IFRS 9 – Approaches to assessing affordability

In its latest ‘Dear CFO’ letter the PRA says it continues to see a need to improve practices on IFRS 9 ECL, making clear its concern that models do not “capture risk associated with affordability” and firms need to “challenge the completeness of post model adjustments”. There are good reasons why models don’t capture the current risks well. There are also ways to overcome this, which can help with year-end reporting.

Background

On the 29th September 2023, Vicky Saporta of the Bank of England (BoE), issued a ‘Dear CFO’ letter covering thematic feedback from the 2022/2023 round of written auditor reports, focussing on IFRS 9 expected credit loss accounting (ECL). In what is becoming a custom, the PRA highlight a need “to embed high-quality practices further”.

What stands out in this letter, against a number of key themes raised, is the central concern of lenders’ ability to assess affordability challenges within their portfolios effectively: the PRA repeatedly point to “the impact of higher inflation and interest rates” on “vulnerable borrowers or sectors”. This speaks to the importance for lenders to critically reflect upon current approaches for measuring customers’ affordability risks. The guidance offered is that consideration should be assessed on all aspects of ECL, including PD, LGD, and enhancing “the quantification of PMAs to capture risks associated with higher inflation and interest rates by moving away from approximate approaches, such as portfolio level scalars”.

Poor historic link between affordability and losses

There is little data history to enable meaningful analysis for establishing a strong relationship between credit risk loss outcomes and the levels of affordability at a sub-portfolio or customer level. Reliable experience of this kind is not common over the last decade or more: the BoE base rate has been at a historically low level since the end of the global financial crisis up until the last two years, driven predominantly by the low inflation levels, around or below the 2% BoE target for that period.

Without appropriate data, both credit loss economic response and the granular correlation between customer loss and customer affordability cannot be confidently quantified.

This difficulty is further compounded by affordability being a key consideration of lenders’ acquisition strategy, often at stressed levels, for example, in mortgage portfolios following the Mortgage Market Review (MMR) requirements. This has led to consumers being assessed to ensure that they are able to continue to meet their payments in the event of significantly higher mortgage rates than were currently on the market. As a result, historically, it would be expected that, out of the accepted population that can be used for modelling, even those at the lower end of the affordability assessment range would still have surplus funds, enabling them to tackle changes in circumstances.

As a consequence of these two challenges, assessment of affordability at the point of acquisition (where most firms have reliable data) and subsequent defaults show little relationship, with default events being driven more by significant life events, e.g., unemployment, or other changes in personal circumstances, e.g., divorce.

Higher inflation and higher interest rates combined will see losses driven by affordability

The current economic situation and outlook is set to most impact those at the lower end of the affordability scale and we can expect to see a stronger relationship emerge. Those with smaller surpluses will gradually be ‘caught’ by the tide of rising outgoings. Specific discussions with lenders and observed industry-level rising mortgage arrears trends suggest an increasing prevalence of customers, either on SVRs or who have recently refinanced to materially higher fixed rates, struggling with the combined effects of higher costs of living and borrowing. The levels currently reported are not considered worrisome by most lenders, but they forecast continued growth in the coming 12-18 months.

An alternative to modelling to arrive at a complete coverage of the risks

If historic data for modelling does not exist, how can lenders address the PRA concerns to ensure risks are captured in the coverage of ECL – whether through overlaying judgement to the models directly or post model adjustments?

This combination of historical data difficulties, the current economic situation, and revised emphasis from the PRA on affordability results in the need for alternate analysis. This analysis needs to be conducted at a sufficiently granular level to allow for variation between portfolios or customer segments, and to more effectively reflect dynamics across multiple scenarios, particularly the combined impact of higher inflation and interest rates.

It is difficult to accurately assess current affordability levels for individual cases, as most information is only sourced at origination. Given changes in total income, essential spending for the individual customer, and data sources for these items, new data-gathering activities are required for each instance.

However, through a combination of available data sources and analytical modelling, it is possible to provide valid estimates of affordability for sub-groups of similar customers. This approach would not attempt to accurately reflect the affordability level of a specific individual but would instead model a probabilistic distribution of what their new affordability could be, thus identifying the likelihood of the customer having difficulties with continuing to service their debt. Hence, whilst this approach is not a panacea for selecting the correct treatment for individual customers, it provides an excellent foundation for post model adjustments for ECL purposes, providing robust means for addressing what the PRA are raising as their main concern.

4most have worked with clients to produce such estimates based on analysing available data: this has demonstrated that there are clear indicators of likely expenditure evolution based on an individual’s age and life situation, such as marriage status and number of dependents. This can be modelled alongside more objective information, such as inflation and or bureau-sourced current individual indebtedness, to provide an updated view of expenditure. In a similar way, income evolution can also be modelled, again based on factors such as age, employment industry and economics, e.g., wage growth.

This approach can be developed leveraging a variety of data sources, depending upon what is available to each institution. Some examples would be: current account data at a transactional level; affordability assessments from multiple applications across differing time periods; or I&E data sourced from collections (once corrected for bias due to the distressed nature of the population under analysis). The advantage of this approach, whereby a model is initially developed, is that the data to calibrate does not need to available for every customer within the ECL model and can be sourced over a period, to ensure sufficient volumes. Additionally, the accuracy of the data on specific cases is less of a concern if it is indicative of the population being modelled on average and consistent over time, e.g. using percentage change in credit turnover as an approximation of percentage change in income.

Once these components have been modelled, the application of the models to the full target population allows assessment of current affordability distributions to enable identification of customers most at risk on a probabilistic basis. This can then be used with more recent analysis, such as roll rates within collections, to determine likely loss outcome by the defined bands.

Conclusion and usage

Re-performing a full affordability assessment on each customer is costly and timely. Few lenders will wish to do this outside of either needing to help individual customers in difficulty or as part of standard affordability checks at the start of a new lending journey. At the same time, there is a real need to find a way to robustly assess the impacts of inflation and interest rate changes on customer affordability.

The approaches outlined above create a platform for closing these gaps and helping to address the PRA’s concerns, as outlined within this ‘Dear CFO’ letter. Enhanced affordability modelling will deliver a greater understanding of the rationale driving adjustments made to the IFRS 9 ECL. This can be done either through the application of PMAs or, over the longer term, the potential to incorporate such analysis within the core model. This is at the core of the PRA’s expectations of firms: an ask to “enhance both the documentation and testing of key model limitations, including the use of sensitivity analysis as part of ongoing model validation to both reassess the impact of using different modelling assumptions and challenge the completeness of PMAs”.

In addition, as an account level assessment, this approach would allow for not only an adjustment to ECL but also the potential to move higher-risk accounts to stage 2; addressing a further ask from the PRA, where they acknowledge that they are “pleased to see firms using PMAs to capture risks that are hard to assess at a loan level, such as affordability, in ECL”, with the caveat that: “we see scope to improve the linkage between the application of PMAs and collective SICR assessments to ensure that risk factors that drive the use of PMAs are also considered for staging purposes”.