Tightening credit standards: the role of accounting quality

A search of Google Scholar indicates that Blume et al. (1998) paper has been cited by 109 studies. In terms of its regulatory impact, the paper was cited by the Federal Reserve Board in its 2001 comment letter on the proposal of Draft Standard and Basis Conclusions-Financial Instruments and Similar Items issued by the International Accounting Standards Committee. It was quoted also by the Federal Reserve Bank in a 2003 research report on the U.S. banking sector.

On the one hand, rating agencies have a financial incentive to accommodate the preferences of bond issuers from which they derive revenues. On the other hand, they have a countervailing incentive to protect their reputation for being independent and objective. Covitz and Harrison (2003) perform empirical tests of these two opposite views, and find that reputation effects dominate. None of this explains why they would tighten credit standards. If direct financial incentives are important, ratings may be too lax. If reputation effects are important, ratings need to be consistent and informative. Furthermore, these effects have been in place since the inception of the credit rating agencies. Any tightening explanation would have to be associated with changes in incentives.

Our examination of common rating determinants such as market beta, idiosyncratic risk, size, leverage and profitability shows that none of these variables meets the two conditions concurrently. So these variables cannot explain the downward drift in ratings. By the same token, neither can accounting conservatism, as the temporal increase in accounting conservatism, if true, should be associated with more favorable ratings over time.

S&P (2003).

From Moody’s web site. Relative to S&P, Moody’s has greater global coverage of banks (1,660 compared to 763) and industrials (3,402 vs. 2,852). See BCBS (2000).

In the U.S. corporate debt market, issuers usually obtain ratings from two rating agencies. A single-rated debt may receive a less favorable price than an otherwise equivalent debt with two similar ratings, as the absence of a second rating is usually viewed by investors as the issuer’s inability to secure another comparable rating.

For example, bond fund managers may use credit ratings to comply with internal by-law restrictions or investment policies that require certain minimum credit ratings for bond investments.

Financial contracts can include “ratings triggers,” which specify that lenders can demand more collateral, or request loan repayment, if the credit rating of the borrower falls below some level. In the case of Enron, for example, ratings triggers were included in trading contracts and contributed to the fast demise of the company.

One such example is the “core earnings” measure published by S&P (2002). The company states that “over the last decade...many members of the investment community expressed concern that earnings reports are becoming harder to understand, more difficult to compare across companies, and less useful to analyst and investors.”

This imprecision can be modeled in other ways. For example, Giesecke (2004) models an imperfectly observed default boundary. Collin-Dufresne et al. (2002) assume that firm values are observed with a lag.

In contrast, several other studies (for example, Francis and Schipper 1999; Landsman and Maydew 2002) find no evidence indicating that the informativeness of financial statements has declined over time. The mixed evidence on changing accounting quality is attributable to many factors such as differences in the definitions of accounting quality, research design, and sample selection.

Moody’s (2001). Ammer and Packer (2000), for example, check for consistency in default rates across sectors and obligor domiciles.

We did not use data on downgrades and upgrades because what matters is default. Some intermediate transition probabilities could be higher or lower, but ultimately the ratings measure is for the cumulative probability to default.

This sample includes non-U.S. firms. We compiled this information from publicly available documents, which contained detailed information for the global sample only. Rating agencies anyway indicate that their ratings should be consistent across geographical locations. Moody’s (2004), for instance, finds similar loss rates across the U.S. and Europe for each rating category. Ammer and Packer (2000) report that they “do not find significant differences in default rates between US and non-US firms.” S&P (2003) also states in its rating criteria that it imposes a consistent common discipline on all across-border analysis so that its ratings are comparable on a single scale. We use 1-year default rates because such choice leads to the largest number of independent observations. Using 10-year default rates would lead to basically two observations only over the period 1981–2003.

This was General American Life Insurance, which defaulted in August 1999 on $6.8 billion in funding agreements. The company had issued this short-term debt with a 7-day put option. In July 1999, Moody’s downgraded the company as a result of deteriorating financial condition. As a result of the downgrade, holders of the funding agreements exercised their puts, triggering a liquidity crisis. This story illustrates how a financial institution can jump from an AA-rating to default within 1 year.

We can apply the usual t-test assuming on an asymptotic normal distribution. The standard error is measured as \(\sigma =\sqrt{p(1-p)/N},\) where pis the default probability and N is the number of observations, assuming constant probability and independent observations.

may be indicative of credit rating loosening, not tightening. However, this effect could be also due to changes in the relative number of firms in the CCC, CC, and C ratings group, e.g., more firms drifting into the C category over time. Unfortunately, this cannot be tested. Due to the small sample size, S&P only reports default rates for the entire CCC/C category.

Interest coverage is the ratio of operating income after depreciation plus interest expense to interest expense. Operating margin is the ratio of operating income before depreciation to net sales. LT debt leverage is the ratio of long-term debt to total assets. Total debt leverage is the ratio of total debt to total assets.

Blume et al. (p. 1398) argue that the interest coverage ratio should enter the equation in a nonlinear fashion. They first set C = 0 for negative entries and C = 100% for entries higher than 100%. The four variables are then set at c1 = C for 0 ≤ C < 5% and c1 = 5 otherwise; c2 = C-5 for 5% ≤ C < 10%, c2 = 0 if C < 5% and c2 = 5 if C = 10%; c3 = C-10 for 10% ≤ C < 20%, c3 = 0 if C < 10% and c3 = 10 if C = 20%; c4 = C-20 for 20% ≤ C and 0 otherwise.

The latter two variables, however, are highly correlated, since total debt includes long-term debt. The collinearity problem may cause instability in the estimated slope coefficients.

declining R-square may reflect lower weights assigned by rating agencies to the included ratios in more recent years. To the extent that reported accounting data become increasingly noisier due to more manipulation of financial reporting, then rating agencies would rationally place lower weights on the reported ratios over time.

The decrease in R-square for investment-grade firms is unlikely due to the increasing sample size, because the number of firms went up even more for the speculative-grade sample.

When operating cash flows data are not available (before 1987), total accruals are calculated as the change in current assets (Compustat #4) minus the change in current liabilities (Compustat #5) minus the change in cash and cash equivalents (Compustat #1) plus the change in current maturities of long-term debt (Compustat #34) minus depreciation and amortization expense (Compustat #14).

The rationale for separate industry estimation is that different industries have different accounting ratios and different abilities to manage earnings. The rationale for separate annual regressions is that earnings and accruals are affected by common industry shocks. A general increase in earnings management, however, will not be picked up by this method. This should bias our results toward not finding evidence of earnings management.

Alternatively, we subtract changes in accounts receivable from changes in sales in the estimation of normal accruals (Eq. 2). This modification makes little difference in our findings.

It is worth noting that the correlation coefficients (Spearman and Pearson) between the yearly medians of discretionary accruals and the R ² in Table 6 for the investment grade firms are equal to  −0.62, significant at the one percent level. This is consistent with the argument that the declining value relevance of accounting data is likely caused by the increasing earnings management activities over time.

It is worth noting that increased accounting conservatism cannot explain the downward drift in ratings, as the temporal increase in accounting conservatism, if true, should be associated with more favorable ratings over time.

We use the OLS regression in place of the probit regression so that the error can be interpreted as residual rating after controlling for the common ratings determinants.

The R ² range from 0.04 to 0.64 with the mean and standard deviation of 0.30 and 0.14, respectively. Not surprisingly, traditional brick-and-mortar industries such as trucking and warehousing, hotels and other lodging, and food stores are among those with the highest R ². On the other end of the spectrum lie many R&D-intensive industries like computer software, transportation equipment, communications, and instruments.

See Standard and Poor’s (2003, p.  55) for a list of the accounting ratios used in its rating process. We use the 8 ratios listed, except EBITDA interest coverage because it is almost perfectly collinear with EBIT interest coverage.

We also use recent coefficient estimates from Begley et al. (1996) to compute Z-score and O-score for our industrials firms and add the Z-score and O-score as a control variable, respectively, to our main regression models. Such alteration, however, does not change our findings.