starts with 5 days before and ends with 5 days after the event date [-5 ; 5]. According to Brown and Warner (1985) is an event window of 11 days sufficient. Figure 1 shows the time line of the event study where L
1= T
1– T
0, is respectively the length of the estimation window and L
2= T
2– T
1the length of the event window (MacKinlay, 1997).
Figure 2: time line of the event study
Brown and Warner (1980) mention in their paper that it is necessary to construct a model generating
‘normal’ returns before abnormal returns can be measured. Three general models are most of the time used in event studies; the mean adjusted return model, the market adjusted return model and the market and risk adjusted return model. Each of the models uses a different way to calculate the
‘normal’ returns. Ex post abnormal returns are obtained as the difference between observed returns
20
of firm i at event day t and the expected return generated by a particular benchmark model. Each model calculates its expected return on a different manner which will be explained below.
Comparing the results of all three models will deliver a more robust event study.
The basic calculation of the abnormal return is for each model the same and described as follows:
AR
it= R
it– E(R
it) (1).
AR
it, R
itand E(R
it) are respectively the abnormal return, the actual return and the normal return for firm i on day t.
3.1.1. Mean adjusted returns
This model is seen as the simplest model of the three. Nevertheless, Brown and Warner (1980, 1985) find that the results of this model yields similar results as the two more sophisticated models explained hereafter. Brown and Warner (1980) describe that this model assumes that the ex ante expected return for a given security i is equal to a constant which can differ across the securities. The predicted ex post return on security i in time period t is equal to
i. The excess return for security i at day t is defined as A
itand is equal to the difference between the actual return, R
it, and the predicted return
i:
A
it= R
it-
i(2).
I
is calculated as:
I
= R
it(3).
This represents the mean of the actual return over the estimation window.
3.1.2. Market adjusted returns
The market adjusted return model differs from the mean adjusted return model because it does not assume that the ex ante expected returns are constant for a given security (Brown and Warner, 1980). The ex post abnormal return on any security i is given by the difference between its return and that on the market portfolio:
A
it= R
it- R
mt(4).
21
R
mtis in this performed event study the return on the MSCI or S&P 500 equally weighted index for day t for the local currency of the security of firm i .
3.1.3. Market and risk adjusted returns
The market and risk adjusted return model, also called the market model or OLS market model is a statistical model which relates the return of any given security to the return of the market portfolio.
The model presumes that some version of the Capital Asset Pricing Model generates expected returns and the model’s linear specification follows from the assumed joint normality of asset returns (MacKinlay, 1997). MacKinlay (1997) argues in his paper that the market model is an improvement over the mean adjusted return model because the variance of the abnormal return will be reduced by removing the portion of the return that is related to variation in the market’s return. The market model is:
R
it= α
i+β
i(R
mt) +ε
it(5).
In this formula are α
iand β
ithe parameters of the market model and ε
itrespectively the zero mean disturbance term.
The abnormal return for the market model is:
A
it= R
it-
i-
i,R
mt(6).
MacKinlay (1997) suggests that under general conditions ordinary least squares (OLS) a consistent estimation procedure is for the market model parameters. The OLS estimators of the market model parameters are:
i =
(7)
with
m(8)
and (9)
where (10)
and (11).
22
Now the estimators are known, the daily abnormal returns, the mean abnormal returns are calculated:
t
= AR
it(12).
In this formula N is presented as the total numbers of events. To calculate the variance, the following formula is allowed to use because L
1is large:
Var(
t)=
2εit(13).
These formulas aggregate the abnormal returns over the event window. MacKinlay (1997) states that for this aggregation it is assumed that there is no clustering. The inferences with clustering will be discussed hereafter. Now the aggregated abnormal returns are known, the cumulative abnormal return for each security i can be calculated with:
CAR (T
1, T
2) =
t(14)
and the variance of the cumulative abnormal returns through:
var(CAR (T
1, T
2)) =
t(15).
The three general models which are used are explained. Nevertheless, some exceptions were applicable in this research.
3.1.4. Crude adjustment
As mentioned earlier, to use formula [13] it is necessary to exclude clustering. Clustering exists when the event windows of the included securities overlap in calendar time. When this is the case, worries about covariance across securities should exist. When the event windows do overlap and the covariance’s between the abnormal returns will not be zero, the aggregated abnormal returns [12]
are no longer applicable (MacKinlay, 1997). Brown and Warner (1980) suggest avoiding the negative effects of clustering by estimating the standard deviation of average residuals from the time series of the average abnormal returns over the estimation period. This adjustment is called crude adjustment which is used for cross-sectional dependence between firm’s average residuals. The crude adjusted standard deviation can be calculated with (Serra, 2004):
S*(
0)= (16)
23 where
* = (17).
Crude adjustment is necessary in this event study because the data is obtained from the same industry (the banking industry) and some events overlap each other
1.
3.2. Data
3.2.1. Data selection and description
This research makes use of data containing downgraded credit ratings of banks from the US and Europe. There are several reasons to select data from companies in the same industry, in this case the banking industry. The first reason was the availability of the data. Banks are already downgraded for years and therefore a long-term period of data is available from this industry. Also the public function of banks and the recent downgrades made the use extra interesting. Finally, selecting data from the same industry makes it easier to compare the conclusions from the different time periods with each other.
There is chosen for a time period from November 2005 till October 2007 which is called the dataset
‘before the crisis’ and a time period from January 2009 till December 2011 which is described as the dataset ‘after the crisis’. This difference is made because in the weekend of 13 & 14 September 2008 it came to light that Lehman Brothers would file for bankruptcy. This is seen in the literature as the beginning of the financial crisis.
Because the downgrades of de credit ratings were not available on the accessible Thomson Datastream database, another method is used to obtain the data. The data from the dataset: ‘after the crisis’ is selected by a research on the websites from Moody’s, Standard & Poor’s and Fitch.
These are the three biggest credit rating agencies. The event date for this dataset is selected on the basis from the first downgrade of one of the three credit rating agencies when several has taken place in the same period. The banks were also selected on the limitation that there was not an earlier downgrade placed in the estimation period (120 days before the event date).
The dataset ‘before the crisis’ was a little bit more difficult to obtain because these ratings were not
1 This is based on the following statement from Brown and Warner (1980) in their paper: ‘However, in an actual event study, a sample of securities whose events are clustered in calendar time may be nonrandom; the sample securities might be drawn from a common industry group having positively correlated performance measures. In such a case, the power of the tests is reduced even if a particular methodology abstracts from the market, and taking into account cross-sectional dependence in order to assure the ‘correct’ proportion of rejections is appropriate in such a case.’
24
available on the website from the rating agencies. The ratings were finally obtained with the Bloomberg database and the same selection criteria were used as by the dataset ‘after the crisis’.
Finally, 41 useful banks were selected for the dataset ‘before the crisis’ as well as for ‘after the crisis’. To investigate the impact of the downgrades, daily return data are used for the event study.
According to Brown and Warner (1985) and MacKinlay (1997) is daily data more sufficient than weekly or monthly data because they provide more explanatory power. The daily return data were obtained from Datastream for every selected bank. The return data were downloaded as closing prices and in local currencies. Local currencies were used to avoid the problem of exchange rates, which could influence the data. The stock indices present several currencies and therefore were several stock indexes used for the market portfolio. The S&P composite is used for the US dollar, the S&P 500 (E) for the Euro, the S&P 500 GBP for the Pond, the MSCI Denmark for the Danish Krone, the MSCI Norway for the Norway krone, the MSCI Switzerland for the Swiss Franc and lastly were the MSCI Sweden used for the banks which use the Swedish Krone.
3.2.2. Descriptive statistics
The descriptive statistics of the dataset before and after the crisis are presented in Table 2 and Table 3. The descriptive statistics are given over the estimation window (-120, -6). In both tables shows the Jarque-Bera test that the samples are not normally distributed. According to Brooks (2008) should the Jarque-Bera give a result below 5.99 to be normally distributed.
market and risk adjusted returns model
Observations 4705
Mean 0,0000
Maximum 0,0792
Minimum -0,0727
standard deviation 0,0115
Skewness 0,1107
Kurtosis 5,7028
Jarque Bera 1442
Table 2: Descriptive statistics of the return in the estimation window of the dataset ‘before the crisis’
The statistics are based on the market and risk adjusted return model discussed by Brown and Warner (1980, 1985) and an estimation window of [-120, -6]. The dataset ‘before the crisis’ is based on the time period November 2005 till October
2007.
25
market and risk adjusted returns model
Observations 4696
Mean 0,0000
Maximum 0,2797
Minimum -0,2511
st.dev 0,0345
Skewness 0,0209
Kurtosis 5,8577
Jarque Bera 1598
Table 3: Descriptive statistics of the return in the estimation window of the dataset ‘after the crisis’
The statistics are based on the market and risk adjusted return model discussed by Brown and Warner (1980, 1985) and an estimation window of [-120, -6]. The dataset ‘after the crisis’ is based on the time period January 2009 till December 2011.
Because the samples are not normally distributed, this would mean that the datasets are not useful for a parametric test. Although Brown and Warner (1980, 1985) state that non-parametric tests will test the robustness of the parametric test. Therefore, both the tests are performed and discussed in the next section. Appendix I contains Table I and Table II with the descriptive of all three models discussed by Brown and Warner (1980, 1985). The data of every model is not normally distributed and a non-parametric test is needed.
3.3. Tests
Several statistical tests are selected and used to test if the abnormal returns significantly differ from zero. The Jarque-Bera test is used to test for normality and the statistical significance of the abnormal returns will be tested using the parametric t-test and the non-parametric Corrado rank test. Finally, the Anova F-test is used to test whether differences between subsample means are significant.
3.3.1. Normality test
Jarque & Bera (1980) developed a method to calculate if a particular dataset is normally distributed or not. They mention in their article that violation of the normality assumption may lead to statements that are not correct. Therefore the data should be tested with the following Jarque-Bera test:
JB = (S
2+ ) (18)
26
With N for the number of observations, S is the sample skewness and K is the sample kurtosis.
Samples that are from a normal distribution have an expected skewness of 0 and an expected excess kurtosis of 0. When the Jarque-Bera test indicates that the sample is from a normal distribution, a parametric test can be used. According to MacKinlay (1997) several alternative approaches are available which are nonparametric. These tests do not have to fulfill the specific normal distribution assumption.
3.3.2. Parametric test
The standard statistic which is most of the time used when the sample is from a normal distribution is the t-test. Brown and Warner (1985) note that the sample should be independent, identically distributed and normal to use this test. When the degrees of freedom are larger then 200, the test statistic can also be assumed as normal. The t-test can be described as follow:
t=
0/ S (
0) (19)
With S(
0) as an estimate of the standard deviation of the average abnormal return. To perform the t-test for the cumulative abnormal returns MacKinlay (1997) presents the following formula:
Φ
1= ~N (0,1) (20)
where var
τ= (21)
In practice, is usually unknown and an estimator must be used to calculate the variance of the abnormal returns. MacKinlay suggest that the usual sample variance measure of from the market model regression in the estimation window an appropriate choice is.
In this event study a two-sided t-test is used to test the first hypothesis because the hypothesis does not define a need for a one-sight t-test. Therefore, the null hypothesis will be rejected for both tail sides of the sample distribution. A significance level of respectively 10%, 5% and 1% will be used for this test. The degrees of freedom are based on N-1.
3.3.3. Non-parametric test
MacKinlay (1997) mentions two common nonparametric tests for event studies, the sign test and
the rank test. When the distribution of abnormal returns is skewed, as can be the case with daily
data, the sign test is maybe not well specified. Therefore, recommends MacKinlay (1997) the rank
27
test. The rank test is developed by Charles Corrado (1989) and therefore also called the Corrado test.
What is important to mention is that the Corrado test will not be used in isolation. It can be used as a robustness check of conclusions based on parametric tests (MacKinlay, 1997). To implement the Corrado rank test, it is necessary to rank all the firms abnormal returns over the both the estimation and the event window. The rank is described as K
il= rank (Ar
il). The test will compare the ranks in the event period for each firm. The test statistic is:
R = (21)
where
S ( ) = (K
it– ))
2(22)
N specifying the number of events and L specifies the length of the estimation and event window.
3.3.4. ANOVA F-test
The last test what is performed is the Anova F-test. This test can be used to compare the means from
two or more sample groups. According to Anderson et al. (2010) is the idea behind ANOVA based on
the development of two independent estimates of the common population variance
2. One of the
estimates of
2is based on the variability of the data within each sample and the other on the
variability among the sample means. Finally, the two estimates
2can be compared and a
conclusion can be made whether the population means are equal. The description of the formula
can be found in the second Appendix II.
28
4. Results and discussions
This chapter presents and discusses the outcomes of the event study with the purpose to analyze if stock prices react differently and if they do, how the react different, on credit rating downgrades from banks before and after the crisis.
The first part will show the results on the market model on the sample ‘before the crisis’ and ‘after the crisis’. The second part deals with the different results on the crude adjusted tests and the tests without adjustment. The chapter finishes with the ANOVA F-test to conclude if the results significantly differ or not.
4.1. Daily abnormal returns
Figure 1 shows the results of the abnormal returns from both the samples ‘before the crisis’ and
‘after the crisis’. The abnormal returns of both the samples are put together in 1 graph to compare the results more easily. The abnormal returns show the impact of the downgrade announcements on the share price of the banks from the samples.
Figure 1: Daily abnormal returns from the dataset ‘before the crisis’ and ‘after the crisis’
The abnormal returns are based on the market and risk adjusted return model discussed by Brown and Warner (1980, 1985) and an estimation window of [-120, -6]. The dataset ‘before the crisis’ is based on the time period November 2005
till October 2007. The dataset ‘after the crisis’ is based on the time period January 2009 till December 2011.
As can be seen has the sample ‘after the crisis’ a very strong negative reaction one day before the downgrade announcement, especially compared with the sample ‘before the crisis’. It is notable that the share prices from ‘before the crisis’ react positive one day before the downgrade announcement and on the announcement day self almost neutral. Appendix 3 shows the results of all three models.
The graphs show that the results of the different models overlap and therefore are robust. The
results of the mean adjusted return model and the market adjusted return model are almost the
29
same and can be founded in Appendix III, Table III and IV. The figures I and II plot the results of these models and be founded in the same Appendix.
4.2. Results with crude adjustment
As explained in the methodology is crude adjustment necessary to exclude clustering from the samples. Figure 2 shows the results of the test which are crude adjusted and the tests which are not adjusted for clustering.
Figure 2: P-values comparing results with crude adjustment and without crude adjustment.
This figure shows the p-values of the market and risk adjusted return model discussed by Brown and Warner (1980, 1985) which are based on an estimation window of [-120, -6]. The dataset ‘before the crisis’ is based on the time period November 2005 till October 2007. The dataset ‘after the crisis’ is based on the time period January 2009 till December
2011.