Research Design: Data and Methodology

The data sources, sample firms, S&P corporate rating methodology, and research methodology used in this study are described in this section.

3.1 Sample and Data

We examined 830 corporations in 53 countries, excluding all financial institutions and banks, that had solicited or unsolicited credit ratings issued by S&P from 1998 to 2003. The sample used in this study consists of corporate issuers that meet the following two conditions. First, the corporate issuers must have had long-term issuer credit ratings (LTRs) in local currency provided by S&P for every January from 1998 to 2003. According to S&P's Corporate Ratings Criteria (2008), an “issuer credit rating” refers to “an opinion of the obligor's overall capacity and willingness to meet its financial obligations as they come due—whether rated or not” (S&P 2008: 10). Second, the issuers must not only have had issuer credit ratings listed in the S&P Global Ratings Handbook, but also have had financial reports that included a Datastream security code provided by the Thomson Reuters Datastream database³ prior to each of the six rating dates. As a result of this two-step screening process, we found that all unsolicited issuer ratings from nonfinancial firms during our study period were from Japan.

Specifically, the dependent variables in the rating determinant equation of the two-step model are the LTRs assigned by S&P as of January 1998, 1999, 2000, 2001, 2002, and 2003, which are reported in the S&P Global Ratings Handbook in February of each of the above years, respectively (S&P 1998a, 1999, 2000a, 2001, 2002, 2003). Data for the financial variables of the sample issuers are from the financial reports of Thomson Reuters Datastream. Some of the financial ratios used in this study (see Table 1 [ PDF 19.2KB | 1 page ] for the complete list and descriptions) are ones that S&P may examine in determining LTRs (S&P Credit Training Services 2000; S&P 2008). The financial ratios measure profitability, capital or debt structure, cash flow protection, liquidity, and firm size. As rating agencies usually examine the relevant financial variables for the previous three to five years, this study uses a similar approach. For example, in the treatment effects model using Wooldridge's instrumental variable method (Wooldridge 2002), the three-year averages of the financial variables in 1999, 2001, and 2002 (if available) were used as the independent variables to explain the January 2003 ratings. The same approach was applied to compute the three-year averages of financial variables of the other years.

In addition, we used the book value of total assets to measure firm size. While S&P does not have a minimum size criterion for any given rating level, the company believes that size and ratings are correlated because size often provides a measure of diversification and/or affects competitive position. In particular, relative size helps determine market position, extent of diversification, and financial flexibility (S&P 2008). Poon and Firth (2005) and Poon, Lee, and Gup (2009) found that larger banks have more incentive to seek credit ratings and that they tend to have higher bank ratings. Because rating agencies consider sovereign credit risk to be important in assessing the credit standing of banks and corporations (S&P 1997, 1998b), S&P's sovereign credit rating (SOV) is included in the rating determinant models to explain LTRs.

3.2 Standard and Poor's Corporate Rating Methodology

S&P considers an issuer's rating as an overall assessment of an issuer's ability and willingness to meet all financial commitments in a timely manner. Its credit analysts study both quantifiable and nonquantifiable factors in determining LTRs. S&P's corporate rating methodology organizes the analysis on a common framework evaluating each issuer's business risk and financial risk. Country risk, industry factors, competitive position, and comparisons of profitability and peer group underlie its business risk assessment. Government risk tolerance and financial policies, accounting, cash flow adequacy, capital structure and asset protection, and liquidity and other short-term factors are the key aspects assessed by S&P in determining the financial risk of an issuer. A preliminary overall company rating is derived from both the business risk rating and the financial risk rating (S&P Credit Training Services 2000; S&P 2008).

Each of the factors used in its rating methodology is evaluated, but S&P claims that it does not have any predetermined weights for these factors and that the significance of specific aspects varies from situation to situation. In fact, S&P usually establishes a team of expert credit analysts to assess information pertinent to the rating. A rating committee of five to seven voting members is always convened to assign a new issuer rating. The rating committee receives financial statistics and comparative analysis in advance and then the lead analyst makes presentations before the committee determines ratings (S&P Credit Training Services 2000; S&P 2008).

3.3 Univariate Tests

As LTR is not a continuous variable and instead follows an ordinal scale, a nonparametric test—the Mann-Whitney U-test—is used to test the following null hypothesis (H1) in addition to the t-test (a parametric test).

The alternative hypothesis to H1 is that there is a significant difference in the distribution of ratings between the two groups, which may imply that unsolicited ratings are, on average, lower or higher than solicited ratings.

3.4 Univariate Tests of Differences in Financial Profiles

Differences between the two groups of ratings may be attributable to differences in financial profiles and/or firm characteristics. T-tests and Mann-Whitney U-tests are conducted to test for differences in financial profiles and firm characteristics between firms with solicited and unsolicited ratings for both the overall sample and the Japanese subsample.

The alternative hypothesis to H2 is that there are significant differences between the financial profiles and firm characteristics of firms with solicited and unsolicited ratings.

3.5 Wooldridge's Two-Step Treatment Effects Instrumental Variable Model

Liu and Malatesta (2006) argued that a firm's decision to seek a credit rating is mainly influenced by firm characteristics such as firm size, profitability, and tangible asset level.⁴ Some of these characteristics, however, are also critical in determining the actual credit rating level of the firm. Hence, some of the characteristics that explain a firm's decision to obtain a credit rating are not independent of the characteristics that determine the credit rating level of the same firm. A financially weak (strong) firm might have less (more) incentive to seek a credit rating because the firm would expect to receive a low (high) credit rating level. Specifically, we need to mitigate the endogeneity of the firm's decision to seek a credit rating. Endogeneity occurs when the characteristics that affect a firm's decision on obtaining a credit rating also determine its credit rating level (i.e., there is a sample-selection bias). The credit rating literature discusses such bias in detail (see Poon 2003; Poon and Firth 2005).

Most previous studies applied Heckman's (1979) two-step estimation method to account for sample-selection bias. The first step in the process is to estimate the selection equation to determine the probability that a firm will seek a credit rating. A sample-selection bias variable (called the inverse Mill's ratio) is estimated in the process. The second step is to estimate the main equation to study the determinants of the credit rating levels by incorporating a set of explanatory variables and the inverse Mill's ratio using a regression model. The challenge of applying Heckman's procedure in the context of credit ratings is that the literature does not provide a theoretical foundation to incorporate specific contributing factors to the credit rating decisions in the rating decision equation. The decisions to include some variables are primarily based on some conceptual arguments or other practical reasons such as data availability. Therefore, the estimated inverse Mill's ratio may change depending on the number of variables used in the selection equation. With a different Mill's ratio, the second step in the Heckman procedure may yield different estimation results.

To circumvent the concern of using Heckman's procedure, we used Wooldridge's (2002) two-step instrumental variable method to account for the sample-selection bias. Similar to Heckman's method, Wooldridge's instrumental-variable approach also uses a probit model to estimate the rating decision equation with a set of firm characteristics in the first step. Then, a fitted probability of seeking a credit rating (Y_hat) is obtained from the estimated probit equation for each firm. The fitted probability is then used as the instrumental variable to replace the dummy variable that measures the effect of a solicited versus unsolicited rating (Y) in the main equation. In the second step (i.e., estimating the main equation), the credit rating determinants are estimated using a set of explanatory variables and the fitted probability instrumental variable. Wooldridge (2002) showed that such an approach does not require a perfect specification of the selection equation. Thus, the concern of specification errors in the first step in Heckman's method is mitigated. Recent studies, such as Faulkender and Petersen (2006) and Lin and Su (2008), also use Wooldridge's approach to deal with the endogenous selection issue. Using Wooldridge's instrumental-variable method, we tested the following null hypothesis:

The alternative hypothesis to H3 is that a sample-selection bias exists in corporate credit ratings. The sample-selection bias may partially explain the downward bias of unsolicited ratings in comparison to solicited ratings. The two-step Wooldridge treatment effects model is illustrated below.

Step 1: Rating decision equation (selection equation based on a probit model)

Y_i^* = Z_iγ+ξ_i (1)

The observed decision is Y_i = 1 if Y_i^* > 0
Y_i = 0 if Y_i^* ≤ 0

Step 2: Rating determinant equation (main equation based on a regression model)

R_i = X_iβ + Y_iδ + ε_i (2)
where

R_i = the observed rating category that is assigned to issuer i;
X_i = a vector of explanatory variables for issuer i in the rating determinant equation;
Y_i = a binary variable representing whether an issuer has solicited or unsolicited ratings from S&P;
Y_i^* = an unobserved continuous latent variable for the selection decision;
Z_i = a vector of explanatory variables in the selection equation;
β,δ, γ = a vector of coefficients or coefficient;
ε_i,ξ_i= the random error terms that follow a bivariate normal distribution with zero mean and correlation ρεξ; and
ρεξ = the correlation between ε_i and ξ_i.

In Step 1, we estimated the rating decision with seven explanatory variables by probit. The explanatory variables of the probit model are: (1) TA = natural logarithm of total assets; (2) FIXTA = fixed asset to total assets ratio; (3) SOV = sovereign credit rating where AAA = 9, AA = 8, A = 7, BBB = 6, BB = 5, B= 4, CCC = 3, CC = 2, and SD/D = 1; (4) ROA = return on assets; (5) MTB = market-to-book ratio; (6) DTA = debt to total asset ratio; and (7) JAPAN = 1 if the firm is based in Japan. We obtained a fitted probability of obtaining a credit rating for each firm in the estimation process. The inclusion of some of these financial variables follows the work in Faulkender and Petersen (2006), Liu and Malatesta (2006), Poon and Firth (2005), and Poon, Lee, and Gup (2009). We expected that a firm would be more likely to seek a credit rating if it were larger, more profitable, and had more tangible assets.

In Step 2 (i.e., rating determinant equation), the dependent variable of the primary regression equation of interest is the S&P LTR, where AAA = 9, AA = 8, A = 7, BBB = 6, BB = 5, B= 4, CCC = 3, CC = 2, and SD/D = 1. The explanatory variables include: (1) TA, (2) SOV, (3) ROA, (4) DTA, (5) FFOTD = funds from operations to total debt, (6) various industry dummy variables, (7) JAPAN, and (8) Y = 1 for a firm having solicited a credit rating in Step 1. A higher value for an explanatory variable (β) suggests a greater probability of a higher credit rating. Because of the sample-selection bias in the decision to seek a credit rating, we used Y_HAT (a fitted probability for the likelihood of a firm having a solicited credit rating in Step 1) as an instrumental variable to replace Y in Equation (2). The testing of the null hypothesis that δ = 0 was used as the test for selectivity bias (H3) and to test whether the solicited rating was higher than the unsolicited rating (H2).

Multicollinearity in Equation (2) may be a concern because financial variables are highly correlated. Therefore, we selected only key financial variables representing the sovereign rating of the country where the firm is located, firm size, profitability, capital structure, and cash flow protection to explain credit rating levels. We also included industry dummy variables to account for industry effects on credit rating levels.

Download this Paper [ PDF 293.6KB| 26 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.