Mastering Your Rating Scale [RMA Journal, The]

The year 2004 stands out as a pivotal time in recent history. It gave us the achievements of American swimmer Michael Phelps, winner of the greatest number of medals in a single Olympics; the longest nine-inning postseason baseball game in recorded history; the reelection of President George W. Bush; the opening of the National World War II Memorial in Washington, D.C.; the groundbreaking for One World Trade Center in New York City; and the emergence of Facebook in Cambridge, Massachusetts. That particular leap year also gave the banking sector a favorable credit environment and record earnings.

Not to be outdone, the international banking regulators in that same year gave us new requirements for capital adequacy. Like most risk managers, the regulators believed banking organizations should have the ability to understand their risks. But to be thoroughly familiar with the risks inherent in a loan portfolio, banks must first have the capacity to measure them.

In the seven years since the Basel Committee on Banking Supervision issued Basel II: International Convergence of Capital Measurement and Capital Standards, more and more institutions have embraced the practices outlined by the Basel accord's internal ratings-based (IRB) approach, shown in Figure 1. Just like Facebook, risk measurement models and scorecards are no longer confined to the complex and sophisticated institutions. These tools have been increasingly adopted by regional and community banks to enhance their ability to effectively manage single-obligor credit risk. Being able to differentiate risk between borrowers, especially in the aftermath of the financial crisis, has become not only a regulatory mandate but also a distinct competitive advantage.

The Role of the Master Rating Scale

Let's begin broadly with the measurement of expected loss (EL), considering that it influences most credit-related activities. Directionally consistent with net charge-offs, EL is a measure of credit risk, and its forecast represents one of the most significant estimates bank management can make.

Unfortunately, the best of intentions and investments can fall short of desired outcomes when the capacity to quantify and articulate risk is limited. From the perspective of the chief risk officer, how are we measuring credit risk? Does our rating scale accurately capture the composition of our portfolio? What about the calibration-the mapping? Does it provide sufficient risk differentiation? The answer to such questions can be found in the design of the bank's master rating scale (MRS).

A bank's MRS provides a common language of risk across the institution, combining unique risk attributes and lending activities into a shared view. It might be a basic scale, consisting of perhaps 10 categories, each with an implied rating and measure of credit risk. Alternatively, it could also be in line with recent practices to bifurcate obligor risk and facility risk. Regardless of the approach, the MRS should facilitate the goal of uniformly discriminating risk for activities such as loan approval, risk rating, loan loss provisioning, pricing, portfolio management, and stress testing.

An ineffective rating scale looks something like the concentrated series in Figure 2. This example asserts that 75% of the loan portfolio presents nearly the same degree of single-obligor credit risk to the bank. This might be the case if the portfolio is heavily concentrated, with few product types, markets, borrowers, etc., but in many cases it is a reflection of a rating scale that has not been designed to distinguish risk between borrowers. The true risk within a portfolio of similar loans can often be imperceptible.

Conversely, an effective rating scale, as shown by the diversified time series, helps bank management identify, monitor, and address asset-quality problems in an accurate and timely manner. Not only is it necessary for credit decisions, it is required by bank regulators. The MRS relies on the output of risk models and scorecards, which are often a score or a probability of default (PD)-consistent with the IRB approach of Basel II. The appropriate solution to derive the output may be a statistically validated model, a judgmentally derived scorecard, or somewhere in between.

Mapping the Risk Rating

Assuming the scorecard's ability to differentiate risk between borrowers, the bank must also examine the calibration of the scorecard. Specifically, it should evaluate how the risk rating output is mapped to its rating scale. Additionally, each rating category of the MRS should correspond to a risk statistic, such as PD, while providing sufficient granularity and reflecting the bank's portfolio and experience.

Figure 3 reflects the average default rate, by Moody's rating, from 1983 to 2010. While this number often serves as a useful starting point for designing a rating scale, it may not be appropriate for most community and regional banks.

Take the first 10 ratings-Aaa through Baa3, which represent credit exposures that are of investment-grade quality. Half the rating scale would be reserved for borrowers, on average, having an annual default rate of around 0.30% or less. Do you know how much of your portfolio would fall into these categories? Perhaps a better question to ask is this: How many borrowers fall within the Ba or B ratings in terms of credit quality? More importantly, are you getting sufficient differentiation where it matters most?

The key is to have the score and PD granularity where there is greater concentration of borrowers. It makes little sense for a bank with limited exposure to investment-grade borrowers to have numerous risk grades covering credits of Aaa to Baa quality. It makes more sense for the differentiation to reside further down the spectrum of credit quality, where the majority of the portfolio is concentrated.

The design of the MRS should also take into account business and regulatory considerations. Loan approval decisions regularly depend on a cutoff at a given risk rating, as do monitoring efforts, loan loss reserve rates, and pricing decisions, just to name a few. Additionally, regulators require differentiation among those assets deemed to be criticized (special mention or worse), and many banks today are seeking to improve the granularity within classified grades, delineating between substandard accrual and nonaccrual.

Designing an Effective MRS

There are numerous approaches to designing an MRS, but the impetus is to make it institution-specific with risk measures 50(such as PDs) increasing in a nonlinear fashion across rating grades. The final shape of the calibration curve would be indicated by the observed or benchmarked default rates, while matching the central tendency of the loan portfolio (that is, the long-run default rate through a complete economic cycle). Observed default rates are preferred, but in the absence of sufficient data, it is common to compare internal estimates against external benchmarks, such as agency ratings, peer banks, and other third-party sources.

For example, the bank could construct a representative sample of the loan portfolio, one that accounts for type, size, quality, sector, and geography. The next step would be to quantify the default risk (PD or score) for each borrower or group of borrowers, and the recovery risk (LGD or score) for each credit facility or group of facilities. Analyzing the descriptive statistics of the estimated PDs or scores will yield some information about the distribution and central tendency of the portfolio. It will direct attention to where concentrations exist and where consideration of score cutoffs is warranted.

In the end, the distribution of borrowers across the rating scale should be aligned with the estimate for the central tendency. This can be refined, over time, by comparing predicted PDs from the scorecard to a study of actual default rates for each rating grade. Ideally, borrowers should fall into a score range with a default rate that is aligned with expectations.

As shown in Table 1, each rating will be assigned a range of PDs and perhaps a midpoint for day-to-day credit processes. These values should remain relatively static as a through-the-cycle measure (being reviewed annually, but refreshed when warranted by historical experience or a material shift in the credit cycle). A borrower risk-rated 5, for example, will have an estimated PD of 1.13%.

Unlike legacy rating scales, where the exposure may reside in a rating category throughout the life of the loan (or close to it), the MRS should provide for rating migration as the credit risk improves or deteriorates. Along those lines, the median probability of default for a B-rated firm is around 3.50% over the past 12 years. During that period, however, it bottomed at 0.30% and peaked at 17%. A rating scale that enables a borrower's credit risk to remain static ignores the reality of such dynamic changes.

Capturing Facility Risk

To this point, the emphasis of the MRS has been on the ratings of the obligor, representing default risk. Although these measures typically account for the greatest contribution to the expected loss, it is necessary to capture the facility risk within the rating scale. The seniority of the loan in the priority of claims, the coverage provided by collateral securing the loan, and the overall structure of the credit facility will influence the LGD estimate (calculated as 1 - recovery).

The definition of loss will often vary as widely as the variance around the calculated average LGD. Most rating scales are designed with the actuality of imprecision in mind-in other words, they are given broad ranges. Similar to the process for obligor default risk, recovery risk is captured in the MRS by applying a floor and a ceiling to the range of LGD estimates for each facility rating category. A senior secured credit facility with cash collateral, for example, may warrant an LGD of 5%, whereas a senior unsecured facility would be closer to 50%.

The estimate should be captured by the scorecard and its output mapped to the rating scale, providing sufficient 51risk differentiation across different facilities. All else being equal, a well-collateralized loan actively monitored by the bank would likely command a facility rating with a lower LGD, compared with an under-collateralized or unmonitored facility to the same borrower.

Table 2 illustrates a hypothetical MRS that also incorporates the recovery risk via the facility rating. As portrayed, PDs are assigned to a borrower rating grade and LGDs to a facility rating grade. The values populating the table are the product of the two: the EL. Ultimately, the design of your MRS should include the following action items:

* Settle on the optimal number of grades, with applicability across the organization.

* Determine the relationship of grade-to-PD and grade-to-LGD.

* Incorporate sufficient pass grades to discriminate between performing borrowers.

* Integrate criticized/classified rating grades consistent with regulatory designations.

* Align grade PDs with internal default expectations and experience.

* Establish a process to assign grades from the preexisting scale to the new scale and then assess the impact.

Conclusion

When it released its new accord in 2004, the Basel Committee set out to have institutions better align risk with capital. But while novel at the time, advanced measurements are no longer considered a nascent view to credit risk. Like the users of online social networking, advanced measurements have moved on. And in the wake of the financial crisis, we as risk managers are tasked with the responsibility to improve how we measure and manage credit risk-specifically, to help the process evolve by means of actionable tools and information. If done well, risk measurement through scorecards calibrated to a thoughtfully designed rating scale can provide that capability and position your bank for the challenges of tomorrow. v

Christian Henkel is a director in the Risk Management Services group within Moody's Analytics. He can be reached at [email protected].