IFRS 9, a new accounting standard on the classification and measurement of financial assets was rolled out in January this year. New expected credit loss (ECL) models for calculating impairment are at the centre of IFRS 9. This has led to regulatory attention on ECL models, its interaction with regulatory capital, risk management, as well as transparency and disclosure.
Now with the IFRS 9 having gone live, the banking industry is shifting gear from preparation to assurance. Within banks, this refocuses their attention from model development to model audit.
Model Audit teams review models from a risk perspective analysing various facets: data, methodology and models, implementation, testing, monitoring, operational controls, use test, and governance. At the heart of all of this is the identification of contentious choices embedded in the model.
IFRS 9 is expected to add further conundrums to the mix – we identify five top concerns based on our discussions with model audit teams at leading banks. Choices over these will have a significant impact on the final provision numbers, ultimately shaping banks’ capital utilisation and financial performance.
- Macroeconomic variables: In IFRS9, the choice of macroeconomic variables drives everything downstream. In industry, we mostly see 3-6 variables being chosen for this with GDP invariably as a variable. The key reasons behind this is data limitations and alignment with Stress Testing models. However, audit teams will challenge the assumptions and assess the validity of this choice as most banks lack a framework for selection of macroeconomic variables.
- Point in Time (PIT) conversion: IFRS9 asks for “adjustment for current conditions” but does not specify a way for it. There is a lack of consensus in the industry to meet this important requirement. This is often termed as ‘Point in Time conversion’ and is mostly achieved through the use of assumptions and benchmarks. Since this is the most material IFRS9 requirement, audit will challenge this both from a methodology selection perspective and will also challenge the overall evidence that the requirement for “adjustment for current conditions” is actually met. For instance, GDP is a coarse metric and 1-2 quarters delayed so PIT adjustment using GDP will be questioned.
- Forecasting: We can forecast the future only when we know where we stand today. Once the ask of “adjustment for current conditions” is met, IFRS9 asks for forecasting the future. The ask is not to be accurate but to be unbiased. Here, audit will specifically challenge the use of probability weights which are attributed to scenarios – this will have a considerable bearing on provisions. Here the key is for a model development team to demonstrate an unbiased methodology for deriving such probability weights which does not look arbitrary from an audit perspective.
- Significant Deterioration: In our opinion, the most arbitrary and least researched element in IFRS9 models is Significant Deterioration. In practice, a coherent selection framework can often be lacking. Instead, we see this requirement being met using multiple data sources and rules created as a patchwork to pick up any distressed credit by one of the five rules of Significant Deterioration. This approach, and the controls around its implementation, however, is being questioned by audit.
- Expert Credit Judgement (ECJ): IFRS9 allows for Expert Credit Judgement at various steps in the process of provision calculation and its use should ideally be linked to self-identified models and process weaknesses. This is interesting for audit as the list of self-identified weaknesses will be analysed for completeness and materiality. Since IFRS9 calls for an unbiased view of losses, ECJ should not be a punitive or conservative add-on. This poses another dilemma for audit to sign-off on the appropriateness of the ECJ after self-identification of the weaknesses.
Another contentious issue overall is audit of vendor models. Auditing such ‘black-box’ solutions poses a challenge due to lack of transparency and replicability which in turn emanates from the vendor’s desire to protect their intellectual property. Since methodology, implementation and controls are already embedded in vendor models, all the above-mentioned concerns get amplified when using vendor models for IFRS9.
Looking ahead, we see 2018 as a year of effective challenge by audit teams, which if addressed properly, promises significant enhancements to IFRS9 models – and to financial stability – in years to come.