Multiple Credit Rating: Triple Rating under the Requirement of Dual Rating in Korea Jin Q. Jeon, and Cheolwoo Lee*
October 2018 ABSTRACT We raise questions on a unique phenomenon where triple rating became prevalent while dual rating is required in Korea. Triple rating may improve information production by introducing increased competition among CRAs (credit rating agencies) while it may exacerbate rating inflation through more rating shopping and rating catering on the ground of greater bargaining power shifted towards the issuer. We examine the effect of triple rating on rating inflation, information production, and rating changes. Triple rating on average has a lower rating and a greater information production effect than dual rating after controlling for endogeneity. Rating inflation, however, is not monotonic in issuer size. The rating level appears to be a significant factor in shaping the future rating mandates in triple rating. The propensity that splits are resolved through rating upgrades in triple rating significantly existed but has noticeably faded away since the strict regulatory changes in 2009. Keywords: credit rating; triple rating; dual rating; rating inflation; rating shopping; rating catering JEL Classification: G24
Jeon (
[email protected]) is at Dongguk Business School, Dongguk University, 3-26 Pil-dong, Chung-gu, Seoul 100-715, South Korea. Lee (
[email protected]) is at College of Business, Ferris State University, 119 South Street, Big Rapids, MI 49307, USA. Jin Q Jeon acknowledges the financial support of the Korea Investors Service and Dongguk University Research Fund. We are responsible for all remaining errors. *
1
Multiple Credit Rating: Triple Rating under the Requirement of Dual Rating in Korea 1. Introduction The roles and functionality of the credit rating industry are paramount in the sense that, when the industry malfunctions, it leads not just only to the collapse of its own industry but also to the market-wide credit crisis, as evidenced in the wake of the recent financial crisis. Credit ratings are also a key determinant for corporate decisions such as investments and capital structure (Graham and Harvey, 2001; He, Qian, Strahan, 2012). Therefore, the quality of credit ratings is key to the proper functioning of the financial system in any market. 1 In addition to issuing credit ratings, CRAs monitor these issuers by publishing ‘outlook report’ that does not involve any ratings or rating changes. The Korean credit rating industry relatively has a short history and is relatively underdeveloped, which leads to negative perception about the credibility of the industry. In an effort to overcome such inferior status of the industry, mandatory dual rating has been enforced for all issuers since 1994 in Korea; this is a unique feature rarely observed in other countries. Aside from this unique feature, triple rating has been increasingly adopted by issuers. This is puzzling because two ratings are required by law. We focus on questions related to this new uncommon phenomenon: triple rating when dual rating is required. Specifically, how is it related to the critical dimensions for systematic bias in the credit rating literature such as rating inflation, information production, probability of financial distress, and probability of upgrade in splits? We also look at the determinants of triple rating. Rating inflation has been one of the hottest topics in the credit rating literature. It means that credit rating agencies (CRAs) purposefully understate the issuer’s risk. The issuer does not want an accurate rating while investors do. A potential problem arises when the payment for credit rating is made by one who wants biased rating and its CRA—that is supposed to maximize profits through winning mandates—succumbs to the interest of the
The quality of the ratings would affect both investors and issuers. Credit ratings are a key channel of information transmission for investors. The ratings affect issuers in the level of cost of capital (Campbell and Taksler, 2003) and capital structure (Kisgen, 2006). However, the ways the bias in credit ratings affects investors and issuers are opposite. 1
2
payer by issuing favorable ratings, not serving as an unbiased information provider. CRAs consequently have a natural incentive to serve the issuer’s interest rather than investors’ simply because revenues come from the issuer. Biased credit ratings distort investment decisions by naïve investors, who presumably heavily rely on these ratings for decision making, and lead to sub-optimal investment decisions for a firm. Rating inflation can occur due to both rating shopping and rating catering. Rating shopping is a process in which issuers shop around in search of better ratings and make only favorable ratings public (Skreta and Veldkamp, 2009; Sangiorgi, Sokobin, and Spatt, 2011; Farhi, Kerner, and Tirole, 2013). It is rational for the issuer to pursue higher rating because higher rating means more money to be raised through better access to financial market and a lower cost of debt. While rating inflation can occur by the issuer’s rating shopping, it may occur also by the CRA that has incentives to issue biased ratings in favor of the issuer to increase their market share and revenues. Bolton, Freixas, and Shapiro (2012) indicate that CRAs have incentives to loosen rating standards for securing market share as the competition intensifies. The mandatory dual rating adopted in the Korean credit rating market targets to reduce rating bias in the presence of rating shopping. Benmelech and Dlugosz (2009) show that the market with multiple rating can alleviate rating shopping, compared to the market where single rating prevails. Using the Korean credit rating data, Kang and Yi (2011) find that the dual rating reduces rating bias and improves information production. Recently, dual rating has lost ground to triple rating that gained popularity in the Korean credit rating market. We take a closer look at this new trend, triple rating. In the univariate analysis, triple rating on average has higher credit rating than dual rating, consistent with Kim (2015). However, this result can be driven by the fact that large firms selectively choose triple rating.2 In the multivariate setting after controlling for issuer characteristics, we find the opposite result; average credit rating is lower for triple rating. In addition to examining individual rating levels, we form and test five portfolios by size. Results show that rating level differentials are not monotonic in size. Credit ratings in the smallest and largest portfolios are higher for triple rating while ratings in the mid-sized portfolios are The main reason to pursue triple rating can be explained by the signal theory. Firms with financial health are more likely to send a good signal by getting an additional rating that will work as certification with more information and ramification of issuer confidence. 2
3
lower. In the analysis after accounting for endogeneity using Treatment Effects Models and Propensity Score Matching (PSM), we find qualitatively the same results. Second, we examine information production differentials between dual and triple rating. On the one hand, an additional rating can lead to a greater certification effect by adding more information and signaling confidence. On the other hand, a greater rating bias can ensue as triple rating promotes the bargaining position of an issuer and consequently facilitates rating shopping with the increased incentive of a CRA for rating catering. To test this question, we examine the probability of defaults for investment grade ratings (BBB- or higher) as a proxy for the degree of information production. We find evidence that triple rating overall has the lower probability of defaults and the strongest result is found in the largest portfolio by size. Finally, we explore rating upgrades in triple rating. If triple rating strengthens an issuer’s bargaining power, CRAs that previously issued relatively lower ratings will have a greater incentive to upgrade. To test this, we analyze whether a CRA upgrades when there was a split rating in the previous quarter. We find that triple rating tends to receive upgrades in the current quarter when there was previously a rating split. Especially, a CRA whose rating was lower than other CRAs in the previous quarter has a great probability of upgrading. This tendency to upgrade, however, became weaker after the Capital Market and Investment Service Provider Act—intended to strengthen work ethic and accountability of the CRAs in Korea—was effective in 2009. In a nutshell, we address rating bias questions in a unique setting where triple rating has become prevalent while dual rating is required by regulations. Under the mandatory dual rating system, triple rating exacerbates rating inflation in partial size groups but it overall promotes information production. After the prevalence of triple rating, CRAs tend to issue inflated ratings to secure credit rating mandates and other ancillary services while this tendency fades away since 2009. The rest of the paper proceeds as follows. In section 2, we review the institutional background of the Korean credit rating market. Section 3 reviews the related literature and develops hypotheses. Section 4 describes data, research design, and sample characteristics. In section 5, we provide empirical results. Section 6 concludes.
4
2. Institutional Background In this section, we provide an overview of the Korean credit rating industry with an emphasis on its unique features.3 The Korean CRAs are regulated by two regulating bodies: the Financial Services Commission and the Financial Supervisory Service. Credit rating was first introduced to Korea in 1985 to assess the eligibility of the issuer for commercial papers. Issuers with a B rating or higher only were allowed to issue. The Korean credit rating industry has since been adopting the issuer-pay model. In the following year, 1986, credit rating expanded the underlying financial securities into corporate and convertible bonds. In 1994, a rule that mandates dual rating for corporate and non-guaranteed bonds became effective, which is unique and different from most countries, including the US credit rating industry. In Korea, dual rating is required by regulations. Korea had gone through the Asian Financial Crisis in the late 1990s. Before the crisis, guaranteed bonds only were issued, which made the role of the CRAs fairly limited. Since then, there was a material change in the composition of bonds in Korean credit market, transitioning from guaranteed to mostly non-guaranteed bonds. The Korean credit rating industry has a different market structure. In the U.S., the recent prominence of the Fitch’s is recognized as the increase in the competition, but the competitive landscape is quite different in Korea. The Korean market has taken a different path, not experiencing meaningful changes in the degree of competition like the U.S. and others. All three major CRAs had been established with similar competitiveness before our sample period and have had the same size of market share with no significant variation over time—each CRA’s market share has consistently been around one-third of all. For example, all three major CRAs’ market shares range between 31.6% and 35.6% for the 2007-2015 period according to the Financial Supervisory Service in Korea, with a trend of being closer to each other over time. These different industry features of the Korean credit rating market provide a unique setting to examine questions regarding triple rating to shed further light on the issue of the bias in credit ratings. Under the different market setting and market Although minor differences are present, the regulations governing the Korean credit market are fundamentally similar in their goals and effectiveness to those of the U.S. and Europe regarding independence and resolution for potential conflict of interest (Kim, 2015). 3
5
structure, we test why the third rating is pursued when two ratings are required by the regulations, and how it affects critical dimensions of credit rating. As an effort to discourage rating shopping and CRA malpractices surfaced in the 2008 financial crisis in addition to the mandatory dual rating, the Capital Market and Investment Service Provider Act became effective in 2009 to enhance work ethic and accountability of the CRAs and to provide protective measures for investors. The Act redefines credit rating business in a broader extent for better investor protection and lawfully requires more stringent standards and internal control rules for CRAs to stay in the business without interference and probation. The internal control rules include a separation of the evaluation unit and sales unit within a CRA, no conflict of interest, no unfair trade involving credit rating, and establishing standardized evaluation protocols for each rating object type. As a continued effort, the Financial Investment Services and Capital Markets Act of 2013 prohibits the shadow rating in which an issuer solicits a pre-rating assessment without making any binding credit rating contract between the issuer and its CRA.
3. Related Literature and Hypothesis Development 3.1.
Related Literature In this section, we review papers tightly related to our questions only.4 A few early
papers explore why the issuer goes for multiple ratings. Hsueh and Kidwell (1988) find that the issuer obtains a second rating to lower their borrowing cost and it provides more information to the market, which is socially desirable. Cantor and Packer (1995) examine but find little evidence that regulatory concerns are the main driver for additional ratings, except some suggestive evidence for junk bonds. Cantor and Packer (1997) also fail to find a link between a third rating and plausible considerations such as information production, rating shopping, and certification while finding evidence for systematically higher ratings for the third ratings. Rating inflation can be driven by either rating shopping or rating catering when we consider the issuer and rater only although there are other factors such as asset complexity For a comprehensive review on credit rating, refer to Graham and Harvey (2001), Campbell and Taksler (2003), and Cornaggia, Cornaggia, and Israelsen (2017). 4
6
(Skreta and Veldcamp, 2009) and the degree of the competition in the credit ratings market (Camanho et al., 2012). Policy and law makers proposed the increased competition as a way to improve rating quality (Becker and Milbourn, 2011). Skreta and Veldkamp (2009) and Bolton, Freixas, and Shapiro (2012) provide theoretical models of rating shopping in relation to rating quality, and their shared conclusion is that competition induces rating shopping. Consistent with these models, Becker and Milbourn (2011) empirically find that increased competition—by the entry of Fitch—leads to inflated ratings and poorer predictive power for defaults. Competition is likely to distort the CRA’s incentive issuing accurate ratings and to induce rating bias in favor of the issuer. On the other hand, a body of studies find no evidence for rating shopping (Cantor and Packer, 1997; Jewell and Livingston, 1999; Covitz and Harrison, 2003; and Bongaerts, Cremers, and Goetzmann, 2012). Griffin, Nickerson, and Tang (2013) find that rating catering is more likely when the issuer is involved in rating shopping, which suggests a close interconnection of rating catering and shopping. This relationship would be more pronounced when the competition among CRAs is fiercer as the catering becomes more important to secure a deal and the shopping therefore becomes easier under the elevated level of catering incentive.
3.2.
Hypothesis Development We aim to examine rating quality or bias between triple and dual ratings through
rating inflation, information production, default probability, and likelihood of upgrade in rating splits. Rating inflation is our main measure for the rating quality. Ratings should ideally not be inflated, accurately predict defaults, and not be used as a tool for resolving splits in the absence of relevant information. Investors want informative ratings that enable them to make appropriate investment decisions while issuers want inflated ratings with which issuers can enjoy benefits, i.e., in the form of a lower cost of capital, better corporate image, and broader investor base.5 Rating tiers are the direct object that we can test rating quality because the tiers will reveal whether the issuer puts pressure on the CRA (Becker and Milbourn, 2011).
Institutional investors—such as pension funds, commercial banks, and insurance companies—face regulatory rules under which they are allowed to include investment grade bonds only in their portfolios in many countries. 5
7
Goel and Thakor (2011) argue that reputation mechanism provides incentives for CRAs to engage in costly information production and penalties are placed on optimistically biased CRAs, not on pessimistically biased CRAs. More ratings would therefore reduce information asymmetry and uncertainty about the issuer for investors. Especially multiple ratings in agreement would increase investor confidence. When the issuer is confident about their quality, they will go for the third rating so investors are more willing to purchase. Because each CRA has their own rating methodologies and processes, the information missing in other CRAs’ reports can be incorporated into the rating by a third CRA. Therefore, more rating is likely more and better information. It can translate into less positively biased ratings given systematic upward bias in ratings. We hence argue that rating levels are lower due to more, better, and complementary information gained in triple rating. Informative or unbiased ratings should predict defaults better. For the same rating tiers, we expect a lower likelihood of financial distress for triple rating. Information-driven rating is also less prone to be used only for resolving splits. We label this line of reasoning as the Information Production Hypothesis: H1a: Rating levels are lower for triple rating H1b: Probability of financial distress is lower for triple rating H1c: Probability of the upgrade in splits is less likely for triple rating In contrast, Mathis, McAndrews, and Rochet (2009) note that the revenue from inflating ratings is sufficiently large to induce rating bias. Other drivers include investor naïveté and the fact that indicative ratings are kept private. Many prior studies find that the issuer-pay model systematically leads to rating inflation (Jiang, Stanford, Xia, 2012; Cornaggia and Cornaggia, 2013). Bolton, Freixas, and Shapiro (2012) show that increased competition through more competitors in the market may create an incentive for CRAs to inflate ratings, in an effort to win the competition that is not present in a monopolistic market setting.6 Such incentive is reinforced by the fact that the issuer can selectively make ratings
This may seemingly be against what economic theory suggests. In a typical economic setting, more competition usually leads to more efficiency in prices and quality. However, it is not clear cut whether competition among CRAs is socially beneficial. What these papers find are not surprising and in line with the theory given that inflated ratings please the consumer (issuer). Flynn and Ghent (2017) mention similar settings where competition aggravates product quality: vehicle emissions testing and auditing of financial statements. In all these settings, payers and consumers are not the same and they have different goals. 6
8
public and CRAs are paid only when their ratings become public. Further, CRAs can advise the architecture of the financial instruments, which provides another source for the potential conflict of interest. Several papers argue that major CRAs have no strong incentive to provide quality ratings because they wield virtually monopolistic power in the market (e.g., White, 2002). Although reputation concerns may suppress the CRA’s incentive to inflate ratings, it may not be always strong enough to abstain from rating inflation. For example, Mathis, McAndrews, and Rochet (2009) note that reputation upholding incentives work only when the fee is sufficiently high. When the expected future rents for inflating ratings are greater than the rents for reputation upkeep, incentives to be accurate would be consequently weak. Bar-Isaac and Shapiro (2013) theoretically show that when reputation damage is small, CRAs have less incentive to be accurate. Becker and Milbourn (2011) argue that increased competition by the entry of Fitch jeopardized reputational incentives. According to the bond rating survey by the Association for Financial Professionals (2002), 25 percent of respondents showed skepticism about the rating accuracy, and long-term and low-rated bonds involve substantial precision issues (Sangiorgi, Sokobin, and Spatt, 2009). A third rating is not mandatory and an issuer has an option not to make the third rating public after seeking it at the third CRA while it is an additional opportunity for the CRA to earn business. Cornaggia and Cornaggia (2013) show that CRAs tend to cater to the issuer’s demand and understate the issuer’s risk and probability to default. This is in part because accurate ratings do not necessarily lead to more profits for CRAs while favorable ratings do through more future credit rating mandates based on the friendly established relationship. This setting naturally gives a bargaining edge to the issuer and consequently strengthens a likelihood of rating shopping and catering. Prior literature frames the increased market presence of Fitch as increased competition (e.g., Becker and Milbourn, 2011). In a way, the emergence of triple rating can be viewed as the manifestation of the increased competition in the market, given that triple ratings mean more mandates for the same number of competitors that split up these additional mandates. Before and after the prevalence of triple rating, the market has been largely equally shared by the three major CRAs. Securing the third rating mandate is up for grab for the third CRA that has not yet rated. Naturally, the issuer sits in the catbird seat when working on the rating contract between them. Under the increased competition with the issuer-pay model, CRAs likely compete not to provide accurate ratings but to please 9
issuers by inflating ratings, as shown in Skreta and Veldkamp (2009) and Bolton, Freixas, and Shapiro (2012). Given that the unique structure of credit rating market and the issuer pay model naturally facilitate rating inflation as shown in the prior literature, ratings are more likely to be inflated when ratings are strategically driven due to increased competition and issuer’s better bargaining position in triple rating. If the third rating is systematically inflated, we should observe consistently higher ratings for triple-rating issuers and these ratings would not accurately predict financial distress because they are not information-driven. More information does not always lead to better information especially when the information is biased. The poor quality of ratings in turn reduces the value of ratings to predict the issuer’s financial distress. It is more likely that upgrades are used to resolve splits because the CRA has a greater incentive to satisfy the issuer’s demand due to the increased competitive pressure. We label this as the Rating Catering Hypothesis: H2a: Rating levels are higher for triple rating H2b: Probability of financial distress is higher for triple rating H2c: Probability of the upgrade in splits is more likely for triple rating
4. Sample, Research Design, and Summary Statistics 4.1 Sample, Data, and Variables For our analysis, we use the credit rating data from 2003 to 2015, provided by the Korean Investors Service (KIS), an affiliate of the Moody’s Investors Service since December 2001. The KIS was the first credit rating agency and introduced the Moody’s advanced rating system into Korea. To be included in our sample, we require firms to be found in the KOSPI (Korean Composite Stock Price Index) and KOSDAQ (Korean Securities Dealers Automated Quotations). The KOSPI is equivalent to S&P 500 and KOSDAQ to NASDAQ (National Association of Securities Dealers Automated System) in the Unites States. We exclude mutual funds, REITs (real estate investment trusts), SPACs (special purpose acquisition companies), and ship investment companies.7 Financial Statements information and Stock 7
Information on the ship investment companies: http://www.koreashipfinance.com/eng/sic/sic01.asp 10
prices are from the KIS Value (http://www.kisvalue.com) and DataGuide (http://www. dataguide.co.kr), respectively. [Insert Table 1 around here] Table 1 shows the yearly breakdown of dual rating and triple rating, further by all ratings and corporate bond ratings. The number of credit ratings had gradually increased from 2003 to 2010 and since then stayed fairly constant, ranging between 1,004 and 1,017. For all the credit ratings including both issuer and bond ratings, the proportion of dual (triple) rating decreased (increased) over time, from 72% to 52% (28% to 48%). Our sample overall has 60% dual ratings and 40% triple ratings. For corporate bond ratings, triple rating increased from 5% to 36% during our sample period. Triple rating has continuously gained popularity during our sample period. All the variables used in our analysis are defined in Table 2. [Insert Table 2 around here]
4.2. Research Design As aforementioned, we examine the quality of credit ratings in terms of dual and triple rating under the regulatory system where mandatory dual rating is in place. First, we estimate the following ordinary least square (OLS) regressions to test the effect of triple rating on the rating inflation with Credit Rating as our dependent variable and Triple Rating as the key independent variable: 𝐶𝑟𝑒𝑑𝑖𝑡 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 = 𝛼 + 𝛽 ∙ 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 + 𝐵 ∙ 𝑋 + 𝑒𝑖𝑡
(1)
where Credit Rating is a rating that is converted into a numerical rating scale as described in Table 2; Triple Rating is an indicator variable that takes the value of one if one receives credit ratings from all three CRAs. Otherwise, zero; X is a vector of control variables that include issuer characteristics such as Size, Sales_Growth, Market to Book, ROA, Leverage, and Volatility, and macroeconomic variables that represent industry- and year-fixed effects. Some issuers are able to put pressure on the CRAs and some CRAs are more susceptible to the pressure. To control for these possibilities, we include issuer- and CRA-specific variables
11
in our estimation. If Triple Rating in equation (1) obtains a negative coefficient with statistical significance, we can conclude that triple rating is more lenient than dual rating. It is plausible that issuers requesting triple rating are more likely to be big firms. In this case, estimates may suffer bias due to potential endogeneity issues between Credit Rating and Triple Rating. To address this potential endogeneity, we estimate the Treatment Effects Model (TEM) and Propensity Score Matching (PSM) Model. In the TEM estimation, we estimate the logit regression with Triple Rating as its dependent variable using the following specification (Maddala, 1983):
∗ 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 = 𝛼 + 𝐵 ∙ 𝑋 + 𝛾 ∙ 𝑧 + 𝑒𝑖𝑡
(2)
𝑤ℎ𝑒𝑟𝑒 𝑇𝑟𝑖𝑝𝑙𝑒 = 1 𝑖𝑓 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔∗ > 0, 𝑜𝑟 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒. where z is a vector of instrumental variables: Annual Avg. Triple and Industry Avg. Triple. The annual average triple rating and industry average triple rating are directly related to the likelihood that one receives triple rating but are not related to the level of credit rating. Therefore, these two instrumental variables would not suffer the weak instrument problem. We estimate the second stage regression after calculating the Inverse of Mill’s Ratio: 𝜙(𝐵̂𝑋 + 𝛾̂𝑧) 𝑖𝑓 𝑇𝑟𝑖𝑝𝑙𝑒 ∗ = 1 ̂ 𝑋 + 𝛾̂𝑧) Φ(𝐵 𝜆̂ = −𝜙(𝐵̂𝑋 + 𝛾̂𝑧) 𝑖𝑓 𝑇𝑟𝑖𝑝𝑙𝑒 ∗ = 0 { Φ(𝐵̂𝑋 + 𝛾̂𝑧)
(3)
𝐶𝑟𝑒𝑑𝑖𝑡 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 = 𝛼 + 𝛽 ∙ 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 + 𝐵 ∙ 𝑋 + 𝜌𝜆̂ + 𝑒𝑖𝑡 In equation (3), 𝜆̂ is the inverse of Mill’s Ratio estimate. We can conclude that endogeneity exists if 𝜌 appears significant. The PSM has become a standard and common approach to make causal inferences based on observational data that are not generated by controlled experimental settings. For ̂𝑖 ) using the the PSM analysis, we estimate the probability that one receives triple rating (𝑃 Probit model as follows (Rosenbaum and Rubin, 1983): ∗ 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 = 𝛼 + 𝐵 ∙ 𝑋 + 𝑒𝑖𝑡
𝑤ℎ𝑒𝑟𝑒 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔 = 1 𝑖𝑓 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔∗ > 0, 𝑜𝑟 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒.
12
(4)
̂𝑖 = 𝑃(𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔 = 1) = Φ(𝑋𝑖′ 𝐵̂) 𝑃 We match triple rating (treated group) with dual rating (control or untreated group) using the propensity scores ensuring that the control group is as similar as possible to the treatment group with respect to observable characteristics. We then calculate the ATT (average treatment effect on the treated) and test the statistical significance using the t-test. The ATT is expressed as: 𝐴𝑇𝑇 = 𝐸(𝐶𝑟𝑒𝑑𝑖𝑡 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑇 − 𝐶𝑟𝑒𝑑𝑖𝑡 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝐶 |𝑋, 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔𝑖 = 1)
(5)
where T stands for treated group and C control group. Next, we examine whether triple rating is superior to dual rating in terms of information production. We estimate the probability of financial distress on the ratings with BBB- or higher for both triple and dual ratings. If the probability for triple rating is lower than that of dual rating, we can conclude that triple rating improves information production. ∗ 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐷𝑖𝑠𝑡𝑟𝑒𝑠𝑠𝑖𝑡 = 𝛼 + 𝛽 ∙ 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔 ∙ 𝐼𝑛𝑣. 𝐺𝑟𝑎𝑑𝑒𝑖𝑡 + 𝐵 ∙ 𝑋 + 𝑒𝑖𝑡
(6)
𝑤ℎ𝑒𝑟𝑒 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐷𝑖𝑠𝑡𝑟𝑒𝑠𝑠 = 1 𝑖𝑓 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐷𝑖𝑠𝑡𝑟𝑒𝑠𝑠 ∗ > 0, 𝑜𝑟 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒. In equation (6), Financial Distress is defined as when an issuer’s TIE (times interest earned; EBIT/Interest) is less than 100% and debt to equity (D/E; total debt/total equity) is higher than 200% (Claessens, Djankov, and Ferri, 1999). If 𝛽 is significantly negative, the financial distress of investment grades rated by triple rating is lower. That is, it translates into a better performance for triple rating, supporting the information production hypothesis. Our third question is whether triple rating affects future credit rating mandates. To address this question, we estimate the multinomial logit model: 𝑃(𝑦𝑖 = 𝐾) = 𝑒𝑥𝑝(𝑥𝑖′ 𝛽𝐾 ) / ∑
𝐽 𝑗=1
𝑒𝑥𝑝(𝑥𝑖′ 𝛽𝑗 )
(7)
In equation (7), there are three cases possible (J): Drop, Stay, and Join. When the CRA who rated issuer i is excluded in the corporate bond rating (Drop), j=1; when the CRA rates the corporate bond (Stay), j=2; when the CRA did not rate issue i but participates in corporate bond rating (Join), j=3. We use Drop (j=1) as our base outcome. 𝑃(𝑦𝑖 = 𝐾) is the probability that a CRA has one of the choices—Drop, Stay, and Join—for issuer i. If 𝛽𝑗 is 13
positive, we can interpret that the probability of Stay or Join will increase as 𝑥𝑖 increases, compared to the base outcome Drop. Finally, we test whether upgrades are more likely for triple rating when rating splits are present. We first examine whether the split is resolved in the following quarter using the following specification: 𝑈𝑝𝑔𝑟𝑎𝑑𝑒𝑖𝑡∗ = 𝛼 + 𝛽 ∙ 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔 ∙ 𝑆𝑝𝑙𝑖𝑡𝑖𝑡 + 𝐵 ∙ 𝑋 + 𝑒𝑖𝑡
(8)
𝑤ℎ𝑒𝑟𝑒 𝑈𝑝𝑔𝑟𝑎𝑑𝑒 = 1 𝑖𝑓 𝑈𝑝𝑔𝑟𝑎𝑑𝑒 ∗ > 0, 𝑜𝑟 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒. To test whether a CRA that previously had a relatively lower rating upgrades, we estimate the following model: 𝐶𝑅𝐴 − 𝑖 𝑈𝑝𝑔𝑟𝑎𝑑𝑒𝑡 = 𝛼 + 𝛽 ∙ 𝐶𝑅𝐴 − 𝑖 𝐿𝑜𝑤𝑡−1 + 𝐵 ∙ 𝑋 + 𝑒𝑖𝑡
(9)
𝑤ℎ𝑒𝑟𝑒 𝐶𝑅𝐴 − 𝑖 𝑈𝑝𝑔𝑟𝑎𝑑𝑒𝑡 = 1 𝑖𝑓 𝐶𝑅𝐴 − 𝑖 𝑢𝑝𝑔𝑟𝑎𝑑𝑒𝑠, 𝑜𝑟 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 In equation (9), CRA-i Upgradet is an indicator variable that takes the value of one if a CRA i upgrades in comparison to its previous rating and zero otherwise. CRA-i is one of CRA-X, CRA-Y, and CRA-Z. CRA-i Lowt-1 is an indicator variable that takes the value of one if a CRA i’s average rating is lower than the average of the other two CRAs’ ratings and zero otherwise. If 𝛽 is significantly positive, it suggests that a CRA having a relatively lower rating in the previous quarter is more likely to upgrade in the following quarter, in line with the rating catering hypothesis.
4.3. Summary Statistics Table 3 summarizes sample characteristics. As shown in Table 1, Panel A of Table 3 shows that 11,071 credit ratings were made during our sample period (2003-2015), 6,560 with dual rating and 4,511 with triple rating. Panel A uses all the credit ratings. Credit ratings with AA- or higher are between 20.6% and 29.9% for dual rating while those are 46.1% for triple rating. Credit ratings with BBB or Higher are also higher for triple rating (between 84.76% and 86.25% vs. 94.26%). Rating Splits are more frequent in triple rating, 12.7%, than in dual rating, 8.35% or lower. The market value of the issuer for triple rating is on average 5.47 Trillion KRW (Korean Won) while those for dual ratings for three combinations of the
14
three CRAs are 1.67 Trillion KRW or lower. The gap between the issuer groups is even wider when comparison is based on total assets. When corporate bond ratings only are considered in Panel B, the qualitatively same trends are observed between triple and dual ratings while rating tiers are higher, splits are lower, and sizes measured by market value and total assets are greater than those in Panel A. [Insert Table 3 around here]
5. Empirical Results In this section, we test the hypotheses we established in section 3.2. First, we examine whether credit ratings are higher for triple rating than for dual rating. We then analyze the likelihood of financial distress to see if triple rating generates greater information effect. Further, we test the relationship between rating levels and future credit rating mandates and assess the probability of upgrades when splits are present.
5.1.
Is Rating Inflation Greater for Triple Rating? Table 4 reports the univariate tests of the differences in credit rating tiers between
dual and triple ratings. To facilitate quantitative analysis, we convert credit rating tiers into a numerical rating scale system following Bongaerts, Cremers, and Goetzmann (2012). The detail on the conversion is described in Table 2. Rating tiers range from 1 (AAA) to 20 (D), with a lower numerical score meaning a higher rating. Panel A of Table 4 uses all the credit ratings. The average ratings for dual rating range between 6.49 and 6.935 while the medians are 6 and 7. Rating tier A (A-) corresponds to 6 (7). On the other hand, the mean (median) for triple rating is 4.947 (5), corresponding to A+ (AA-). The differences in means and medians are statistically significant in the t-tests and rank-sum tests, respectively. Triple rating on average has higher credit rating, which is also found in Pane B when corporate bond ratings are considered in isolation. [Insert Table 4 around here] Panel C of Table 4 reports the differences between dual and triple rating groups by portfolio size. We classify the sample issuers into five portfolios by the market value in each quarter, with the first largest 20% in portfolio Largest and the smallest 20% in Smallest. 15
Panel C shows the means and medians of those five groups again by dual and triple rating groups. Triple rating is clustered at the Largest (largest 20%) and P80 (second largest 20%) issuer groups while dual rating at Smallest (smallest 20%) and P40 (second smallest 20%). Triple rating in the Largest and P80 portfolios is 66% of all (=2,982/4,511 where 2,982=1,241+1,741). The rating tiers are statistically higher for triple rating in the Largest and Smallest portfolios while the tiers are higher for dual rating in the other mid-sized portfolios—P40, P60, and P80. This is interesting because rating inflation occurs in the largest and smallest market value groups. However, these results are based on simple univariate tests, not controlling for other dimensions that may critically affect the degree of rating inflation. To pin down the differences in the degree of rating inflation, one needs regression analysis in the multivariate framework after controlling for relevant factors that affect the degree, which we do in the following section. Further, testing the question requires that we address the issue of the endogeneity raised in section 3.2 since factor(s) may simultaneously affect rating tiers and triple rating. The conclusion established earlier is valid only when triple rating is exogenous to rating tiers. To control for the endogeneity, we use the Treatment Effects Model (TEM) and Propensity Score matching (PSM) analysis. [Insert Table 5 around here] Table 5 reports the multivariate regression results. The univariate OLS estimation in Model (1) of Panel A with Triple Rating, an indicator variable for triple rating, as its only independent variable shows the same results as in Table 4. In Model (2) where the estimation is controlled for the control variables, however, Triple Rating switches the sign while significant at the 1% level. The rating tiers for triple rating appear to be lower after accounting for the firm characteristics. Among the control variables, rating tiers are positively correlated with Size and ROA, and negatively with Debt Ratio and Volatility. In Model (3) of Panel A in Table 5, we estimate the TEM using Annual Avg. Triple— the annual average rating of triple ratings—and Industry Avg. Triple—the average rating of triple ratings within the industry—as the instrumental variables (IVs). we estimate the logit model with Triple Rating as its dependent variable in the first stage regression. These two IVs are significantly correlated with Triple Rating, suggesting that our specifications do not suffer the weak instrument problem. The statistical significance of Lambda suggests that analysis based on the OLS is biased and problematic due to the existence of endogeneity. The TEM results have qualitatively the same implications but Triple Rating has more than ten 16
times greater coefficient for that in the OLS estimation of Model (2). To put this coefficient into perspective, triple rating has three-notch lower crediting rating tiers than does dual rating, e.g., AA- for triple rating vs. AAA for dual rating). As shown and expected from the results in Table 4, Size is a critical determinant for triple rating. Panel B of Table 5 reports the PSM results. We estimate the probability of triple rating by running the logit model to generate propensity scores and facilitate matching. The treated group (triple rating) has the average rating of 4.903 while the control group (dual rating) has that of 4.681, showing the similar relative difference in rating tiers between two groups. The ATT (average treatment effect on the treated) is 0.22 that is significant at 10% level Results to date suggest that triple rating has more conservative ratings after accounting for financial attributes of the issuer and endogeneity, which is against the rating catering hypothesis and lends support to the information production hypothesis. [Insert Table 6 around here] Table 6 shows the results on how triple rating affects rating upgrade by portfolio size. To conserve space, we report the second stage results only in Panel A. Triple Rating in Models (1) and (5) where the smallest and largest 20% of the sample issuers are considered has higher rating (negative coefficients with statistical significance), respectively. On the other hand, the other portfolios—P40, P60, and P80—have lower rating. Triple Rating is significant at the 1% level in all specifications. Panel B using the PSM shows the qualitatively same while the result on portfolio P80 is not statistically significant. Rating inflation exists in two groups by market value, Largest and Smallest. In sum, triple rating is overall more conservative when all sample ratings unconditional on size are considered but the conservativeness differs in the size of the issuer, with rating inflation in two extreme size groups.
5.2.
Does Triple Rating Improves Information Production? In this section, we test whether triple rating provides greater information production
than does dual rating. In a market where rating inflation exists, if CRAs employ their strict standards in issuing credit rating, more information production will entail as more CRAs rate the issuer. If credit rating is affected more by the interest in revenues and market shares
17
rather than by internally-set rating standards, the relationship between the number of CRAs and information product will be insignificant or even negative. To estimate the degree of information production, we estimate the probability that investment grade (IG) issuers—issuers with BBB- or higher—get into financial distress in the future. If the IG issuer has a higher probability of being in financial distress in the future, it will represent a lower level of information production. An issuer is defined as in Financial Distress when it experiences a debt to equity ratio (D/E; total debt/total equity) higher than 200% and a times interest earned (TIE; EBIT/interest where EBIT is earnings before interest and taxes) lower than 100% for four quarters after the rating (Classens, Djankov, and Ferri, 1999). [Insert Table 7 around here] “Financial Distress=1” in Table 7 means that an issuer experiences Financial Distress as defined in the previous paragraph. Financial distress occurred in 914 ratings with 525 in dual rating and 391 in triple rating for all rating tiers. When IG ratings only are considered, dual rating had 295 while triple had 288 occurrences. The means and medians for the issuers in financial distress are 10.01 (9) for dual rating and 8.02 (8) for triple rating. [Insert Table 8 around here] Table 8 reports the logit estimations on how IG ratings with triple rating affect the probability of being in financial distress. The dependent variable is Financial Distress, an indicator variable that takes a value of one if an issuer experiences financial distress for the four quarters after receiving a rating. Our key variable of interest in Table 8 is the interaction variable, 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔 × 𝐼𝑛𝑣. 𝐺𝑟𝑎𝑑𝑒, which is a product of the triple rating dummy and the investment grade dummy. 𝑇𝑟𝑖𝑝𝑙𝑒 𝑅𝑎𝑡𝑖𝑛𝑔 × 𝐼𝑛𝑣. 𝐺𝑟𝑎𝑑𝑒 appears significant with a negative sign in portfolios Smallest, P60, and Largest. This interaction variable is positive for P40 and P80 but not statistically significant. The probability of IG ratings getting into financial distress is lower for triple rating, suggesting that triple rating provides higher information production. The results on Table 8 are suggestive evidence for the information production hypothesis and may be a sign of reputation upholding by CRAs.
5.3.
Is Upgrade More Likely when Splits are Present?
18
In this section, we examine whether a CRA that issued a lower rating is more likely to upgrade in the subsequent rating change when rating split is present. It is plausible that an issuer with a split attempts to resolve it by hiring a new CRA as the issuer is in a greater bargaining position in the market. In order to examine this possibility, we analyze whether upgrades are more prevalent in triple ratings with splits. We also test whether a CRA that previously assigned a relatively lower rating upgrades in the following rating. Table 9 shows the probability of upgrade when triple rating has a split using a Logit regression. The coefficient for Triple Rating×Split is positive and statistically significant. Triple rating with a split tends to have an upgrade in the next quarter. It implies that split is resolved by an upgrade when a CRA assigned a relatively lower rating in the previous quarter and a greater bargaining power of the issuer under the peculiar market structure influences its CRA to issue a more lenient rating. [Insert Table 9 around here]
Table 10 compares the average ratings by upgrading CRAs to those by the other CRAs. When CRA-X upgrades, its average rating is lower than the average ratings by the other CRAs (7.06 vs. 7.00). When CRA-Y (CRA-Z) upgrades, the average rating changes from 7.08 (7.04) to 5.95 (5.87). When each of the CRAs does not upgrade, the rating of the incumbent CRA was consistently higher than that of the other CRAs. [Insert Table 10 around here] Table 11 shows the logit regression results on how the rating level in the previous quarter affects the likelihood of upgrade in the current quarter. The dependent variable is an indicator variable that takes the value of one if each CRA upgrades and zero otherwise. The key explanatory variable is CRA-i Lowt-1 where i is an ith CRA, an indicator variable that takes the value of one when a rating of an incumbent CRA is lower than other CRAs in the previous quarter and zero otherwise. For all three CRAs, the likelihood of upgrade is significantly higher when each has a lower rating in the previous quarter than other CRAs. This finding is in line with the notion that splits are resolved through the upgrade in triple rating and triple rating in pursuit of split resolution exacerbates rating shopping. [Insert Table 11 around here]
19
The Capital Market and Investment Service Provider Act became effective in 2009. This regulation intends to strengthen work ethic and accountability of the CRAs. To control for this effect, we create an indicator variable—After 2009—that takes the value of one if a rating is issued after 2009 and zero otherwise. The coefficient for After 2009 is negative and significant for all CRAs. The regulation has significantly reduced upgrades in triple rating, suggesting its effectiveness.
6. Conclusion Rating shopping is one of the reasons that credit rating receives low credibility from investors. In Korea, dual rating is mandatory in order to limit the bias due to rating shopping. While dual rating is required, issuers increasingly elect to obtain a third rating. Triple rating may enhance information production by introducing increased competition among CRAs and complementary information channels while it may aggravate rating inflation through more rating shopping and rating catering due to the issuer’s advantageous bargaining power. To explore the quality of triple rating and its consequences on the critical dimensions of credit rating, we examine the determinants of triple rating, rating inflation, probability of defaults, and splits between dual and triple rating. Triple rating overall has a lower rating and a greater information production effect than dual rating after controlling for endogeneity. Rating inflation, however, appears to be not monotonic in issuer size. The rating level appears to be a significant factor in shaping the future rating mandates in triple rating and, consequently, triple rating provides an incentive for CRAs to generate inflated credit ratings. The propensity that splits are resolved through rating upgrade in triple rating meaningfully existed in the Korean credit rating market but has noticeably faded away since the strict regulatory changes in 2009.
20
References Association for Financial Professionals. (2002). Rating agencies survey: accuracy, timeliness, and regulation. Bethesda, Maryland. Bar-Isaac, H., & Shapiro, J. (2013). Ratings quality over the business cycle. Journal of Financial Economics, 108(1), 62-78. Becker, B., & Milbourn, T. (2011). How did increased competition affect credit ratings? Journal of Financial Economics, 101(3), 493-514. Benmelech, E., & Dlugosz, J. (2009). The alchemy of CDO credit ratings. Journal of Monetary Economics, 56(5), 617-634. Bolton, P., Freixas, X., & Shapiro, J. (2012). The credit ratings game. The Journal of Finance, 67(1), 85-111. Bongaerts, D., Cremers, K. J., & Goetzmann, W. N. (2012). Tiebreaker: Certification and multiple credit ratings. The Journal of Finance, 67(1), 113-152. Campbell, J. Y., & Taksler, G. B. (2003). Equity volatility and corporate bond yields. The Journal of Finance, 58(6), 2321-2350. Cantor, R., & Packer, F. (1995). The credit rating industry. The Journal of Fixed Income, 5(3), 10-34. Cantor, R., & Packer, F. (1997). Differences of opinion and selection bias in the credit rating industry. Journal of Banking & Finance, 21(10), 1395-1417. Claessens, S., Djankov, S., & Ferri, G. (1999). Corporate distress in East Asia: Assessing the impact of interest and exchange rate shocks. Emerging Markets Quarterly, 3, 8-14 Cornaggia, J., & Cornaggia, K. J. (2013). Estimating the costs of issuer-paid credit ratings. The Review of Financial Studies, 26(9), 2229-2269. Cornaggia, J., Cornaggia, K. J., & Israelsen, R. D. (2017). Credit ratings and the cost of municipal financing. The Review of Financial Studies. Forthcoming. Covitz, D. M., & Harrison, P. (2003). Testing conflicts of interest at bond rating agencies with market anticipation: Evidence that reputation incentives dominate. Working Paper, Federal Reserve Board, Washington, DC. Farhi, E., Lerner, J., & Tirole, J. (2013). Fear of rejection? Tiered certification and transparency. The RAND Journal of Economics, 44(4), 610-631. Flynn, S., & Ghent, A. (2017). Competition and credit ratings after the fall. Management Science. Forthcoming. Goel, A. M., & Thakor, A. V. (2011). Credit ratings and litigation risk. Working paper. Washington University in St. Louis. 21
Graham, J. R., & Harvey, C. R. (2001). The theory and practice of corporate finance: Evidence from the field. Journal of Financial Economics, 60(2), 187-243. Griffin, J. M., Nickerson, J., & Tang, D. Y. (2013). Rating shopping or catering? An examination of the response to competitive pressure for CDO credit ratings. Review of Financial Studies, 26(9), 2270-2310. He, J. J., Qian, J. Q., & Strahan, P. E. (2012). Are All Ratings Created Equal? The Impact of Issuer Size on the Pricing of Mortgage‐Backed Securities. The Journal of Finance, 67(6), 2097-2137. Hsueh, L. P., & Kidwell, D. S. (1988). Bond ratings: are two better than one?. Financial Management, 46-53. Jiang, J. X., Stanford, M. H., & Xie, Y. (2012). Does it matter who pays for bond ratings? Historical evidence. Journal of Financial Economics, 105(3), 607-621. Kim, P., 2015, A Study on the improving functions and role of credit rating agencies in Korea. Korean Capital Market Institute Report 15-13. Kang, K-H., & Yi, J. (2011) The effects on rating shopping of the requirement of multiple credit rating in Korea. The Journal of Money and Finance, 25(3), 93-121. Kisgen, D. J. (2006). Credit ratings and capital structure. The Journal of Finance, 61(3), 1035-1072. Macey, J. Wall Street Versus Main Street: How Ignorance, Hyperbole, and Fear Lead to Regulation’(1998). University of Chicago Law Review, 65, 4-1500. Maddala, G. S. (1983). Limited-dependent and qualitative variables in econometrics (No. 3). New York: Cambridge University Press. 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. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 41-55. Sangiorgi, F., Sokobin, J., & Spatt, C. (2009). Credit-rating shopping, selection and the equilibrium structure of ratings. Working Paper, Stockholm School of Economics and Carnegie Mellon University. Skreta, V., & Veldkamp, L. (2009). Ratings shopping and asset complexity: A theory of ratings inflation. Journal of Monetary Economics, 56(5), 678-695. White, L. (2002). An Industrial Organization Analysis of the Credit Rating Industry, in M.K. Ong, ed., Credit Ratings: Methodologies, Rationale, and Default Risk, Risk Books.
22
Table 1. Distribution of Dual and Triple Ratings by Year The table shows the yearly breakdown of credit ratings into dual rating and triple rating for both all and corporate bond ratings. For our analysis, we use the credit rating data from 2003 to 2015, provided by the Korean Investors Service (KIS). We exclude mutual funds, REITs (real estate investment trusts), SPACs (special purpose acquisition companies), and ship investment companies.8 Financial Statements information and Stock prices are from the KIS Value (http://www.kisvalue.com) and DataGuide (http://www.dataguide.co.kr), respectively. All ratings include both issuer and bond ratings. An issuer or a bond is classified as dual (triple) rating when two (three) CRAs (credit rating agencies) rate it. All Ratings Dual Rating
8
Triple Rating
Corporate Bond Ratings Only Total
Dual Rating
Triple Rating
Total
Year
No.
%
No.
%
No.
%
No.
%
No.
%
No.
%
2003
404
72%
156
28%
560
5%
333
95%
18
5%
351
6%
2004
445
67%
220
33%
665
6%
318
95%
17
5%
335
6%
2005
471
66%
239
34%
710
6%
234
83%
47
17%
281
5%
2006
475
66%
247
34%
722
7%
169
67%
82
33%
251
4%
2007
488
64%
271
36%
759
7%
258
86%
42
14%
300
5%
2008
509
63%
302
37%
811
7%
285
63%
170
37%
455
8%
2009
529
61%
343
39%
872
8%
344
85%
62
15%
406
7%
2010
526
56%
405
44%
931
8%
286
66%
148
34%
434
8%
2011
580
58%
424
42%
1004
9%
351
70%
153
30%
504
9%
2012
555
55%
451
45%
1006
9%
427
60%
288
40%
715
12%
2013
530
52%
480
48%
1010
9%
242
40%
366
60%
608
11%
2014
521
52%
483
48%
1004
9%
349
69%
159
31%
508
9%
2015
527
52%
490
48%
1017
9%
399
64%
222
36%
621
11%
Total
6560
60%
4511
40%
11071
100%
3995
73%
1774
27%
5769
100%
Information on the ship investment companies: http://www.koreashipfinance.com/eng/sic/sic01.asp 23
Table 2. Description of Variables Variable Definition Dependent Variable Credit Rating Financial Distress
Drop Stay Join Upgrade
Credit Rating is a rating that is converted into a numerical rating scale as follows: AAA=1, AA+=2, AA=3, AA−=4, A+=5, A=6, A−=7, BBB+=8, BBB=9, BBB−=10, B+=11, BB=12, BB−=13, B+=14, B=15, B−=16, CCC=17, CC=18, C=19, D=20. An indicator variable that takes the value of one if an issuer experiences a debt to equity ratio (D/E; total debt/total equity) higher than 200% and a times interest earned (TIE; EBIT/interest where EBIT is earnings before interest and taxes) lower than 100% for four quarters after the rating and zero otherwise. When the CRA who rated issuer i is excluded in the corporate bond rating. In the multinomial logit regression, j=1 When the CRA who rated issuer i rates the corporate bond. In the multinomial logit regression, j=2 When the CRA did not rate issue i but participates in corporate bond rating. In the multinomial logit regression, j=3 An indicator variable that takes the value of one if a credit rating is an upgrade from the prior quarter rating. Otherwise, zero.
Key Independent Variables Triple Rating Dual Rating Inv. Grade Split Rating Difft-1 Rating Difft CRA-i Lowt-1 CRA-i Upgradet
An indicator variable that takes the value of one if one receives credit ratings from all three CRAs. Otherwise, zero An indicator variable that takes the value of one if one receives credit ratings from two CRAs. Otherwise, zero An indicator variable that takes the value of one if one receives credit rating that is BBB- or higher. Otherwise, zero. An indicator variable that takes the value of one if one receives credit ratings that are not identical across CRAs. Otherwise, zero. The difference between a CRA’s rating and the rating average by the other two CRAs in the previous quarter. The difference between a CRA’s rating and the rating average by the other two CRAs in the previous quarter. An indicator variable that takes the value of one if a CRA’s average rating is lower than the average of other two CRAs’ ratings in the previous quarter and zero otherwise. i includes CRA X, CRA Y, and CRA Z. An indicator variable that takes the value of one if a CRA upgrades in comparison to its previous rating and zero otherwise. i includes CRA X, CRA Y, and CRA Z.
Control Variables Size Sales_Growth Market to Book
Log (market value) where market value is market share price times the number of shares outstanding at the end of the quarter. A proxy for the size of a firm. Bolton, Freixas, and Shapiro (2012) suggest that CRAs more likely to inflate large issuers. 𝑆𝑎𝑙𝑒𝑠𝑡 − 1. t and t-1 represent two subsequent quarters in the chronological order. 𝑆𝑎𝑙𝑒𝑠𝑡−1
Market value of equity divided by book value of equity where the book value of equity is defined as stockholders’ equity divided by the number of shares outstanding.
ROA
Return on Assets that is net income divided by total assets.
Debt Ratio
Total debt divided by total asset.
Volatility
Daily return volatility for one year.
Instrumental Variables Annual Avg. Triple
The average of all triple ratings for the year.
Industry Avg. Triple
The average of all triple ratings for the industry to which a ratee belongs for the year.
24
Table 3. Summary Statistics by Dual and Triple Ratings The table shows the summary of credit ratings by dual and triple ratings. All ratings in Panel A include both issuer and bond ratings. An issuer or a bond is classified as dual (triple) rating when two (three) CRAs (credit rating agencies) rate it. Market value and total assets are expressed in Korean Won. One thousand Korean won (1,000 KRW) is approximately equivalent to $1 USD. We denote three CRAs as X, Y and Z to preserve anonymity.
Panel A: All Ratings
No
AA- or Higher
BBB or Higher
Split
Market Value (Trillion in KRW)
Total Assets (Trillion in KRW)
Dual Rating CRA-X & CRA-Y
2,124
635
29.90%
1,832
86.25%
83
3.91%
1.50
2.26
CRA-Z & CRA-X
1,904
487
25.60%
1,649
86.61%
159
8.35%
1.67
2.18
CRA-Y & CRA-Z
2,532
522
20.60%
2,146
84.76%
95
3.75%
1.29
3.01
4,511
2,081
46.10%
4,252
94.26%
573
12.70%
5.49
14.08
11,071
3,725
33.60%
9,879
89.23%
910
8.22%
3.13
7.24
Triple Rating Total
Panel B: Corporate Bond Ratings Only
No
AA- or Higher
BBB or Higher
Split
Market Value
Total Assets
(Trillion in KRW)
(Trillion in KRW)
Dual Rating CRA-X & CRA-Y
1,256
572
45.50%
1,197
96.36%
13
1.04%
3.88
17.44
CRA-Z & CRA-X
1,139
564
49.50%
1,102
96.75%
16
1.40%
4.44
20.33
CRA-Y & CRA-Z
1,600
640
40.00%
1,514
94.63%
28
1.75%
3.26
21.42
Triple Rating
1,774
1,510
85.10%
1,766
99.55%
14
2.61%
7.31
77.59
Total
5,769
3,286
57.00%
5,579
96.71%
71
2.61%
4.93
37.64
25
Table 4. Univariate Tests; Dual vs. Triple Ratings All ratings include both issuer and bond ratings. An issuer or a bond is classified as dual (triple) rating when two (three) CRAs (credit rating agencies) rate it. We label three CRAs as X, Y, and Z to preserve anonymity. Credit ratings are converted into integers as follows: AAA=1, AA+=2, AA=3, AA−=4, A+=5, A=6, A−=7, BBB+=8, BBB=9, BBB−=10, BB+=11, BB=12, BB−=13, B+=14, B=15, B−=16, CCC=17, CC=18, C=19, D=20. Panel A (B) compares triple rating with three CRA combinations of dual rating using all rating (bond ratings only). In Panel C, portfolios are formed by size for the dual vs. triple rating comparison. We classify the sample issuers into five portfolios by the market value in each quarter, with the first largest 20% in portfolio Largest, 21%-40% in P40, 41%-60% in P60, 61%-80% in P80, and the smallest 20% in Smallest. All the variables are defined in Table 2 in section 3.2. tor z-statistics using robust standard errors are reported in parentheses. *, **, *** indicate significance levels at the 10%, 5%, and 1%, respectively. Panel A: All Ratings Mean
t-test (Dual-Triple)
Median
Rank-Sum test (Dual-Triple)
Stdev
Dual Rating CRA-X & CRA-Y
6.49
20.3438***
6
19.367***
2.999
CRA-Z & CRA-X
6.516
19.6643***
6
17.446***
3.132
CRA-Y & CRA-Z
6.935
27.6313***
7
26.101***
3.019
Triple Rating
4.947
5
2.827
Total
5.967
6
3.081
Panel B: Corporate Bond Ratings Only Mean
t-test (Dual-Triple)
Median
Rank-Sum Test (Dual-Triple)
Stdev
Dual Rating CRA-X & CRA-Y
4.976
27.26***
5
25.42***
2.706
CRA-Z & CRA-X
4.597
23.15***
5
22.54***
2.526
CRA-Y & CRA-Z
5.269
31.71***
5
29.46***
2.78
Triple Rating
2.608
2
2.072
Total
4.255
4
2.751
Panel C: All Ratings by Portfolio Size Size Dual vs. Triple Dual Rating Smallest Triple Rating P40 P60 P80 Largest
N 1880
Mean 8.772
304
7.497
Dual Rating
1879
6.582
Triple Rating
329
7.296
Dual Rating
1321
5.838
Triple Rating
896
6.294
Dual Rating
967
5.247
Triple Rating
1241
5.433
Dual Rating
513
4.105
Triple Rating
1741
3.017
26
t test 5.30*** -6.67*** -7.44*** -1.91** 7.00***
Median 8 7 6 7 6 6 5 5 4 2
Rank-Sum Test 5.043*** -8.326*** -7.491*** -2.739*** 8.39***
Table 5. Full Sample Analysis: Dual vs. Triple Ratings Panel A shows the ordinary least square (OLS) and Treatment Effects model estimations. Lambda (λ) in Model (3) is the inverse of the Mill’s ratio. Panel B reports the Propensity Score Matching (PSM) estimations. All the variables are defined in Table 2 in section 3.2. Year and industry dummies are included in the estimations but not reported to conserve space. t- or z-statistics using robust standard errors are reported in parentheses. *, **, *** indicate significance levels at the 10%, 5%, and 1%, respectively. Panel A: OLS and Treatment Effect Model Regressions Treatment Effects Model Regression
OLS
1st Stage
Model (1) Dependent Variable: Triple Rating
(2)
Credit Rating -1.723
[-30.49]***
Size
2nd Stage (3)
Credit Rating
Triple Rating (=1 if Triple Rating)
Credit Rating
0.318 [7.32]*** -1.245 [-70.94]***
3.409 [8.54]*** 0.573 [47.35]***
-1.792 [-24.61]***
Sales_Growth
0.000 [1.01]
0.000 [-2.66]***
0.001 [2.02]**
Market to Book
0.016 [1.73]*
0.002 [0.45]
0.014 [1.91]*
ROA
-0.082 [-13.64]***
-0.011 [-4.76]***
-0.070 [-19.14]***
Debt Ratio
0.002 [7.03]***
0.000 [-6.03]***
0.002 [22.71]***
Volatility
0.005 [3.77]***
0.001 [3.35]***
0.004 [10.46]***
Annual Avg. Triple
0.912 [3.84]***
Industry Avg. Triple
0.673 [3.00]***
Intercept
6.669 [176.93]***
39.11 [84.99]***
-20.274 [-25.16]***
54.908 [27.64]***
Lambda
-1.847 [-7.83]*** Industry and Year dummies are included
No. of Observations
11,071
10,099
10,099
10,099
F/ Wald tests
929.4
2853
13,618
13,617
R2
0.076
0.61
Panel B: Propensity Score Matching
Credit Rating
Treated
Controls
(Triple Rating)
(Dual Rating)
4.903
4.681
27
ATT
t stat
0.22
1.84 *
Table 6. Dual vs. Triple Ratings by Portfolio Size Panel A shows the Treatment Effects model estimations by portfolio size. The dependent variable of the second stage of the model is Credit Rating. To conserve space, we report the second stage results only. Panel B reports the Propensity Score Matching (PSM) estimations. We classify the sample issuers into five portfolios by the market value in each quarter, with the first largest 20% in portfolio Largest, 21%-40% in P40, 41%-60% in P60, 61%-80% in P80, and the smallest 20% in Smallest. All the variables are defined in Table 2 in section 3.2. Year and industry dummies are included in the estimations but not reported to conserve space. t- or z-statistics using robust standard errors are reported in parentheses. *, **, *** indicate significance levels at the 10%, 5%, and 1%, respectively. Panel A: Treatment Effects Regressions Dependent Variable: Credit Rating Model (1)
(2)
(3)
(4)
(5)
Size Portfolio:
Smallest
P40
P60
P80
Largest
Triple Rating
-3.135***
2.954***
1.460***
2.005***
-0.957***
[-17.34]
[19.26]
[4.46]
[7.07]
[-3.55]
-2.039***
-1.073***
-0.763***
-0.503***
-1.015***
[-19.85]
[-6.13]
[-4.00]
[-2.66]
[-14.40]
0.001
-0.001
0.000
0.001
-0.001**
[1.14]
[-1.10]
[0.13]
[0.84]
[-2.10]
-0.041*
0.029***
-0.185***
0.042
-0.149***
[-1.77]
[4.40]
[-9.11]
[1.21]
[-3.67]
-0.062***
-0.079***
-0.056***
-0.095***
-0.025***
[-9.01]
[-14.18]
[-8.82]
[-9.66]
[-4.53]
0.003***
0.003***
0.002***
0.004***
0.001***
[10.12]
[13.23]
[9.62]
[15.84]
[10.33]
0.002***
0.028***
0.015***
0.014***
0.040***
[5.83]
[16.50]
[8.04]
[6.80]
[11.48]
62.911***
31.879***
27.035***
16.037***
32.676***
Size Sales_Growth Market to Book ROA Debt Ratio Volatility Intercept
[22.88]
[6.23]
[5.03]
[2.90]
[15.89]
No. of Observations
2,022
1944
1996
2066
2071
Wald X2
2,390
2301.65
1987.47
1315.95
3407.43
Panel B: Propensity Score Models Treated
Controls
ATT
t stats
Portfolios
(Triple Rating)
(Dual Rating)
Smallest
7.808
8.988
-1.181
[-5.98]
***
P40
7.311
6.313
0.998
[7.14]
***
P60
6.478
5.77
0.708
[6.98]
***
P80
5.253
5.095
0.158
[1.41]
Largest
3.042
3.604
-0.562
[-5.01]
28
***
Table 7. Summary Statistics: Probability of Financial Distress The table presents the summary statistics of credit ratings by Financial Distress, and further by Dual and Triple Ratings. Financial Distress is an indicator variable that takes the value of one if an issuer experiences a debt to equity ratio higher than 200% and a times interest earned lower than 100% for four quarters after the rating and zero otherwise. P5 and P95 represent the 5th percentile and 95th percentile. All the variables are defined in Table 2 in section 3.2. Financial Distress=0
Financial Distress=1
Dual
Triple
Total
Dual
Triple
Total
N
6,037
4,120
10,157
523
391
914
N (Rating>BBB-)
5,332
3,964
9,296
295
288
583
6.38
4.65
5.68
10.01
8.02
9.16
P5
3
1
1
6
4
5
Median
6
4.33
5
9
8
9
11.5
9
10
17
12
16
Mean
P95
29
Table 8. Logit Regression: Probability of Financial Distress The table shows the logit regression results on how investment grade ratings with triple rating affect the probability of financial distress. Financial Distress is an indicator variable that takes the value of one if an issuer experiences a debt to equity ratio higher than 200% and a times interest earned lower than 100% for four quarters after the rating and zero otherwise. We classify the sample issuers into five portfolios by the market value in each quarter, with the first largest 20% in portfolio Largest, 21%-40% in P40, 41%-60% in P60, 61%-80% in P80, and the smallest 20% in Smallest. All the variables are defined in Table 2 in section 3.2. Year and industry dummies are included in the estimations but not reported to conserve space. t- or z-statistics are reported in parentheses. *, **, *** indicate significance levels at the 10%, 5%, and 1%, respectively. Dependent Variable: Financial Distress (=1 if financial distress occurs for 4 quarters after rating) Smallest Triple Rating × Inv. Grade
-3.64 [-1.81]*
Triple Rating
0.081 [0.13]
Inv. Grade Size Sales_Growth
-0.557 [-1.75]* 0.662 [2.76]*** 0.009 [4.90]***
P40
P60 -1.524 [-1.85]*
0.272 [0.36]
-5.889 [-2.53]**
-0.365 [-0.52]
1.602 [2.06]**
0.82 [1.20]
6.527 [2.84]***
-1.081 [-1.58]
19.75 [10.44]***
-2.606 [-4.99]***
-0.886 [-1.26]
3.108 [5.13]***
-0.185 [-0.26]
-1.252 [-5.36]***
0.000 [0.02]
-0.01 [-3.90]***
-0.012 [-3.69]***
0.013 [1.83]*
0.025 [1.11]
0.387 [5.56]***
-0.079 [-0.30]
ROA
-0.104 [-5.60]*** -0.174 [-9.51]***
Intercept
0.634 [2.19]**
-0.009 [-1.03]
-0.194 [-1.07]
Volatility
Largest
0.897 [1.07]
Market to Book Debt Ratio
P80
-0.412 [-12.64]** -0.393 [-10.37]*** -0.346 [-5.54]***
0.009 [2.73]***
0.005 [2.43]**
0.002 [5.28]**
-0.002 [-3.27]***
0.012 [3.80]***
-21.325 [-3.65]*** -89.76 [-5.40]***
0.002 [7.88]***
0.002 [5.96]***
0.035 [4.52]**
0.008 [3.00]***
0.022 [1.28]
2.12 [0.10]
-22.43 [-2.62]***
12.69 [1.60]
Industry and Year dummies are included No. of Observations
1,882
1,837
1,737
2,061
1,636
Wald tests
192.43
420.15
430.8
295.28
302.9
Pseudo R2
0.5188
0.5702
0.53
0.3831
0.451
30
Table 9. Probability of Upgrade in Rating Splits for Triple Rating The table shows the logit regression results on the probability of upgrades when splits are present in the prior quarter. The dependent variable is an indicator variable that take the value of one if a rating is an upgrade and zero otherwise. All the variables are defined in Table 2 in section 3.2. Year and industry dummies are included in the estimations but not reported to conserve space. z-statistics are reported in parentheses. *, **, *** indicate significance levels at the 10%, 5%, and 1%, respectively.
Dependent Variable: =1 if upgraded Triple Rating×Split Triple Rating
0.538
[2.12]
**
-0.085
[-0.75]
Split
1.131
[6.22]
Size
-0.017
[-0.44]
Sales_Growth
-0.001
[-0.47]
Market to Book
-0.011
[-0.62]
0.057
[7.54]
***
0
[3.20]
***
ROA Debt Ratio
***
Volatility
-0.001
[-1.87]
**
Intercept
-2.413
[-2.24]
**
Year and industry dummies are included No. of Observations
9,845
Wald test
272.8
R2
0.0716
Pseudo
31
Table 10. Comparison of Prior Credit Ratings by Upgrade The table presents average ratings of the current and prior quarters for each CRA by upgrade. It also compares the average ratings of the incumbent CRA and other CRAs in terms of upgrade in the previous quarter. We label three CRAs as X, Y, and Z to preserve anonymity. Rating CRA CRA-X
CRA-Y
CRA-Z
Upgrade
Average Ratings in Quartert-1
t
t-1
This CRA
Others
No
4.91
4.87
4.86
4.91
Yes
5.81
7.08
7.06
7
No
4.9
4.87
4.85
4.9
Yes
5.95
7.09
7.08
7.08
No
4.92
4.89
4.87
4.91
Yes
5.87
7.05
7.04
6.95
32
Table 11. Effect of Prior Credit Rating on Upgrade for 3 CRAs The table shows the logit regression results for the effect of the prior credit ratings on upgrades on the following ratings. The dependent variable—an indicator variable that takes the value of one if an CRA upgraded a rating in comparison to the prior rating. is an indicator variable that takes CRA-X (CRA-Y, CRA-Z) Lowt-1—is an indicator variable that takes the value of one if the CRA’s rating in the prior quarter is lower than other CRAs’ ratings and zero otherwise. After 2009 is an indicator variable that takes a value of one if a rating is issued after 2009 and zero otherwise. Definitions for other control variables are found in Table 2 in section 3.2. Year and industry dummies are included in the estimations but not reported to conserve space. z-statistics are reported in parentheses. *, **, *** indicate significance levels at the 10%, 5%, and 1%, respectively. Dependent Variable (CRA-i Upgradet): CRA-X Lowt-1
=1 if CRA-X upgraded 0.607
=1 if CRA-Y upgraded
[2.08]**
CRA-Y Lowt-1
0.882
[3.37]***
CRA-Z Lowt-1 Size
=1 if CRA-Z upgraded
[3.41]***
-0.142
[-2.24]**
-0.097
[-1.46]
-0.108
Sales_Growth
0.001
[1.69]*
0
[0.14]
0.001
[1.76]*
Market to Book
0.053
[2.36]**
0.029
[1.32]
0.015
[0.66]
ROA
0.056
[5.23]***
0.056
[5.07]***
0.062
[5.75]***
Debt Ratio
0
[1.19]
0
[-1.66]*
0.902
[0.83]
0
Volatility
0.002
[3.05]***
0.012
[5.18]***
0.001
After 2009
-0.589
[-3.41]***
-0.699
[-4.09]***
-0.453
Intercept
-0.429
[-0.23]
-0.669
[-0.36]
0.581
Year and industry dummies are included No. of Observations Wald
X2
Pseudo R2
3,925
3,912
3,914
81.09
114.85
85.24
0.058
0.0656
0.053
33
[2.39]** [1.51] [-2.55]** [0.32]
