- Omotunde Johnson
- Published Date:
- April 2002
The New Capital Adequacy Framework
The Basel Committee’s proposals for a new capital adequacy framework aim to align regulatory capital more closely with underlying economic risk. The Committee has proposed a new standardized approach to the risk weighting of banking book assets, which would place greater reliance than the current Accord on external credit risk assessment. The Committee has established the requirements that specialized agencies must meet for their ratings to be used in determining capital ratios for banks. The standards refer to agencies’ credibility and independence as well as to the objectivity and transparency of their judgments. Rating agencies must also show a satisfactory track record.
The standardized approach is mainly tailored to banks lending to large firms or operating in countries where financial markets are highly developed and the external assessment of borrowers’ creditworthiness is widely accepted and solicited. The main advantages are the standing of rating agencies within financial markets and the fact that credit ratings should be promptly revised when significant changes occur in the financial conditions of borrowers or in their sectors of activity.
However, the use of external ratings has some serious shortcomings that are mainly related to the differences between “market economies” and “banking economies” (Mayer, 1994). In banking economies, the financing of the corporate sector is mainly in the form of bank credit. A small number of nonfinancial entities are listed and rated by official agencies; only a small fraction of private liabilities are traded on official markets. As a consequence, a large number of safe, mainly small and medium-sized firms could be treated as risky simply because they do not raise funds in the capital market.
In Italy, for instance, the capital market is still small in relation to the size of the real economy. At the end of 1999, there were fewer than 300 listed companies, as against 1,000 in both France and Germany. Their number is so low mainly because of the very large number of small firms that do not raise any funds in the market and the low propensity of medium-sized firms to seek listings.
As an alternative to the standardized approach, the new capital adequacy framework envisages the use of internal ratings, in order to allow individual banks to exploit the large amount of quantitative and qualitative information they possess (Internal Ratings-Based Approach).
External Versus Internal Ratings: Issues for Financial Institutions
Developing internal rating systems appears to be the only way for many banks to comply with capital requirements closely related to credit risk. For this reason, the paper by David Stephen and Michael Fischer is of special interest from the point of view of banking economies.
The paper describes the Credit Suisse First Boston (CSFB) internal rating model. It provides a wide coverage of the issues that are relevant in developing an internal rating system and managing the risk associated with individual counterparties.
As clearly described in the paper, the minimum elements of an internal rating system include a classification procedure with separate borrower grades; a methodology for quantifying the probability of default (PD) of borrowers falling in any specific grade; and an estimate of the amount of loss-given default (LGD). For each loan to individual customers, the product of PD and LGD provides the expected loss rate on the exposure.
The paper focuses on three key issues:
- the requirements for consistent ratings;
- the validation of internal rating process; and
- the importance of processes to manage credit risk.
My comments will deal with the methodologies and validation procedures necessary for ensuring consistent internal ratings (points 1 and 2). I will first briefly summarize how this consistency is achieved in the CSFB’s internal rating process and then examine some implications of this approach.
Consistency in the CSFB’s Internal Rating Process
The CSFB’s internal rating process is based on a statistical rating model, the results of which are quantitatively validated and qualitatively assessed in order to ensure consistency. According to this approach, all ratings within the institution—even if they come from different rating scales—should be comparable on a similar basis and against externally recognized benchmarks.
The reason for referring to external ratings, according to Stephen and Fischer, is that they are the only global benchmarking tool and the fact that the credit risk management functions of many financial institutions have been built on the basis of methodologies comparable to the major external agencies. Therefore, according to them, the goal of the rating models within CRS (i.e., the credit rating system developed by Credit Suisse First Boston) is to replicate the rating methodology of the major rating agencies.
The model itself is based on linear regression techniques run on financial and market data and sector-specific variables; elements of both nonlinear and expert rule-based systems are also incorporated. The model classifies borrowers into buckets that reflect the rating scales used by the specialized agencies.
As regards quantitative validation, some arguments are raised in the paper: the model can be estimated on a sample of borrowers that also have an external rating; the ratings obtained on the internal portfolio are statistically tested against those of the benchmark portfolio; and as an alternative, historical default rates for each internal rating grade can be compared with agencies’ published default rates. This method can be used on the whole portfolio; however, it requires a large amount of internally generated historical data on defaults.
Implications of the CSFB Approach to Consistency and the Issue of Internal Versus External Ratings
A first and very general preliminary remark concerns the basic assumption that seems to underlie this approach—that is, that external ratings are the best proxy for the financial soundness of individual firms. In a sense, this means that if external rating agencies had access to the information on banks’ lending relationships, this would add little to their assessment of borrowers’ credit quality. It nonetheless seems hard to believe that banks’ ability to extract and elaborate information is inferior to that of rating agencies.
Italian experience suggests that individual banks’ ability to perceive deterioration in their customers’ financial situation may be diminished in the case of large firms having multiple credit relationships. It has been shown that the probability of a company’s insolvency increases as its banking relationships become more fragmented (Foglia, Laviola, and Marullo Reedtz, 1998). However, under normal circumstances and for the great majority of borrowers (especially medium-sized and small firms), the relationship is almost exclusive. Banks can count on their direct relationship with borrowers and the continuous monitoring of their deposit and loan accounts to make accurate evaluations of their financial conditions.
A second remark concerns the underlying assumption of full comparability of external and internal ratings. To begin with, caution is needed in referring to external ratings, especially as regards the following.
Comparability of ratings issued by different agencies. The rating industry includes not only the four large global rating agencies but also some regional agencies that adopt rating schemes different from those of the largest agencies and that are usually considered less conservative.
Comparability of ratings referring to bank loans’ market debt. The methodology of rating agencies has been developed for issuers of bonds in the United States and on international markets. The application of this methodology to different types of bank borrowers, including small and medium-sized companies, can be impeded by the different incentives faced by each counterpart, depending on the type of liability considered. Financially distressed companies may be induced to meet their obligations to the capital market in order to safeguard their reputation. As regards their credit relationships, they might accumulate substantial arrears and then ask for some rescheduling of their debts. Therefore, the probability of default attached to each rating level may differ for the same company.
As a matter of principle, solicited ratings are superior to unsolicited ratings in evaluating a company’s actual financial conditions. In the case of solicited ratings, the company will be willing to disclose more detailed information to the rating agency than it discloses publicly.
Moreover, a series of implications should be carefully considered, such as the following.
- Potential inconsistencies can arise between the through-the-cycle approach, typically used by rating agencies, and the point-in-time approach, typically used by banks, in assigning grades.
- Rating agencies claim that they rate on a relative, rather than an absolute, scale. Their intention is not to assign probabilities of default to the issuer/issue, but to construct an ordinal ranking of the ability to service debt; this means that the default probabilities of companies with identical ratings are actually different according to the country/sector of activity. Therefore, the validation of the internal rating model through the default track record of the rating agencies would require agencies’ default data per country/sector.
A third remark concerns the estimation process of the internal rating systtbem aimed at reproducing the external agency benchmarks. The authors recognize that a key issue is that significant proportions of agency-rated counterparties are large, whereas smaller firms tend not to be externally rated. Accordingly, we control for these factors.
To control these factors, Stephen and Fischer need to have a sample of bank borrowers who also have an external rating: they argue that there must be sufficient sample overlap between the internal and benchmark portfolios to give meaningful comparisons.
For this reason, Stephen and Fischer’s paper would benefit from the inclusion of some explanations regarding the methodologies that have been developed to control for the selection bias problem. It would be helpful to have some figures to evaluate
- the share of borrowers externally rated within the bank’s portfolio;
- the share of borrowers that, although not externally rated, are of comparable size/sector of economic activity/geographic region as those externally rated; and
- the share of unrated, small to medium-sized borrowers.
A fourth remark concerns the interpretation of the final results. The point is the extent to which it is possible to use the default frequency rates provided by the rating agencies in correspondence with each grade as a proxy for the default probability of each borrower. Recent analyses have focused on the high variability of the default rates within each rating class published by the rating agencies and across the rating classes. This means that it is difficult to distinguish the credit quality of individual bank borrowers within the same risk bucket. For instance, 65 percent of the loans granted by 26 large U.S. banks is concentrated in buckets corresponding to ratings BB and BBB whose PD rates looking one year ahead are 0.21 and 0.91 percent, respectively (Treacy and Carey, 2000).
The need to develop additional models to be applied to private counterparties is testified by recent research in the field by rating agencies, such as issuer ratings, which expand the universe of rated firms beyond those that have issued public debt, and bank loan ratings, which adjust for the differences in expected recoveries often observed for bank loans in default relative to bonds in default.
Within the same research field, various alternative validation methodologies have been identified and developed. In fact, the authors themselves mention other approaches that can be applied to estimate credit rating models, such as those based on accounting variables and market data.
Results from the Bank of Italy
Research carried out at the Bank of Italy since the early 1990s assesses the quality of bank credit portfolios. Bank borrowers are classified according to their riskiness on the basis of logit models run on a sample of healthy and distressed firms using both financial ratios and credit variables. The input data are drawn from the Balance Sheet Register run by the Cerved company and the Central Credit Register run by the Bank of Italy.
As regards credit variables, the share of the credit facility actually drawn down at each bank by individual firms and the amount of credit used in excess of the credit line have been introduced among the regressors. These variables are usually considered to be indicators of stringent liquidity needs. Their inclusion substantially increases estimation accuracy, compared with an alternative version of the model that only includes the accounting indicators (Borgioli, 1999).
The distinction between insolvent and normally operating firms has been made on the basis of the classification of bad debts in the Credit Register: 3,343 firms were classified as bad borrowers for the first time in 1995 and 1996. The “bad” firms were divided into four economic sectors and four geographic areas. A third of them were used for out-of-sample checks. For each set of distressed firms, a matched sample of healthy firms was randomly drawn from the Cerved Register.
The 16 models that have been estimated allow one to produce credit risk scores for 160,000 companies. For 56 of Italy’s largest banks, these firms account for 60 percent of total loans to the corporate sector.
A link with rating agencies’ evaluations can be established using the actual default rates to classify the borrowers in the same buckets identified on the basis of the default rates published by the agencies for each rating level.
The result of the exercise is shown in Table 3.6. Thirty-three percent of loans are granted to firms with a BB rating and 53 percent to firms with a rating of BBB or better. A quick comparison with corresponding figures in the Unites States is provided in Figure 3.14.
|Total||100||100||100||100|Figure 3.14.Loans Distribution per Economic Sector and S&P’s Rating Class
1 Source: Treacy and Carey (2000).
This example shows that for Italian banks, benchmarking to rating agencies does not allow a sufficient level of granularity to be reached within ranges of evaluation. An alternative approach, relying on historical loss data, allows more finely diversified rating scales to be obtained. In fact, internally rated firms can be grouped into classes of expected default frequency (PD) according to their actual historical default rate record.
This approach assumes that a large historical data set on the rate of default recorded in lending to different sectors and regions is available at each individual bank as well. For our research we used the data on bad loans contained in the Credit Register.
To simplify the presentation, PDs were computed for only four sectors of activity and a small number of risk buckets was considered.
This rating scale is more finely grained than that based on the default track record of rating agencies, however.
In Table 3.7, bank loans to firms with an expected default frequency between 0.1 and 0.9 percentage point (the PD corresponding to the S&P’s BB rating) have been divided into four classes. It is worth noting that 24 percent of total loans falls into the class with the lowest default frequency, and 23 percent falls into the second class with a 0.27 PD. The risk buckets with the lowest PD collect a lower portion of loans to the services sector and to the construction industry.
The evidence on the relative riskiness of economic sectors, emerging from both classification exercises, is consistent with what could be expected on the basis of the statistics on bad debts.
BorgioliS.1999 “An Exercise to Predict Corporate Insolvencies Using Balance Sheet Ratios and Credit Relationships Indicators” (unpublished; Rome: Bank of Italy).
FogliaA.S.Laviola and P.Marullo Reedtz1998 “Multiple Banking Relationships and the Fragility of Corporate Borrowers,” Journal of Banking and Finance Vol. 22 (October) pp. 1441–56.
MayerColin1994 “The Assessment: Money and Banking: Theory and Evidence,”Oxford Review of Economic Policy Vol. 10 (Winter) pp. 1–13.
TreacyWilliam and MarkCarey2000 “Credit Risk Rating Systems at Large U.S. Banks,” Journal of Banking and Finance Vol. 24 (January) pp. 167–201.