The impact of split bond ratings on accounting research results: some additional evidence and some preliminary insights.

Tallapally, Pavani ; Luehlfing, Michael S. ; Cochran, James J. 等

INTRODUCTION

Dandapani and Lawrence (2007) indicate that an investor uses bond ratings to measure the relative credit risk of bonds. Additionally, they state that bond ratings affect a firm's access to capital as well as its cost of capital. Further, they suggest that two major credit rating agencies dominate the market in rating publicly traded bonds--Moody's Investors Service (Moody's) and Standard & Poor's (S&P). When Moody's and S&P (or some other credit rating agencies) disagree on the rating of a particular bond issue, a "split rating" is said to have occurred.

Livingston et al. (2007) indicate that about 20% of U.S. corporate and municipal bonds have letter split ratings (e.g., Ba versus B; and, BB versus B) while approximately 50% of notch-level ratings are split (e.g., B2 versus B3; and, B versus B-). Unfortunately, accounting studies that employ bond ratings (as either independent or dependent variable measures) do not address the issue of split bond ratings because such studies do not use more than one bond rating source (e.g., Khurana and Raman, 2003; and, Brandon et al., 2004). Thus the results of such accounting studies may be influenced by bond rating agency bias. Stated otherwise, we believe that the use of only one bond rating source may bias (to some unknown degree) the results of accounting studies employing bond ratings since such accounting studies most likely include split rated bond issues. Accordingly, to gain insights regarding the potential influence of split ratings on the results of previous accounting studies and to provide guidance with respect to future accounting studies employing bond ratings, we use the following strategy. First, we identify and summarize recent accounting studies that employ bond ratings. Second, we review the split bond rating literature. Third, we provide additional evidence to support the majority view that Moody's bond ratings are more conservative than S&P bond ratings. Finally, we discuss the potential limitations of not using multiple bond rating sources when conducting accounting studies employing bond ratings.

RECENT ACCOUNTING STUDIES

We noted two recent studies that we believe represent the current methodological state of accounting research employing bond ratings--Khurana and Raman (2003) and Brandon et al. (2004). Key methodological characteristics of these two studies are summarized in Table 1. Khurana and Raman (2003) use S&P bond ratings (n = 667) as one of their independent variables and "yield to maturity" as their dependent variable in a regression analysis. In contrast, Brandon et al. (2004) use Moody's bond ratings (n = 333) as their dependent variable; they employ logistic regression since bond rating data are polychotomous. Both studies employ Compustat accounting data as their data source.

Khurana and Raman (2003) find "... the fundamentals [detailed financial accounting information] to be priced in the market for new bond issues as indicators of expected future earnings and to be value-relevant in enabling the market to discern differences in bond credit quality over and above the published bond ratings." In turn, the results of Brandon et al. (2004) "indicate that the amount of nonaudit services provided by a firm's external auditors is negatively associated with that client's bond rating." While we believe that both of these studies contribute to the accounting literature, we also believe that the use of only one bond rating source represents a limitation in both studies because the use of only one bond rating source masks the potential influence, if any, of split ratings on the results of these studies. The remainder of our study documents the logic underpinning our concern.

PREVIOUS RESEARCH ADDRESSING SPLIT RATINGS

Billingsley et al. (1985) and Liu and Moore (1987) assert that when split ratings exist, investors primarily focus their attention on the lower of the two ratings. In contrast, Hsueh and Kidwell (1988) and Reiter and Ziebart (1991) find that investors primarily focus on the higher of the two ratings when determining the market price of a bond. While Cantor and Packer (1997b) find that the price of a split-rated bond reflects the average of the two ratings, the results of Jewell and Livingston (1998) on this issue are inconclusive. Livingston et al (2007, p. 61) suggest that "... split-rated bonds should be priced to offer additional risk premiums to compensate investors for the uncertainty about the issuing firm's fundamentals."

To varying degrees, uncertainty about the issuing firm's fundamentals relates to asset opaqueness (i.e., the inability to determine the value of an asset). On this issue, Morgan (2002) states that split ratings are the result of the asset opaqueness of some firms--especially banks. Somewhat similarly, Livingston et al. (2007) attribute split ratings to the level of asset opaqueness (informational asymmetry) of a firm.

Moon and Stotsky (1993) and Cantor and Packer (1997a) suggest that split ratings result from differences in rating scales (e.g., standards, cut off points and/or weights associated with rating determinants). However, Dandapani and Lawrence (2007) find that differences in rating scales are not the only explanation for split ratings. They (p. 79) indicate that their results "... suggest that about one-third of the bond split ratings can be due to the differences in ratings scales, while the remaining two-thirds are due to other reasons such as information asymmetry, judgmental differences, and randomness."

While Ederington (1986) asserts that split ratings are caused by random errors associated with a particular credit rating agency, Livingston et al. (2007) provide evidence that split ratings are not always associated with such random errors. Specifically, they (p. 50) find that "... split ratings are not symmetric between the two rating agencies. Instead, split ratings are more lopsided, with Moody's consistently on the downside." Stated otherwise, Livingston et al. (2007) find that Moody's bond ratings are generally more conservative (lower) that S&P bond ratings.

While the results of Livingston et al. (2007) are consistent with the results of Horrigan (1966), Morton (1975), and Perry (1985), the results are inconsistent with Cates (1977) who finds that S&P bond ratings are more conservative than Moody's bond ratings. This inconsistency may be associated with Cates' (1977) focus on bank holding companies (see Morgan, 2002). Before commenting on the possible influence of split bond ratings on accounting research results, we provide additional evidence to support the "majority view" that Moody's bond ratings are generally more conservative than S&P bond ratings.

DATA COLLECTION

There are 333 new bonds included in our study. These new bonds were issued from January 2004 through June 2006 and were rated by both Moody's and S&P. We collected Moody's bond ratings from the Long Term Debt Section of Mergent Online (www.mergentonline.com) and S&P bond ratings from the Credit Ratings Search Section of the S&P website (www.standardandpoors.com). As indicated in Table 2, the Moody's bond ratings range from Caa3 (greatest credit risk) to Aaa (least credit risk). All Moody's bond ratings except for Aaa are modified by the addition of a 1, 2, or 3 to show relative standing within the category, where the highest within the rating is 1 and the lowest is 3. The equivalent symbols used by S&P to designate its bond ratings are also provided in Table 2.

Hollander and Wolfe (1999) suggest that the expected frequency of observations in each category should be at least five observations when performing statistical analyses. Since there are fewer than five observations for the Moody's bond ratings designated Aaa, Aa1, and Aa2, we group the observations in these categories into a single category that we identify as "Aa2 and above" as shown in Table 3. We group Caa1, Caa2 and Caa3 for similar reasons and identify the resulting group as "Caa1 and below." Similar aggregation is done for categories with less than five observations with respect to the S&P bond ratings; that is, AAA, AA+ and AA are grouped and identified as "AA and above" and CCC+, CCC and CCC- are grouped into a category represented as "CCC+ and below." As a result of this aggregation, fifteen bond rating categories remain for both Moody's and S&P as shown in Table 3.

DATA ANALYSIS

Of the 333 bonds included in our study, 172 (52%) were assigned the same rating (i.e., the same equivalent rating) by both Moody's and S&P. Of the 161 split rated bonds, 104 (65%) are rated more conservative by Moody's than by S&P. In contrast, 57 (35%) of the 161 split rated bonds are rated lower by S&P than by Moody's. The extent of the differences between the split rated bonds are displayed in Table 4 using Moody's bond ratings as a benchmark.

As indicated in Table 4, of the 104 Moody's bond ratings which were below the related S&P bond ratings, 79 (76%) were one rating below the related S&P bond rating, 24 (23%) were two ratings below the related S&P bond rating, while only one (1%) was three ratings below the related S&P bond rating. As also indicated in Table 4, of the 57 Moody's bond ratings which were above the related S&P bond ratings, 50 (88%) were one rating above the related S&P bond rating while 7 (12%) were two ratings above the related S&P bond rating. The median Moody's bond rating is Ba2, while the median S&P notch-level rating is BB+. Since Ba2 and BB+ are not equivalent ratings, we statistically evaluate the following null hypothesis and alternate hypothesis using the Wilcoxon signed-rank test.

[H.sub.o]: The median bond ratings assigned by Moody's is greater than or equal to the median bond ratings assigned by S&P.

[H.sub.a]: The median bond ratings assigned by Moody's is less than the median bond ratings assigned by S&P.

Based on the results of the Wilcoxon signed-rank test, we reject the null hypothesis in favor of the alternative hypothesis (Wilcoxon signed-rank S = -2292.5; p < 0.00005, one-tail). Stated otherwise, with respect to the sample of bond ratings included in our study, the results of the Wilcoxon signed-rank test support the conclusion that Moody's notch-level bond ratings are more conservative than the S&P notch-level bond ratings. Thus our results are consistent with the majority of the prior research comparing bond rating agencies.

As previously indicated, 161 (48%) of the bonds included in our study were assigned split ratings. Given that the bond ratings included in our study were captured at the notch-level, and given that Livingston et al. (2007) find that approximately 50% of notch-level ratings are split, the number of split ratings included in our sample is consistent with previous research. However, since the accounting studies identified above (i.e., Khurana and Raman, 2003; Brandon et al., 2004) use letter ratings--not notch-level ratings--we re-perform the above procedures using letter ratings. To accomplish this, we collapse our fifteen bond rating categories into six bond rating categories (e.g., B1, B2, and B3 were collapsed into category B for the Moody's bond ratings, etc.). The resulting bond rating distribution is reported in Table 5.

The use of six bond rating categories (in lieu of fifteen bond rating categories) yields somewhat different results with respect to the number of split ratings. Of the 333 bonds included in our study, 263 (79%) were assigned the same letter rating by both Moody's and S&P using the six bond rating categories noted above. Consistent with the findings of Livingston et al. (2007), split letter ratings were assigned to 70 (21%) of the bonds included in our study. Of the 70 letter split rated bonds, 42 (60%) are rated lower by Moody's than by S&P by one letter rating category. The remaining 28 (40%) split rated bonds are rated lower by S&P than by Moody's by one letter rating category. The median Moody's bond rating is Ba, while the median S&P rating is BB. Please note that these median bond ratings are equivalent, whereas median bond ratings associated with the fifteen bond rating categories for Moody's and S&P were different.

Similar to the results associated with the fifteen bond rating categories, the Wilcoxon signed-rank test results associated with the six bond rating categories also support the conclusion that the Moody's bond ratings are more conservative than the S&P bond ratings (Wilcoxon signed-rank S = -248.5; p = 0.0475, one-tail) albeit at the [alpha] = .05 level of significance versus the [alpha] = .00005 level of significance (associated with the fifteen bond rating categories). To gain insights regarding the difference in significance levels, we performed a sensitivity analysis using a series of Wilcoxon signed-rank tests with respect to both the fifteen and six bond rating categories. The results are summarized in Table 6.

For the fifteen bond rating categories, the Wilcoxon signed-rank test results are significant at the [alpha] = .01 level for all but one of the time horizons (sub-samples) indicated in Table 6 (i.e., "Year 2004" with p = 0.0646, one-tail). Thus, in all but one instance, the results of the sensitivity analysis using fifteen bond rating categories support the conclusion that Moody's bond ratings are more conservative than the S&P bond ratings. In contrast to the fifteen bond rating category results, the six bond rating category results support the conclusion that the Moody's bond ratings are more conservative than the S&P bond ratings in only one instance at the [alpha] = .01 significance level (i.e., "Years 2005 and 2006" with p = 0.00705, one-tail). Generally speaking, we believe that these contrasting results primarily stem from the difference in the categorical widths employed in the sensitivity analysis. Stated otherwise, the use of a broader categorical width to measure each of the six bond rating categories (versus the narrower categorical width used to measure each of the fifteen bond rating categories) masks the extent of the significance of split ratings and, in turn, masks the extent that Moody's bond ratings are more conservative than S&P bond ratings. Additionally, the results reported in Table 6 suggest that Moody's bond ratings become increasingly more conservative versus S&P bond ratings during the period of our study.

IMPLICATIONS

Dandapani and Lawrence (2007) suggest that split ratings have significant financial consequences--especially at the mid range level. For example, they (p. 65) state that "... [regulators restrict many investment firms from investing in securities that do not receive investment ratings from at least two major rating agencies." If regulators employ at least two bond rating sources in making decisions, it would seem that accounting researchers should also consider using at least two bond rating sources--or at least state why they do not consider it necessary to use at least two bond rating sources. Use of only one bond rating source by accounting researchers is especially problematic in view of the fact that notch-level split ratings occur approximately 50% of the time (Livingston et al., 2007). Admittedly, while the extent, if any, of the potential bias associated with using only one data source has yet to be quantified, it appears that some bias exists given that a majority of the research to date suggests that Moody's bond ratings are more conservative than S&P bond ratings.

It is noteworthy that both recent accounting studies noted above (e.g., Khurana and Raman, 2003; and, Brandon et al., 2004) used letter ratings (not notch-level ratings), that is, both studies employed the broader categorical variable widths associated with letter ratings in lieu of the narrower categorical variable widths associated with notch-level ratings. Thus both of these "letter rating" studies implicitly reduced the extent of the potential bias, if any, associated with the use of only one bond rating source since letter split ratings occur approximately 20% of the time--not approximately 50% of the time as is the case for notch-level split ratings (Livingston et al., 2007). Admittedly, it would be interesting to know the extent to which differences, if any, would have resulted if both of the recent accounting studies noted above had employed notch-level ratings (versus letter rating) and/or had re-performed their analyses using another bond rating source. We suggest that accounting researchers who employ bond ratings in their studies should consider performing their analyses using multiple bond rating sources.

We note another methodological issue which could possibly bias the results of an accounting study employing bond ratings--data mixing. For example, while Brandon et al. (2004) employ Compustat accounting data to measure certain independent variables, they use Moody's bond ratings--not S&P bond ratings--as their dependent variable. Given that S&P owns Compustat, it is logical to assume that S&P employees involved in assigning the S&P bond ratings utilize data from the S&P accounting database (i.e., Compustat). Similarly, given Moody's previously affiliation with Moody's Industrial Manual (now known as Mergent's Industrial Manual), it is logical to assume that Moody's employees involved in assigning the Moody's bond ratings utilize accounting data from Mergent if for no other reason than tradition (and the fact that S&P is their competitor). However, there is a more interesting argument against mixing data in bond rating studies--Compustat standardizes its accounting data while Mergent does not, that is, Mergent employs "as reported" data in their accounting databases. We provide the following relevant excerpt regarding Compustat's standardized accounting data (www.compustat.com):

 Our internal research team rigorously examines original company
 sources by carefully extracting financial information, removing
 reporting biases and reconciling data discrepancies. After
 collecting data from diverse sources, we standardize it by
 financial statement and by specific data item definition, preparing
 information that is comparable across companies, industries, time
 periods and sectors. This standardized presentation makes it easier
 to identify companies with similar characteristics, such as capital
 structure and operating performance and is designed to complement
 how the data [are] used. Additionally we analyze financial
 statement notes to provide detailed breakouts to gain additional
 insight overlooked by other companies.

In summary, while Mergent accounting data proxy for the "as reported" accounting data found in EDGAR, this is not necessarily the case with respect to Compustat accounting data (as indicated immediately above). While the use of "standardized" versus "as reported" data in accounting studies is an issue beyond the scope of this study, we concur with the thoughts of Kern and Morris (1994) who warn that (p. 284)

 ... analysts and researchers need to exercise great care when
 selecting databases and variables from those databases. These
 choices can affect the results of and inferences drawn from
 empirical research in ways more than is anticipated by researchers.

CONCLUSION

The empirical results of our study, which are consistent with the majority of the previous bond rating research results, indicate that Moody's bond ratings are generally more conservative than S&P bond ratings. This finding leads us to suggest that accounting researchers employ multiple bond rating sources when conducting research on bond ratings to minimize the potential for bond rating agency bias. Additionally, we suggest that accounting researchers avoid mixing data sources (e.g., Compustat accounting data should be employed in studies where S&P bond ratings are employed).

Extensions of this research could focus on quantifying the extent of the bias, if any, associated with the use of only one bond rating source. Additionally, future research could focus on quantifying the extent of the bias, if any, associated with mixing data sources. Future research could also focus on quantifying the extent of the bias, if any, associated with the use of standardized versus "as reported" accounting data.

REFERENCES

Brandon, D. M., A. D. Crabtree & J. J. Maher (2004). Nonaudit fees, auditor independence, and bond ratings. Auditing: A Journal of Practice & Theory, 23(2), 89-103.

Billingsley, R.S., R.E. Lamy, M.W. Marr & G.R. Thompson (1985). Split ratings and bond reoffering yields. Financial Management, 14(2), 59-65.

Cantor, R. & F. Packer (1997a). Differences of opinion and selection bias in the credit rating industry. Journal of Banking and Finance, 21(10), 1398-1417.

Cantor, R. & F. Packer (1997b). Split ratings and the pricing of credit risk. Journal of Fixed Income, 7(3), 72-82.

Cates, D.C. (1977). Questions concerning the debt rating system. The Banker's Magazine, 160(2), 58-63.

Dandapani, K. & E. R. Lawrence (2007). Examining split bond ratings: Effect of rating scale. Quarterly Journal of Business and Economics, 46(2), 65-82.

Ederington, L. H. (1986). Why split ratings occur. Financial Management, 15(1), 37-47.

Horrigan, J. O. (1966). The determination of long-term credit standing with financial ratios--empirical research in accounting. Journal of Accounting Research. 4(supplement), 44-62.

Hollander, M. & D.A. Wolfe (1999). Nonparametric statistical methods (Second Edition). Wiley Series in Probability and Statistics.

Hsueh, P. & D. Kidwell (1988). Bond ratings: Are two better than one. Financial Management. 17(1), 46-53.

Jewell, J. & M. Livingston (1998). Split ratings, bond yields, and underwriter spreads for industrial bonds. Journal of Financial Research, 21(2), 185-204.

Kern, B. B & M. H. Morris (1994). Differences in the COMPUSTAT and expanded value line databases and the potential impact on empirical research. Accounting Review. 69(1), 274-284.

Khurana, I. K. & K. K. Raman (2003). Are fundamentals priced in the bond market? Contemporary Accounting Research, 20(3), 465-494.

Liu, P. & W. Moore (1987). The impact of split bond ratings on risk premia. Financial Review, 22(1), 71-85.

Livingston, M., A. Naranjo & L. Zhou (2007). Asset opaqueness and split bong ratings. Financial Management, 36(3), 49-62.

Morgan, D.P. (2002). Rating banks: risk and uncertainty in an opaque industry. American Economic Review, 92(4), 874-888.

Moon, C.G. & J. G. Stotsky (1993). Testing the differences between the determinants of Moody's and Standard & Poor's ratings. Journal Of Applied Econometrics, 8(1), 51-69.

Morton, G. (1975). A comparative analysis of Moody's and Standard and Poor's bond ratings. Review of Business and Economic Research, 11(2), 74-81.

Perry, L. G. (1985). The effect of bond rating agencies on bond rating models. The Journal of Financial Research. 8(4), 307-315.

Reiter, S. A. & D. A. Ziebart (1991). Bond yields, ratings, and financial information: evidence from public utility issues. The Financial Review, 26(1), 45-73.

Pavani Tallapally, Slippery Rock University

Michael S. Luehlfing, Louisiana Tech University

James J. Cochran, Louisiana Tech University

Gene H. Johnson, University of Hawaii-Hilo

Table 1: Summary of Recent Accounting Studies Employing Bond Ratings

 Item Khurana and Raman (2003) Brandon et al. (2004)

Subject Area Financial Accounting Auditing
Bond Rating Source S&P Ratings Moody's Ratings
Dependent Variable Yield to Maturity Bond Ratings
Data Source for
 Independent
 Variables Compustat Compustat
Sample Size 667 333
Statistical Procedure Regression Logistic Regression

Table 2: Initial Bond Rating Distribution

Moody's Ratings Number of Issues S&P Ratings Number of Issues

Aaa 3 AAA 3
Aa1 0 AA+ 0
Aa2 4 AA 9
Aa3 10 AA- 8
A1 13 A+ 13
A2 18 A 21
A3 15 A- 21
Baa1 20 BBB+ 21
Baa2 35 BBB 30
Baa3 28 BBB- 26
Ba1 15 BB+ 18
Ba2 27 BB 18
Ba3 28 BB- 23
B1 32 B+ 34
B2 29 B 44
B3 39 B- 25
Caa1 14 CCC+ 13
Caa2 2 CCC 6
Caa3 1 CCC- 0
Total 333 Total 333

Table 3: Bond Rating Distribution-Fifteen Categories

Moody's Ratings Number of Issues S&P Ratings Number of Issues

Aa2 and Above 7 AA and Above 12
Aa3 10 AA- 8
A1 13 A+ 13
A2 18 A 21
A3 15 A- 21
Baa1 20 BBB+ 21
Baa2 35 BBB 30
Baa3 28 BBB- 26
Ba1 15 BB+ 18
Ba2 27 BB 18
Ba3 28 BB- 23
B1 32 B+ 34
B2 29 B 44
B3 39 B- 25
Caa1 and Below 17 CCC+ and Below 19
Total 333 Total 333

Table 4: Split Ratings Analysis--Fifteen Categories

Panel A: Moody's Bond Ratings Below S&P Bond Ratings

Total Ratings Below One Rating Below Two Ratings Below S&P Rating
S&P Rating Below S&P Rating

 104 79 24
 100% 76% 23%

Panel B: Moody's Bond Ratings Above S&P Bond Ratings

Total Ratings above One Rating Above Two Ratings Above S&P Rating
S&P Rating Below S&P Rating

 57 50 7
 100% 88% 12%

Table 5: Bond Rating Distribution--Six Categories

Moody's Ratings Number of Issues S&P Ratings Number of Issues

Aa and Above 17 AA and Above 20
A 46 A 55
Baa 83 BBB 77
Ba 70 BB 59
B 100 B 103
Caa and Below 17 CCC and Below 19
Total 333 Total 333

Table 6: Wilcoxon Signed-Rank S Statistics (with Significance
Levels, One-Tail)

 15 Bond Rating 6 Bond Rating
 Year(s) Categories Categories

2004, 2005, -2292.5 -248.5
and 2006 (<0.00005) (0.0473)

2004 -186 13.5
 (0.0646) (0.3515)

2005 -235 -37.5
 (0.0059) (0.1141)

2006 -328 -52.5
 (<0.00005) (0.0207)

2004 and 2005 -862 -51
 (0.0021) (0.2884)

2004 and 2006 -1065 -94
 (<0.0001) (0.1212)

2005 and 2006 -1114.5 -180
 (<0.00005) (0.0071)