The Capital Requirements Directive/Basel II internal risk-based (IRB) approach to calculating probabilities of default (PD), loss given default (LGD) and exposure at default (EAD) is often viewed as relying heavily on internal data and statistical techniques. However, a number of sectors, that make up a significant proportion of banks’ portfolios, lack sufficient internal data to calibrate and validate models in a robust statistical manner with an appropriate degree of granularity and conservatism. Given these limitations, the qualitative aspects of model selection and validation, such as the model design and construction, quality of data and internal use of the models, play a much bigger role.
Low default portfolios or sectors typically include exposures to sovereigns, local and regional governments, large corporations and banks and non-bank financial institutions such as insurance companies and hedge funds, where no or very few defaults have been observed over time. They can also include somewhat higher default sectors, which are highly heterogeneous and make the simple use of historical default experience inappropriate for statistical validation (e.g. project finance, leveraged finance, etc). Sectors where the credit characteristics and business dynamics have changed significantly over time, such as utilities, pose additional challenges in this respect as historical data, even if available, may no longer be relevant.
Even where market data or data pooled among a number of institutions is used for statistical analysis, the same challenges regarding the statistical model construction and validation are likely to remain. One key issue in this respect is the comparability of the internal rating systems, criteria and default definitions of all the banks in the pool. Equally, given the intrinsic nature of these portfolios, it is likely that the future will not provide enough additional default data.
Despite the data limitations, the regulatory and market expectations are that the low default portfolios should still be subject to the minimum standards for IRB regarding the accuracy and conservatism of PD estimation and use of all relevant and material data.
Under appropriate circumstances and taking into account specific banks’ profile and portfolio characteristics, the use of third party default data can prove a valuable way to augment internal loss data. The challenge in this case becomes the ability to demonstrate that the external information is an appropriate benchmark for the portfolio concerned and the robustness of the mapping process from the external information.
Standard & Poor’s (S&P) ratings demonstrate good stability and consistency over time by sector, geography and rating category, attributes like any suitable choice of external benchmark. In addition, S&P’s ratings practices and procedures all explicitly seek to monitor and evolve ratings criteria to maintain this consistency in a constantly changing environment. The criteria utilised are explicitly forward looking to reflect rating factors that address this specifically and which have a big impact on the outcome. Importantly, these attributes are what permit any historical set of default data to be used in a forward-looking context.
Mapping Process
Under this approach, many banks utilise the S&P historically observed default frequencies to assign PD rates to their internal ratings categories via a mapping process. Despite the apparent simplicity, a robust, codified mapping process that can be easily documented and validated poses significant challenges.
This has led to a recent interest from banks and regulators, specifically regarding the circumstances where a mapping of a bank’s internal ratings to the S&P ratings categories supports the use of the provider’s ratings default and migration statistics based on the total universe of ratings to supplement the historical dataset available internally.
The regulatory expectation is as follows: “Mappings must be based on a comparison of internal rating criteria to the criteria used by the external institution and on a comparison of the internal and external ratings of any common borrowers.” (Revised framework no.462).
This implies the important assumption that the bank’s internal ratings system reaches sufficiently similar conclusions to that of S&P’s ratings. This requires confidence that the bank’s internal ratings system and S&P’s ratings methodologies are sufficiently comparable so as to produce similar results.
Additionally, a comparison of outcomes based solely on the overlap between the publicly available universe and the bank’s portfolio may not be sufficiently representative, as the rated universe may not be a good proxy for the banks’ portfolio or may be too limited. The banks also need to ensure the comparability and relevance between the exposures in its portfolio and those in the reference dataset within a well-defined and documented selection process.
Since ratings are forward looking, reflect important qualitative elements and their criteria constantly evolves to try to ensure today’s ratings perform consistently with the historical expectation, it would seem logical that a bank’s internal ratings system that maps to historical ratings default data needs to reflect these dynamics.
Typically, this type of internal ratings system comprises a list of risk drivers that are weighted and calibrated, leading to exposures being allocated a score. Given the nature of S&P’s ratings, qualitative factors can have a weight of more than 50% of the total rating analysis. Examples of qualitative factors include industry risk and trends, market position, competitive advantages and disadvantages; management quality; expected parent or state support and legal and financial structure.
Some fundamental issues that impact alignment are as follows:
- The definition and objectification of the qualitative factors.
- The relative weighting of quantitative and qualitative factors.
- Capturing non-linear or rule based approaches in the ratings methodology.
- The consistent and objective evaluation of the impact of parent or state support on the rated universe and its application within the IRS.
- The need for regular review, update and maintenance to address criteria evolution.
The treatment of qualitative factors in a model, and application of these to the segments of low default portfolios where external ratings are not available, present particular challenges that can easily lead to significant divergence in outcomes between a bank’s internal ratings system and S&P ratings or credit estimates on unrated counterparties.
In practice, adapting an existing bank internal ratings system methodology to become more aligned with the S&P methodology does not typically require radical change or loss of valuable work already established. It does, however, need detailed and precise gap analysis between the methodologies. With a transparent, aligned methodology as the base, bank specific customisations can typically be easily executed while retaining high confidence in the use of S&Ps ratings default data.
Regular (typically annual) reviews of the alignment of the methodology are key to using ratings default data in an internal ratings system with confidence as, in order to keep long-run ratings performance stable, ratings methodologies continuously evolve to reflect changes in the credit drivers in any given industry or segment.
Conclusion
The issue of methodological alignment is not a question of one method being viewed as superior to another, as no single internal risk system or methodology can be considered best in a low default portfolio. Rather, it reflects the need to validate the suitability of use of S&P ratings default data when calculating internal ratings system default probabilities in the absence of internal historical loss data.
Key Factors in Developing, Enhancing and validating Internal Ratings Systems for Internal Ratings Based Approaches Under Basel II Using Standard & Poor’s Ratings Default Data – Standard & Poor’s Risk Solutions – Jun 2006