Credit Ratings and Quality of Risk Disclosure in the US Banking Industry

(1)

Credit Ratings and Quality of Risk Disclosure in the US Banking

Industry

Jason Richard Wever (Student number: 1693948)

University of Groningen, Faculty of Economics and Business Master Thesis Accountancy

(wever.j.r@gmail.com)

University Supervisor Dr. Bo Qin

July 2011

Abstract

This paper examines the relation between credit ratings and risk disclosure quality of US banks. I measure risk disclosure quality using a framework which captures the relative quantity, the density of risk disclosure in the Management Discussion & Analysis (MD&A) part of annual report, and the density of specific risk factors discussed. Using a sample of 126 firm-year observations from 2005 to 2009, I find that credit ratings and risk disclosure quality are negatively related and that the relative quantity aspect of risk disclosure quality determines much of the relation between credit ratings and risk disclosure quality. The results also show that for banks with negative earnings the relation is stronger than for banks with positive earnings. These findings suggests that the higher the risk disclosure quality of a bank, the lower its credit rating. In addition, I provide evidence that when a bank has a low credit rating it will try to compensate by disclosing more qualitative information in order to reduce cost of capital, since low credit ratings usually leads to high cost of debt capital.

(2)

EXECUTIVE SUMMARY

Credit rating agencies have played a key role in the financial markets for various years (White, 2001). Their success has not come without criticism. The scandals of 2000-2002 and the recent financial crisis saw a massive criticism from investors and regulators which questioned both the integrity and the quality of the credit rating agencies (Frost, 2006; Hunt, 2008). This study looks at the relationship between credit ratings and risk disclosure quality of banks in the United States with a sample of 26 banks listed on the New York Stock Exchange from the years 2005 to 2009.

The results from this research indicate that the higher the risk disclosure quality the lower the rating. The voluntary disclosure theory predicts that more disclosure of information leads to better informed investors which lead to better assessment of the value of a specific firm. It is believed that firms will be rewarded when disclosing more information. However, in this study the results indicate otherwise.

I argue that there are several reasons which may lead to these results. Firstly, more qualitative information about risk can be perceived as “negative” information which informs credit rating agencies more about risks that banks are exposed to. If this is the case, I advise all banks to give less (negative) information about risk and risk management in their annual reports as it has a negative effect toward their credit rating. The second possibility is that credit ratings determine risk disclosure quality. When a bank has a perceived low credit rating and it wants to better its rating or maybe compensate for the low rating, it will disclose more qualitative information about its risks and risk management. This second possibility was tested and the results support my argument.

In short, this paper provides evidence that more disclosure quality about risk does not lead to better credit ratings. Risk disclosure quality is influenced by credit ratings and firms with low credit ratings would try to compensate by disclosing more qualitative and

(3)

, on the issue of examinations report of credit rating agencies they stated: "We've uncovered serious shortcomings at these firms, including a lack of disclosure to investors and the public, a lack of policies and procedures to manage the rating process, and insufficient attention to conflicts of interest". This report on shortcomings was published before the Lehman Brothers downfall of September 2008.

Credit rating agencies (CRAs) provide the service of giving opinions on

creditworthiness of entities, countries, and some financial instruments. The information provided by these CRAs is used by investors, investment banks and dealers, debt issuers and regulators. The large CRAs play key roles in capital markets by spreading

information to all market participants and they facilitate contracting between creditors and debtors (Gonzales et al., 2004).

Even though the service provided by CRAs may be seen as vital for the functioning of the financial markets, they are not fully regulated when you compare them to other

important “gatekeepers” like auditors (Partnoy, 2006). There has been some criticism towards the informational value of CRAs‟ ratings since the accounting and auditing scandals of 2000-20022. Conflict of interest, anticompetitive or unfair practices, lack of diligence and competence, and inadequate disclosure are some of the criticism towards CRAs3. The recent financial crisis has seen further criticism towards the methods used by CRAs on the ratings of structured debt products such as CDOs4 (Hunt, 2008). As a result

1 _{United States Securities and Exchange Commission, “The summary report of issues identified in the}

commission staff’s examination of select credit rating agencies “(2008)

2

For example, see U.S. Senate [2003], and SEC[2003]

3_{See Frost (2006) for a broader view on the criticism towards CRAs} 4

Collateralized Debt Obligations: a type of structured asset-backed security whose value and payments are derived from a portfolio of fixed-income underlying assets.

(5)

2 of the financial crisis, there were various regulatory reports5 that indicated the credit

ratings on structured products were used to convince investors in purchasing toxic instruments. CRAs gave some CDOs the same rating as they gave government bonds6, giving the false perception that these products were of the highest grade and are low risk investments. These products did not have the same low risk of default as other highly rated products, which lead investors to lose confidence in the ratings since they did not correlate with the associated risk7. Taking these criticisms aside, one can not

underestimate the role that CRAs played in the last decades of the growing international financial markets. The CRAs play a vital role in financial markets by assisting in the reduction of information asymmetry, by acting as an agent providing reliable and independent opinions about a default risk of a company (White, 2001). Less informed investors can use this information to better assess their risk when making their portfolios. Credit ratings also play an important role as tools for corporate governance and as tools for mitigating principal-agent problems (Altman and Rijken, 2004; Cantor, 2004; Löffler, 2004). They help mitigate principal-agent problems because investors impose ratings-based portfolio and indirectly force the management of firms to issue credit ratings and/or try to influence it. Even though academic researchers have been skeptical about CRAs added value, they have an implicit contract and monitoring relationship with the firms (Boot et al, 2006).

Before Lehman Brothers collapsed in September of 2008, the firm had at least an “A” rating from the three largest CRAs. Lehman Brothers boasted in their annual report 2008, of having a “culture of risk management at every level of the firm”. Their risk

management failed as the company went bankrupt. Although literature prior to the financial crisis was generally positive about CRAs (e.g., Altman and Rijken, 2004; Amato and Furfine, 2003; Blume et al., 1998; Covitz and Harrison, 2003; Ederington et al., 1987), recent criticism of CRAs cannot be overlooked. Prior studies on the quality of credit ratings may need to be revisited and their rating system must receive a closer look.

5_{For example, see SEC’s Improvements to the regulation of Credit Rating Agencies, July 2010} 6

Government bonds usually receive the highest grade of AAA (S&P) or similar rating (Utzig, 2010)

7

(6)

3 CRAs use financial information and related disclosures8 of a firm as an instrument to

partially determine their ratings. Moody‟s Investor Services released a document in 20079

on rating methodologies for banks, which states that they consider the risk disclosure of banks when assessing a credit rating. One of the five fundamental factors is risk

positioning10. Moody‟s look to see if the extent risk discipline is aligned with the bank‟s strategy. When evaluating what makes for excellent risk management practices, it states that the bank needs to be highly aware of risks and establish risk appetite on an annual basis. They further mention that their needs to be clear disclosure about credit risk, market risk, operational risk and other related risks. In a special comment they stated that in their observations of risk disclosure of banks they found that the quality varies greatly amongst banks and that some banks need to provide more in-depth disclosure about the magnitude and exposure of their risks11. Standard & Poor‟s12 issued a special comment that states that they will use risk disclosure, as required by the Basel II accord,

extensively in their risk assessment. They will use both internal and external risk

disclosure in their assessment. Banks have to provide extensive disclosure on their credit risk. They will also have to disclose the breakdown of their exposure at default for each Basel II asset class and major geographic region, as well as the associated capital requirements and methodologies applied. Although CRAs claim that they consider risk disclosure in their rating, no prior research has studied the relation between risk

disclosure and credit rating.

Since a credit rating measures the probability of a default, a bank‟s risk exposure must certainly be taken into account when considering default risk. Since a bank is more likely to know its own specific risk, it can disclose these risks to other parties in various ways. The financial crisis of the 21st century saw the failures of several banks worldwide. The subprime mortgage crisis that originated in the United States had a domino effect for almost all banks. The financial subprime mortgage crisis could have been prevented if the banks properly reported on their risks. The failures of Washington Mutual, IndyMac,

8 _{Standard & Poor’s (2006) “Corporate Ratings Criteria”} 9

Moody’s Investor Relations (2007) “Bank financial strength ratings”

10

Moody’s state that the factor risk positioning in mature markets has an overall weight of 20% and 40% weight as qualitative factors.

11

Moody’s Investor Relations (2006) Risk disclosures of banks and securities firms

12

(7)

4 Wachovia, HBOS, show that banks were not properly aware of their risk and had

inadequate risk management. Banks and other financial institutions such as insurance companies, operate day-to-day on assessing risks. Because their core business is

concentrated on risks, they may teach us the most about credit ratings and risk disclosure, since CRAs have to take internal and external information about risks more into account than non-financial firms. A bank has the option to provide everyone the same information via annual reports and press releases or it can give specific information to specific

groups13. We know that disclosure quality influences the risk premium a firm receives (Botosan, 1997) and the cost of capital (e.g., Bertomeu et al., 2008; Easley and O‟Hara, 2004; Lambert et al., 2007) but we do not know if risk disclosure quality has any influence on credit rating.

We also know that corporate management are attentive about their firm‟s credit ratings and some firms even set objectives to achieve a certain rating grade (Graham and Harvey, 2001; Kisgen, 2006,2007). Management may have an incentive to disclose more information, particularly on risk, if there is a probable upgrade in credit rating or to avoid a downgrade. There is also the possibility that the credit rating affects the quality of disclosure. After receiving a downgrade a firm might react by disclosing more risk information in the following year in order to reverse the downgrade. A higher quality of risk disclosure may better inform the CRAs of the risks that the firm is exposed to. If a firm discloses low quality risk information and does not provide adequate private information to the CRAs, there is a chance that CRAs base their rating on inaccurate information, instead of real information that could have been disclosed by the firm (Spence, 1973). Seeing that credit rating gives an indication of creditworthiness, I am of the opinion that risks are essential in determining the final grade. My research question is: does a relationship exist between risk disclosure quality and the credit rating which a bank receives?

This study contributes to two streams of literature. First, this study provides to

empirical evidence concerning the role of CRAs in corporate decision making. There has

13_{In 2000 the SEC introduced the Regulation Fair Disclosure (FD), which prohibits U.S. public firms to}

disclose non-public information to financial professionals. The exception was rating agencies, but only if the information is used to prepare credit rating

(8)

5 been prior research which question the quality of the credit ratings from CRAs (e.g.,

Blume et al, 1998; Cantor and Packer, 1997; Covitz and Harrison, 2003; Poon, 2003). Further research focused on the informational role that credit ratings have (e.g., Ederington et al, 1987; Goh and Ederington, 1993; Kliger and Sarig, 2000) and on the relation between analysts and the credit ratings (e.g., Cheng and Subramanyam, 2008; Ederington and Goh, 1998). Even though there is ample literature on credit ratings from a financial and economic perspective, not much research has looked at credit ratings from an accounting perspective. We know that top managers care about credit ratings (Kisgen, 2007) and that ratings influence both financial decisions and CEO compensation (Kisgen, 2006; Kuang and Qin, 2011). It will be equally important to examine if risk disclosure quality is also affected by or affects the rating.

Second, this paper contributes to the growing literature on risk disclosure. There has been prior research on the informativeness of risk disclosure (e.g., Dobler, 2005; Jorion, 2002; Kravet & Muslu, 2010; Rajgopal, 1999), research on the attitudes of investors toward risk disclosure (Solomon et al, 2000) and the market reactions towards risk disclosure (e.g., Linsmeier et al., 2002; Rajgopal, 1999; Welker, 2002). Jorgensen and Kirschenheiter (2003) developed an analytical model of discretionary risk disclosure, where they found positive effects of risk reporting on a firm‟s share price, its beta and its risk premium. There has been research which examined determinants of risk reporting in the banking industry (Helbok & Wagner, 2005), but it only focuses on operational risk14 reporting and does not test for other important risks such as credit risk, interest rate risk and market risk. While operational risk focuses on internal processes which a bank can influence, other risks such as market risk cannot be influenced. There has also been research on the effects of risk disclosure for the banking industry. Homolle (2003) found that risk disclosure may see a rise in a bank‟s risk exposure. Cordella and Levy Yeyati (1998) found that less risk disclosure in bad times may improve chance of survival and that when a bank doesn‟t have control over its risk portfolio, more risk disclosure increase the probability of a bank failure. Although prior studies tackled some issues related to risk disclosure, little is known about the quality of risk disclosure.

14_{Operational risk is the risk of loss resulting from inadequate or failed internal processes, people and}

systems.Source: Basel II: International Convergence of Capital Measurement and Capital Standards: a Revised Framework

(9)

6 This paper also contributes with a unique hand-collected data from the MD&A part of annual reports. This is based on an original multi-dimensional measurement of risk disclosure quality. Future research regarding risk disclosure quality for banks can use a similar measurement method.

Lastly, since credit rating and cost of debt capital are highly correlated (e.g.,

Ederington et al., 1987) and there is prior evidence that disclosure quality has influenced the cost of debt capital (e.g., Botosan, 1997; Sengupta, 1998), this study will extend the investigation of the consequences of risk disclosure quality and the cost of debt capital.

The rest of this paper is organized as follows. Section 2 presents background information on CRAs and on risk disclosure. Section 3 reviews prior literature and develops the hypothesis. Section 4 will describe the research design implemented for this research. Section 5 discusses the empirical findings. The last section contains the

conclusion of this paper.

2. INSTITUITIONAL BACKGROUND 2.1 Credit ratings

A credit rating is an opinion of the general creditworthiness of an obligor, or the creditworthiness of obligor with respect to a particular debt security or other financial obligation, based on relevant risk factors (Standard & Poor‟s, 2006).15

Most credit rating agencies base their rating on the probability of default. The rating consists of a letter, which is not universally used between all CRAs. There are various levels of rating, starting from prime to non-investment grade and down to substantial risks. Some CRAs also give a comment on the rating, giving their opinion on the outlook of the rating. CRAs business model is built around their reputation, as it is highly exposed and sensitive to reputation risk (Covitz and Harrison, 2003). If a CRA publishes inaccurate ratings, its credibility may vanish and it would face costly legal damages. Since CRAs are an important player in the financial markets for dissemination of information and in

15

(10)

7 facilitating contracting (Frost, 2007), there has been interference of the Securities and

Exchange Commission (SEC). This interference has strengthened the role which CRAs have in the financial market, by establishing a regulatory entity, NRSRO16. This

interference also saw a rise of dominance of the big three CRAs17, whom are responsible for 98% of all outstanding ratings issued (Hunt, 2008).The scandals of 2000-200218 caused much damage to the reputation of CRAs as they were unable to properly rate the firms which led to the market shock. Moody‟s staff even acknowledged that they should have been more diligent in their rating procedure of Enron (Morris, 2002). However, Frost and Kim (2006) found that CRAs have become more stringent on downgrades since the Enron incident. Frost (2007) also argues that the Enron scandal cannot be solely blamed on the CRAs as they are not in the position to accumulate reliable information as Enron‟s lawyers and independent auditors were supposed to provide.

Following large scandals and corporate failures such as Enron, the US congress passed the “Credit Rating Agency Reform Act”, giving SEC power to regulate the NRSROs internal processes and guard against conflict of interest. The financial crisis has seen the CRAs suffer from huge reputation loss; the SEC has expanded its regulation in

supervision in the rating process and imposing greater accountability for the CRAs19. This expansion in supervision requires NRSROs to maintain more robust internal supervision of the ratings process and comply with a number of additional procedures designed to increase transparency and mitigate conflicts of interest.

2.2 Risk disclosure

Lack of useful disclosures about a firm‟s risks and uncertainties have been a long-time criticism of financial reporting (Schrand and Elliott, 1998).The purpose of risk disclosure is to give stakeholders an early warning of possible and detrimental

developments. Ernst & Young found in their survey (Basel II: The business impact) that, a majority of investors identified risk information as a top priority in considering initial investment. The SEC‟s Financial Reporting Release (FFR) 48 of 1997also requires

16

Nationally recognized statistical rating organizations are Credit Rating Agencies which is permitted, according to the SEC, to issue rating which can be used by financial firms for regulatory purposes.

17_{Fitch Ratings, Moody’s Investors Services and Standard & Poor’s} 18

For example, see U.S. Senate [2003], and SEC[2003]

19

(11)

8 companies to disclose market exposures of their financial assets and liabilities. On an

international level the Basel Committee20 introduced pillar 3 of the Basel II Accord21 in order to improve market reporting. The Basel Committee had a fundamental objective to “develop a framework that would further strengthen the soundness and stability of the international banking system, while maintaining sufficient consistency that capital adequacy regulation will not be a significant source of competitive inequality among internationally active banks”. The first pillar deals with the maintenance of regulatory capital requirements, the second pillar deals with the supervisory review of the first pillar. The third pillar, market discipline, contained disclosure requirements that are to be met by banks to make a bank‟s capital and its risks transparent for the market partakers. Pillar 3 has introduced new disclosure requirements, which include information about capital structure, capital adequacy, risk management and risk measurement. The Basel II Final rule (Basel II Rules) was published on December, 2007. It established the requirements for the implementation for US banks and includes the disclosure requirements under Pillar 3. Banks have to successfully complete a qualification period before receiving regulatory approval. Furthermore there is a Basel III accord which aims to further enhance the function of the three pillars, but is not fully developed. The Basel III accord proposed the implementation of broader and deeper disclosure standards, ensuring that market participants are able to better assess the risk which a bank may be exposed to. The SEC has also been giving plenty of attention to investigate the inadequate risk disclosures in corporate filings, since the financial crisis (Johnson, 2010).

20

The Basel Committee on Banking Supervision provides a forum for regular cooperation on banking supervisory matters. Its objective is to enhance understanding of key supervisory issues and improve the quality of banking supervision worldwide. It seeks to do so by exchanging information on national supervisory issues, approaches and techniques, with a view to promoting common understanding (http://www.bis.org/bcbs/)

21

Basel II is the second of the Basel Accords, which are recommendations on banking laws and regulations issued by the Basel Committee on Banking Supervision

(12)

9 3. LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT

3.1 Prior literature on credit ratings

The existences of Credit Rating Agencies are highly dependent on the quality of the service they provide. Their aim is to provide information about the probability of default of a bond or other financial security. Their ratings are intended to provide informational value to the financial markets. There are different views on how CRAs contribute by providing informational value. Hickman (1958) was one of the first researchers that found a positive relationship between ratings and default rate, which indicates that CRAs accurately rated the bonds that went in default. Through the 1980‟s and 1990‟s various researchers have shown that credit ratings are strongly correlated with credit spreads22 and equity prices (e.g., Ederington et al., 1987; Cantor et al, 1997; Goh and Ederington, 1993; Kao and Wu, 1990)23. Their results showed that credit ratings had informational value for market participants as they are on par with the market reactions. However, contradicting results were found when Kliger and Sarig (2000) tested the informational value of bond ratings; there was evidence that rating information did not affect firm value. Their results showed that upgrades (downgrades) of a credit rating leads to a rise (fall) in debt value while the equity value falls (rise) at almost an equivalent value. Further research concerning informational value of credit ratings (Perraudin and Taylor, 2004) shed some light on the contradicting results found in previous research. They found that although at first instance credit rating and market-based credit opinions vary, the difference disappears when one controls for non-credit factors, such as risk premiums, liquidity or tax effects. They concluded that credit rating and bond prices share similar attributes to credit risk and any difference are often temporary. There is also research that solely focused on the informational value of negative announcements or downgrades. There is evidence that negative announcements or downgrades causes significant (share)price or spread movements (Hand et al., 1992; Ederington and Goh, 1998) and even before the rating is expected the bond and equity prices decline or rise sharply in

22

The spread between Treasury securities and non-Treasury securities that are identical in all respects except for quality rating.

23

For a broad review of all empirical evidence between credit rating and market reactions see the overview of related studies in the paper of Norden and Weber (2004)

(13)

10 anticipation, as investors may be concerned about the same factors that determine ratings (Cantor, 2004). When the relation between earnings announcements and credit ratings is examined, Dichev and Piotroski (2001) found that there are differences in returns at subsequent earnings announcements of stocks, but are partly due to incomplete adjustment of information. Their research showed that stocks with rating upgrades

outperformed stocks with rating downgrades for up to a year, but no difference was found after that year. Their results indicate that bond rating changes has no effect on the long-run stock returns. However, upgrades and downgrades of ratings may influence a firm‟s financing decisions, showing that firm‟s may have great incentive to influence the rating (Kisgen, 2006). Jorion et al (2005) found that even though earlier research showed insignificant market reactions to rating upgrades, the reaction became significant after Regulation FD24 came into effect. Ederington et al (1998) explored the difference in information supplied to the market from CRAs and earnings analysts. They concluded that the information available which influenced downgrades, were partially available to analysts and the market. However, downgrades are still new information for the market as there is evidence of negative returns by announcements of downgrades. These results show that even though it appears that analyst have the same information as CRAs, the market (or a portion of the market) still see credit ratings as new information. Evidence from the CDS25 market show that negative actions26 against a rating has significant effect on the price of a CDS, but only when there is a negative review and not for rating

changes or outlook changes (Hull et al, 2004; Norden and Weber, 2004). Because CDS can be seen as the market price for the risk which a credit rating is intended to measure (Cantor, 2004), the results from the two above mentioned papers show strong evidence that credit ratings are information rich.

24

Regulation Fair Disclosure: a regulation that was promulgated by the SEC in August 2000. The rule mandates that all publicly traded companies must disclose material information to all investors at the same time. An exception was made for the information a firm may provide to a credit rating agency, only if this information is to help assess the credit rating.

25

Credit Default Swap: Is a swap contract in which the buyer is protected against the case of a credit instrument experiencing a credit event. Their price is solely based on credit risk, unlike equity and bond prices which are affected by credit risk and non-credit risk.

26

(14)

11 Although some of the above mentioned literature shows that credit ratings provide

informational value, some researchers have questioned the quality of CRAs‟ assessments. Some researchers did find evidence which suggests that CRAs do not provide the

qualities that are expected of them; however some are of the opinion that some critiques are unjust. Ang and Patel (1975) found that Standard & Poor‟s ratings had weak powers in predicting financial distress (e.g. incidence of default or loss of market value due to deterioration of the firm‟s prospects). Some researchers analyzed possible biases between solicited and unsolicited credit ratings and found that unsolicited ratings are indeed lower even after adjusting for weaker financial profiles (Poon, 2003; Poon and Firth, 2005; Roy, 2006). These results against CRAs may be attributed because they use more soft information27 when there is a relationship between the CRA and issuer. This soft

information is mostly only given to CRAs and not disclosed openly (Butler and Rodgers, 2003). Blume et al (1998) showed that through the 1970‟s to the 1990‟s, CRAs have improved their standards for the better, resulting in a prevalence of downgrades over upgrades in the late 1990‟s. When testing for possible conflict of interest (which is a common critique towards CRAs) Covitz and Harrison (2003) concluded that rating changes do not appear to be influenced by conflicts of interest but rather that CRAs are motivated by reputation-related incentives. Butler and Rodgers (2003) also found that there is no indication for conflict of interest between bond issuers and the CRAs. There has been some criticism towards CRAs about reacting slowly towards changes in credit rating quality, but Altman and Rijken (2004) found that CRAs base their rating on a “through the cycle” methodology28

which looks at the long-term perspective. They believe that because investors are short-term oriented, they perceive that the CRAs do not timely adjust their ratings. Amato and Furfine (2003) also concluded from their research, based on credit ratings for all firms from Standard and Poor‟s, that little evidence is found of pro-cyclicality29 in ratings. They also found evidence that CRAs take the long-run perspective in their rating. However, they did find evidence that when the CRAs do

27 _{Soft information is relevant information that comes from personal interaction with issuers that is not}

readily available for unsolicited ratings (Butler and Rodgers, 2003).

28

Is a rating methodology that is based on stress scenario. The rating is insensitive to the credit quality cycle and is long termed oriented (Carey and Hrycay, 2001).

29

Is a condition of positive correlation between an economic indicator and the overall state of the economy.

(15)

12 change their ratings, the rating overreacts. This overreaction is positively correlated with the state of the economy. For example, when they downgrade during a weakened

economy, the downgrade is over exaggerated.

Further research has looked at which factors affect credit ratings, since a concrete description of the methods used is not available. There has been research which focused on the effect of governance mechanisms and analysts have on credit ratings. Bhojraj and Sengupta (2003) found evidence that governance mechanisms may lead to higher bond ratings and lower bond yields. However, concentrated institutional ownership structure has an adverse impact on bond yields and bond rating. Ashbaugh-Skaife et al. (2004) investigated whether credit ratings were influenced by strong or weak governance. Using Standard & Poor‟s framework for evaluating corporate governance structures, they found that a variety of governance mechanisms do help explain credit rating. They provide evidence that a firm with optimal governance nearly doubles its chances of receiving an investment grade credit rating, which can lead to debt cost savings. Kuang and Qin (2011) revealed that a CEO compensation package affects a firm‟s credit rating. The more risk-taking incentives in the CEO compensation package, the lower the credit rating. By using a sample of 8,189 firm year‟s observation they concluded that rating agencies include risk-taking incentives in their risk assessments of the firm. They also show that the firm itself will try to influence their rating by decreasing direct risk-taking incentives, when large corporations have great concern about their rating. When

exploring the effect that analysts have on credit ratings, Cheng and Subramanyam (2008) found that analyst following is negatively associated with a firm‟s credit rating. Firms with a strong information environment and controls are less affected by analysts following. This result illustrates that “equity analysts following” has a major impact on credit rating when information quality is low.

Much is known about credit ratings and credit rating agencies, but there is no research which focused on the effect of risk disclosure quality on credit ratings. In the following paragraph I will give a brief summary of what is known about risk disclosure and then proceed with developing my hypothesis.

(16)

13 3.2 Prior literature on risk disclosure

Ever since the separation between ownership and management there have been agency problems. According to the agency theory an agent and the principal have

different interests, which lead to the problem of asymmetric information between the two Jensen & Meckling (1976). This conflict of interest may lead to the agent taking actions that are in conflict with the interest of the principal (Fama & Jensen, 1983; Shleifer & Vishny, 1997). This agency problem brings with it various agency costs (e.g., costs of producing financial statements and costs associated with aligning interests). According to Verrecchia (1983), information asymmetry between firm‟s management and shareholders creates a demand for disclosure and provides an incentive for firms to disclose

information because the value of additional information is great. It is believed that corporate disclosure is critical for the functioning of an efficient capital market (Healy & Palepu, 2001), but according to Akerlof (1970) the market is not efficient and there is a chance for „lemons‟ in the market. Management can distinct themselves as not being „lemons‟ by disclosing qualitative information. A side from mandatory disclosures, firms can distinct themselves by providing voluntary disclosure in their description of their risks. There are various reason why firms may voluntary disclose information; 1) improves stock liquidity (e.g., Diamond & Verrecchia, 1991; Kim & Verrecchia, 1994); 2) reduces cost of capital (e.g., Botosan, 1997); and 3) increases information

intermediation (e.g., Lang & Lundholm, 1999).

One can argue that evidence on disclosure literature would also count for risk disclosure since it‟s a specific kind of disclosure and when considering the banking industry, risk is probably the most important topic. Although risk disclosure has not been thoroughly researched compared to general disclosure, there is some research which contributed to literature on risk disclosure. There a few research that looked at reactions from investors on the subject of risk disclosure. Rajgopal (1999) examined the

relationship between oil and price sensitivity and proxies for commodity risk exposure. The paper provided evidence that investors will have difficulty in comparing results of the risk management activities when firms choose different disclosure formats. Although there are differences between the banking sector and oil and gas sector, this evidence provides insight into the investors‟ reaction. This also provides some evidence that there

(17)

14 is a chance that firms use impression management in their risk disclosure. The problem of different formats could be resolved by implementing a framework for mandatory

disclosure. One can argue that it would be beneficial to investors because one would expect similar disclosure formats across the industries. However, Solomon et al (2000) found that institutional investors do not favor regulated risk disclosure and preferred voluntary disclosure. The respondents of their research did agree stated that risk disclosure would help them in their portfolio investment decisions, thus revealing the importance of risk disclosure. They further argue that risk management and risk

disclosure must be illuminated in the same context, seeing that a risk management system is continuously active with identifying risks and determine their possible impact in order to respond accordingly. Jorgensen and Kirschenheiter (2003) also concluded that

mandatory risk disclosure lowers the firm‟s share prices relative to share prices in a voluntary disclosure regime.

Another stream of research focused on the various factors that determined risk disclosure. Helbok and Wagner (2006) focused on the determinants of operational risk reporting for banks. They found that banks with lower equity/assets ratio and or

profitability ratio disclose more information whereas those with higher ratio disclose less. The sample period that the paper took into account, was at a time that operational risk was not mandatory to disclose. The findings provide insight into what affects disclosing on operational risk. The lower a bank‟s leverage and the higher its profitability ratio, the less it will disclose. This may indicate that when a bank takes more risk in taking on more leverage, the bank may have an incentive to disclose about the risks, while when a bank is less profitable it may have more incentive to give more information about its risk to compensate for the less than desired results. The study shows that banks will use risk disclosure as a means to justify or compensate for other results than expected. In another study Dobler (2005) looked at the informativeness of risk disclosure and manager

incentives. He argues that risk reporting is subjective and is subject to manipulation from management. Nonetheless, he argues that risk management system provides the

information for external reporting and thus risk management determines risk disclosure. He also argues that risk report may be used as instrument for risk handling, given that managers may manipulate the information for other purposes than informing. Thus

(18)

15 management‟s may have an incentive to manipulate what it will disclose about risk and risk management.

There is also research that focused on exploring if risk disclosure had positive or negative effects. On the negative side, Homölle (2003) found that risk reporting does not lead to less risk exposure and bank run probability. Risk disclosure may also see a rise in the bank‟s exposure, it may even reduce welfare. Cordella and Levy Yeyati (1998) found in their study that when a bank doesn‟t have control over its risk portfolio, risk disclosure increases the probability of a bank failure. They argue that when risk is exogenous, disclosure no longer affects risk-taking behavior but still induces negative feedback on the chance of a bank failure by allowing deposit rates to change. Disclosing less in bad times may therefore improve the chance of survival.

There are also researches which point out the positive effect of more disclosure. Welker (1995) investigated the relation between disclosure policy and liquidity in equity markets and found that well-regarded disclosure policy reduces information asymmetry and increases liquidity in equity markets. He proved that disclosure is good for the overall welfare as it helps with liquidity in markets. Another study by Kravet and Muslu (2010) also found positive effect of risk disclosure. They found that risk disclosures are

associated with increases in the number of earnings forecast, increases in trading volume and improved forecast accuracy. They concluded that overall, disclosing on risk provides useful information about a firm‟s uncertainties. Linsley and Shrives (2000) found in their study that reporting on risks may lead to more accurate risk management for firms. This illustrates that reporting helps a firm assess internally in greater detail about the risks and thus helping them to create measures to mitigate those risks. Further results in the study by Cordella and Levy Yeyati (1998) showed that if a bank has control over its risk portfolio, risk disclosure reduces risk-taking incentives focused on how risk disclosure affected risk taking incentives. Disclosure level and risk disclosure also benefits firms with their share prices and cost of capital. Botosan (1997) examined the association between the level of disclosure and the cost of equity capital, based on voluntary disclosure. The research found a negative relationship between disclosure measure and cost of equity for firms having a low analyst following. However, no relation was found for firms having a high level of analyst following. The findings indicate that when a firm

(19)

16 discloses more information to the public, it is rewarded by receiving lower cost of equity capital. Jorgensen and Kirschenheiter (2003) also found similar results from their study when they looked at risk disclosure in particular. They found that the risk premium of a firm is negatively related to the amount of disclosure. This shows that firms that disclose risk have higher share prices than those that do not. This result also shows the market appreciates such information and rewards those that disclose more. Equity prices are not the only security that is affected by risk disclosure. Sengupta (1998) looked at the link between a firm‟s overall disclosure quality and its cost of debt financing and found evidence that firms with higher ratings on disclosure quality from financial analysts, receive lower effective interest i.e. lower cost of debt. The results show that firms that consistently provide quality informative disclosures have lower chance of withholding relevant information, which leads to a lower risk premium. In a more recent study Tassel (2011) found that the incentive for a bank to disclose information is negatively related to a bank‟s capital ratio. He argues that a bank holding private information may choose to disclose information about its clients (information about credit risk) in order to convince external investors that it has a low risk-profile. Tassel‟s findings about banks are in line with findings from Botosan (1997) and Sengupta (1998).

In summary, we know that CRAs base their ratings partially on information about market risk, credit risk and operations risk (Standard & Poor‟s, 2005). Homölle (2003) and Cordella and Levy Yeyati (1998) argued that more risk disclosure may lead to more risk exposure which may lead to a negative relation between risk disclosure quality and credit ratings. However, this paper argues that since credit ratings give an indication of risk default, risk disclosures are an important factor which CRAs may take into

consideration when evaluating their rating. As described above, there is evidence that illustrates a positive relation between disclosure quality and cost of debt (more disclosure leads to less costs). Furthermore, since cost of debt is positively correlated with credit ratings I will expect the same relation between credit ratings and risk disclosure. Based on the above discussion, my hypothesis is as follows:

(20)

17 H1: There is a positive relation between the credit rating that a bank receives

and the quality of the bank’s risk disclosure.

4. RESEARCH DESIGN 4.1 Variables

Credit Ratings

This study investigates the relationship between credit ratings and risk disclosure quality for banks. Prior literature (e.g., Ashbaugh et al., 2006; Bhojraj and Sengupta, 2003; Cheng and Subramanyam, 2008; Kisgen, 2006, 2007; Kuang and Qin, 2011) which focused on credit ratings as a variable, used Standard & Poor‟s issuer credit ratings to proxy default risk. The credit ratings from S&P that I will use are derived from

COMPUSTAT. To avoid relying solely on one rating agency I will also use credit ratings provided by Moody‟s30_{. For both S&P and Moody‟s I will look at the Long-Term}

Domestic Issuer Credit rating. Similar to Francis et al. (2005), Cheng and Subramanyam (2008) and Kuang and Qin (2011) I compile the possible ratings into the range of 1 to 20. Table 2 and Table 3 summarize the recoding of the credit ratings and also provide the distribution for both CRAs. For S&P the rating A (16.7%) has the greatest weight in the full sample while for Moody‟s, the greatest weight is the rating A3 (20.3%). This distribution is not consistent with prior studies (e.g., Cheng and Subramanyam, 2008; Kuang and Qin, 2011). The inconsistency in distribution may be attributed to the fact that this research is focused on banks and prior studies focused on non-financial firms. There is also a summary of descriptive statistics of over the years in Table 7, for the CRAs. From the table we can see that on average for each year, S&P gave out lower ratings than Moody‟s. The table also illustrates that ratings on average fell between the years 2005 and 2009. This fall in credit ratings was probably caused by the financial crisis or through more stringent assessment by the CRAs.

30

(21)

18 Risk Disclosure Quality

Banks, like most financial institutions, have different risk exposure than non-financial firms. Linsley and Shrives (2006) defined financial risk, operations risk, strategic risk etc. as important risk factors. Banks differ from other firms because they are risk management entities and can make significantly different types of risk disclosure (Bessis, 2002). Quality is a subjective interpretation and there has been various ways which disclosure quality has been quantified. Some define disclosure quality as the ease of readability for the investors (e.g., Hopkings, 1996), while some have argued that quantity is a proxy for quality (Belkaoui and Kapik, 1989; Cooke, 1992; Richardson and Welker, 2001).

However, Beretta and Bozzolan (2004) argues that because disclosure of risk is

intrinsically narrative, the quantity of disclosure is not a satisfactory proxy for the quality of disclosure. They are of the opinion that quality of disclosure depends on both the quantity of information and the richness of its content.

Since this study looks at risk disclosure quality I will look at the richness of the content which banks disclosed about their risks as well as the quantity disclosed.

Therefore I measure risk disclosure quality (RDQ) by taking the average between RQTi, DENi and DPTi. The factor RQT is a proxy for the quantity of information about risk disclosed and the factor DEN proxies the ratio in which risk information is disclosed compared with other information disclosed in the MD&A section of an annual report. The last factor DPT looks at the amount of topics discussed about the specific risk factors. This measurement method is primarily based on the model used by Beretta and Bozzolan (2004) except it does not include the fourth factor they use which is outlook profile index for a company. The fourth factor is excluded because the factor DPT already looks at some topics which are based on looking forward information (e.g., risk mitigation). Further details on the estimation of RDQ are presented in Appendix A.

Table 5 shows the descriptive statistics of the measure of risk disclosure quality. Panel A shows that the DPT index has the largest mean (0.58) but also has the largest standard deviation (0.20) and the DEN index has the smallest mean (0.29). The RDQ index has a mean of 0.41 and a standard deviation of 0.12. Panel B illustrates information about RDQ and its components through the years. All of the components saw a rise through the years and there was an evident growth in the RQT index (66% between 2005

(22)

19 and 2009). Panel C illustrates the descriptive statistics of DPT and its components. The sample firms disclosed more information about their interest rate risk (mean=3.34) than the other risk factors31_{and the least about operational risk (mean=1.56). Moreover, panel}

D shows the descriptive statistics for DPT and its components over the years. The results show that through the years, the five risk factors saw a rise in the items being disclosed.

4.2 Research design

This study is aimed at finding a relation between credit ratings and risk disclosure quality. In order to find a relationship between the two variables I run the following regression model:

(1)

where RATING proxies credit ratings from S&P and Moody‟s and it is measured based on that higher (lower) value of RATING signifies higher (lower) default risk, which is based on prior research (e.g., Cheng and Subramanyam, 2008; Kuang and Qin, 2011);

QUALITY proxies the quality of the information which firm i discloses about risk. In this

research QUALITY will be measured by using RDQ and its components RQT, DEN and

DPT. The variable CONTROL in the equation is a proxy of control variables.

In the hypothesis I predict to be negatively related to SPRATING and MRATING, expecting that when a bank discloses more qualitative information about their risk they will receive a higher rating. The hypothesis is based on the argument that more disclosure leads to lower cost of capital (Goh and Ederington, 1993; Botosan, 1997; Kirschenheiter, 2003; Kravet and Muslu, 2010). In order to develop a strong regression model I will use various control variables. These control variables, which are described in Table 1, are derived from prior literature that found significant relations with credit ratings (Poon and Firth, 2005; Ashbaugh et al., 2006; Cheng and Subramanyan, 2007; Kuang and Qin, 2011). In a variety of prior studies, there were evidence of positive relations between credit ratings and proxies for performance (ROA), size (LNASSETS), loans loss reserve ratio (LLR/GL) and equity to assets ratio (ETA). Prior studies also found a negative

31

(23)

20 relation between loans and total assets (LTA) and credit ratings. Poon and Firth (2005) also introduced the control variable net interest margin (NIM), this variable measures how much a bank earns on their loans. I expect that NIM will have a positive relation with credit ratings, thus the larger the interest margin on a bank‟s loan the more profitable a bank will be which can lead to a better rating.

4.3 Sample and descriptive statistics

The sample of this research consists of the years from 2005 to 2009 for banks from the United States. To avoid sovereign risk and the effect of culture on disclosure methods, I will only take banks that originate from the United States of America. The Basel II Accord was first published in June 2004, thus I expect that banks would start disclosing according to the third pillar of the accord, no later than fiscal year 2005. If I take a sample for years prior to the first publication of the accord, I would expect variety in the methods banks would disclose about their risks. There are a total of 169 different banks which have a S&P32 rating in the United States. This study looks at banks which are partially dependent on customer deposits, so I do not consider investment banks in our sample since they have a different business model and their exposure to various risks will differ. To avoid culture differences I will only take banks that are headquartered in the United States and thus are not subsidiaries of non US financial holdings. To quantify risk disclosure quality I will need to be able to have access to publicized annual reports and will thus take only publicly listed banks in my sample. Since Chung and Jo (1996) argued that the New York Stock Exchange (NYSE) has greater visibility and more stringent listing requirements and analysts tend to follow NYSE listed firms more intensively, this paper will focus on NYSE listed banks. After considering the above mentioned criteria, I have a sample of 26 banks and 126 firm years. Table 23 provides a list of the US banks which are in the sample.

In Table 6 I report descriptive statistics and in Table 8 correlation matrix between the relevant variables used in the regression. The descriptive statistics table show that

SPRATING has a mean (median) of 7.09 (7) while MRATING has a mean of 6.29 (6), which represents an average rating of A- and A3 rating respectively. The variable RDQ

32

(24)

21 has a mean (median) of 0.41 (0.44) and ranges from 0.14 to 0.69 with a standard

deviation of 0.14. Although the control variables deviate somewhat from Poon and Firth (2005), the difference is not that significant. Poon and Firth (2005) looked at banks from different countries which may be the reason for that difference.

To avoid the phenomenon of multicollinearity between the independent variables, which may cause coefficient estimates to change erratically in response to small changes in the model of the data, I use the correlation matrix in Table 8 to look for signs of highly correlated variables. The two independent variables SPRATING and MRATING are highly correlated (0.890) and thus will not be used simultaneously in the same regression. The control variables all are correlated with value of less than 0.70, which in general does not raise the possibility of a multicollinearity problem (Newbold et al., 2007). When I analyze the results of the regression I will test the regression model, using the variable inflation factor (VIF), to see if I need to remove a specific variable from the equation. The variable RDQ is positively correlated with both SPRATING and MRATING, but is not significantly related.

5. EMPIRICAL RESULTS

5.1 The impact of risk disclosure quality on credit ratings

In this paper I examine the relation between the risk disclosure quality of US listed banks and the credit ratings they receive from CRAs. The regression for Equation (1) is illustrated in Table 9. The variable of interest is RDQ, which represents the risk

disclosure quality of the banks from the sample. The dependent variable is SPRATING and MRATING, which are credit ratings from Standard & Poor‟s and Moody‟s. The other variables in the table represent the control variables.

The results in Table 9 show that adjusted R-squared is high with a value of 0.751 (SPRATING) and 0.675 (MRATING). As mentioned in Section 4, I will test for the possibility of a multicollinearity problem using Variance Inflation factor (VIF)33. Table 10 illustrates the tolerance and VIF value for all of the dependent variables and we can

33

(25)

22 conclude that there is no evidence of a possible multicollinearity problem34_{. However the}

variable LNASSETS has a large impact on SPRATING and MRATING, with a standardized coefficient of -0.682 and -0.604 respectively (see table 10). This result shows that the size of a bank greatly determines the credit rating, and can dilute too much explanatory power of other independent variables. Therefore, I compiled a second table of results, without the control variable of LNASSETS, in table (12). By comparing table (9) and table (12) I will reduce the chance that the independent variable of interest (RDQ) is too heavily influenced by the proxy size. The R-squared of the equation without

LNASSETS is still high with a value of 0.460 (SPRATING) and 0.444 (MRATING). Although it may seem controversial to consider a model without LNASSETS after we see that R-squared is largely reduced, I need to clarify that I will look at both models (with and without LNASSETS) to assess if the proxy size does not dilute the explanatory power of RDQ. I will not ignore the model which includes LNASSETS but I will simply use both models to see if the variable RDQ still has the same relation with the dependent variables.

The regression results from both Table 9 and Table 12 show that risk disclosure quality has a significant impact on credit ratings. The coefficients on RDQ in both

SPRATING and MRATING model specifications are significantly. When LNASSETS is in the equation, the p-value for the coefficients on RDQ in both SPRATING and

MRATING models are smaller than 0.01. While when LNASSETS is not in the equation, the p-value is <0.10 and <0.05 respectively. The results find evidence that contradicts my prediction. According to the results, the higher the quality of risk disclosure from banks, the lower the rating which a bank would receive. In table 10 we can see that RDQ has a standardized coefficient of 0.189 and 0.200 for SPRATING and MRATING respectively when including LNASSETS in the equation. When excluding LNASSETS, RDQ has standardized coefficient of 0.098 and 0.126 for SPRATING and MRATING, respectively (see Table 11). These results suggest that there isn‟t a big difference between the two CRAs when considering the risk disclosure quality of banks.

34 _{O’Brien (2007) argues that the rule of thumb for the tolerance value must be greater than 0.1 and the}

rule of thumb for VIF value is that the value must be smaller than 5, otherwise there is the possibility of a multicollinearity problem

(26)

23 When observing the control variables (see Table 9 and 12), the results are mostly

consistent with prior findings (e.g., Poon and Firth, 2005). There is a significantly negative association between ROA and the rating proxies (p-value<0.01), in both the equation with and without LNASSETS. Those results conclude that a bank receives a higher rating when a firm performs better. The control variable that measures the ratio between equity and total assets (ETA) is significantly positive related to the S&P credit ratings with p-value<0.01 and significantly positive related to Moody‟s credit ratings with p-value<0.05 when LNASSETS is out of the equation. This result suggests that the larger the ratio between equity and total assets a bank receives, the lower its credit rating. The results further show that the variable looking at the ratio between loan loss reserves and gross loans (LLR/GL) only has a significant negative relation with S&P credit ratings with a p-value<0.10, which does not suggest a strong relation with credit ratings.

However, the ratio between loans to total assets (LTA) does show significant and positive relations with both SPRATING and MRATING with a p-value<0.01 for both dependent variables. This indicates that how larger a portion loans take in the total assets of a bank, the lower its credit rating. Lastly, the control variable of net interest margin (NIM) shows a negative relation with the credit ratings but does not have significant results.

In summary, the variable of interest (RDQ) found significant results in relation with Standard & Poor‟s credit ratings and Moody‟s. However, the results contradict my hypothesis, by showing a negative relation between the quality of risk disclosure and the credit rating of a bank. Suggesting that the higher the quality of risk disclosure the lower the rating of which a bank receives. A possible explanation for these findings can be attributed to the possibility that credit ratings affect risk disclosure quality thus a reverse causality. Banks which are of the opinion that they have a low credit rating would try to compensate for that low rating by disclosing more information and therefore make up for the higher cost of capital (Botosan, 1997). I will conduct a Granger causality test in order to try to verify this relation in the next section.

5.2 Impact of the components of Risk Disclosure Quality on credit ratings

Even though a significant result was found for RDQ, it will be interesting to analyze further how different aspect of risk disclosure quality affect SPRATING and MRATING

(27)

24 separately. I employed three dimensions of RDQ (i.e., RQT, DEN and DPT) which

determined the risk disclosure quality of a bank. Table 13 shows the results when LNASSETS is a part of the equation, while Table 14 shows the results for the equations when LNASSETS is excluded as a control variable. When observing the results for the relative quantity index (RQT) I found a strong significant positive relation with both SPRATING and MRATING. When looking at the density index (DEN) I found a positive relation with SPRATING (p-value<0.01) and MRATING (p-value<0.05) but only when the equation includes LNASSETS. When LNASSETS is taken out of the equation, there are no significant results. The variable that measures the depth index (DPT) show mixed results. When LNASSETS is included in the equation, there is a significant

(p-value<0.05) positive relation with MRATING but no significant relation with

SPRATING. However, when excluding LNASSETS there is a significant (p-value<0.05) negative relation with SPRATING and no significant results with MRATING.

In conclusion, the results indicate that RQT shows a consistent relation with SPRATING and MRATING. The amount disclosed about risk, corrected by size, is negatively related to credit ratings. The more a bank discloses in its MD&A part of its annual report about risk, the lower the credit rating that it will receive. The factor DEN found significant positive relation with SPRATING and MRATING only when

LNASSETS is included in the equation. A positive relation means that the more reported about risks and risk management in the MD&A, the lower a bank‟s credit rating. When looking at the factor DPT I found evidence that Standard & Poor‟s credit ratings are positively related with the depth of which a bank discloses about risk, meaning that when a bank discloses about a greater variety of topics, the lower its rating. Although there are significant results for DPT, it is not consistent when you include and exclude the variable LNASSETS. Since RQT is the only factor which found significant results, I conclude that it is the key factor for determining the negative relation between risk disclosure quality and credit ratings.

5.3 Robustness analysis

In what follows, I examine whether the previous finding are robust to changing estimation method and considering moderator effect of negative earnings. In my previous

(28)

25 model I used a linear regression model to see if there is a relationship between credit

ratings and risk disclosure quality. I used the values 1 through 20 in order to rank the credit rating scores, following prior research (e.g., Francis et al., 2005; Cheng and

Subramanyam, 2008; Kuang and Qin; 2011). I will use an ordered logit model in order to relax the assumption that the distances between the values of the credit ratings are

uniformly distributed.

Table 15 shows the results when including LNASSETS and Table 16 shows the results excluding the variable. When we look at Table 15 we see that the results are significantly in line with the results from the original linear regression for both SPRATING and MRATING. Table 16, which illustrates the ordered logit model excluding LNASSETS, shows no significant results regarding the variable of interest RDQ for both SPRATING and MRATING. Although no significant results were found when excluding LNASSETS, the R2 is significantly smaller and thus the model is less likely to predict the dependent variables when compared to the model which includes LNASSETS.

The second test is to adjust for the possible effects that negative earnings (loss) may have on both credit ratings and risk disclosure. Prior studies on credit ratings looked at the variable of negative earnings as a possible determinant of credit ratings (e.g., Ashbaugh-Skaife et al., 2006; Cheng and Subramanyam, 2008; Kuang and Qin, 2011). Therefore, I will add a dummy variable (DLOSS) to the equation and include an interaction term (DLOSSRDQ) of DLOSS and RDQ.

The results are presented in Table 17 and table 18. Table 17 shows that even when controlling for DLOSS, RDQ still has a significantly positive coefficient which is in line with results found in previous tests. However, as can be seen in Table 18 when excluding LNASSETS from the equation, there was no significant results regarding RDQ. The positive coefficient on DLOSSRDQ means that firm with negative earnings have a stronger relationship between credit ratings and risk disclosure quality compared to firms with positive earnings.

(29)

26 The third test will see if the negative relation between credit ratings and risk

disclosure quality is driven by banks that have negative earnings or have positive

earnings. Two subsamples are used, the first is a subsample for firms which had negative earnings and the second one for firms with positive earnings.

The results of the regression analyses are presented in Table 19 and Table 20. When including LNASSETS as a variable, banks with negative earnings show a significant positive relation between RDQ and MRATING (11.282). Banks with positive earnings also show a significant positive relationship between RDQ and with both dependent variables. SPRATING shows a coefficient of 3.384 while MRATING shows a coefficient of 2.541. The results show that the effect of RDQ on credit ratings is larger for banks with negative earnings than banks with positive earnings. When excluding LNASSETS as a control variable, the results for banks with negative earnings show a positive relation between RDQ and with both SPRATING and MRATING. The subsample of banks with positive earnings shows no significant results. As concluded in the previous test, banks with negative earnings have a stronger relationship between credit ratings and risk disclosure quality when compared to banks with positive earnings.

In summary, the robustness tests support the previous findings. The ordered logit regression showed the same results as the linear regression. When testing for the possible effect of negative earnings and adding a interaction term between negative earnings firm and risk disclosure quality, I also found results supporting the findings from Equation (1). The test also found that banks with negative earnings have stronger relationship between the variables RATING and RDQ. When using subsample for banks with negative

earnings and another subsample for banks with positive earnings, I still found significant positive results between RDQ and RATING. However, stronger results were found for the subsample of banks with negative earnings which shows that banks with negative earnings are the ones driving the positive relation between credit ratings and risk disclosure quality.

(30)

27 6. ADDITIONAL ANALYSIS

6.1 Causality Test

In the previous section I found results which contradicted my hypothesis and argued that there is a possibility that its credit ratings which affects risk disclosure quality because firms have the incentive to remedy low ratings by disclosing more (quality) information about risks. In order to control for lead-lag relations to establish causality between the variables RATING and QUALITY I will use Granger Causality test. Granger causality tests determine whether lagged information on a variable QUALITY provides any statistically significant information about variable RATING. Since the previous results found a negative relation between credit ratings and risk disclosure quality and I argue that it there may be a reverse causality, I will conduct the following test.

(2)

The results of the regression of Equation (2) are presented in panel A of Table 21. The F-test rejects the null hypothesis of no Granger causality between risk disclosure quality and credit ratings. The significant coefficients on the lagged value of SPRATING and MRATING thus show Granger causality in regard to RDQ. The result for both SPRATING and MRATING show that risk disclosure quality is negatively influenced by credit ratings (assuming a lower score for good ratings) and thus supports the argument that when a firm receives a low credit rating it tries to remedy by disclosing high quality risks information. I also tested for Granger Causality test with the variable RATING as the dependent variable and did not find the same strong significant results. The results can be seen in panel B of Table 21.

6.2 Rating Downgrades

To further support my argument that credit ratings affect risk disclosure quality because firms have an incentive to fix low ratings, I will look if rating downgrades have

(31)

28 an effect on risk disclosure quality in the subsequent years. Kisgen (2006) showed that

firms are concerned with their ratings and they adapt their capital structure decisions. Kuang and Qin (2011) also showed that rating troubled companies will adjust executive compensation by reducing CEO risk-taking incentives in option grants. This paper argues that firms concerned with their ratings will also try to change their risk disclosure in order to compensate or aim for a better rating. To support this argument I measure a bank‟s rating concern as SPDOWN and MDOWN. SPDOWN is coded one if the firm

experienced a downgrade from Standard & Poor‟s in the prior year and zero otherwise. MDOWN is coded one if the firm experienced a downgrade from Moody‟s in the prior year and zero otherwise. My analysis of downgrade concerns employs the following model:

(3)

where is are presented by RDQ and RQT which are proxies for quality and quantity of risk disclosure of firm i; RATINGDOWN are proxies for rating downgrades from S&P and Moody‟s for the a year prior and two year prior and CONTROL is a proxy for control variables which some were used in the paper from Miihkinen (2011). The paper looked at what drives risk disclosure quality and it found a relation between firm performance, size and leverage with the level of risk disclosure quality.

The regression for Equation (3) is illustrated in Table 22. The independent variables of interest are SPDOWN and MDOWN, which represent credit rating downgrades for two credit rating agencies in prior years. The dependent variables are RDQ, RQT, DEN and DPT which measures the risk disclosure quality of a bank and measures the relative quantity of risk information disclosed. I test all four factors to see if credit rating

downgrades have an effect on the quality of risk disclosure and/or the relative quantity of risk information disclosed. The results in Table 22 show that rating downgrades from Moody‟s show significant positive relationship for all four factors. The factor RQT shows highly significant relation with the credit rating. A positive relation between the variables shows that when credit rating downgrades occurs in prior years, banks disclose more information and more qualitative information about their risks and how they

Credit Ratings and Quality of Risk Disclosure in the US Banking Industry

Credit Ratings and Quality of Risk Disclosure in the US Banking

Industry

Table of Contents