You are on page 1of 43

Moody's and S&P Ratings: Are They Equivalent?

Conservative Ratings and Split Rated Bond Yields

Miles Livingston University of Florida and Erasmus University P.O. Box 117168 Gainesville, FL 32611-7168 Phone: (352) 392-4316 Fax: (352) 392-0301

Jie (Diana) Wei* Office of the Comptroller of the Currency Washington, DC 20219 Phone: (202)874-6532 Fax: (202) 874-5394

Lei Zhou Northern Illinois University College of Business Department of Finance DeKalb, IL 60115-2897 Phone: (815) 753-7882 Fax: (815) 753-0504

February, 2010

The views expressed here are those of the individual authors alone and do not necessarily reflect those of the Office of the Comptroller of the Currency or the Department of the Treasury.

Electronic copy available at: http://ssrn.com/abstract=1567665

Moody's and S&P Ratings: Are They Equivalent? Conservative Ratings and Split Rated Bond Yields Abstract
We examine the relative impact of Moodys and S&P ratings on bond yields and find that at issuance yields on split rated bonds with superior Moodys ratings are, on average, 8 basis points lower than yields on split rated bonds with superior S&P ratings. This pattern suggests that investors differentiate between the two ratings and assign more weight to the ratings from Moodys, the more conservative rating agency. Moody's ratings become relatively more conservative after 1998 and the impact of a superior Moody's rating upon split rated bond yields is stronger. In addition, the differential impact of the two ratings on split rated bond yields is more pronounced for Rule 144A issues, which are more opaque.

Electronic copy available at: http://ssrn.com/abstract=1567665

Moody's and S&P Ratings: Are They Equivalent? Conservative Ratings and Split Rated Bond Yields
I. Introduction Most publicly issued corporate bonds in the US receive ratings from two major rating agencies, Moodys and Standard & Poors (S&P). About 50% of the time, the Moody's and the S&P ratings are different at the notch level, resulting in so-called split ratings, with Moody's more likely to have the conservative (lower) rating (Morgan, 2002, Van Roy, 2005).1 We show that about 56% of split rated bonds have conservative Moodys ratings and investors differentiate between these two ratings, especially during the period 1998-2008. When Moody's has the superior (higher) rating, bond yields are approximately 8 basis points lower than when S&P has the superior rating. Thus investors appear to hold Moodys reputation in higher regard than S&Ps. Split ratings and the differential impact of Moody's versus S&P are important for two reasons. First, while virtually all public issues of bonds receive ratings from both rating agencies, financial regulators generally do not differentiate between these ratings.2 Most rating-based regulations require only one rating without specifying a particular rating agency (Cantor and Packer, 1995, BIS, 2000). If the bond issue has multiple ratings, then either the highest or the second highest rating is usually used regardless of which rating agency assigns the highest or second highest rating (Cantor and Packer, 1995).3 In addition, under the Basel II framework,

See also Ederington (1986), Pottier and Sommer (1999), and Livingston et al. (2007). For a rating to be acceptable to regulators, the rating agency must be designated by the SEC as a nationally recognized statistical rating organization, or NRSRO. Moodys and S&P are both NRSROs. 3 A notable exception to this general rule is that of the National Association of Insurance Commissioners (NAIC). In the case of split ratings, the NAIC can choose either the superior or inferior rating based on its own analysis (Cantor and Packer, 1996).
2

banks can choose to use either one or more rating agencies to risk-weight their credit exposures. If a bank chooses to use one rating agency, it makes no difference whether it is Moodys or S&P. If two rating agencies are used, the inferior of the two ratings determines the risk weights, regardless of which agency assigns the inferior rating (Van Roy, 2005). Second, many academic studies treat these ratings as interchangeable in spite of the fact that Moody's and S&P ratings frequently disagree. Some studies use Moodys ratings as a proxy for default risk (for example, Kidwell et al., 1984) while others use S&P ratings (for example, Avramov et al., 2007). A few studies take advantage of both Moodys and S&P ratings, but use the average of the two when they differ without regard to which rating is superior (for example, Fenn, 2000). Another common practice is to use the Moodys (S&P) rating when it is available and supplement with the S&P (Moodys) rating when the Moodys (S&P) rating is not available, implicitly assuming that these two ratings are equivalent and interchangeable (for example, Yu, 2005, Butler, 2008). This study uses a sample of 6,652 newly issued, split rated, non-financial US corporate bonds from 1983 to 2008 to examine the relative impact of the two ratings on bond yields. By focusing on split rated bonds, our analysis highlights the differences between Moody's and S&P ratings. First, we confirm findings in earlier studies that Moodys is more likely to give a conservative (or inferior) rating than S&P when these two differ. Second, when Moody's has the superior rating, investors require lower yields, about 8 basis points on average, for split rated bonds.4 Bond investors appear to differentiate between these two ratings and assign greater

The 8 basis points yield difference is statistically significant. While the sample size is very large, the statistical significance is not just driven by the sample size. In a subsample of 1,291 Rule 144A bond issue, the yield difference is also significant at the 1% level. In addition, we estimate the yield difference for each of the 33 split rating categories. The yield differences are significant at the 1% or 5% level in 10 categories. The sample size for these rating categories ranges from 26 to 643, indicating that the statistical significance is not a result of large sample size.

weight to the ratings from the more conservative agency. In addition, the differential impact of the two ratings on bond yields is more pronounced for Rule 144A bond issues. As Rule 144A issues are not registered with the SEC and have lower standards for disclosure, information about them tends to be more opaque. Bond investors appear to be particularly concerned about the information opaqueness of Rule 144A bond issues and therefore rely more heavily on the ratings from the conservative rating agency in their assessment of default risk. Third, Moodys ratings are relatively comparable to S&P ratings prior to 1998, but become more conservative than S&P thereafter. Accordingly, the yield difference between bonds with a superior Moodys rating and bonds with a superior S&P rating is minimal prior to 1998, but significantly larger (about 13 basis points) between 1998 and 2008.5 This evidence suggests that bond investors, mostly institutional investors, are sufficiently sophisticated to detect evolving differences between the two major rating agencies and act accordingly. Furthermore, we also compare the yields for the sample of 6,652 split rated bonds to the yields for a sample of 7,201 non-split rated bonds. Livingston and Zhou (2010) find that split rated bonds average a 7-basis-point yield premium over non-split rated bonds of similar credit risk and they attribute this yield premium to information opacity of split rated bonds. However, they do not distinguish between splits with superior Moodys ratings and splits with superior S&P ratings. We find that the yield premium is different in these two cases. When Moodys (S&P) assigns the superior rating, the yield premium is much smaller (larger). The main findings of the paper are summarized in Figure 1, which illustrates the yield premiums assigned to split rated bonds over non-split rated bonds. When no distinction is made

Blume et al. (1998) find that S&P tightens rating standards and its ratings become more stringent from 1978 to 1995. However, the study does not compare S&P ratings with Moodys ratings. Furthermore, the Blume et al. study covers a time period from 1978 to 1995, while we find Moodys becomes more conservative (relative to S&P) after 1998. Thus, our findings do not contradict those of Blume et al. (1998).

between splits with superior Moody's ratings and splits with superior S&P ratings, split rated bond issues have an average yield premium of approximately 6.5 basis points above the average of yields for non-split superior rated and non-split inferior rated bonds, similar to findings in Livingston and Zhou (2010). When Moody's has the superior rating, the yield premium is only 2 basis points. Conversely, the yield premium is 10 basis points if S&P has the superior rating. The difference in the yield premium, 8 basis points, indicates that yields on split rated bonds with superior Moodys ratings are, on average, 8 basis points lower than those with superior S&P ratings. In order to put the 8-basis-point yield spread difference into perspective, we estimate that the average yield spread difference between two adjacent notch rating categories is about 30 basis points. Thus, a yield difference of 8 basis points is approximately equivalent to a difference of one fourth of a notch rating. Furthermore, for a typical split rated bond in our sample, the 8-basis-point difference in yield spread translates to a difference of $1.42 million in price.6 This study has two important implications. First, it calls into question the common practice by financial regulators and by academic studies to treat the two major ratings equally. While we do not advocate using one rating agency versus the other, regulators should continuously monitor the performance of each rating agency. Although regulatory agencies in many countries set criteria for eligible credit rating agencies for regulatory purposes, only three countries (France, Italy and Japan) have ongoing monitoring of the recognized rating agencies (BIS, 2000). In the US, once a rating agency is recognized as an NRSRO by the SEC, there is no monitoring of the performance of the rating agency and no formal mechanism to strip its

For the 1998-2008 subsample period, the yield spread difference is about 13 basis points, equivalent to a difference of half a notch rating or $2.3 million in price difference for a typical split rated bond.

NRSRO status. To increase the competition in the rating industry, the SEC has significantly increased the number of NRSROs from three in 1975 to currently ten.7 The papers findings on the evolving relative conservativeness of different rating agencies suggest that rating agencies may not consistently maintain the quality of their ratings. Conceivably, some rating agencies might, after being designated as an NRSRO, sell their SEC-sanctioned ratings to maximize current profits. Thus, the SEC should not only be the gate-keeper of the NRSRO status, but should also be the rater of the raters. Second, the findings in this paper suggest that bond investors are quite sophisticated and differentiate between the two major rating agencies. Our findings also highlight the importance for rating agencies to protect their reputational capital. If one rating agency consistently becomes more lenient (conservative) over time, bond investors will gradually adapt to such changes by putting less (more) weight on that agencys rating in their assessment of default risk. This finding supports the argument that rating agencies have strong incentives to issue honest and accurate ratings to protect their reputation capital (Smith and Walter, 2002, Covitz and Harrison, 2004). Such concerns for reputation capital can offset or mitigate the potential conflicts of interest when rating agencies are paid by issuing firms. The remainder of the paper is organized as follows. Section II summarizes the previous literature, and Section III describes the data and provides summary statistics. Section IV analyzes the differential impacts of Moody's and S&P ratings upon yields of split rated bonds. Section V compares yields of split rated and non-split rated bonds. Section VI concludes the paper.

The SEC website lists the ten NRSROs (http://www.sec.gov/answers/nrsro.htm).

II.

Background and Literature Review Given the dominance of the two major rating agencies, several studies compare Moodys

and S&P ratings. Ederington (1986) examines determinants of the two ratings for US corporate bonds from 1975 to 1980. His study finds that the two rating agencies use a common set of factors such as firm size and leverage ratio to estimate credit risk, and they assign similar weights to these factors. In addition, there is no systematic difference in the rating scales used by the two major rating agencies. Thus, Ederingtons (1986) findings suggest that the two ratings are basically equivalent, implying that differences in the two ratings are merely random errors. According to Ederington, the respective positions of the two agencies could easily have been reversed; i.e., on another day or with a slightly different set of analysts, either agency might assign a different rating. We call this view the rating equivalence hypothesis. An implication of the rating equivalence hypothesis is that yields on split rated bonds should be the same regardless of which rating agency assigns a superior rating. In other words, a bond rated A+ by Moodys and A by S&P should have the same yield as another bond rated A by Moodys and A+ by S&P. Other studies, however, suggest that there are systematic differences between the two rating agencies. Moon and Stotsky (1993) find that the two rating agencies use different economic variables to determine ratings on municipal bonds. In addition, they find that the rating scales of the two agencies differ systematically.8 Similarly, in a study of financial strength ratings of property liability insurance companies by A.M. Best, Moodys and S&P, Pottier and Sommers (1999) find that the three rating agencies use different factors in modeling an insurers

Note that these results should be treated with caution. Moon and Stotsky (1993) use outstanding municipal bonds in their study. Observed split ratings for outstanding bond issues might be caused by asynchronous changes in ratings over time rather than by systematic difference between the two rating agencies.

financial strength, suggesting that different ratings convey different information and may not be equivalent. In addition to the empirical comparison of the two rating agencies, a survey study by Baker and Mansi (2002) investigates the attitude of issuing firms and investors toward the rating agencies. The study reports that issuing firms believe S&P ratings are more accurate than Moodys, while investors do not see a difference. This finding is not a surprise since S&P tends to assign superior ratings compared to Moodys. On the other hand, investors believe that corporate bond prices more closely follow Moodys rating than S&P rating.9 This study suggests that both investors and issuing firms perceive these two ratings differently. Consistent with Baker and Mansis (2002) survey results, Gttler and Wahrenburg (2007) find that Moodys updates its ratings to reflect changing default risk in a timelier manner than S&P. Recent studies on split ratings also indicate that there might be systematic differences between the two rating agencies. Morgan (2002) suggests that split ratings are caused by asset opacity of issuing firms. He further argues that split ratings are likely to be lopsided; that is, one rating agency will be more likely to assign a superior rating than the other, if asset opacity is the cause of the split rating. On the other hand, split ratings should be symmetric if split ratings are merely random errors. Morgan's idea is that the two rating agencies may not worry equally about over-rating. When severe asset opacity problems exist, the more conservative rating agency (or the agency that worries more about over-rating than under-rating) tends to assign an inferior rating. Indeed, Morgan (2002) finds that Moodys ratings for banking firms are more likely to be inferior compared to S&P ratings when the two differ. Livingston et al. (2007) and Pottier and Sommers (1999) report similar patterns for corporate bond ratings and financial
The survey asks the following question: How would you rank the following rating agencies in terms of how closely their ratings of corporate bonds compare to where these bonds actually trade? It only asks investors this question.
9

strength ratings of insurance companies respectively. These findings suggest that Moodys is more conservative and less likely to over-rate a bond issue. An empirical study of rating accuracy by Gttler (2005) supports this view. The study finds that Moodys ratings are slightly better at predicting default than S&P ratings. We call this view the systematic difference hypothesis.10 An implication of the systematic difference hypothesis is that split rated bonds with superior ratings from the more conservative rating agency should have lower yields if bond investors differentiate between these two ratings. Thus, in examining the yields on split rated bonds, we perform a joint test that 1) there is a systematic difference between the Moodys and S&P ratings, and 2) bond investors differentiate between these two ratings. This study is also linked to another line of research on the impact of split ratings on bond yields. Some studies find that inferior ratings determine yields on split rated bonds (Billingsley et al., 1985, Liu and Moore, 1987, Perry et al., 1988); others find that the superior ratings set bond yields (Hsueh and Kidwell, 1988, Reiter and Ziebart, 1991). None of the studies, however, distinguish split rated bonds with superior Moodys ratings from those with superior S&P ratings. More recent studies find that both the superior and inferior ratings have an impact on yields (Cantor et al., 1997, Jewell and Livingston, 1998). In addition, Livingston and Zhou (2010) find that split rated bonds average a 7-basis-point yield premium over non-split rated bonds of similar credit risk. Again, these studies do not test whether a superior rating from one particular rating agency has a different impact on yields than a superior rating from another rating agency.

10

Note that the systematic difference between the two rating agencies might be caused by 1) different rating scales, 2) different weights assigned to risk factors, 3) relative costs of over-rating vs. under-rating. Regardless of the underlying cause, systematic difference hypothesis suggests that one rating agency will be more conservative, that is, more likely to assign inferior ratings than the other.

III.

Data Collection and Description We collect data on fixed rated, US domestic, non-financial public and Rule 144A

corporate bond issues from the Thomson Financial SDC database. We use the data on original bond issues for two reasons. First, bond ratings on new issues reflect the most up-to-date information about the issuers.11 Second, split ratings on existing bond issues can be caused by asynchronous changes in ratings by the two rating agencies in response to changes in underlying default risk.12 Thus, split ratings on outstanding bonds may merely be the result of slow updating by one agency instead of a systematic difference between the two ratings. The sample period covers 1983 to 2008. We start in 1983 because Moodys began issuing notch ratings after April 1982, in addition to letter ratings.13 Perpetual bonds, bonds with credit enhancements, and putable bonds are excluded. We also exclude several issues that are rated CCC- by at least one rating agency.14 The final sample consists of 13,853 bond issues comprising 7,201 non-split rated issues and 6,652 split rated issues. All split rated bonds in the sample have Moodys and S&P ratings that differ by one or two notches.15 Table 1 describes the sample. Non-split rated bonds have average maturity similar to split rated bonds but slightly smaller issue size.16 The non-split rated sample has an average original yield to maturity (YTM) of 7.86%, slightly lower than that of the split rated sample. To make yields comparable between bonds of different maturities, we subtract the yield for Treasury
11

Some research has found that rating agencies tend to lag financial markets in reflecting new information (Holthausen and Leftwich, 1986, Ederington and Goh, 1998). 12 Gttler and Wahrenburg (2007) document a lead-lag relation between the ratings of Moodys and S&P. 13 S&P started to issue notch ratings in 1974 (Cantor and Packer, 1995). 14 We exclude these issues as a data quality filter. It is highly unlikely that firms with a CCC- rating are able to access the public bond market. Inclusion of these issues in our sample does not, however, change our results. 15 We exclude issues with three or more notches of split ratings because there are very few observations in each of these distinct rating combination categories. 16 Investment grade bonds have an average maturity of 13.9 years while below-investment grade bonds have an average maturity of 9.4 years. This suggests that high quality firms are more likely to issue longer term bonds.

securities of similar maturity from the yield to maturity of each bond to calculate the Treasury Spread. The average Treasury Spread is slightly lower for the non-split rated sample, but there is no difference between the superior S&P rating subsample and the superior Moodys rating subsample. To summarize the credit quality of the sample, we create two numerical rating variables: Moodys Rating and S&P Rating. Both variables range from 1 (for CCC) to 18 (for AAA). The Average Rating is the average of Moodys and S&P Ratings. The Average Rating for non-split rated bonds is 10.33, slightly above BBB, higher than the Average Rating for split rated bonds. This is consistent with findings in Livingston et al. (2007). The split rated sample has an average Moodys Rating of 9.87, lower than the average S&P Rating of 10.01, meaning that Moody's gives an inferior (lower) average rating. For each bond issue, we also calculate the Rating Difference, or Moodys Rating minus S&P Rating. The average Rating Difference of the split rated sample is -0.14, statistically significant at the 1% level. In other words, Moodys Rating is, on average, 0.14 notches below the S&P Rating when the two ratings differ. In addition, 55.5% of the split rated issues in our sample (3,691) have superior S&P ratings and 44.5% of the issues (2,961) have superior Moodys ratings. These findings suggest that Moodys is more conservative than S&P, consistent with Morgan (2002) and Livingston et al. (2007).

IV.

Empirical Analysis of Split Rating Sample In this section, we examine the yields on split rated bond issues. Specifically, we

compare yields on split rated bonds with superior Moodys ratings to the yields on split rated bonds with superior S&P ratings.

10

1. Univariate Analysis The summary statistics suggest that Moodys is, on average, more likely to assign a more conservative rating than S&P. This section examines whether this pattern holds true for bonds of different credit quality. First, we create 33 distinct rating categories, ranging from AAA/AA+ to CCC+/CCC, based on the combination of the two ratings.17 For each rating category, we distinguish between those with superior Moodys ratings and those with superior S&P ratings. Figure 2 depicts the number of split rated bonds with superior Moodys ratings and the number of splits with superior S&P ratings in each rating category. In 25 out of the 33 rating categories, there are more issues with superior S&P ratings than issues with superior Moodys ratings. Interestingly, for the 8 rating categories with more issues of superior Moodys ratings, 5 are below-investment grade, consistent with our earlier finding that the Average Rating for issues with superior Moodys ratings is slightly lower. This pattern might be explained by the fact that the Moodys rating incorporates both the probability of default and the expected recovery rate, while the S&P rating is strictly a measure of default risk (BIS, 2000). This difference has little impact on investment grade bonds, but might create systematic differences between the two ratings for below-investment grade bonds. Figure 2 and the summary statistics confirm the findings in previous studies that Moodys is more likely to give conservative ratings than S&P when these two differ. The next logical question is whether this conservative tendency in Moodys rating is recognized by bond investors and reflected in lower bond yields. Table 2 reports the average treasury spread for each rating category of the split rated sample, as well as two sub-samples: those with superior Moodys ratings and those with superior S&P ratings. In 20 out of the 33 rating categories, the
17

Moodys and S&P use slightly different rating notation but they are generally thought to be equivalent. For example, S&Ps BBB+ corresponds to Moodys Baa1. Following other studies, we use the S&P rating notation in this paper.

11

superior Moodys rating sub-samples have lower treasury spreads than the superior S&P rating sub-samples, and the differences are significant in 9 rating categories. For example, in the A+/A split rating category, bonds with superior S&P ratings (S&P rated A+, Moodys rated A) average a treasury spread of 94.14 basis points, while bonds with superior Moodys ratings (S&P rated A, Moodys rated A+) average a treasury spread of 82.78 basis points, an 11-basis-point difference with high statistical significance. Only in three rating categories do superior Moodys rating subsamples have significantly higher treasury spreads. These findings indicate that bond investors differentiate between these two bond rating agencies and require lower yields when Moodys assigns a superior rating.

2. Multivariate Analysis This section examines yields on split rated bonds using multivariate regressions. The dependent variable is the Treasury Spread (that is, the yield to maturity minus Treasury yield). The explanatory variables include rating dummy variables and control variables. The test variable is a dummy variable, SUPMOODY, set equal to one for issues with a superior Moodys rating and zero otherwise. Thus, the base case is split rated bonds with a superior S&P rating. If bond investors require lower yields for issues with a superior Moodys rating, the coefficient on SUPMOODY should be significantly negative.

a. Control Variables

12

Previous studies have shown that many factors other than default risk can have an impact on treasury spreads.18 Thus, three sets of control variables are included in addition to the bond ratings: controls for different bond features, controls for registration types, and controls for market conditions. To control for differences in bond features, we include the MATURITY, PROCEEDS, SENIOR dummy, CALL dummy, and UTILITY dummy variables in the regression models. MATURITY is the natural log of years to maturity. Previous studies find that yield spreads vary with bond maturity (for example, Kidwell et al., 1984, Chaplinsky and Ramchand, 2004).19 PROCEEDS is the total dollar proceeds of the bond issue. Large issues of bonds are usually more liquid than small issues and consequently investors may require lower rates of return for large issues of bonds. Hence, we expect the coefficient for PROCEEDS to be negative. A SENIOR dummy is set equal to one if the issue is senior debt, and zero otherwise. Since senior debt is less risky than subordinated debt, the coefficient for the SENIOR dummy is anticipated to be negative. The CALL dummy variable is set equal to one if the bond is callable and zero otherwise. We expect the coefficient for this variable to be positive because callable bonds are riskier for investors. The UTILITY dummy equals one if the issuer is a utility firm and zero otherwise. We also include SHELF dummy and R144A dummy variables to control for different methods of bond registration. SHELF equals one for shelf registered issues and zero otherwise. R144A equals one for Rule 144A issues and zero otherwise. Kidwell et al. (1984, 1987) show that shelf-registered bonds have lower yields. Thus, we expect the coefficient for the SHELF
For example, Houweling et al. (2005) finds a liquidity premium on corporate bonds. Fenn (2000) shows that issue size and maturity have an impact on bond yield. Kidwell et al. (1984, 1987) document lower yields for shelfregistered bond issues. 19 As a robustness check, we also run the regression models without the MATURITY variable. The results do not change materially. Different specifications of the variable do not affect the results either.
18

13

dummy to be negative. Studies on Rule 144A issues find higher yields on Rule 144A issues than non-Rule 144A issues (see Livingston and Zhou, 2002, Chaplinsky and Ramchand, 2004). Thus, we include this dummy as a control variable. Finally, since the bond default risk premium fluctuates with overall market conditions, a RISKPREM variable is included in the regression to control for changes in the market default risk premium. The RISKPREM is defined as the difference between the Moodys AAA Corporate Bond Index Yields (obtained from the Federal Reserve website) and yields for 10-year Treasury securities. The coefficient for this variable is expected to be positive. Other variables to control for market conditions are a series of zero/one year dummy variables with 2008 as the base case.

b. Main Regression Results In Table 3, Model 1 uses the sample of split rated bonds to study whether splits with a superior Moodys rating have a lower treasury spread than splits with a superior S&P rating. The coefficient on SUPMOODY is -8.38 and significant, indicating that yields on bonds with a superior Moodys rating average about 8 basis points lower than yields on bonds with a superior S&P rating.20 The coefficients on most control variables have the expected signs with the exception of the SENIOR dummy variable.21 In Model 2, we distinguish one-notch splits and two-notch splits by creating two test dummy variables: SUPMOODY1 and SUPMOODY2. SUPMOODY1 (SUPMOODY2) equals one for one-notch (two-notch) split rated bonds that have a superior Moodys rating and zero
20

Multiple bond issues by the same issuing firm may create a clustering problem (Wooldridge, 2002, 2003). We use the Cluster option in STATA to adjust for the potential clustering problem and report the cluster-robust p-values. 21 The significantly positive coefficient for the SENIOR dummy variable indicates that senior bonds have higher yields, a counter intuitive result that may be caused by a tendency for rating firms to give senior bonds unjustified higher ratings (John et al., 2010). Fridson and Garman (1997) and Fenn (2000) have similar findings.

14

otherwise.22 Livingston and Zhou (2010) find that multiple-notch split ratings are stronger signals of information opacity and bond investors require higher yield premiums on multiple splits than on one-notch splits. With greater information opacity and uncertainty, bond investors are likely to be concerned about the accuracy of ratings and more worried about over-rating. If Moodys is recognized by investors as a more conservative rating agency, superior Moodys ratings will lower yields for two-notch split rated bonds more than for one-notch split rated bonds. The empirical results confirm this conjecture. The coefficient on SUPMOODY1 notch is -7.64 while the coefficient on SUPMOODY2 is -12.60; both of them are statistically significant. Model 3 distinguishes between notch split rated bonds and letter split rated bonds by creating two other test dummy variables: SUPMOODY_Notch and SUPMOODY_Letter. SUPMOODY_Notch (SUPMOODY_Letter) equals one for notch (letter) split rated bonds that have a superior Moodys rating and zero otherwise.23 Rating splits at the letter level may indicate greater uncertainty and investors may be more worried about rating accuracy. As a result, we expect that a superior Moodys rating may lower yields more when the two ratings differ at the letter level than at the notch level. The coefficient on SUPMOODY_Letter, -13.75, is much larger in magnitude than the -5.89 coefficient on SUPMOODY_Notch. In addition, the coefficient on SUPMOODY_Notch is only marginally significant, while the coefficient on SUPMOODY_Letter is significant at the 1% level. Finally, to check if the results are driven by only a few credit rating categories, the Model 1 regression is run for each split rating category without the rating dummy variables. Table 4 reports the coefficient on SUPMOODY for each regression. The coefficient is negative in 26 out

22 23

Among the split rated sample, 83.87% (5,579) are one-notch splits and 16.13% (1,073) are two-notch splits. There are 2,158 letter split rated issues. Among them, 694 are two-notch splits and 1,464 are one-notch splits. Among the 4,494 notch split rated issues, 379 are two notch splits and 4,115 are one-notch splits.

15

of the 33 rating categories and is significant at the 1% or 5% level in 10 categories.24 In addition, 6 of the 10 significant rating categories are letter split ratings, consistent with the earlier finding that a superior Moodys rating causes a larger reduction in yields for letter splits. On the other hand, the coefficient is positive and marginally significant in only one rating category (BB+/BB).25 These findings indicate that investors consistently require lower yields for bonds with superior Moodys ratings.

c. Evolution of Conservative Ratings and its Impact on Treasury Spread Our sample period spans 26 years. It is very plausible that the relative conservative tendencies of the two rating agencies may evolve during such a long time period. This section investigates the changing relative rating conservativeness and its impact on bond yields. We first break the split rated sample into two time period sub-samples: 1983-1997 and 1998-2008.26 For the earlier period, the average Moodys Rating and S&P Rating are virtually the same: 10.458 and 10.463 respectively. In addition, 49.6% of issues have conservative (lower) Moodys ratings. On the other hand, for the latter period, the average Moodys Rating is 9.275, significantly lower than the average S&P Rating of 9.551. Furthermore, the percentage of issues with conservative (lower) Moodys ratings increases to 61.4%. Thus, the Moodys ratings are not systematically higher or lower than the S&P ratings in the earlier period, but become more conservative in the late 1990s and 2000s.

24

6 of the 10 rating categories with significant coefficients on SUPMOODY have fewer than 100 observations. This suggests that the statistical significance for the main results in Table 3 is not just driven by the large sample size. 25 To save space, we do not report regression results for control variables. Results are available upon request. 26 We choose 1998 as the cutoff year so that the two sub-samples have roughly the same number of observations. We have also broken the sample into two 13-year periods by using 1996 as the cutoff year and the results are qualitatively the same.

16

If bond investors are cognizant of the evolving relative conservativeness of the two rating agencies, the earlier (latter) period should exhibit smaller (greater) yield differences between issues with superior and inferior Moodys ratings. To check this hypothesis, we run the main regression (Model 1) for each of the two time period sub-samples and report the results in the first two columns of Table 5. The coefficient on SUPMOODY is -4.57 and only marginally significant for the 1983-1997 sub-sample. On the other hand, the coefficient on SUPMOODY is -13.37 and highly significant for the 1998-2008 sub-sample, suggesting that our main results are largely driven by the latter sample period, where Moodys is significantly more conservative than S&P. In addition to the sub-sample analysis, we also run the main regression on the full split rated sample with an interaction term between the SUPMOODY and a Time Trend variable, which is defined as issue year minus 1982. With the interaction term in the regression, the coefficient on SUPMOODY becomes positive (3.94) though not significant. The coefficient on the interaction term is -0.83 and significant at the 1% level.27 This finding indicates that bond investors gradually differentiate between the two rating agencies as one agency becomes more conservative than the other over time.28 The preceding results suggest that bond investors are sophisticated and can detect subtle differences between the two major rating agencies and act accordingly. Thus, these findings also highlight the importance for the rating agencies to protect their reputational capital. If one rating agency consistently becomes more lenient (conservative) over time, bond investors will

For the sake of brevity, we do not report the complete regression results, but they are available upon request. We also run separate regressions for each year and the results are consistent with the main findings. For the 15 years prior to 1998, the coefficient is negative and significant at the 1% or 5% level in only 2 years (1992 and 1995). For the 11 years in the latter period, the coefficient is significantly (at the 1% or 5% level) negative in 5 years. In addition, the magnitude of the coefficients is larger in more recent years than in earlier years.
28

27

17

gradually adapt to such changes by putting less (more) weight on that agencys rating in their assessment of default risk.

d. Conservative Ratings and Rule 144A An interesting development in the US bond market since the early 1990s is the significant growth of Rule 144A issues. Rule 144A issues are not registered with the SEC but can be traded among large institutional investors, or Qualified Institutional Buyers (QIB).29 Due to the lack of registration and lower standards for disclosure, there may be greater information opacity problems for Rule 144A bond issues.30 Indeed, Livingston and Zhou (2002) report that over half of Rule 144A bonds in their sample are issued by firms accessing the bond market for the first time and almost a quarter of the issuing firms do not file periodic disclosures with the SEC. Thus, bond investors are forced to rely more heavily on bond ratings to assess the default risk of Rule 144A issues. In addition, information opacity problems may also decrease the accuracy of bond ratings on Rule 144A issues and, as a result, bond investors will tend to be more concerned about over-rating if the two major ratings split. Hence, we expect that more conservative ratings will have a greater impact on yields of split rated Rule 144A issues. To check this hypothesis, we utilize the 1998-2008 subsample of split rated bonds.31 We break the sample into Rule 144A and non-Rule 144A issue subsamples and estimate the main regression (Model 1) for the two subsamples separately. The regression results are reported in the last two columns of Table 5. For Rule 144A issues, the coefficient on SUPMOODY is -22.81,
29 30

See Fenn (2000) and Livingston and Zhou (2002) for more details about Rule 144A issues. Lack of disclosure or lower quality of disclosure may lead to information opacity problems. Alternatively, firms with information asymmetry problems may prefer to issue in the Rule 144A market. 31 Rule 144A was first adopted in 1990 and the market grew significantly in the late 1990s. Indeed, most Rule 144A issues in our sample are from the 1998-2008 sub-sample. We have also used the full sample of split rated bonds to examine this issue and the results are similar. However, using the full sample introduces bias because most Rule 144A bonds are issued in the latter time period, a period where, as documented in the previous section, Moodys is much more conservative than S&P and the impact of rating conservativeness on bond yields is more significant.

18

much larger in magnitude than its counterpart for the non-Rule 144A bonds, -7.02. This evidence suggests that bond investors are particularly concerned about rating accuracy on bond issues with severe information opacity problems and place greater trust in the ratings from the conservative rating agency.

e. Economic Significance While the previous sections find the yield difference between bonds with superior Moodys and superior S&P ratings statistically significant, this section estimates the economic significance of the yield difference. We take two approaches to gauge the economic significance. First, we compare the 8-basis-point (13-basis-point) yield difference for the whole sample (the 1998-2008 subsample) to the average yield difference between two adjacent notch rating categories. We estimate, in Section V, that the average yield difference between two adjacent notch rating categories is about 30 basis points. Thus, a yield difference of 8 (13) basis points is approximately equivalent to a difference of 1/4 (1/2) of a notch rating. Second, we estimate how much the bond price will differ due to the 8-basis-point (13-basis-point) yield difference for a typical split rated bond in our sample. Table 1 reports that the average size of split rated bonds in our sample is $235 million, with an average maturity of 12.6 years and yield to maturity of 8.18%. The modified duration for the average bond is approximately 7.5. Thus, a yield difference of 8 (13) basis points translates to a price difference of 0.6% (0.98%), or $1.42 ($2.30) million.

19

V.

Conservative Ratings and Information Opacity Premiums The previous section compares the yields on split rated bonds with superior Moodys

ratings to those with superior S&P ratings. A follow up question is how the yields on these two different types of split rated bonds compare to non-split rated bonds. Livingston and Zhou (2010) document a 7-basis-point yield premium on split rated bonds over non-split rated bonds of similar credit quality and they attribute the yield premium to the information opacity of split rated bonds. However, they do not distinguish between those with superior Moodys ratings and those with superior S&P ratings. Our findings of lower yields on splits with superior Moodys ratings suggest that the information opacity premium varies between splits with superior Moodys ratings and splits with superior S&P ratings. Using the methodology of Livingston and Zhou (2010), we estimate the information opacity premiums for split rated bonds with superior Moody's ratings and for split rated bonds with superior S&P ratings. The methodology is described in detail in Appendix A. This methodology uses two treasury spread regression models to estimate the information opacity premium. Each model is run with both split rated and non-split rated bonds, allowing a measure of the impact of split ratings on treasury spreads. In the first model, called the superior rating model, treasury spreads (for the full sample of non-split rated bonds and split rated bonds) are regressed against rating dummy variables, control variables, and a split rating dummy variable. For split rated bonds, the superior of the two ratings is used to construct the rating dummy variables. Thus, the split rating dummy variable reflects the fact that a split rated bond has an inferior second rating not captured by the rating dummy variables. The coefficient on the split rating dummy variable can be interpreted as

20

the difference between the actual treasury spreads of split rated bonds and the estimated treasury spreads of these bonds if both rating agencies had assigned the same superior rating. The process is reversed in the second regression model, called the inferior rating model. For the full sample of non-split rated and split rated bonds, the treasury spread is regressed against rating dummy variables, control variables, and a split rating dummy variable. For split rated bonds, the inferior of the two ratings is used. Thus, the split rating dummy variable reflects the fact that a split rated bond has a superior second rating compared to the non-split rated bonds. The coefficient on the split rating dummy variable can be interpreted as the difference between the actual treasury spreads of split rated bonds and the estimated treasury spreads if both rating agencies had assigned the same inferior rating. Finally, the difference of the absolute values of the coefficients on the split rating dummy for the superior rating model and the inferior rating model is divided by two to arrive at the information opacity premium. The logic of this is explained in the Appendix. We first replicate Livingston and Zhous (2010) main findings on our whole sample (of both split rated bonds and non-split rate bonds) in Table 6.32 The coefficients on the split rating dummy variable from the superior and inferior rating models are 24.80 and -11.76 respectively, suggesting an information opacity premium of 6.5 basis points (that is, [24.80-11.76]/2). This is very similar to the 7-basis-point premium reported in Livingston and Zhou (2010). Next, we distinguish the cases of split rated bonds with superior Moodys and superior S&P ratings. In the first set of regressions, we exclude splits with superior S&P ratings. Thus, the yields of split rated bonds with superior Moodys ratings are compared with yields of nonsplit rated bonds, and the results are reported in the first two columns of Table 7. The
32

Our sample is slightly different from that of Livingston and Zhou (2010) for two reasons. First, their sample covers 1983 to September 2008, while ours extends to the end of 2008. Second, our sample excludes bonds with three or more notches of split ratings.

21

coefficients on the split rating dummy variables from the superior and inferior rating models are 20.01 and -16.22 respectively, indicating an information opacity premium of 2 basis points (that is, [20.01-16.22]/2). Then, we exclude splits with superior Moodys rating and compare the yields of split rated bonds with superior S&P ratings and yields of non-split rated bonds in the second set of regressions. The results are reported in the last two columns of Table 7. The 28.34 and -8.50 coefficients from the superior and inferior rating models imply an information opacity premium of 10 basis points (that is, [28.34-8.5]/2). This evidence indicates that the information opacity premium is significantly smaller when Moodys assigns a superior rating (2 basis points), and higher (10 basis points) when S&P assigns the superior rating. See Figure 1. The difference in the information opacity premium between the two groups of split rated bonds is 8 basis points, which is in line with the estimated average yield difference between the two groups as reported in Section IV. Based on the regression results in Table 7, we also estimate the average difference in yield spreads between two adjacent notch rating categories. For example, from the first column of Table 7, the coefficients on AA- and A+ are -63.79 and -51.98 respectively, suggesting a 12 basis points difference in the yield spreads between the two adjacent rating categories. For each rating category, we find the absolute yield spread difference from its adjacent higher rating category. Then, we weight the absolute yield spread difference by the number of observations in the rating category.33 The weighted average yield spread difference between all the adjacent rating categories is about 30 basis points.34

33

A simple average is biased because the yield spread difference between below investment grade rating categories are much larger but there are fewer observations. For below investment grade rating categories, the average yield difference is 67 basis points but they only account for 25% of the sample. For investment grade rating categories, the average yield difference is much smaller, about 16 basis points. 34 Using the coefficients reported in the other three columns of Table 7, we obtain similar estimates.

22

VI.

Discussion and Conclusion While most public corporate bond issues have ratings from both Moodys and S&P,

regulators and many academic studies do not differentiate between the two ratings and treat them as equivalent and interchangeable. In other words, a bond with an A+ rating from Moody's and an A- rating from S&P is thought to be the same as another bond with an A+ from S&P and an A- from Moody's. This wide perception of equivalence of the two major ratings is particularly troublesome when previous studies have consistently shown that Moodys is a more conservative rating agency and is more likely to assign a conservative rating than S&P when the two differ. This study performs a joint test that 1) there is a systematic difference between the Moodys and S&P ratings and 2) bond investors differentiate between these two ratings. We find that split rated bonds with superior Moodys ratings have lower yields than similar bonds with superior S&P ratings, suggesting that investors differentiate between these two situations. The yield difference is both statistically and economically significant. In addition, the differential impact of the two ratings on bond yields is more pronounced for Rule 144A bond issues, which have greater information opaqueness problems. This evidence indicates that bond investors are particularly concerned about opaque bond issues and rely more heavily on the ratings from the conservative rating agency in their assessment of default risk. Furthermore, the relative conservativeness of the two rating agencies is not static, but evolving. Accordingly, bond investors differentiate between the two rating agencies only when there is a systematic difference between them. This evidence highlights the importance for the rating agencies to protect their reputational capital. If one rating agency consistently becomes more lenient (conservative) over time, bond investors will gradually adapt to such changes by putting less (more) weight on that agencys rating in their assessment of default risk. This

23

finding supports the argument that rating agencies have strong incentives to issue honest and accurate ratings to protect their reputation capital. In addition, the finding suggests a need for the regulatory agencies, such as the SEC, to monitor and evaluate the performance of the nationally recognized rating agencies whose ratings are used as regulatory tools.

24

Appendix A We follow Livingston and Zhous (2010) methodology to estimate the information opacity premium of split rated bonds. Specifically, two treasury spread regression models are used. The first regression model compares treasury spreads for split rated bonds with non-split rated bonds with superior ratings. The superior ratings are used to create seventeen rating dummy variables: SUP_RATINGj (j = 1 to 17). To distinguish the split rated and non-split rated bond issues, a dummy variable for split rating, SPLIT, is included in the regression. The SPLIT dummy variable reflects the fact that a split rated bond has an inferior rating (in addition to its split superior rating) not captured by the rating dummy variables. The regression model is as follows:
TSi s *SPLITi j *SUP_RATING ji j *Control Variable ji j *YEAR ji . (A1)
j 1 j 1 j 1 17 8 25

Equation (A1) is called the Superior Rating Model. If yields for the split rated bonds are determined by the superior rating alone and the second rating has no impact, S should be insignificant. Alternatively, if investors price the inferior rating as well, S should be significantly positive; that is, the inferior second rating should increase yields since it conveys additional negative information. Thus, the coefficient for SPLIT, S, can be interpreted as the difference between the actual treasury spreads of split rated bonds and the estimated treasury spreads of these bonds if both rating agencies had assigned the same superior rating. The procedure is then reversed. In the second regression model, yields for split rated bonds are compared with yields for the inferior rating. Specifically, the inferior ratings are used

25

to create the rating dummy variables: INF_RATINGj (j = 1 to 17).A1 Thus, the SPLIT variable reflects the fact that a split rated bond has a superior rating not captured by the rating dummy variables. The regression model is as follows:
TSi I *SPLITi j * INF_RATING ji j * Control Variable ji j * YEAR ji . (A2)
j 1 j 1 j 1 17 8 25

Equation (A2) is called the Inferior Rating Model. The coefficient for SPLIT, I, can be interpreted as the difference between the actual treasury spreads of split rated bonds and the estimated treasury spreads of these bonds if both rating agencies had assigned the same inferior rating. The final step compares the two coefficients, S and I. Let N be the actual treasury spreads of split rated bonds and S (I) be the estimated treasury spreads if both rating agencies had assigned the same superior (inferior) rating. Then, as illustrated in Figure A1: S = N |S| I = N + |I| Let A be the average of S and I, or: A = (S+I)/2 = (N |S| + N + |I|)/2 = N + (|I| |S|)/2. Thus, the information opacity premium of split rated bonds is: PREM = N A = N (N + (|I| | S|)/2) = (|S| |I|)/2 (A3)

(A4)

That is, the information opacity premium (PREM) is the difference of the absolute values of the two coefficients divided by two. The information opacity premium is shown in Figure A1. If there is no information opacity premium on split rated bonds, then, per Equation (A4), the absolute values of the two coefficients, |S| and |I|, should be same. Conversely, if there is an information opacity premium on split rated bonds, then |S| should be larger than |I|.
A1

For non-split rated bonds, there is no difference between the SUP_RATING and INF_RATING variables.

26

Reference Avramov, D., T. Chordia, G. Jostova, and A. Philipov, 2007, Momentum and Credit Rating, Journal of Finance 62, 2503-2520. Bank for International Settlements, Basel Committee on Banking Supervision, 2000, Credit Ratings and Complementary Sources of Credit Quality Information, Working Papers No. 3. Billingsley, R.S., R.E. Lamy, M.W. Marr, and G.R. Thompson, 1985, Split Ratings and Bond Reoffering Yields, Financial Management 14, 59-65. Baker, H.K, and S.A. Mansi, 2002, Assessing Credit Rating Agencies by Bond Issuers and Institutional Investors, Journal of Business Finance & Accounting 29, 1367-1398. Blume, M., F. Lim, and C. Mackinlay, 1998, The Declining Credit Quality of U.S. Corporate Debt: Myth or Reality? Journal of Finance 53, 1389-1413. Butler, A.W., 2008, Distance Still Matters: Evidence from Municipal Bond Underwriting, Review of Financial Studies 21, 763-784. Cantor, R., and F. Packer, 1995, The Credit Rating Industry, Journal of Fixed Income 5, 10-34. Cantor, R., and F. Packer, 1996, Discretion in the Use of Ratings: the Case of the NAIC, Journal of Insurance Regulation 15, 234-255. Cantor, R., F. Packer, and K., Cole, 1997, Split Ratings and the Pricing of Credit Risk, Journal of Fixed Income 7, 72-82. Chaplinsky, S., and L. Ramchand, 2004, The Impact of SEC Rule 144A on Corporate Debt Issuance by International Firms, Journal of Business 77, 1073-1097. Covitz, D, and P. Harrison, 2004, Testing Conflicts of Interest at Bond Rating Agencies with Market Anticipation: Evidence that Reputation Incentives Dominate, Federal Research Working Paper No. 2003-68. Ederington, L., 1986, Why Split Ratings Occur? Financial Management 15, 37-47. Ederington, L., and J. Goh, 1998, Bond Rating Agencies and Stock Analysts: Who Knows What When? Journal of Financial and Quantitative Analysis 33, 569-585. Fenn, G.W., 2000, Speed of Issuance and the Adequacy of Disclosure in the 144A High-Yield Debt Market, Journal of Financial Economics 56, 383-405. Fridson, M., and M. Garman, 1997, Valuing Like-rated Senior and Subordinated Debt, Journal of Fixed Income 7, 83-93. 27

Gttler, A., 2005, Using a Bootstrap Approach to Rate the Raters, Financial Markets and Portfolio Management 19, 277-295. Gttler, A., and M. Wahrenburg, 2007, The Adjustment of Credit Ratings in Advance of Defaults, Journal of Banking and Finance 31, 751-767. Holthausen, R., and R. Leftwich, 1986, The Effect of Bond Rating Changes on Common Stock Prices, Journal of Financial Economics 17, 57-89. Houweling, P., A. Mentink, and T. Vorst, 2005, Comparing Possible Proxies of Corporate Bond Liquidity, Journal of Banking and Finance 29, 1331-1358. Hsueh, P., and D.S. Kidwell, 1988, Bond Ratings: Are Two Better Than One? Financial Management 17, 46-53. Jewell, J., and M. Livingston, 1998, Split Ratings, Bond Yields, and Underwriter Spreads for Industrial Bonds, Journal of Financial Research 21, 185-204. John, K., S.A. Ravid, and N. Reisel, 2010, The Notching Rule for Subordinated Debt and the Information Content of Debt Rating, Financial Management, forthcoming. Kidwell, D.S., W. Marr, and G.R. Thompson, 1984, SEC Rule 415: the Ultimate Competitive Bid, Journal of Financial and Quantitative Analysis 19, 183-195. Kidwell, D.S., W. Marr, and G.R. Thompson, 1987, Shelf Registration: Competition and Market Flexibility, Journal of Law and Economics 30, 181-206. Liu, P., and W. Moore, 1987, The Impact of Split Bond Ratings on Risk Premia, Financial Review 22, 71-85. Livingston, M., and L. Zhou, 2002, The Impact of Rule 144A Debt Offerings upon Bond Yields and Underwriter Fees, Financial Management 31, 5-27. Livingston, M., A. Naranjo, and L. Zhou, 2007, Asset Opaqueness and Split Bond Ratings, Financial Management 36, 49-62. Livingston, M., and L. Zhou, 2010, Split Bond Ratings and Information Opacity Premium, Financial Management, forthcoming. Moon, C.G., and J.G. Stotsky, 1993, Testing the Differences between the Determinants of Moodys and Standard & Poors Ratings, Journal of Applied Econometrics 8, 51-69. Morgan, D.P., 2002, Rating Banks: Risk and Uncertainty in an Opaque Industry, American Economic Review 92, 874-888. 28

Perry, L.G., P. Liu, and D.A. Evans, 1988, Modified Bond Ratings: Further Evidence on the Effect of Split Ratings on Corporate Bond Yields, Journal of Business Finance & Accounting 15, 231-241. Pottier, S.W., and D.W. Sommer, 1999, Property-liability Insurer Financial Strength Ratings: Differences across Rating Agencies, The Journal of Risk and Insurance 6, 621-642. Reiter, S., and D. Ziebart, 1991, Bond Yields, Ratings, and Financial Information: Evidence from Public Utility Issues, Financial Review 26, 45-73. Smith, R.C., and I. Walter, 2002, Rating Agencies: Is There an Agency Issue? in R.M. Levich, G. Majnoni, and C. Reinhart, Eds., Rating, Rating Agencies and the Global Financial System, Boston, Kluwer Academic Publishers. Van Roy, P., 2005, Credit Ratings and the Standardised Approach to Credit Risk in Basel II, European Central Bank Working Paper No. 517. Wooldridge, J.M., 2002, Econometric Analysis of Cross Section and Panel Data, Cambridge, MA: MIT Press. Wooldridge, J.M., 2003, Cluster-Sample Methods in Applied Econometrics, American Economic Review 93, 133-138. Yu, F., 2005, Accounting Transparency and the Term Structure of Credit Spreads, Journal of Financial Economics 75, 53-84.

29

Treasury Spread (Basis Points) I: Non-Split Inferior Rating I S: Non-Split Superior Rating A: Average of Non-Split Superior and Inferior Ratings L:
30

SP
8 10 6.5 2

L M A

All Splits

SP: Split-S&P Higher M: Split-Moodys Higher

Superior

Inferior

Rating

Figure 1. Split Rated Bond Risk Premium. This figure illustrates the risk premium on split rated bonds over nonsplit rated bonds. The difference of 30 basis points between I and S is the average difference in treasury spread between two adjacent ratings and is derived in Section V. The difference of 10 basis points between SP and A and the 2-basis-point difference between M and A are derived from Table 7 and discussed in Section V. The difference of 6.5 basis points between L and A are derived from Table 6 and also discussed in Section V.

30

450

400

350

300
Numer of Issues

250

200

150

100

50

0
AA-/A+ BBB/BB+ A-/BBB+ AAA/AA+ B/CCC+ A-/BBB BB/B+ BBB+/BBB BBB+/BBBCCC+/CCC AA+/AA AA-/A BBB-/BB+ AA/AAAA/A+ BB+/BB AAA/AA BBB/BBBAA+/AABBB-/BB B-/CCC+ A/BBB+ BB-/B+ A+/A BB/BBB+/B BB+/BBB-/CCC A/AA+/ABB-/B B+/BB/B-

Superior S&P Rating Superior Moody's Rating

Figure 2. Number of Split Rated Issues in Each Rating Category. This figure depicts the number of split rated bonds with superior S&P ratings and the number of splits with superior Moodys ratings in each split rating category.

31

Information Opacity Premium

Figure A1. Illustration of Information Opacity Premium. I stands for the estimated treasury spreads on split rated bonds if both rating agencies had assigned the same inferior rating. S stands for the estimated treasury spreads on split rated bonds if both rating agencies had assigned the same superior rating. A is the average of I and S. N stands for the actual treasury spreads of split rated bonds. The difference between N and A is the information opacity premium (PREM).

32

Table 1 Summary Statistics


This table reports the descriptive statistics of the Non-Split Rated sample, Split Rated sample, the Superior S&P Rating sub-sample, and the Superior Moodys Rating sub-sample. The Moodys Rating and the S&P Rating are two numerical variables ranging from 1(for CCC rating) to 18 (for AAA rating). Average Rating is the average of the Moodys Rating and the S&P Rating.

Non-Split Rated Sample Proceeds (in million dollars) Maturity Yield to Maturity Treasury Spread (in basis points) Moodys Rating S&P Rating Average Rating Rating Difference (Moodys Rating S&P Rating) % of Senior Bond % of Shelf-registration % of Utility Issues % of Rule 144A Issues % of Callable Bonds No. of Obs. 223.78 12.35 7.86% 197.91 10.33 10.33 10.33 0.00 87.46% 66.42% 37.65% 22.09% 33.90% 7,201

Split Rated Sample 235.46 12.60 8.18% 215.73 9.87 10.01 9.94 -0.14 86.83% 60.70% 40.33% 23.08% 35.75% 6,652

Superior S&P Rating 243.80 11.86 7.97% 215.17 9.43 10.60 10.01 -1.17 89.65% 61.64% 37.06% 26.28% 31.94% 3,691

Superior Moodys Rating 225.07 13.52 8.45% 216.43 10.42 9.27 9.84 1.15 83.32% 59.54% 44.41% 19.08% 40.49% 2,961

33

Table 2 Mean Treasury Spreads of Split Rated Bonds by Rating Category


This table reports the mean treasury spreads of each split rating category for the Split Rated sample, Superior S&P Rating sub-sample, and Superior Moodys Rating sub-sample.

AAA/AA+ AAA/AA AA+/AA AA+/AAAA/AAAA/A+ AA-/A+ AA-/A A+/A A+/AA/AA/BBB+ A-/BBB+ A-/BBB BBB+/BBB BBB+/BBBBBB/BBBBBB/BB+ BBB-/BB+ BBB-/BB BB+/BB BB+/BBBB/BBBB/B+ BB-/B+ BB-/B B+/B B+/BB/BB/CCC+ B-/CCC+ B-/CCC CCC+/CCC

Split Rated Sample 60.19 101.34 62.70 104.49 66.77 81.41 89.09 109.04 88.81 118.13 113.11 135.90 137.85 126.00 150.42 168.74 175.08 196.98 201.32 233.39 288.43 333.31 299.53 363.19 367.12 437.13 423.94 461.08 486.03 569.09 583.64 528.64 716.84

Superior S&P Rating 54.48 110.64 51.90 106.86 67.41 81.37 92.00 120.22 94.14 113.19 111.69 112.51 143.99 120.42 151.30 157.91 173.67 183.73 191.80 240.33 277.18 334.11 300.10 401.68 369.06 484.05 446.13 491.04 496.84 552.74 626.89 558.95 819.24

Superior Moodys Rating 65.37 88.01 65.60** 92.65 66.11 81.46 82.16 81.74*** 82.78*** 124.09 115.82 159.73*** 132.91** 132.46 149.47 191.67 177.13 237.76* 217.82 214.17 309.12 331.07 298.29 311.12*** 365.43 366.74*** 394.60*** 415.78*** 475.26* 574.73 553.78*** 515.81 599.81

***,**, * indicate the difference between the Superior S&P Rating sub-sample and the Superior Moodys Rating sub-sample is significant at the 1%, 5% or 10% level.

34

Table 3 Treasury Spread Regressions for Split Rated Sample


This table reports the treasury spread regression results for the Split Rated sample. The dependent variable is the treasury spread in basis points. The control variables include 32 split rating dummy variables with BBB+/BBB as the base case. MATURITY is the natural log of the number of years to maturity. PROCEEDS is the gross proceeds of the bond issue in millions of dollars. SENIOR equals 1 for senior bonds, 0 otherwise. CALL equals 1 for callable bonds, 0 otherwise. UTILITY equals 1 for utility issues, 0 otherwise. R144A equals 1 for Rule 144A issues, 0 otherwise. SHELF equals 1 for shelf registered issues, 0 otherwise. RISKPREM is the difference (in basis points) between Moodys AAA Bond Index Yield and 10-Year Treasury yield. The regressions also include 25 year dummies with 2008 as the base case. In Model 1, the test variable is SUPMOODY, equal to 1 if Moodys assigns a superior rating and 0 otherwise. In Model 2, the test variables are SUPMOODY1 and SUPMOODY2. SUPMOODY1 (SUPMOODY2) equals 1 if Moodys rating is one (two) notch above the S&P rating. In Model 3, the test variables are SUPMOODY_Notch and SUPMOODY_Letter. SUPMOODY_Notch equals 1 if Moodys rating is above the S&P rating but they are in the same letter rating category, 0 otherwise. SUPMOODY_Letter equals 1 if Moodys rating is in a letter category superior to the S&P rating, 0 otherwise. The p-values (in the parenthesis) have been adjusted for potential clustering problems that might arise from multiple bond issues by the same firm.

Intercept SUPMOODY SUPMOODY1 SUPMOODY2 SUPMOODY_Notch SUPMOODY_Letter AAA/AA+ AAA/AA AA+/AA AA+/AAAA/AAAA/A+ AA-/A+ AA-/A A+/A A+/AA/AA/BBB+ A-/BBB+ A-/BBB BBB+/BBBBBB/BBBBBB/BB+ BBB-/BB+ BBB-/BB BB+/BB BB+/BB-

Model 1 58.19 (0.00) -8.38 (0.00)

Model 2 57.60 (0.00) -7.64 (0.01) -12.60 (0.03)

Model 3 56.58 (0.00)

-81.35 (0.00) -75.22 (0.00) -72.70 (0.00) -55.64 (0.00) -67.20 (0.00) -49.91 (0.00) -59.45 (0.00) -53.38 (0.00) -55.37 (0.00) -47.32 (0.00) -45.72 (0.00) -9.83 (0.00) -24.83 (0.00) -15.62 (0.00) 27.95 (0.00) 26.02 (0.00) 60.54 (0.00) 64.22 (0.00) 101.13 (0.00) 140.26 (0.00) 197.94 (0.00)

-81.41 (0.00) -73.33 (0.00) -72.91 (0.00) -54.60 (0.00) -67.22 (0.00) -47.83 (0.00) -59.30 (0.00) -51.85 (0.00) -55.34 (0.00) -45.05 (0.00) -45.62 (0.00) -7.40 (0.00) -24.89 (0.00) -13.34 (0.00) 29.66 (0.00) 26.09 (0.00) 61.98 (0.00) 64.35 (0.00) 102.66 (0.00) 140.40 (0.00) 199.47 (0.00)

-5.89 (0.07) -13.75 (0.00) -77.46 (0.00) -72.06 (0.00) -73.54 (0.00) -54.86 (0.00) -67.32 (0.00) -46.52 (0.00) -56.64 (0.00) -50.65 (0.00) -55.31 (0.00) -47.27 (0.00) -45.40 (0.00) -5.98 (0.00) -20.70 (0.00) -11.93 (0.00) 28.35 (0.00) 26.23 (0.00) 63.07 (0.00) 67.43 (0.00) 103.76 (0.00) 140.70 (0.00) 198.64 (0.00)

35

BB/BBBB/B+ BB-/B+ BB-/B B+/B B+/BB/BB/CCC+ B-/CCC+ B-/CCC CCC+/CCC MATURITY PROCEEDS SENIOR UTILITY CALL R144A SHELF RISKPREM Year dummies No. of Obs. R-squared

137.79 (0.00) 212.26 (0.00) 229.76 (0.00) 280.53 (0.00) 289.28 (0.00) 329.39 (0.00) 360.50 (0.00) 434.04 (0.00) 463.65 (0.03) 417.29 (0.06) 594.53 (0.01) 17.05 (0.00) -0.01 (0.09) 59.68 (0.00) -12.48 (0.00) 13.93 (0.00) 8.85 (0.07) -16.12 (0.00) 1.04 (0.00) Yes 6,652 0.81

137.99 (0.00) 214.51 (0.00) 229.80 (0.00) 282.68 (0.00) 289.41 (0.00) 331.49 (0.00) 360.56 (0.00) 437.58 (0.00 463.62 (0.03) 420.64 (0.06) 594.62 (0.02) 17.07 (0.00) -0.01 (0.09) 59.62 (0.00) -12.40 (0.00) 13.89 (0.00) 8.88 (0.07) -16.02 (0.00) 1.04 (0.00) Yes 6,652 0.81

138.29 (0.00) 215.85 (0.00) 233.92 (0.00) 284.01 (0.00) 289.51 (0.00) 329.73 (0.00) 360.60 (0.00) 439.41 (0.00) 468.25 (0.03) 422.37 (0.06) 594.79 (0.01) 17.08 (0.00) -0.01 (0.09) 59.70 (0.00) -12.27 (0.00) 13.90 (0.00) 8.96 (0.07) -16.02 (0.00) 1.04 (0.00) Yes 6,652 0.81

36

Table 4 Treasury Spread Regressions for Split Rated Sample for Each Rating Category
The dependent variable is the treasury spread in basis points. Model 1 (from Table 3) treasury spread regressions without the split rating dummy variables are estimated for each split rating category. The coefficients on SUPMOODY are reported with the cluster-robust p-values. The last two columns report the number of observations and the R-squared value for each regression. Some control variables are dropped from the regressions when there is only one level of variation.

AAA/AA+ AAA/AA AA+/AA AA+/AAAA/AAAA/A+ AA-/A+ AA-/A A+/A A+/AA/AA/BBB+ A-/BBB+ A-/BBB BBB+/BBB BBB+/BBBBBB/BBBBBB/BB+ BBB-/BB+ BBB-/BB BB+/BB BB+/BBBB/BBBB/B+ BB-/B+ BB-/B B+/B B+/BB/BB/CCC+ B-/CCC+ B-/CCC CCC+/CCC

SUPMOODY 9.77 -94.54 8.59 -28.54 -3.91 -30.79 -13.07 -30.14 -12.63 16.86 -5.45 -25.22 -6.24 -0.12 -1.30 -34.11 -2.89 -33.83 19.41 -31.49 32.35 -19.56 22.45 59.68 -18.38 -122.32 -39.38 -28.30 -12.82 -23.66 -77.70 -168.44 -116.00

p-value 0.26 0.02 0.30 0.01 0.16 0.00 0.02 0.04 0.00 0.18 0.14 0.06 0.15 0.99 0.77 0.05 0.70 0.47 0.12 0.14 0.07 0.77 0.26 0.53 0.21 0.04 0.00 0.38 0.22 0.79 0.02 0.22 0.55

No. of Obs 42 26 123 36 285 77 338 86 643 106 599 109 559 123 685 81 492 53 186 49 142 53 187 40 197 55 386 103 543 39 142 37 30

R-squared 0.84 0.91 0.58 0.93 0.75 0.86 0.50 0.82 0.65 0.75 0.68 0.76 0.66 0.84 0.66 0.84 0.53 0.77 0.70 0.88 0.69 0.75 0.59 0.80 0.55 0.83 0.47 0.69 0.42 0.78 0.42 0.82 0.58

37

Table 5 Treasury Spread Regressions for Split Rated Sub-Samples


The dependent variable is the treasury spread in basis points. This table reports results of the Model 1 (from Table 3) treasury spread regressions for several sub-samples. The first two columns report the results for two sub-samples of different time periods: 1983 1997 and 1998 2008. The third column excludes Rule 144A issues. The fourth column contains the Rule 144A sample. The p-values (in parentheses) have been adjusted for potential clustering problems that might arise from multiple bond issues by the same firm.

Intercept SUPMOODY AAA/AA+ AAA/AA AA+/AA AA+/AAAA/AAAA/A+ AA-/A+ AA-/A A+/A A+/AA/AA/BBB+ A-/BBB+ A-/BBB BBB+/BBBBBB/BBBBBB/BB+ BBB-/BB+ BBB-/BB BB+/BB BB+/BBBB/BBBB/B+ BB-/B+ BB-/B B+/B B+/BB/BB/CCC+ B-/CCC+ B-/CCC CCC+/CCC MATURITY

1983 -1997 -70.61 (0.00) -4.57 (0.07) -78.21 (0.00) -68.23 (0.00) -63.71 (0.00) -58.29 (0.00) -59.23 (0.00) -47.20 (0.00) -58.03 (0.00) -54.72 (0.00) -46.28 (0.00) -32.46 (0.00) -33.89 (0.00) -7.86 (0.30) -21.74 (0.00) -16.20 (0.03) 29.40 (0.04) 22.60 (0.00) 65.51 (0.00) 50.29 (0.00) 92.56 (0.00) 148.42 (0.00) 207.02 (0.00) 166.73 (0.00) 247.79 (0.00) 259.79 (0.00) 282.59 (0.00) 323.70 (0.00) 376.45 (0.00) 395.10 (0.00) 454.34 (0.18) 486.12 (0.87) 408.76 (0.00) 490.20 (0.00) 18.53 (0.00)

1998-2008 -11.55 (0.65) -13.37 (0.00) - 85.66 (0.00) - 79.61 (0.00) -89.02 (0.00) - 54.48 (0.00) - 81.84 (0.00) - 47.74 (0.00) -60.15 (0.00) - 50.12 (0.00) -63.45 (0.00) - 56.95 (0.00) - 59.03 (0.00) - 13.92 (0.14) - 29.09 (0.00) - 12.68 (0.08) 37.09 (0.01) 29.27 (0.00) 59.63 (0.00) 75.72 (0.00) 138.77 (0.00) 134.51 (0.00) 195.08 (0.00) 127.78 (0.00) 181.61 (0.00) 213.23 (0.00) 282.66 (0.00) 270.98 (0.00) 295.84 (0.00) 343.86 (0.00) 432.25 (0.00) 456.81 (0.00) 460.72 (0.00) 619.58 (0.00) 18.74 (0.00)

1998-2008 Non-Rule 144A 28.90 (0.36) -7.02 (0.12) -82.98 (0.02) -70.55 (0.00) -79.47 (0.00) -53.45 (0.00) -76.21 (0.00) -62.54 (0.00) -49.28 (0.00) -49.43 (0.00) -59.96 (0.00) -47.43 (0.00) -52.18 (0.00) -16.20 (0.10) -20.43 (0.00) -8.68 (0.21) 28.30 (0.03) 28.12 (0.00) 61.81 (0.00) 87.11 (0.00) 115.34 (0.01) 131.83 (0.00) 140.09 (0.00) 111.02 (0.00) 78.61 (0.20) 190.47 (0.00) 363.98 (0.00) 236.21 (0.00) 303.57 (0.00) 287.02 (0.00) n.a. 488.50 (0.00) 289.62 (0.00) n.a 19.10 (0.00)

1998-2008 Rule 144A Issue 66.97 (0.28) -22.81 (0.01) n.a -61.15 (0.00) -180.78 (0.00) -24.82 (0.19) -176.99 (0.00) n.a. -97.21 (0.00) -3.25 (0.91) -16.01 (0.42) -67.59 (0.00) -48.31 (0.01) 3.39 (0.89) -9.81 (0.58) -10.93 (0.59) 65.12 (0.01) 34.50 (0.07) 58.90 (0.03) 90.47 (0.00) 184.36 (0.00) 149.61 (0.00) 232.89 (0.00) 166.74 (0.00) 204.87 (0.20) 237.76 (0.00) 287.10 (0.00) 297.12 (0.00) 320.07 (0.00) 370.07 (0.00) 468.65 (0.00) 500.55 (0.00) 512.70 (0.00) 670.22 (0.00) -9.85 (0.19)

38

PROCEEDS SENIOR UTILITY CALL R144A SHELF RISKPREM Year dummies No. of Obs. R-squared

-0.02 (0.08) 75.58 (0.00) -3.74 (0.20) 9.78 (0.00) -10.18 (0.13) -10.90 (0.01) 0.38 (0.00) Yes 3,339 0.88

-0.01 (0.22) 51.11 (0.00) -20.43 (0.00) 13.62 (0.12) 29.20 (0.01) -0.24 (0.98) 1.40 (0.00) Yes 3,313 0.75

0.01 (0.07) 15.29 (0.43) -7.25 (0.12) 13.19 (0.22) 1.16 (0.91) 1.16 (0.00) Yes 2,022 0.70

-0.07 (0.00) 65.17 (0.00) -38.55 (0.00) 4.38 (0.74) n.a 2.02 (0.01) Yes 1,291 0.66

39

Table 6 Information Opacity Premium Regressions for Full Sample


The dependent variable is the treasury spread in basis points. The base case is BBB+ rated bonds. SPLIT is equal to 1 for split rated bonds and 0 otherwise. MATURITY is the natural log of the number of years to maturity. PROCEEDS is the gross proceeds of the bond issue in millions of dollars. SENIOR equals 1 for senior bonds, 0 otherwise. CALL equals 1 for callable bonds, 0 otherwise. UTILITY equals 1 for utility issues, 0 otherwise. R144A equals 1 for Rule 144A issues, 0 otherwise. SHELF equals 1 for shelf registered issues, 0 otherwise. RISKPREM is the difference (in basis points) between Moodys AAA Bond Index Yield and 10-Year Treasury yield. The regressions also include 25 year dummies with 2008 as the base case. In the Superior (Inferior) Rating Model, we use the superior (inferior) rating of split rated bonds to construct the rating dummy variables. Thus, the coefficient for SPLIT measures the impact of the Inferior (Superior) second rating on the treasury spreads of split rated bonds. The p-values (in parentheses) have been adjusted for potential clustering problems that might arise from multiple bond issues by the same firm.

Intercept SPLIT AAA AA+ AA AAA+ A ABBB BBBBB+ BB BBB+ B BCCC+ CCC MATURITY PROCEEDS SENIOR UTILILTY CALL R144A SHELF RISKPREM YEAR Dummies No. of Obs. R-square

Superior Rating Model 36.72 (0.00) 24.80 (0.00) -84.21 (0.000 -74.41 (0.00) -66.32 (0.00) -61.06 (0.00) -52.25 (0.00) -37.26 (0.00) -21.89 (0.00) 17.33 (0.00) 48.50 (0.00) 137.20 (0.00) 157.14 (0.00) 219.25 (0.00) 282.39 (0.00) 349.28 (0.00) 423.03 (0.00) 546.11 (0.00) 461.52 (0.00) 17.38 (0.00) -0.01 (0.07) 63.42 (0.00) -11.86 (0.00) 14.87 (0.00) 7.67 (0.09) -17.18 (0.00) 0.98 (0.00) Yes 13,853 0.81

Inferior Rating Model 43.86 (0.00) -11.76 (0.00) -90.33 (0.00) -65.57 (0.00) -60.28 (0.00) -53.29 (0.00) -42.96 (0.00) -35.12 (0.00) -21.33 (0.00) 13.35 (0.00) 42.28 (0.00) 93.21 (0.00) 161.49 (0.00) 183.85 (0.00) 252.58 (0.00) 318.80 (0.00) 392.65 (0.00) 488.51 (0.00) 507.13 (0.00) 17.58 (0.00) -0.01 (0.00) 60.41 (0.00) -10.79 (0.00) 15.27 (0.00) 10.18 (0.02) -15.72 (0.00) 0.98 (0.00) Yes 13,853 0.81

40

Table 7 Information Opacity Premium Regressions for Two Subsamples


The dependent variable is the treasury spread in basis points. The base case is BBB+ rated bonds. SPLIT is equal to 1 for split rated bonds and 0 otherwise. MATURITY is the natural log of the number of years to maturity. PROCEEDS is the gross proceeds of the bond issue in millions of dollars. SENIOR equals 1 for senior bonds, 0 otherwise. CALL equals 1 for callable bonds, 0 otherwise. UTILITY equals 1 for utility issues, 0 otherwise. R144A equals 1 for Rule 144A issues, 0 otherwise. SHELF equals 1 for shelf registered issues, 0 otherwise. RISKPREMIUM is the difference (in basis points) between Moodys AAA Bond Index Yield and 10-Year Treasury yield. The regressions also include 25 year dummies with 2008 as the base case. In the Superior (Inferior) Rating Model, we use the superior (inferior) rating of split rated bonds to construct the rating dummy variables. Thus, the coefficient for SPLIT measures the impact of the Inferior (Superior) second rating on the treasury spreads of split rated bonds. The p-values (in parentheses) have been adjusted for potential clustering problems that might arise from multiple bond issues by the same firm.

Superior Moodys Splits and Non-Splits Superior Inferior Rating Model Rating Model Intercept 37.59 (0.01) 40.15 (0.00) SPLIT 20.01 (0.00) -16.22 (0.00) AAA -86.87 (0.000 -90.33 (0.00) AA+ -79.37 (0.00) -66.97 (0.00) AA -67.03 (0.00) -61.37 (0.00) AA-63.79 (0.00) -56.52 (0.00) A+ -51.98 (0.00) -47.14 (0.00) A -36.33 (0.00) -36.73 (0.00) A-22.71 (0.00) -19.95 (0.00) BBB 14.26 (0.00) 12.56 (0.00) BBB43.49 (0.00) 41.70 (0.00) BB+ 131.07 (0.00) 107.63 (0.00) BB 175.92 (0.00) 176.81 (0.00) BB210.51 (0.00) 197.38 (0.00) B+ 268.30 (0.00) 258.46 (0.00) B 347.71 (0.00) 320.27 (0.00) B419.78 (0.00) 401.99 (0.00) CCC+ 521.10 (0.00) 487.62 (0.00) CCC 464.65 (0.00) 456.23 (0.00) MATURITY 17.74 (0.00) 17.86 (0.00) PROCEEDS -0.00 (0.29) -0.01 (0.19) SENIOR 65.95 (0.00) 65.44 (0.00) UTILILTY -11.92 (0.00) -10.58 (0.00) CALL 14.40 (0.00) 15.95 (0.00) R144A 1.93 (0.71) 7.88 (0.12) SHELF -17.35 (0.00) -14.41 (0.00) RISKPREM 0.94 (0.00) 0.94 (0.00) YEAR Dummies Yes Yes No. of Obs. 10,162 10,162

Superior S&P Splits and Non-Splits Superior Inferior Rating Model Rating Model 34.61 (0.01) 44.45 (0.00) 28.34 (0.00) - 8.50 (0.00) -83.74 (0.00) -89.16 (0.00) -67.16 (0.00) -69.70 (0.00) -65.43 (0.00) -63.21 (0.00) -60.50 (0.00) -57.47 (0.00) -49.52 (0.00) -43.98 (0.00) -37.42 (0.00) -35.45 (0.00) -18.68 (0.00) -22.59 (0.00) 14.23 (0.00) 10.13 (0.00) 44.35 (0.00) 38.32 (0.00) 128.84 (0.00) 89.31 (0.00) 153.08 (0.00) 157.83 (0.00) 217.74 (0.00) 181.93 (0.00) 286.73 (0.00) 255.08 (0.00) 345.47 (0.00) 328.64 (0.00) 423.72 (0.00) 400.23 (0.00) 558.49 (0.00) 515.55 (0.00) 460.13 (0.00) 569.75 (0.00) 18.16 (0.00) 18.38 (0.00) -0.01 (0.06) -0.01 (0.07) 65.37 (0.00) 60.03 (0.00) - 9.51 (0.00) -8.48 (0.00) 14.01 (0.00) 12.71 (0.00) 9.98 (0.07) 7.58 (0.14) -16.76 (0.00) -18.20 (0.00) 0.96 (0.00) 0.97 (0.00) Yes Yes 10,892 10,892

41

You might also like