When calculating expected credit losses, accuracy is paramount. This is a challenging task, but there are specific steps financial institutions can take to build meaningful risk ratings that lead to more precise loss calculations and better, more informed decisions.
Financial institutions have begun preparations to comply with the Current Expected Credit Loss (CECL) model. Many are looking for solutions to help them generate expected credit loss for various financial instruments. However, in working with financial institutions of all sizes and types, I have found that they are often overlooking one important thing: the need to provide accurate inputs into the calculation of expected credit loss.
The quality of inputs into a CECL process is vitally important to deriving meaningful and reliable outputs that can be used to calculate allowances and to inform business decisions. After all, the allowance for credit losses still represents one of the most significant estimates in a bank's financial statements.
Let's first evaluate what is being done today. Allowance methodologies at many commercial banks begin with segmenting the loan portfolio into pools with common credit risk characteristics.
During this process, there is a tradeoff between having too many segments where historical loss data is insufficient or having segments that are too broad and fail to capture different risk behaviors within a segment. Fortunately, the availability of industry default and loss data has expanded considerably in the last decade, improving the ability to quantify credit risk.
After a portfolio has been segmented, historical loss rates are calculated for loans within each segment. For example, commercial real estate (CRE) loans in the portfolio may have either a historical loss rate or an average annual net charge off rate of one percent that is applied to pass-rated CRE loans. Sometimes a multiplier is applied to higher-risk loans to assign a higher loss rate to loans that are adversely rated.
Management then applies one of several methodologies to convert the annual loss rates to lifetime credit measures for CECL. Examples of these methodologies are “remaining life,” “snapshot” or “vintage,” the details of which are beyond the scope of this article. Management may also apply qualitative adjustments to incorporate current conditions and economic forecasts.
While this approach is conceptually sound, and generally permissible from an allowance adequacy perspective, it provides limited accuracy. Credit decisions have a financial impact and are made at the loan level, so why not make these loan-level decisions more informed?
Qualitative Approach: Pros and Cons
Let's explore this question with an example of two CRE loans. Loan A is an amortizing loan secured by a stabilized income-producing office building in Cleveland, with 1.10x debt service coverage (DSC) and 80 percent loan-to-value (LTV). Loan B is an amortizing loan secured by an income-producing apartment complex in downtown San Francisco, with similar DSC and LTV ratios.
It is reasonable to assume that the two loans are close in risk rating and therefore have similar (if not the same) credit risk parameters feeding into the allowance calculation. But should they? Is the expected credit loss really the same? Should we not be accounting for the difference in property types, markets and sensitivities to macroeconomic factors as well?
Introducing qualitative factors is one approach to accounting for these differences. The qualitative reserve provides incremental reserves for risks not adequately captured in the quantitative reserve. However, therein lies part of the problem.
CECL brings better alignment of credit risk measurement and financial accounting than the incurred loss model. Common drivers of the qualitative reserve can now be captured in the internal risk rating and fed into the quantitative component of the allowance. Although the allowance for credit losses is an estimate that requires judgment, and qualitative adjustments will still be necessary, the allowance can be more reasonable and supportable if it is derived from a more sophisticated analysis.
Challenges with Internal Risk Ratings
Internal risk ratings are an institution's common language of risk for loans in their portfolio. These ratings should serve as inputs into the CECL calculation and should be conceptually sound and accurate for producing estimates of credit loss for individual financial instruments and segments.
However, I have found that many commercial banks employ a subjective methodology for assessing and assigning their internal risk rating to a customer or loan. It usually uses a numeric scale, such as 1-10, and often places most of the loans into just a few rating categories. Moreover, it is often not very clear what these numbers truly mean. Is the number linked to something, like a charge off or a measure of default risk? Is there consistency in loss rates for the same rating across the portfolio?
There are several tools in the market that will help an institution produce scenario-conditioned measures of risk that are based on call report data or the institution's internal ratings. Yet there is often a large disconnect between what a bank includes in those internal metrics and what actually influences credit loss. In addition to tools, workflows and solutions for CECL readiness, institutions must focus more on the foundation of the risk assessment: their internal risk ratings.
Simply put, without an accurate and meaningful risk rating as the starting point for the allowance calculation under CECL, the outputs will be of little use for making sound business decisions.
There are a number of ways to define accurate. One definition could be that the rating methodology is conceptually sound and derived from data and information appropriate for the portfolio of loans it will be used to score. Another could be that the rating comes from a model or scorecard that can be rank-ordered and differentiates the credit risk of obligors and loans in the portfolio.
Still another definition for an accurate risk rating could be when the rating is linked to a quantifiable risk measure, such as a probability of default (PD) or loss given default (LGD), whose level is aligned with the bank's default or loss experience (or an appropriate proxy, where data is limited). Lastly, accurate can be defined as when the resulting risk measure performs well in back-testing analysis, like comparisons between predicted and actual losses.
Building Meaningful Risk Ratings
The first step in the process for building meaningful risk ratings for CECL is to segment the portfolio into homogenous groups with similar risk characteristics for accurate and intuitive risk measurement. These groups should be based on criteria such as sector and subsector, size, geography and materiality.
It is worth noting that CECL does not dictate specific modeling methodologies for credit loss estimation. The choice largely depends on data availability, segmentation and, if appropriate, the availability of relevant third-party models for each segment. The size and composition of the sub-portfolios weigh heavily on methodology selected. Smaller, less material portfolios may be well suited for the simple approach described earlier.
Data collection for model building, or for testing the applicability of a vended model, should be closely aligned with the segmentation and methodology decisions, given how intertwined these tasks are. Independent of the methodology, the rating ought to be based upon a model or scorecard comprised of factors that are predictive of credit risk measures (e.g., PDs, LGDs, expected loss and net charge offs) to produce an output that adheres to the aforementioned accuracy criteria.
Continuing with the CRE example, there are models and methodologies that include macroeconomic factors in the risk rating. For instance, for a certain forecast of unemployment, you can assess the scenario's impact on real estate variables such as rents and capitalization rates. These variables cascade into a forecast of financial ratios like DSCR and LTV, which are among the most prevalent factors used in assigning risk ratings to CRE loans.
Including economic factors enables banks to incorporate the CECL-required current conditions and scenario-conditioned forecast across the remaining term of the loan. This approach reduces or negates the need for certain qualitative factors, because these are now woven into the more transparent and objective risk rating that feeds the quantitative component of the allowance.
Internal risk ratings are commonly derived from scorecards that blend statistically derived PD or LGD credit measures with more qualitative information to produce a single, borrower-specific PD rating or facility-specific LGD rating. This output usually maps to a master rating scale designed to help standardize risk across an institution.
Scorecards can take a variety of forms, including (1) a vended model with a bank-specific qualitative overlay; (2) an internally developed statistical model with a qualitative overlay; or (3) an expert-judgment scorecard, which can be used for a less material portfolio segment or in cases where there is insufficient data for modeling.
As institutions subject to CECL evaluate their systems and processes, it is vital that they also take into account how their risk ratings and modeling methodologies will be impacted by measuring expected credit loss.
A first step in completing the CECL puzzle should be to ensure that your firm has a risk rating framework that can serve as the foundation for its allowance process, providing “one version of the truth” for quantifying credit risk. Otherwise, your firm may face inconsistency and potentially inaccurate reporting in its financial statements.
By putting quality risk rating numbers into the calculation, you will be assured that CECL outputs are truly meaningful and strategically informative.
Chris Henkel is a senior director in the enterprise risk solutions group at Moody's Analytics. He leads a global team of risk consultants who work closely with banks, insurers and other financial institutions to improve how they measure and manage financial risk. He has previously served as credit risk instructor for the graduate banking school at Southern Methodist University, and has expertise in credit risk modeling; commercial credit and financial analysis; portfolio management; asset quality; allowance for credit losses; stress testing; credit administration; and safety and soundness examinations.
This article was originally published on GARP’s Risk Intelligence website on August 17, 2018.