Credit Rating Agencies: Is competition beneficial?
Empirical evidence from structured finance in the U.S.
MSc Economics: Markets & Regulation
Author:
Max C. Kuipers 10659110
Supervisors:
Prof. Maarten Pieter Schinkel Timo Klein
Statement of Originality
This document is written by Max C. Kuipers who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements
I would like to thank my supervisors, Timo Klein and Prof. Maarten Pieter Schinkel for their guidance on writing this paper and providing the necessary feedback and comments.
Abstract
JEL-codes: G24, L13, L14, L15
Key words: Credit ratings, competition, conflict of interest, panel data, reputation.
“Credit-related and other analyses, including ratings, and statements in the Content are
statements of opinion as of the date they are expressed and not statements of fact.”
- Disclaimer Standard & Poor’s Global Ratings, 2018
Credit rating agencies decrease the information asymmetry between investors and issuers by providing credit ratings. However, these credit rating agencies were criticized for contributing to the recent financial crisis by providing ratings that were too high for the actual risk of the asset. With increased competition and thereby lower future rents, the reputation-based incentives might not be overriding, resulting in these inflated ratings. This paper uses the entrance of DBRS into the U.S. market of credit ratings for structured finance to empirically assess the effect of increased competition on the level of credit ratings provided by the incumbents. We find strong significant evidence indicating an increase in the level of credit ratings when competition is fiercer. Our results suggest that increased competition in the market for credit rating agencies might not be beneficial, as it possibly increases the information asymmetry even further.
1 INTRODUCTION 6
2 THE MARKET OF CREDIT RATINGS 8
2.1 PROPERTIES OF THE MARKET OF CREDIT RATINGS 8
2.2 CONFLICT OF INTEREST FOR CRAS 10
2.3 COMPETITION AMONGST CRAS 12
2.4 STRUCTURED FINANCE 13
3 METHODOLOGY 14
3.1 DATA SPECIFICATION 14
3.2 BASELINE MODEL 17
3.3 ALTERNATIVE MEASURES FOR COMPETITION 18
3.4 ORDERED PROBIT 19
4 EMPIRICAL RESULTS 20
4.1 DESCRIPTIVE STATISTICS 20
4.2 RESULTS MODEL 1 22
4.3 RESULTS MODEL 2 AND 3 24
4.4 RESULTS MODEL 4 26
5 DISCUSSION AND LIMITATIONS 27
6 CONCLUSION 29
REFERENCES 31
APPENDICES 33
APPENDIX I: TYPE OF STRUCTURED FINANCE AND TYPE OF COLLATERAL 33
APPENDIX II: MARGINAL EFFECTS MODELS 4.1-4.3 35
1 Introduction
Credit rating agencies (CRAs) play an important role in the financial market. They provide ratings on financial products that indicate the probability of default. With credit ratings, the information asymmetry between the investor and the issuer can be reduced. Furthermore, some regulatory measures regarding financial institutions are dependent on the ratings provided by officially licensed CRAs. Accordingly, CRAs also have certain responsibility and authority within the financial market aside from constituting a key channel of information distribution alone.
Nevertheless, in the years after the financial crisis, CRAs have received considerable critique for inaccurately rating financial assets and consequently contributing to the subprime crisis (Mathis et al, 2009). Considering most of the revenues of CRAs are coming from the issuer of credit, a possible conflict of interest for the CRA arose. Namely, in the event of a low rating, the issuer could opt not to pay for the rating and search for a more favourable rating. As a consequence of this concern of losing income, CRAs mighthave an incentive to rate with an upward bias. This in turn increases the information asymmetry again and consequently diminishes the quality of the credit rating.
Even though CRAs depend on their reputation to rate correctly, this dependence could be reduced when future rents are lower due to increased competition (Becker and Milbourn, 2011). As a consequence, the effect of competition on the quality of credit ratings has been an emerging area of research.1 For example, Becker and Milbourn (2011) use the fact that Fitch gained
significant market share in the early 2000’s to empirically assess this effect of increased competition on the level of firm and bond credit ratings of the incumbents (Moody’s and Standard & Poor’s). Nevertheless, they do not examine the effect for structured finance, because Fitch was already substantially large in this market. Interestingly, the market for structured finance has grown significantly and rapidly prior to the financial crisis of 2007. Even more so, the ratings of structured finance were particularly called into question after the financial crisis.
For decades, the market for credit ratings has been dominated by three firms: Fitch, Moody’s and Standard & Poor’s (S&P). However, Dominion Bond Rating Service (DBRS) was granted the aforementioned official license by the Securities and Exchange Commission (SEC) in
1 E.g.: Doherty et al. (2012) assessed the effect of competition on information content of ratings of insurers’
ratings;Bolton et al. (2013) theoretically model the market for CRAs and assess, amongst others, the efficiency in the market for monopoly and duopoly. Section 2.3 elaborates more on this.
2003. This allowed DBRS to enter the market specifically for credit ratings of structured finance, as most investors in this particular market are often subject to credit ratings-based regulation. Figure 1 illustrates the growth of DBRS market share - based on output - from 2004 to 2016. With the entrance of DBRS, the competition in the market of credit ratings of structured finance changes, allowing us to assess the effect of this increased competition on the quality of credit ratings of the incumbents: Fitch, Moody’s and S&P.
Figure 1: Market share of DBRS based on output from 2004-2016.
Note: Market shares are calculated as the percentage of DBRS ratings of the total amount of ratings in a year. Calculated with data
obtained from Bloomberg.
By following Becker and Milbourn (2011), we use a ratings scale ranging from 0 to 28 (28 being the lowest credit risk, 0 the highest). With this scale, the effect of increased competition on the level of credit ratings by the incumbents can be examined. In this analysis, we consider a general increase in credit ratings level (ratings inflation) as a decrease in quality of the ratings, since credit ratings should have a stable interpretation, especially for naïve investors.
With the use of panel data on credit ratings of structured finance obtained via Bloomberg, we empirically examine the effect of increased competition on the level of credit ratings provided by the incumbents: Fitch, Moody’s and S&P. By using a scale assigning a numerical value to each rating category, our results suggest an increase in ratings level with increased competition, which
is initially measured by the market share of DBRS based on output. We extend the analysis with two alternative measures for competition: the Hirschmann-Hirfandahl Index (HHI) and the market share of DBRS based on issuance volume in dollars. Finally, we perform an ordered probit regression, to allow the difference between rating categories to vary across the ratings scale, instead of treating every step equally. Again, we observe increased levels of ratings when competition is fiercer.
This study contributes to the existing literature with respect to competition and CRAs. Prior research mainly focussed on corporate bonds and firm credit rating, whereas this study concentrates on structured finance. However, we observe some limitations in this analysis. Firstly, our measures for competition might be noisy and this possibly biases our estimates. Furthermore, there is the possibility of omitted variable bias. If a factor influences both higher credit ratings and a higher market share of DBRS, our estimates are biased.
The rest of this analysis is organized as follows. Section 2 first elaborates on some properties of the market of credit ratings as well as existing literature regarding competition and credit ratings. Section 3 describes the methodology and variables used in the analysis followed by Section 4, which presents our results. Some limitations of this research are discussed in Section 5. Finally, we conclude in Section 6.
2 The Market of Credit Ratings
This section discusses the relevant literature for this analysis. First, some general properties of the market of credit ratings are explained. Secondly, literature with respect to the conflict of interest for CRAs is summarized, followed by research regarding competition and credit ratings. Finally, structured finance is discussed.
2.1 Properties of the Market of Credit Ratings
In finance, there is the issue of information asymmetry between the issuer and the investor about the risk associated with a security or bond. Particularly, the investor often has substantially less information about the probability of repayment than the issuer. CRAs reduce this information asymmetry by gathering information on these financial products about the likelihood of default (White, 2010). Following a scale, where each rating implies a different credit risk, the quality of
the bonds can be interpreted (see Table 1 for the different ratings). Therefore, the ratings provided by CRAs, are a key channel of information dissemination in financial markets (Becker and Milbourn, 2011).
The importance of CRAs is also mentioned by Boot et al. (2006). They state that credit ratings play an essential role as “focal points” for firms and investors. Moreover, they theoretically show that credit ratings serve as a coordinating mechanism for the beliefs of investors. Likewise, Bannier and Hirsch (2010) state that the economic role in the financial market of CRAs is extended even more with the introduction of the “watch list procedure” in 1991.2
Regulators acknowledge this key role for CRAs, considering the regulatory measures in financial markets rely heavily on these credit ratings. From 1936, financial institutions in the U.S. such as pension funds and banks are prohibited from investing in “speculative” or “junk” bonds. Thus, they are only allowed to hold “investment-grade” financial assets, which is equal to for example BBB or higher on the S&P scale (White, 2010; Coval et al., 2009).3 Others, such as
insurance companies, have capital requirements based on the ratings of financial products they hold. More recently, the “Credit Rating Agency Reform Act” (2006) and the Dodd-Frank Act (2010) supplemented the regulation for CRAs. Both Acts mostly address the transparency issues of CRAs such as their rating methodologies and results and the potential conflicts of interest (White, 2013). Additionally, the Dodd-Frank Act attempts to change the liability of the CRAs. The reason for this is that CRAs mostly mention that their credit ratings are merely “an opinion”, which is indicated in the disclaimer of most credit ratings.4
This current ratings-based regulation is theoretically modelled by Opp et al. (2013). They explicitly show that regulation favouring highly rated bonds increases the quantity of highly rated bonds. Moreover, they argue that for sufficiently large benefits resulting from preferential regulation, CRAs prefer to inflate credit ratings.
In order for this rating-contingent financial regulation to be more transparent, the SEC created a new classification for CRAs in 1975: “Nationally recognized statistical rating organization” (NRSRO). Only with credit ratings from NRSROs are these financial institutions
2 Aside from the credit rating, most CRAs provide watchlists containing outlooks and reviews of ratings,
giving indications about future credit rating changes.
3 For the rating scale of S&P, see Table 1.
4 So far, CRAs mostly won in court using the First Amendment argument (which entails freedom of
allowed to interpret if a security was “investment-grade” or not. Hence, these NRSROs were basically granted the force of law by the SEC. With this classification, a significant barrier of entry arose: any potential CRA would likely remain small without the designation NRSRO. When a new CRA enters the market without having this classification, financial institutions (those demanding bonds) would partly ignore this CRA, since they cannot use its ratings to interpret the credit risk of the bond. As a consequence, issuers (selling bonds) would not buy credit ratings from this unlicensed CRA (White, 2010).5
In the early 1970’s the market for CRAs incurred a significant and structural change from “investor pays” to “issuer pays”, where the revenues no longer came from the investors requiring a rating, but from the issuer itself.6 According to White (2010), a possible reason was the only
recently widespread use of photocopy machines at that time. This allowed investors to free-ride on the copies of rating manuals from fellow investors. In fear of losing revenues coming from these manuals, CRAs converted to the issuer-pays business model. As a consequence, a potential conflict of interest was introduced in the market for credit ratings.
In particular, the phenomenon of “issuer shopping” arose. Skreta and Veldkamp (2009) define issuer shopping as shopping around for ratings and ultimately only disclose the most positive rating(s). By fearing that the issuer will not pay in the event of a low rating and thereby losing revenues, CRAs might have an incentive to rate bonds with an upward bias. With this conflict of interest, the quality (or accuracy) of the rating is possibly reduced, which consequently increases the aforementioned information asymmetry between investors and issuers.
2.2 Conflict of Interest for CRAs
The potential conflict of interest mentioned in Section 2.1 is countered by Cantor and Packer (1994). They argue that CRAs have an overriding incentive to rate correctly in order to maintain their reputation. They mention that for each rating, the name, integrity and credibility are at stake, which is confirmed by industry sources too. In 2010, the CEO of Moody’s at the time, Raymond McDaniel testified for the SEC that “Moody’s success depends on our reputation for issuing
5 Note: some smaller CRAs managed to survive without this title NRSRO, such as A.M. Best Company,
DBRS and Egan-Jones Rating Company. See “Action needed to improve Rating Agency Registration Program and Performance-Related Disclosures” by the SEC, 2010, retrieved from: https://www.gao.gov/assets/310/309849.pdf.
6 Not all CRA changed to “issuer-pays”. Some CRAs still employ an “investor-pays” or “subscriber-pays”
objective and accurate ratings”.7 Additionally, empirical evidence from Covitz and Harrison
(2003) suggests CRAs are mainly influenced by reputation-based incentives.
However, some research concerning this reputation mechanism and the possible conflict of interest has been conducted too. Bolton et al. (2012) construct a theoretical model between two firms where they allow for, amongst others, issuers shopping and reputation concerns of CRAs. Moreover, they allow a part of the investors to be trusting or “naïve”. They show that increased competition (i.e. from monopoly to duopoly) generally decreases efficiency. Even though the investor could obtain more information in a duopoly, the issuer has an increased opportunity to shop around for ratings. Additionally, they argue that when the expected reputation costs are lower, ratings might be inflated.
Furthermore, Bolton et al. (2012) show that a CRA is more prone to inflate ratings when the issuer is of importance to the CRA. If an issuer is a frequent costumer and/or a large issuer8,
the CRA generates substantial revenues from this issuer. Findings by He, Qian and Strahan (2010) suggest that ratings are more favourable for large structured product issuers, especially during booms. In addition, Faltin-Traeger (2009) finds that issuers are more prone to stick with the same CRA when they received more favourable ratings.
Mathis et al. (2009) also theoretically show that the reputation argument only holds when a sufficient amount of the revenue of a CRA comes from sources other than rating complex products9. Additionally, they present empirical evidence indicating an increase in the proportion
of AAA-rated residential-mortgage-backed-securities (RMBS) for the main three agencies from 2001 to 2007.
Moreover, Skreta and Veldkamp (2009) confirm this by showing that issuer shopping is more likely to occur when products are complex. Due to the complexity of the asset, ratings differ enough to incentivize issuers to shop around for ratings. Even more so, the process and complexity of assets, require the CRAs to basically become part of the “underwriting team” (Mason and Rosner, 2007). After creating the complex assets in collaboration with the issuer, the CRA provides a rating of that same asset, increasing the conflict of interest even further.
7 Testimony of Raymond W. McDaniel, Chairman and CEO of Moody’s Corporation, April 23rd 2010,
retrieved from: https://www.hsgac.senate.gov/imo/media/doc/STMTMcDANIELRaymondMoodys0.pdf, p.5.
8 Which is often the case in structured finance.
Finally, Becker and Milbourn (2011) argue that competition potentially reduces future rents and consequently weakens reputation incentives. This smaller concern for reputation possibly diminishes quality provision of the ratings. Also, Skreta and Veldkamp (2009) argue that increased competition leads to lower prices and as a consequence lower costs for issuer shopping. Thus, with increased competition the reputation-based incentives could diminish and consequently credit ratings might be inaccurate with an upward tendency.
2.3 Competition amongst CRAs
Considering the quality of a rating is important for the proper functioning of the financial market as it decreases the information asymmetry, research has been done regarding factors possibly influencing this quality. Becker and Milbourn (2011) assess the effect of increased competition on the quality of bond credit ratings in the US from 1996-2006, by using Fitch’s market share as a measure for competition.10 They empirically show that increased competition, measured by Fitch’s
market share, raises the level of bond ratings.
Inflated ratings, defined as a “general increase in the ratings level”, are considered a decrease in quality, because the interpretation of ratings should be stable (Becker and Milbourn, 2011). A stable interpretation of credit ratings is necessary when there are naïve investors. If rating levels are not stable over time, a rating of AAA today would not have the same associated risk as an AAA-rating ten years ago. With the presence of naïve investors - those who are not able to filter out the ratings inflation and arguably need ratings the most - the information asymmetry increases with ratings inflation and consequently reduces the quality of the rating.
Nevertheless, Becker and Milbourn (2011) specify Fitch’s market share as a fraction of all credit ratings that is provided by Fitch. This market share based on output is not the ideal measure for competition, as they mention themselves too. Market shares based on revenues for example, are preferred. Another competition measure is the Hirschman-Hirfandahl-Index (HHI), which is equal to the summation of the quadratic market shares. Davis and Garces (2009) state that alongside the K-firm concentration ratio11, the HHI is most commonly used as indicator for
concentration (and thus competition).
10 In addition, the effect of increased competition on firm credit rating and the yield spread is examined
(Becker and Milbourn, 2011).
In addition, Becker and Milbourn (2011) investigate this only for corporate bonds and not for structured finance, since Fitch’s presence was already substantial early on in this submarket. However, issuer shopping is less present in the market for corporate bonds, as both Moody’s and S&P have a policy of rating and disclosing all corporate taxable U.S. bonds, even in the event of no payment. Moreover, it is important to note that for structured finance, investors rely even more on the ratings because of the complexity of the products (Mitchel and Fender, 2005).
From the previous subsection, it is evident that the reputation-based incentives for CRAs are reduced with increasing complexity of the assets, such as structured products. Mathis et al. (2009) show that the reputation argument decreases with a larger part of CRA income coming from complex products. Interestingly, Coval et al. (2009) mention that in 2006, Moody’s reported that rating structured finance amounts 44 percent of its revenues, while 32 percent of revenues comes from rating corporate bonds. With almost half of the revenues coming from structured finance and thus complex products, the fear of losing revenues might be overriding the reputation-based incentives.
Therefore, Cohen and Manuszak (2013) empirically examine the effect of increased competition on the subordination of Commercial-Mortgage-Backed Securities (CMBS) for the U.S. from 2002-2007. Using a slightly similar methodology to Becker and Milbourn (2011), they too use Fitch’s market share as a measure of competition and find that increased competition yields lower subordination, which suggests more lax ratings. However, structured finance consists of more than solely CMBS.
2.4 Structured Finance
From the previously discussed literature, it is evident that structured finance is a complex form of an asset. This subsection provides some general insights in structured finance as well as a brief overview of the development of structured finance over the past decades, especially prior to the financial crisis of 2007.
Structured finance can be defined as “the pooling of economic assets like loans, bonds, and mortgages, and the subsequent issuance of a prioritized capital structure of claims, known as tranches, against these collateral pools” (Coval et al., 2009, p.3). Due to this design of prioritization and lower risk, many of the created tranches have a higher rating than the average rating of the
assets in the underlying pool. This process increases the complexity of the asset and consequently makes structured finance more prone to inflated ratings.
It has been stated that structured finance played a significant role in the financial crisis of 2007 (Coval et al., 2009; Mathis et al., 2009). In the years before the financial crisis, structured finance grew substantially and rapidly. Moreover, structured finance often received high ratings with more attractive yields compared to other financial assets such as corporate bonds. For example, Fitch Ratings (2007) reported that about 60 percent of all outstanding ratings for structured finance were “AAA”, whereas only 1 percent of corporate and financial institutions obligors received the same rating.12 Together with the rating-contingent regulation, the demand and supply for structured finance increased immensely in the early 2000’s, reaching its peak with an issuance value of 100 billion dollar in the second quarter of 2007 alone (Coval et al. 2009).
However, in the first two quarters of 2008 the issuance value of structured finance dropped to only 5 billion dollars. The premium ratings of structured products dropped and the apparent high yields were in fact too low for the associated risk of the asset (Coval et al. 2009). Thus, the ratings were inaccurate and therefore contributed to the subprime-mortgage crisis. As a consequence, CRAs were heavily criticized.
3 Methodology
This section elaborates on the methodology of this analysis. First the variables are specified, followed by the baseline model of this study. Finally, the extensions to the baseline model are discussed.
3.1 Data Specification
The dataset used is mostly provided by the Bloomberg Terminal at the Erasmus University in Rotterdam. It contains credit ratings of long-term structured finance in the U.S. for the period of 2004-2016, where DBRS gained market share and competition for the incumbents possibly increased. For each credit rating, the previous rating of that same asset by the same agency, is also included. Considering most ratings are letter-based, we need to assign a numeric value to each
12On June 30th, 2007. See “Inside the Ratings: What Credit Ratings Mean” by Fitch Ratings (2007),
rating. By using the same ratings scale as Becker and Milbourn (2011), we can numerically distinguish the level of credit risk, ranging from 0 to 28. Table 1 displays the ratings scale assigning a “score” for each rating with the associated credit risk for the incumbent CRAs (Rating and
lagged Rating). Finally, some credit ratings in the database are indicated by NR (“not rated”), WR
(“withdrawn rating”) or simply do not match any of the specifications of Table 1. All these ratings are dropped from the database.
Aside from the credit ratings the Bloomberg database also provides the type of the structured product, where they distinguish between 5 categories: mortgage-backed securities (MBS), asset-backed-securities (ABS), collateral debt obligations (CDO), whole loans and “Other”. With these 5 groups, we can control for every type by including type fixed effects. Additionally, the collateral type of each asset is specified, resulting in over 60 different collateral types ranging from “student loans” to “commercial-mortgage-backed securities”.13 This allows us
to control the collateral type of the asset, but consequently makes the type fixed effects of structured products redundant.
Furthermore, the different measures for competition are obtained from separate databases. The market share of DBRS based on output can be calculated using the credit ratings dataset constructed from the Bloomberg terminal. Specifically, competition is determined by the percentage of ratings provided by DBRS of all ratings (DBRSqshare). Secondly, the HHI is retrieved from (annual) reports by the SEC, which is also based on output.14 Unfortunately, data
on the HHI was only available from 2007 onwards, shrinking our database slightly.
Lastly, a third measure for competition, issuance volume market share of DBRS (DBRSishare), is collected from the Asset-Backed Alert database.15 The sum of these market
shares often amounts to more than 100 percent, since some financial assets might be rated multiple times by different agencies in the same year. Therefore, we adjust these market shares by calculating the share of DBRS as a fraction of the sum of all issuance volume shares. The data for the issuance volume market shares are from 2007 onwards.
13 For the complete list of types and collateral types, see Table 8 and 9 in Appendix I, respectively. 14 The SEC reports the inverse HHI, which is equal to 10,000
𝐻𝐻𝐼 . Retrieved from:
https://www.sec.gov/rules/final/2014/34-72936.pdf and https://www.sec.gov/ocr/reportspubs/annual-reports/2017-annual-report-on-nrsros.pdf
Table 1: The ratings scale, following Becker and Milbourn (2011).
16Retrieved from: https://www.standardandpoors.com/en_US/delegate/getPDF?articleId=2017758&type=
COMMENTS&subType=REGULATORY.
The table defines categories of ratings on long-term credit with the numerical value assigned to each category ranging from 0 to 28. For categories with multiple numerical values, each single rating level represents the number assigned to ratings with a + qualifier, no qualifier and – qualifier, respectively. The source for ratings definitions is Standard and Poor’s Rating Definitions from 26th of June, 2017 (Standard and Poor, 2017, p.3)16.
Rating Group
Rating Agency Numerical
value assigned (“score”)
Category Definition
Moody’s S&P and Fitch Investment
grade AAA AAA 28
The obligator’s capacity to meet its financial commitment on the obligation is extremely strong.
Aa AA 24, 25, 26
The obligor’s capacity to meet its financial commitment on the obligation is very strong.
A A 21, 22, 23
Somewhat more susceptible to the adverse effects of changes in circumstances and economic conditions than obligations in higher-rated categories. However, the obligor’s capacity to meet its financial commitments on the obligation is still strong.
Baa BBB 18, 19, 20
Exhibits adequate protection parameters. However, adverse economic conditions or changing circumstances are likely to lead to weakened capacity of the obligor to meet its financial commitment on the obligation.
Ba BB 15, 16. 17 Obligations rated ‘BB’, ‘B’, ‘CCC’, ‘CC’ and ‘C’ regarded as having significant speculative characteristics. ‘BB’ indicates the least degree of speculation and ‘C’ the highest. While such obligations will likely have some quality and protective characteristics, these may be outweighed by large uncertainties or major exposures to adverse conditions.
B B 12, 13, 14
Caa CCC 9, 10, 11
Ca CC 7
C C 4
Default D D 0 An obligation in payment default. The ‘D’ rating category is
used when payments on an obligation are not made on the date due even if the applicable grace period has not expired, unless Standard & Poor’s believes that such payments will be made during such grace period. The ‘D’ rating also will be used upon filing of a bankruptcy petition or the taking of a similar action if payments on an obligation are jeopardized.
Accordingly, an (unbalanced) panel dataset is constructed for structured finance in the U.S. over the period of 2004-2016. With these data specifications, we can introduce a model similar to Becker and Milbourn (2011) to assess the effect of increased competition, due to the entry of DBRS into the market, on the level of credit ratings provided by the incumbents: Fitch, Moody’s and S&P.
3.2 Baseline Model
By following Becker and Milbourn (2011), we can construct the models of our research. With the ratings scale of Table 1, we now have a numerical value for all ratings ranging from 0 to 28. As mentioned in the previous subsection, an increase in the general level of ratings (ratings inflation) is considered a deterioration of the quality of the rating. Becker and Milbourn (2011) argue that a decrease in quality of the credit ratings translates to on average higher ratings, because ratings closer to the upper end of the scale (e.g. AAA) are preferred by issuers, as it decreases their cost of capital.
However, the decrease in quality is also based on the assumption that not all investors are sophisticated. When there are naïve investors, an increase in the general level of ratings cannot be filtered out by all investors. Especially considering the notion that naïve investors possibly rely even more on credit ratings to reduce the information asymmetry. Thus, ideally ratings have a stable interpretation and predict default accurately.
Consequently, the first model to assess the effect of increased competition on the level of ratings of the incumbents can be formulated, resulting in Model 1:
𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 = 𝛽0+ 𝛽1𝐷𝐵𝑅𝑆𝑞𝑠ℎ𝑎𝑟𝑒𝑡+ 𝛽2𝐿𝑎𝑔𝑔𝑒𝑑 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡+ 𝑌𝑒𝑎𝑟𝑡+ 𝑇𝑦𝑝𝑒𝑖+ 𝜀𝑖,𝑡 (1)
where 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 equals the numeric value (from Table 1) of the credit ratings of product i in year
t of the incumbents, 𝐷𝐵𝑅𝑆𝑞𝑠ℎ𝑎𝑟𝑒𝑡 is the output-based market share of DBRS as defined in Section 3.1 in year t, 𝐿𝑎𝑔𝑔𝑒𝑑 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 is the numeric value of the previous rating an asset i by
the same agency.17 Alongside these variables and a constant (𝛽
0), some year fixed effects are
included as well as type fixed effects.18 Finally, an error term 𝜀
𝑖,𝑡 is added to the regression.
This straightforward regression model (Model 1) is the baseline for the other models presented in this paper. From the literature discussed in Section 2, we hypothesize the effect of increased competition to have a positive effect on the level of credit ratings of the incumbents. For Model 1, an increase in market share of DBRS is assumed to increase competition for the incumbent firms (Fitch, Moody’s and S&P). Hence, we hypothesize the coefficient 𝛽1 to be positive.
However, we recognize that this is not the optimal measure for competition and it might even be noisy, which is confirmed by Becker and Milbourn (2011). Market shares based on revenues instead of output would be preferred. Especially considering it is argued that the loss of revenues might induce CRAs to inflate their ratings incorrectly. Unfortunately, data on revenues were not available. Nevertheless, some alternate measures for competition have been gathered and computed, which is elaborated on in the next subsection.
3.3 Alternative Measures for Competition
With the baseline model (Model 1) now specified, some extensions and adjustments can be made to further examine the effect of increased competition. Firstly, alternative measures for competition are included in the model. This results in two similar models, where the independent variables of interest are now the HHI and the market share of DBRS based on issuance value in dollars, respectively:
𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 = 𝛾0 + 𝛾1𝐻𝐻𝐼𝑡+ 𝛾2𝐿𝑎𝑔𝑔𝑒𝑑 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡+ 𝑌𝑒𝑎𝑟𝑡+ 𝑇𝑦𝑝𝑒𝑖 + 𝜇𝑖,𝑡 (2)
𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 = 𝛿0+ 𝛿1𝐷𝐵𝑅𝑆𝑖𝑠ℎ𝑎𝑟𝑒𝑡+ 𝛿2𝐿𝑎𝑔𝑔𝑒𝑑 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡+ 𝑌𝑒𝑎𝑟𝑡+ 𝑇𝑦𝑝𝑒𝑖+ 𝜔𝑖,𝑡 (3)
17 In case there is a previous rating by the same agency of that asset i.
18 This is incorporated in STATA by including dummy variables for T-1 years and dummy variables for
Type-1 types. Moreover, collateral type fixed effects are also included using dummy variables, making type fixed effects redundant.
Where 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 again represents the numeric value of the credit rating for product i in year t of the incumbents. 𝐻𝐻𝐼𝑡 is the Hirschman-Hirfandahl-Index based on output and 𝐷𝐵𝑅𝑆𝑖𝑠ℎ𝑎𝑟𝑒𝑡 is the market share of issuance volume in dollars, both in year t. Again, the Lagged Rating is included in the model. Additionally, we control for fixed effects for years as well as the type of structured finance fixed effects.19 Lastly, error terms 𝜇
𝑖,𝑡 and 𝜔𝑖,𝑡 are added for product i in year t.
These adjustments to Model 1 allow us to further assess the effect of competition on the level of the ratings of the incumbents. As we assume increased competition to be detrimental to the quality of ratings by inflating ratings, we hypothesize the effect of 𝐻𝐻𝐼𝑡 the to be negative. This is based on the rationale that the HHI is defined as (Davis & Garces, 2009):
𝐻𝐻𝐼 = ∑𝑁 𝑠𝑖2
𝑖=1 (4)
where 𝑠𝑖 is the market share of firm i. This measure decreases when the market concentration increases. Thus, increased competition leads to a lower HHI and as a consequence we expect the coefficient 𝛾1 to be negative. Alternatively, competition for the incumbents increases when the issuance volume-based market share of DBRS increases. Hence, we expect the coefficient 𝛿1 to be positive, following the same rationale from Model 1 in section 3.2. It is important to note that these alternative measures could be noisy too. Therefore, the estimations of this model must be interpreted carefully (more on this in Section 5).
3.4 Ordered Probit
Looking at Models 1 to 3, a considerable critique arises still. Particularly, the ratings scale is designed such that almost every step between ratings (e.g. from AA to AA+ or from BB to BB-) is treated equally.20 It is unlikely however, for this to be the case. There is no reason to assume that
credit ratings increase linearly with the increase in credit risk. As our previous models do not allow for rating levels to differ across the ratings scale, our estimations are possibly biased.
As the credit rating (and thus Rating) is a measure of quality of a financial product, our dependent variable is ordinal: the outcome of credit ratings is ordered, yet the numeric values have
19 Alternatively, we include collateral type fixed effects, just as in Model 1.
20 Note: From AA+ to AAA, CCC to CC, CC to C and C to D, the steps are bigger different, looking at
no further meaning (Heij et al., 2004). Therefore, a possible solution is the ordered probit regression. An ordered probit estimates cut-off points, allowing the ratings to vary over different parts of the ratings scale.
The ordered probit model assigns probabilities to each category of the ratings scale, conditional on independent variables, using the cumulative normal distribution. More specifically:
Pr(𝑅𝑎𝑡𝑖𝑛𝑔 =score |𝐷𝐵𝑅𝑆𝑞𝑠ℎ𝑎𝑟𝑒, 𝐿𝑎𝑔𝑔𝑒𝑑 𝑅𝑎𝑡𝑖𝑛𝑔, 𝑌𝑒𝑎𝑟, 𝑇𝑦𝑝𝑒)
(5) = Φ(𝛽0+ 𝛽1𝐷𝐵𝑅𝑆𝑞𝑠ℎ𝑎𝑟𝑒𝑡+ 𝛽2𝐿𝑎𝑔𝑔𝑒𝑑 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡+ 𝑌𝑒𝑎𝑟𝑡+ 𝑇𝑦𝑝𝑒𝑖)
where all variables are defined as in the previous section. The probability of a certain credit rating (score or numerical value) is Φ, which is the cumulative normal distribution. Additionally, since the ordered probit is a nonlinear function of the coefficients, OLS cannot estimate our coefficients. Instead, the ordered probit model uses the maximum likelihood estimation (Stock and Watson, 2014).
With this alternative estimation, we can assess the effect of competition allowing for steps between credit ratings to vary across the scale of ratings. This method is applied on all three previous models, where we hypothesize the effect of DBRS market shares based on output and issuance volume to be positive, whereas the effect of HHI is expected to be negative, similar to the previous regressions.
4 Empirical results
The results of this research are shown in this section. The first subsection displays some descriptive statistics. Thereafter, we present the results from the models from our empirical analysis.
4.1 Descriptive Statistics
After combining and cleaning the datasets, we have a sample of approximately 840,000 observations of credit ratings of structured finance over the period 2014-2016. Due to availability of the data on the HHI, the number of observations decreases with roughly 18,000 to over 820,000. The number of credit ratings varies substantially for every year. In 2004, the number of credit
ratings is only 4,764, whereas this is over 173,000 in 2009. The average number of credit ratings each year is approximately 65,000. Considering we only assess the effect on competition on the level of credit ratings of the incumbents, we have fewer observations in our regression results, which are discussed in the next subsection.
Table 2 shows the different CRAs and their share of the sample. Our dataset consists of ratings from 7 different CRAs, where approximately 95 percent is from either Moody’s, Fitch or S&P. KBRA and A.M Best only take a small fraction of the sample, whereas DBRS and Morningstar amount to roughly 5 percent.
Table 2: CRAs and their percentages
CRA Frequency Percent
A.M. Best 100 0.01 DBRS 29,909 3.56 Fitch 163,433 19.46 KBRA 624 0.07 Moody's 303,519 36.13 Morningstar 13,458 1.60 S&P 328,975 39.16 Total 840,018 100.00
For a general overview of the data gathered in our dataset, some descriptive statistics of the variables of interest in this analysis are specified. Table 3 summarizes these variables and provides some basic insights on the dataset. The average numerical value for credit ratings equals 16.406, amounting to for example BB on the ratings scale of S&P. The previous rating (Lagged
Rating) is on average higher than the current credit rating, with a numerical value of 19.090 or
BBB on the S&P scale. Evidently, the minimum is zero and the maximum is 28, as conducted by the ratings scale.
The first measure for competition, the market share of DBRS on output, shows an average of 3.561 percent, with a standard deviation of 4.421 percent. The maximum of DBRSqshare was in 2014 and equals 18.133 percent, which could possibly be an outlier. This in turn could violate
the OLS assumption that large outliers are unlikely (Stock and Watson, 2014). Therefore, we also perform all regressions of Model 1 without this possible outlier as a check for robustness.21
Table 3: Summary of statistics
Notes: Rating and Lagged Rating correspond to the numerical value assigned by Table 1. DBRSqshare is the annual fraction of
DBRS ratings of the total ratings denoted in percentages. DBRSishare is defined as the fraction of total yearly issuance volume that is rated by DBRS, again in percentages. The HHI is based on output.
For the HHI, the average is 3087.50, which gives an inverse HHI of approximately 3.24 uniformly sized companies.22 The maximum of the HHI was in 2007, whereas the minimum was
in 2016. Our final measure for competition, the market share based on issuance volume in dollars (DBRSishare), has a mean of 7.595 percent with a standard deviation of 3.736. The maximum for this measure was in 2010, amounting to 14.147 percent.
4.2 Results Model 1
The results of Model 1, in which we estimate the effect of the output market share of DBRS on the level of credit ratings of structured finance, are presented in Table 4. When time fixed effects and type fixed effects are implemented in the model (Column 2, Model 1.2), we observe a positive and strongly significant coefficient of 0.304, as hypothesized in Section 3. A one percentage point increase of DBRS market share inflates ratings on average by 0.304. This equals an increase of the credit rating of roughly one in every three securities. Similarly, a 4.421 percentage point increase – which is the standard deviation of DBRSqshare (see Table 3) – in competition, increases the
21 These estimations are not reported. The results are significant at a 1 percent level and are, with the
exemption of Model 1.1, slightly smaller than the results in Section 4.2. In total, we dropped approximately the 49,000 observations from 2014.
22 Inverse HHI = 10,000
𝐻𝐻𝐼 , which theoretically indicates the number active firms if all are equally large.
Variable Observations Mean Std. Dev. Min Max
Rating 840018 16.406 8.174 0.000 28.000
Lagged Rating 840018 19.090 7.274 0.000 28.000
DBRSqshare 840018 3.561 4.421 0.000 18.133
HHI 822176 3087.50 238.20 2717.39 3690.04
average level of credit ratings by 1.34, which approximately amounts to a one step increase (e.g. AA to AA+) for every bond. Hence, these results suggest that competition leads to a higher level of credit ratings.
Table 4: Results Model 1
Dependent variable: Rating (1) (2) (3) Model 1.1 (OLS) Model 1.2 (OLS) Model 1.3 (OLS) DBRSqshare 0.255 *** (0.0010) 0.304*** (0.0079) 0.223*** (0.0081) Lagged Rating 0.945 *** (0.0007) 0.997*** (0.0007) 0.973*** (0.0007) Constant -2.55 *** (0.0130) -1.4577*** (0.0573) 0.5673*** (0.0839)
Year fixed effects No Yes Yes
Type fixed effects No Yes No
Collateral type fixed effects No No Yes
R2 0.686 0.723 0.730
Observations 795927 795927 795927
Notes: Dependent variable is the numerical value from the credit rating of the incumbents (Rating). DBRSqshare is in
percentages. Model 1.1 is an OLS regression without fixed effects. Model 1.2 includes time fixed effects and type fixed effects by including t-1 dummies with type-1 dummies. Model 1.3 excludes type fixed effects, but controls for collateral type fixed effects by including collateral type-1 dummies. All models include a constant. Robust standard errors in parentheses. *, **, *** represent a significance level of 0.10, 0.05, 0.01, respectively.
Looking at Table 2, it is evident DBRS is not the only competitor for the incumbents (Moody’s, Fitch and S&P), while we use DBRSqshare as a proxy for competition in Model 1. However, we also performed Model 1 with the annual output share of all “outsiders” combined (all CRAs, except Moody’s, Fitch and S&P) as a competition measure.23 The results are similar in
significance (p-values < 1%) and somewhat smaller than the estimations presented in Table 4.
23 This is defined as the fraction of all ratings that are provided by A.M. Best, DBRS, KBRA and
4.3 Results Model 2 and 3
With two alternative measures for competition, we again estimate the effect of increased competition on the level of credit ratings of structured finance. Table 5 shows the results from Model 2, where the measure for competition is the HHI. The coefficient for HHI for Model 2.2, where we control for time and type fixed effects, equals -0.00299 and is significant at a 1% level. As hypothesized, the level of ratings increases when the HHI decreases (or competition increases). With a change of the standard deviation of the HHI (238.20), the level of ratings would decrease with approximately 0.71 when the HHI increases (i.e. when competition decreases), meaning that for seven out of every ten bonds, the credit rating decreases by one step (e.g. AA to AA-).
Table 5: Results Model 2
Dependent variable: Rating (1) (2) (3) Model 2.1 (OLS) Model 2.2 (OLS) Model 2.3 (OLS) HHI -0.00345 *** (0.0000) -0.00299*** (0.0000) -0.00214*** (0.0000) Lagged Rating 0.953 *** (0.0008) 0.996*** (0.0007) 0.973*** (0.0007) Constant 8.768 *** (0.0644) 8.966*** (0,1065) 8.0723*** (0.1247)
Year fixed effects No Yes Yes
Type fixed effects No Yes No
Collateral type fixed effects No No Yes
R2 0.679 0.722 0.729
Observations 778095 778095 778095
Notes: Model 2.1 is an OLS regression, Model 2.2 includes time and type fixed effects. Model 2.3 includes time and collateral type fixed effects. All models include a constant. Robust standard errors in parentheses. *, **, *** represent a significance level
of 0.10, 0.05, 0.01, respectively.
In addition, Table 6 depicts the results from Model 3, where the measure for competition is the market share of DBRS based on issuance volume in dollars in that year. We again observe a strongly significant coefficient for DBRSishare equal to 0.214 (Column 2, Model 3.2), which is
positive, as expected. A one percentage point increase of DBRSishare increases credit ratings level by 0.214 on our scale, suggesting an increase of one step for approximately one in every 5 bonds. With a standard deviation change (3.736 percent), credit ratings levels increase by 0.80. Thus, with a standard deviation change in the market share of DBRS based on issuance volume, 8 out of every ten bonds sees an increase in the level of ratings of one step. Thus, both measures alternative to Model 1 suggest ratings inflation with increased competition.
Table 6: Results Model 3
Dependent variable: Rating (1) (2) (3) Model 3.1 (OLS) Model 3.2 (OLS) Model 3.3 (OLS) DBRSishare 0.0815* *** (0.0013) 0.214*** (0.0056) 0.157*** (0.0057) Lagged Rating 0.931 *** (0.0008) 0.997*** (0.0007) 0.973*** (0.0007) Constant -2.5450 *** (0.0195) -2.8506*** (0.0408) -0.3720*** (0.0714)
Year fixed effects No Yes Yes
Type fixed effects No Yes No
Collateral type fixed effects No No Yes
R2 0.671 0.723 0.730
Observations 795927 795927 795927
Notes: Model 3.1 is an OLS regression, Model 3.2 includes time and type fixed effects. Model 3.3 includes time and collateral
type fixed effects. All models include a constant. Robust standard errors in parentheses. *, **, *** represent a significance level
of 0.10, 0.05, 0.01, respectively.
However, as mentioned before, the measures for competition (DBRSqshare, HHI and
DBRSishare) are not ideal and likely noisy. The measurement errors in the independent variables
possibly bias the estimated coefficients towards zero. With this attenuation bias, the estimated effect of competition might even be underestimated (Stock and Watson, 2014). Section 5 elaborates more on this.
4.4 Results Model 4
Alternative to the OLS, an ordered probit regression is performed to allow the steps between ratings to vary over the ratings scale. Table 7 depicts the results from Model 4 for all three measures for competition with time and type fixed effects. Looking at the market share of DBRS based on output we observe a smaller, but positive coefficient of 0.058 at a 1% significance level. This implies that a higher market share of DBRS and thus more competition, increases the probability of a particular asset receiving a high rating from the incumbents (more to the AAA end of the scale).
Table 7: Results Model 4
Dependent variable: Rating (1) (2) (3) Model 4.1 (Ordered probit) Model 4.2 (Ordered probit) Model 4.3 (Ordered probit) DBRS Market Share (q) 0.0767*** (0.0018) HHI -0.000714*** (0.0000) DBRSishare 0.0539*** (0.0013) Lagged Rating 0.230*** (0.0003) 0.230*** (0.0003) 0.229*** (0.0003)
Year fixed effects No Yes Yes
Type fixed effects No Yes Yes
Collateral type fixed effects No No No
Observations 795927 778095 795927
Notes: Dependent variable is the numerical value from the credit rating of the incumbents (Rating). Model 4.1 is an ordered probit including DBRS market share of output as competition measure. Model 4.2 uses the HHI, whereas Model 4.3 includes the market share of DBRS based on issuance volume in dollars. All models do not include intercepts. Cut-off points are not reported. Robust standard errors in parentheses. *, **, *** represent a significance level of 0.10, 0.05, 0.01, respectively.
However, we have to interpret these coefficients with prudence, since the steps between the rating categories are not equal anymore. Therefore, we look at the marginal effects of
DBRSqshare for every outcome (Rating)24, where we observe increasing probabilities of receiving
a rating above BBB+ with increased competition by one percentage point, whereas the probabilities decrease for receiving a rating from BBB+ and below. All marginal effects are significant at a 1% level.
Similarly, Model 4.3 indicates a significant positive coefficient for the market share of DBRS based on issuance volume (p-value<0.01). This again implies that an increase in
DBRSishare increases the likelihood of receiving a higher rating from the incumbents. The
significant marginal effects (all p-values <0.01) of DBRSishare are similar to those of Model 4.2, where a one percentage point increase in DBRSishare increases the probability of receiving a rating from the incumbents above BBB+ and decreases for ratings from BBB+ and below.
The coefficient of the HHI is negative at a significance level of 1% too. For the marginal effects, we observe that the probability of receiving a rating above BBB+ decreases when the HHI increases (or less competition), whereas probabilities increase for ratings from BBB+ and below. Again, the marginal effects are significant at the 1% level. Thus, all three measures indicate an increase in the level of credit ratings when competition increases, looking at both the coefficient and the marginal effects.
Table 7 only shows the ordered probit results with year and type fixed effects. However, when using collateral type fixed effects, the coefficient changes only slightly while the sign remains the same for all three measures. Likewise, these results are significant at a 1 percent level.
5 Discussion and Limitations
This part discusses some limitations and notes with respect to this analysis. Firstly, it is mentioned in the previous section that the measures for competition might be noisy. Due to this measurement error in the independent variable, our estimates are possibly biased and inconsistent. Nevertheless, this bias is towards zero and is called attenuation bias. We can illustrate this with a simplified example of Model 2. Suppose Model 2.1 is redefined as:
𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 = 𝛽0+ 𝛽1𝑋̃𝑡+ [𝛽1(𝑋𝑡− 𝑋̃𝑡)]+𝜇𝑖,𝑡 (5)
where 𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 is still the numerical value of asset i in year t. 𝑋̃𝑡 is the HHI on output in year t, which is a mismeasurement of competition (𝑋𝑡) itself and 𝜇𝑖,𝑡 is the error term in year t for asset i. Rewriting this, yields:
𝑅𝑎𝑡𝑖𝑛𝑔𝑖,𝑡 = 𝛽0+ 𝛽1𝑋̃𝑡+ 𝜈𝑖,𝑡 (6)
where 𝜈𝑖,𝑡 = 𝛽1(𝑋𝑡− 𝑋̃𝑡) + 𝜇𝑖,𝑡. Thus, the error term includes a measurement error, which is the
difference between 𝑋̃𝑡 and 𝑋𝑡. When this difference is correlated with the 𝑋̃𝑡, the error term is
correlated with 𝑋̃𝑡 and our coefficient 𝛽1 is biased and inconsistent. Suppose that 𝑋̃𝑡 is the actual measure plus a completely random element, 𝜂𝑡, with a mean zero and variance 𝜎𝜂. With the assumption that this error is uncorrelated with both 𝑋𝑡 and 𝜇𝑡, we get the classical measurement error model, with 𝑋̃𝑡= 𝑋𝑡+ 𝜂𝑡.25 With this model, the probability limit of 𝛽̂
1 is (Stock and Watson, 2014)26: 𝛽̂1 𝑝 → 𝜎𝑋2 𝜎𝑋2+𝜎𝜂2𝛽1 (7) and since 𝜎𝑋 2
𝜎𝑋2+𝜎𝜂2 is below 1, the bias is towards zero.
In addition, Hausman (2001) mentions that if the variance of the noise (𝜎𝜂2) increases
relative to the variance of “competition” (𝜎𝑋2), the bias towards zero increases. Moreover, the use
of panel data with fixed effects – which this analysis does – generally increases the variance of the measurement error (𝜎𝜂2) relative to the variance of competition (Griliches and Hausman, 1986).
Due to this attenuation bias, the estimated coefficient might be smaller than the actual effect of competition on the level of credit ratings.
Secondly, our specification might have endogeneity issues. If DBRS would enter in a market where ratings are high, our estimates are biased due to reverse causality. However, if issuers already receive high ratings, the incentive to shop around for ratings is decreased. It seems unlikely issuers would ask for a second rating from another CRA, when they already received a high rating.
25 Note: 𝑐𝑜𝑟𝑟(𝜂
𝑡, 𝑋𝑡) = 0 and corr(𝜂𝑡, 𝜇𝑖,𝑡) = 0.
Therefore, it is not probable for reverse causality to explain our estimations. Moreover, the use of controls such as the previous ratings as well as fixed effects control partly for endogeneity.
However, there is still another problem of endogeneity, particularly omitted variable bias. If there is a determinant that influences higher ratings as well as the market share of DBRS, our measure for competition is correlated with the error term. When demand for credit ratings increases substantially, which seems to be the case for structured finance in our dataset, the incumbent CRAs might not have the capacity to supply all demand. Consequently, it might be easier for DBRS to enter a market that is booming. This endogeneity possibly biases the estimations in this analysis. To solve this endogeneity issue, an instrumental-variable regression could be performed. If there is data available, an instrument (𝑍𝑖,𝑡) could be obtained that is only correlated with the
measure(s) for competition and not with the error term of the level of credit ratings. With an exogenous (corr(𝑍𝑖,𝑡, 𝜀𝑖,𝑡)) and relevant instrument (corr(𝑍𝑖,𝑡, 𝑋𝑖,𝑡)), we could estimate the market share of DBRS or other competition measures. An example of an instrument could be the expenses of DBRS on marketing and/or costumer relations. These expenses affect the market share of DBRS, but do not affect the level of the credit ratings of the incumbents.
6 Conclusion
This paper analyzes competition amongst CRAs and its effect on the quality of the credit ratings. With increased competition, CRAs might be induced to inflate their ratings and thus it might lower the quality of the ratings itself. Specifically, ratings inflation could increase the information asymmetry between issuers and investors even further. The entrance of the rating agency DBRS into the market of credit ratings for structured finance provides us with the opportunity to research the effect of this increased competition on the quality of credit ratings. Using dataset of credit ratings of structured finance of the U.S. from 2004-2016, we empirically assess the effect of three measures for competition on the level of credit ratings of the three incumbents.
Our results suggest, for all three measures, an increase in the level of credit ratings of the incumbents when competition increases. We control for time and type fixed effects, as well as the previous rating of the same asset. Additionally, we perform an ordered probit regression to allow the steps between credit ratings to vary on our scale of ratings. Again, we observe inflated ratings of the incumbents when competition is fiercer.
Nevertheless, some issues might still arise in this study. Even though we use three different measures for competition, they are still likely to be noisy. This measurement error potentially biases our estimations, but only towards zero. Therefore, the true effect might be underestimated. A different and more accurate measure for competition, such as revenue-based market shares, could reduce this attenuation bias. Another problem is endogeneity, where some determinant drives both the measure of competition and the level of credit ratings. This omitted variable bias remains unsolved, but an instrumental variable approach possibly solves this endogeneity issue. Future work could focus on solving this problem of endogeneity to control for the bias.
The results suggest an increase in the level credit ratings of the incumbents with increased competition, possibly due to the conflict of interest CRAs face. In terms of regulation, full disclosure of all credit ratings could diminish this conflict of interest, since it eliminates the possibility of an issuer to shop around for a more favourable rating. Moreover, regulation promoting competition in the market for credit ratings might lower the reputation-based incentives of CRAs and thereby decrease the quality of credit ratings.
References
Bannier, C. E., & Hirsch, C. W. (2010). The economic function of credit rating agencies–What does the watchlist tell us?. Journal of Banking & Finance, 34(12), 3037-3049.
Becker, B., & Milbourn, T. (2011). How did increased competition affect credit ratings?. Journal of Financial Economics, 101(3), 493-514.
Bolton, P., Freixas, X., & Shapiro, J. (2012). The credit ratings game. The Journal of Finance, 67(1), 85-111.
Boot, A. W., Milbourn, T. T., & Schmeits, A. (2005). Credit ratings as coordination mechanisms. The Review of Financial Studies, 19(1), 81-118.
Cantor, R., & Packer, F. (1994). The credit rating industry. Quarterly Review, (Sum), 1-26. Cohen, A., & Manuszak, M. D. (2013). Ratings competition in the CMBS market. Journal of
Money, Credit and Banking, 45(s1), 93-119.
Coval, J., Jurek, J., & Stafford, E. (2009). The economics of structured finance. Journal of Economic Perspectives, 23(1), 3-25.
Covitz, D. M., & Harrison, P. (2003). Testing conflicts of interest at bond rating agencies with market anticipation: Evidence that reputation incentives dominate.
Davis, P., & Garcés, E. (2009). Quantitative techniques for competition and antitrust analysis. Princeton University Press.
Doherty, N. A., Kartasheva, A. V., & Phillips, R. D. (2012). Information effect of entry into credit ratings market: The case of insurers' ratings. Journal of Financial Economics, 106(2), 308-330.
Faltin-Traeger, Oliver, 2009, Picking the right rating agency: Issuer choice in the ABS Market, Working paper, Columbia Business School.
Fender, I., & Mitchell, J. (2005). Structured finance: complexity, risk and the use of rating. Griliches, Z., & Hausman, J. A. (1986). Errors in variables in panel data. Journal of
econometrics, 31(1), 93-118.
Hausman, J. (2001). Mismeasured variables in econometric analysis: problems from the right and problems from the left. Journal of Economic perspectives, 15(4), 57-67.
He, J., Qian, J., & Strahan, P. E. (2011). Credit ratings and the evolution of the mortgage-backed securities market. American Economic Review, 101(3), 131-35.
Heij, C., De Boer, P., Franses, P. H., Kloek, T., & Van Dijk, H. K. (2004). Econometric methods with applications in business and economics. OUP Oxford.
Mason, J. R., & Rosner, J. (2007). Where did the risk go? How misapplied bond ratings cause mortgage backed securities and collateralized debt obligation market disruptions.
Mathis, J., McAndrews, J., & Rochet, J. C. (2009). Rating the raters: are reputation concerns powerful enough to discipline rating agencies?. Journal of monetary economics, 56(5), 657-674.
Opp, C. C., Opp, M. M., & Harris, M. (2013). Rating agencies in the face of regulation. Journal of Financial Economics, 108(1), 46-61.
Skreta, V., & Veldkamp, L. (2009). Ratings shopping and asset complexity: A theory of ratings inflation. Journal of Monetary Economics, 56(5), 678-695.
Spatt, C., 2009. Discussion of ratings shopping and asset complexity: a theory of ratings inflation. Journal of Monetary Economics 56, 696–699.
White, L. J. (2010). Markets: The credit rating agencies. Journal of Economic Perspectives, 24(2), 211-26.
White, L. J. (2010). Credit-rating agencies and the financial crisis: less regulation of CRAs is a better response. Journal of international banking law, 25(4), 170.
White, L. J. (2013). Credit rating agencies: An overview. Annual Review Financial Economics, 5(1), 93-122.
Appendices
Appendix I: Type of Structured Finance and Type of Collateral
Table 8: Type of Structured product and percentages
Type Number of Ratings Percent
ABS 223,411 26.60
CDO 169,142 20.14
CMBS 97,981 11.66
Other 286 0.03
Whole 349,198 41.57
Note: Types of Structured Finance and their fraction in the sample, presented in percentages. Table 9: Collateral types and percentages
Collateral Type Number of Ratings Percentage
AB-Ref 3,557 0.42 ASSET 2,655 0.32 AUTO 6,703 0.80 AltA10 79 0.01 AltA15 6,799 0.81 AltA20 353 0.04 AltA30 74,05 8.82 AltAAR 19,933 2.37 BOAT 50 0.01 CARD 1,012 0.12 CDO 42,254 5.03 CFCDO 54,015 6.43 CFCLO 45,747 5.45 CLO 16,208 1.93 CM-Ref 723 0.09 CMBS 96,241 11.46 CNSMER 62 0.01 EQUIP 1,286 0.15 FGLMC 7 0.00 FN 7 0.00 FNCL 3 0.00 HBCDO 1,697 0.20 HBCLO 1,826 0.22
HELOC 4,411 0.53 HLTHCR 1 0.00 HOMEEQ 90,366 10.76 HOMEIP 129 0.02 HOTEL 30 0.00 LENDNG 1,423 0.17 MANUF 5,03 0.60 MULTI 887 0.11 MVCDO 466 0.06 MVCFO 17 0.00 MVCLO 134 0.02 N.A. 71 0.01 OFFICE 82 0.01 PLANE 1,946 0.23 RESB/C 91,266 10.86 RETAIL 18 0.00 RV 154 0.02 SF-Ren 239 0.03 SNCDO 6,722 0.80 SNCLO 254 0.03 STUDNT 12,725 1.51 Sm Bus 28 0.00 TAX 51 0.01 TRADE 275 0.03 UTLTY 42 0.00 VA 2,877 0.34 WH10 1,026 0.12 WH10BL 3 0.00 WH15 15,488 1.84 WH15BL 494 0.06 WH15R 9 0.00 WH20 827 0.10 WH25 28 0.00 WH30 118,229 14.07 WH30R 1,252 0.15 WH5Bln 8 0.00 WHARM 106,789 12.71
WHBln 46 0.01
WHOLE 908 0.11
Notes: Collateral types and the fraction of the sample they represent in percentages. Total number of different collateral types: 62.
Appendix II: Marginal Effects Models 4.1-4.3
Table 10: Marginal Effects Model 4.1
Score Rating (S&P) DBRSqshare Std. Error
28 AAA 0.0109*** (0.000261) 26 AA+ 0.00148*** (3.63 x 10^5) 25 AA 0.00145*** (3.54 x 10^5) 24 AA- 0.000455*** (1.15 x 10^5) 23 A+ 0.000427*** (1.09 x 10^5) 22 A 0.000297*** (8.49 x 10^6) 21 A- 7.17 x 10^6*** (1.86 x 10^6) 20 BBB+ -9.28 x 10^5*** (3.17 x 10^6) 19 BBB -0.000338*** (8.58 x 10^6) 18 BBB- -0.000255*** (6.31 x 10^6) 17 BB+ -0.000314*** (7.68 x 10^6) 16 BB -0.000480*** (1.16 x 10^5) 15 BB- -0.000243*** (6.02 x 10^6) 14 B+ -0.000256*** (6.33 x 10^6) 13 B -0.000474*** (1.16 x 10^5) 12 B- -0.000415*** (1.02 x 10^5) 11 CCC+ -0.000247*** (6.23 x 10^6) 10 CCC -0.00122*** (2.97 x 10^5) 9 CCC- -0.000733*** (1.81 x 10^5) 7 CC -0.00250*** (6.08 x 10^5) 4 C -0.00250*** (6.10 x 10^5) 0 D -0.00498*** (0.000119) Observations 795,927 -
Notes: Marginal effects of DBRSqshare per rating category. Based on estimations from Table 7, computed by STATA. *, **, *** represent a significance level of 0.10, 0.05, 0.01, respectively. Rating is according to the S&P scale. Column 3 indicates the change in probability of receiving that particular rating when DBRSqshare increases by one percentage point; i.e. with an increase of one percentage point of DBRSqshare, an asset is approximately 1 percent (0.0109) more likely to receive an “AAA” rating.
Table 11: Marginal Effects Model 4.2
Score Rating (S&P) HHI Std. Error
28 AAA -0.000102*** (1.22 x 10^6) 26 AA+ -1.35 x 10^5*** (1.79 x 10^7) 25 AA -1.33 x 10^5*** (1.74 x 10^7) 24 AA- -4.16 x 10^6*** (6.13 x 10^8) 23 A+ -3.93 x 10^6*** (5.96 x 10^8) 22 A -2.80 x 10^6*** (5.54 x 10^8) 21 A- -1.02 x 10^7*** (1.73 x 10^8) 20 BBB+ 7.82 x 10^7*** (2.32 x 10^8) 19 BBB 2.99 x 10^6*** (4.48 x 10^8) 18 BBB- 2.25 x 10^6*** (3.08 x 10^8) 17 BB+ 2.80 x 10^6*** (3.66 x 10^8) 16 BB 4.30 x 10^6*** (5.43 x 10^8) 15 BB- 2.15 x 10^6*** (2.94 x 10^8) 14 B+ 2.27 x 10^6*** (3.11 x 10^8) 13 B 4.21 x 10^6*** (5.59 x 10^8) 12 B- 3.67 x 10^6*** (5.00 x 10^8) 11 CCC+ 2.21 x 10^6*** (3.22 x 10^8) 10 CCC 1.11 x 10^5*** (1.45 x 10^7) 9 CCC- 6.79 x 10^6*** (9.29 x 10^8) 7 CC 2.35 x 10^5*** (3.00 x 10^7) 4 C 2.35 x 10^5*** (3.04 x 10^7) 0 D 4.73 x 10^5*** (5.72 x 10^7) Number of Observations 778,095 -
Notes: Marginal effects of HHI per rating category. Based on estimations from Table 7, computed by STATA. *, **, ***
represent a significance level of 0.10, 0.05, 0.01, respectively. Rating is according to the S&P scale. Column 3 indicates the change in probability of receiving that particular rating when HHI increases by one percentage point.
