The PRA has recently completed a review of auditor reports and findings in relation to expected credit loss (ECL) calculations under the IFRS9 accounting standard. In this blog post, Carl Ireland reflects on the implications of their review, and sets out some advice that all impacted lenders should consider.
In October 2019, the PRA sent a letter to lenders outlining expectations for the high-quality practices that should be developed to improve the accuracy and consistency of Expected Credit Loss frameworks. This included model and data limitations, use of multiple economic scenarios, definition and application of significant increase in credit risk (SICR), recovery strategies and assumptions made about the lifetimes of exposures.
The latest review highlights positive progress but identifies four areas of concern that firms need to address:
1. Model Risk Management
The PRA reported significant progress compared to 2019 in relation to model risk controls, supported by testing and validation that provided a far greater level of assurance over levels of ECL. However, firms fell short in the following areas:
- Not all material models used in the calculation of ECL were appropriately covered by model risk controls (e.g. stress testing).
- Model monitoring was often based on data older than six months.
- The limits set as to what constituted acceptable performance. The approaches for grading models were inconsistent and often not in line with broader board risk appetite.
- Models were generally only assessed individually, meaning information on ECL performance at a portfolio or group level was not well represented.
- Lack of sensitivity analysis as part of continuous model risk management.
- Weak model performance and models defined as needing improvement following validation has led to the extensive application of post-core model adjustments (PMAs) - exacerbated by the impact of Covid-19 and the breakdown of model performance related to the unexpected/extreme movements in economic variables.
What does it mean? Firms need to further refine their model risk frameworks and:
- Reassess model definitions and model materiality thresholds.
- Enhance testing and validation processes to capture missing items.
- Redefine model monitoring and reporting to ensure useful information is captured that can be used to improve modelled outcomes. Reporting must be at the appropriate level and in line with risk appetite.
- Assess their current capability to operate their model risk and control framework under conditions where model performance is more volatile, focusing longer-term model redevelopment plans on the need to reduce the reliance on PMAs.
2. Economic Scenarios
Here, the PRA identified significant gaps, worsened by the pandemic, as firms try to cope with the impact of economic volatility:
- Sensitivity analysis is too limited with focus on the impact of adjusting the weightings of existing scenarios rather than exploring different scenarios. The application of these further scenarios required significant time and manual intervention making them impractical to use for regularly managing levels of impairment.
- Benchmark activity tends to be performed on changes to underlying models and not to the impact on ECL.
- Lack of clarity on how economic uncertainty is reflected in ECL estimations.
- Due to the significant time taken to run new economic conditions within a firm’s control structure, many have increased the use of tactical solutions to measure the impact of ever-changing economic dynamics and to avoid out-of-date data.
- Significant disparity emerged between the number and magnitude of firm’s downside scenarios. Firms using just one downside scenario were outliers prior to Covid, but this is now more pronounced and it is apparent that one downside scenario is insufficient.
What does it mean?
- Without the assessment of benchmark or sensitivity in ECL terms it is difficult to justify that a scenario is stressing a firm to an acceptable level. Clear justification that a scenario is suitably challenging a firm’s portfolios is necessary.
- Firms must enhance capabilities and governance to enable more frequent and varied economic scenarios without so much reliance on PMAs.
3. Significant Increase in Credit Risk
While progress against the 2019 high-quality practices was observed, several issues remained:
- Testing and measurement of the performance of SICR was broadly insufficient.
- SICR measurement is broadly focused on loan-level changes to PD, while other qualitative risk factors should be considered. The PRA notes that firms making good use of qualitative factors have relatively higher levels of stage 2 exposures.
- Collective assessments should be considered to capture events not well reflected in PD estimations (e.g. emerging risks or specific localised conditions).
What does it mean?
- Testing and validation thresholds should be reflecting a firm’s view of what levels SICR should be, considering industry-standard metrics.
- Qualitative factors should be considered when assessing SICR (e.g. forbearance, over-indebtedness, negative affordability, use of payday loans, breach of lending policy, and interest-only-loans approaching maturity without a confirmed repayment vehicle).
- Firms should look at specific events in assessing SICR and consider collective assessments where appropriate. Emerging risks such as Brexit, covid-19 and climate change may require additional thought and assessment as the impact of these may not be well reflected in current economic data.
4. Lifetime of an Exposure
The PRA observed:
- Differences in approaches to retail and commercial revolving facilities.
- A risk of bias as lives are cut-off at the point when substantially all defaults are expected to have occurred.
- Some firms were using credit review dates as a proxy for lifetime, although review standards were inconsistent and unsuitable to determine whether lives extend beyond the next review date.
What does it mean?
- Firms need to ensure they are assessing the limitations of modelling assumptions and simplifications on an ongoing basis, especially when these have material impact on ECL estimations.