Insights from the PRA roundtable on wholesale modelling

On 15 July, the Prudential Regulation Authority (PRA) held a roundtable meeting on the internal ratings based (IRB) approach for wholesale exposures, for PRA-regulated firms with, or interested in seeking, an IRB permission for these asset classes. This provided clarification on specific topics as requested by the industry and presented on related cross-firm modelling issues.

Our thoughts on the key highlights

Here are the 8 key insights from the roundtable:

1. Definition of wholesale
The PRA define wholesale (wave 1a) to include small, medium, and large corporates and financial institutions. The definition provides clarity and confirms our assumption that the expectations regarding wholesale apply to both smaller businesses in addition to their large global counterparts. Consistent perimeters are required and modelled portfolios should not overlap, meaning one obligor to one portfolio. The PRA are clear that appropriate segmentation must be applied within these perimeters – one size cannot fit all, and the level of homogeneity needs to be fully justified.

• Banks should think about their model segmentation. Segmentation policies must ensure full alignment with PRA expectations, with a clear internal definition of an obligor.

2. Human involvement
There are concerns about the level of judgements and decisions within the credit risk lifecycle and their relationship with the modelling process. The PRA have outlined that all judgements should be reviewed to understand whether they are accompanied by an increase in uncertainty, and therefore require a Margin of Conservatism (MoC). There is an expectation of greater human judgement in low data availability scenarios, accompanied by higher MoCs versus more quantitative approaches.

• Banks need to carefully consider how human judgement is applied and show the uncertainty in these judgements is clearly captured. This should include assessing the impact of alternative judgements and we would expect the use of a clear example-based framework.

3. Rating models

New rating models are not mandatory if existing models are already compliant with regulation and supported by a clear positive feedback loop. The PRA have outlined that rating models do not necessarily need to be built on a downturn period, but performance must be tested on both the current portfolio and the downturn. This testing must be at both portfolio level and within material segments. The models should be driven by data (where available) and include qualitative aspects where appropriate. The assessment of a consistent downturn period is therefore fundamental to both risk rating and calibration.

• PRA expectation is that Banks are incentivised to redevelop models when not working for users. We think a strong model use assessment programme that is robustly enforced is fundamental.

4. Low Default Techniques
To estimate Probability of Defaults (PDs) there are requirements to thoroughly understand alternative methods and assumptions with justification and analysis of sensitivity to these. We think it is worth completing a Pluto Tasche (PT) assessment, as the PRA will use this to inform a benchmark. Assumptions within Low Default Portfolio (LDP) methodologies are important because they are so influential on the PD within these methodologies. This includes a clear tangible and evidence-based definition of the number of years of the internal portfolio existing and portfolio size.

• Banks should have PT and at least one alternative approach for their LDP estimation process and justify the final calibration with consideration of the alternative outcomes.
• Banks need to clearly define when LDP processes should be used and be consistent in terms of use (input parameters, choice of PT verses other approaches).
  

5. Cycle length
Complete wholesale data availability is expected to be scarce. However, the selection of the long-run period (cycle length) should not automatically be all data available for modelling purposes. The selection of the period should involve a sufficient mix of good and bad years, be informed with reference to internal default experience, applicable economics and heavily tested alternative time periods. Data should be relevant and representative of the overall portfolio perimeter and not heavily biased by sub-segments.

• Banks should have a framework for identifying the cycle length across the rating system and drivers for this being subject to update as well as models being re-calibrated. We think this could involve leveraging industry trends and economics to pick up evolving vulnerable segments. This should also include a consistent view on downturn periods to inform appropriate model testing (point 3 above).

6. Master Grading Scale 

Master scales are tried, tested, and acknowledged to provide the benefit of being able to compare risks if set up appropriately. This enables consistent comparison and better conversations to evaluate risks across business units. The PRA are clear that their use can be continued, but the number of grades must have homogeneity and heterogeneity at its core and can be changed over time if suitably justified.

• Banks need to assess their Master Scale approaches, and if they have identified limitations, be clear on the supporting evidence for change. The PRA commentary suggests that changing these will require significant evidence into the future.

7. Margins of Conservatism (MoC) 

Fundamental to the MoC assessment process is identification and adjustment for the deficiency.  MoCs should be simple, justified and subject to appropriate governance and approval. This means that a MoC should not be dynamically removed or reduced without governance. Whilst simplicity is desirable, complexity is not precluded providing it is well supported.

• Banks should review their MoC frameworks to ensure they clearly set out how deficiencies are identified, standardised approaches for deficiency adjustment and MoC estimation. Taking account of the greater level of human judgement that are often needed, the frameworks should include robustly governed exit criteria within their MoC frameworks.

8. The target variable

Internal default, external default, agency rating and expert rank are all viable options for the target variable, but all relevant information should be considered. Default predictor models are stated as the preference where internal default experience is rich, but guidance on the minimum number of defaults has not been provided. The target variable should be qualitatively and, where possible, quantitatively assessed to confirm it is appropriate for all obligors within the portfolio.

• Banks should set clear parameters for identifying when defaults are scarce to the extent that alternative approaches to internal defaults should be considered, either as challenger or champion approaches. This will allow a strong understanding of the limitations inherent in any approach where default data is limited.

What wasn’t mentioned…

While the roundtable provided valuable insights into the PRA’s expectations, it is important to note some areas that were not addressed within the summary paper – these include:

• Number of performing grades: Due to the nature of wholesale portfolios having lower default volumes and target customer profiles, it can be a struggle to achieve the recommended amount of performing grades (7, as per in CRR Article 170) expected for a submission. However, the PRA didn’t provide any additional guidance to suggest a tolerance for a lower amount of performing grades for wholesale portfolios. 
• Definition of obligor: There remains variability between firms in the definitions of obligor, groups of connected clients and how model perimeters should be assigned within connected groups.
• Approaches to treatment of third-party support: These are significant risk mitigants within wholesale portfolios, and application of rules for full and partial support continues to be challenging with these models.

What’s next?

A key take-away from the PRA’s update is their expectation for a thorough understanding of alternatives and utilisation of strong sensitivity testing to back up robust decision-making. This is relevant across all areas raised. 

Get in touch If you would like to explore how we can support your approach to wholesale modelling and meeting the PRA’s requirements under CRR – info@4-most.co.uk.