With greater clarity on the path out of COVID-19 restrictions in most parts of Australia, the market is gradually gaining confidence that the worst of the pandemic is behind us. However, the outlook remains uncertain on the back of possible new variants, the pace of economic recovery and the potential for the pandemic to have left ‘structural’ or permanent³ effects on the economy in areas such as consumer spending and organisational profitability.

In this blog, we examine the implications for expected credit loss (ECL) calculations and discuss some of the trends that organisations should consider in calculating their ECL under AASB 9^[1].

Given macroeconomic and model uncertainty, many organisations have increased the number and size of overlays, post-model adjustments (PMAs) and in-model adjustments (IMAs). These overlays typically require significant management judgement. It is important that the rationale for overlays is sufficiently granular by documenting the particulars of the risk that is not captured in the model results which gave rise to the overlay as well as how this was calculated. Granularity is also important in ensuring that there is appropriate apportionment and allocation within reporting tables in an entity’s financial statements.

Adjustments should also be subject to robust challenge and governance (e.g. formal challenge from committees evidenced in minutes of meetings). Organisations should also consider whether overlays are aligned to management’s business views and actions. For example, it may be difficult to justify an overlay due to expected increase in default risk, if management’s internal forecasts do not reflect this.

As the economic recovery emerges and is evident in model data, some overlays will be incorporated in model results and others may no longer be required. The original rationale for the overlay should inform the timing of the release.

Questions organisations should be asking include:

• Have we justified our overlays with sufficient rationale?
• Have we calculated our overlays with sufficient granularity to enable stage apportionment in the financial statement disclosure?
• What is the likely duration of the overlay and the timing for release?
• Have our overlays been subject to a sufficient level of governance and challenge?
• Are the overlays consistent with our internal business views going forward?

Although the worst of the pandemic does seem to be behind us, uncertainty remains high in comparison with historic levels. Other macroeconomic concerns have emerged; such as skills shortages, the spectre of inflation and the potential for global shocks from sources such as the Evergrande default.

Organisations should ensure that the extent of uncertainty is reflected in their ECL estimates to account for the possibility of different forward-looking scenarios, rather than just the best estimate. Although the possibility of a downside scenario could be low, in some cases the loss associated with the outcome could be high, which would not be captured if only a best estimate was accounted for.

Organisations should also consider whether there is broad alignment between the scenarios included in an entity’s stress testing methodology (i.e. severe but plausible adverse conditions) and the downside scenarios included as part of a broader range of scenarios in the overall ECL estimate. The approach to weighting also requires careful consideration.

Questions organisations should be asking include:

• How are we addressing uncertainty in our AASB 9 framework (e.g. scenario definition / scenario weighting / other sensitivity testing), how has it been quantified, and how is the risk allocated in the loan book?
• Is there sufficient governance over areas where judgment is required in the estimate?
• Is there broad consistency across our internal and adjacent forecasting processes and AASB 9?
• Have we sufficiently catered for different outcomes in our ECL calculation, rather than only the best estimate?

As COVID-19 has emerged, some organisations responded to the economic uncertainty by adjusting their risk tolerance, which curtailed lending in certain segments of their business. This has resulted in a change to the mix of portfolios, where the historical risk profile differs from the present. This may require revisiting the segmentation or risk drivers within ECL models (including risk grading scorecards) to ensure they adequately cater for the risk profile of the loan book.

Another aspect for organisations to consider is whether other additional risk drivers have adequately been catered for, such as specific sector or industry risks. Historically, industry segmentation has not always been viewed as a significant driver of ECL in some portfolios. However, consideration of industry segmentation is more relevant in the current economy, particularly with respect to the timing recovery.

For some Australian sectors, the 18-month long pandemic is still far from over: in arts and recreation, and accommodation and food services, the number of people employed is currently more than 20% below the pre-pandemic peak. At the same time, the financial and insurances services industry has seen significant growth in employment throughout much of the COVID-19 period, with employment rising by more than 10% since March 2020.

The differing experiences across Australian industries throughout COVID-19 hints at the potential recovery paths in 2022. While for some industries it will be ‘steady as she goes’, other segments of the economy will enjoy a rapid rebound. Understanding the industry composition of the loan book has perhaps never been more important from an ECL perspective.

Questions organisations should be asking include:

• What additional segmentation or risk drivers may be required either in model or through overlays?
• Do we have accurate and complete data to enable additional risk differentiation, including at an industry level?

In many organisations, the actual default rate observed has been lower than that predicted by their Probability of Default (PD) models. This was caused by lower observed default rates from changing customer behaviour (spending less), government support and repayment deferrals[2] suppressing loan deterioration.

 Globally, some organisations froze Point-in-Time calibrations because management assessed the low observed default trends to be giving an artificially benign view of risk. Going forward, as the impact of government support dissipates, there is some uncertainty regarding the timing and extent of a ‘catch up’ of defaults that were previously avoided or delayed. This will create challenges for organisations as they consider the calibration of their models going forward. In addition to the calibration of models, organisations should also consider the effect of suppressed default history in their model monitoring (e.g. back-testing).

Questions organisations should be asking include:

• How do we treat the calibration data ‘window’ to accurately predict the default rate going forward?
• Which monitoring mechanisms should we employ to assess model performance and whether additional action is required (e.g. recalibration)?

Most ECL models used by banks incorporate forward-looking macroeconomic data as a means of incorporating the impact of the broad economic environment on credit defaults. Historically, economic downturns and periods of higher unemployment have been correlated with higher levels of default.

However, the COVID-19 economic shock unbuckled some of these historical correlations. In large part, that result was caused by the size of the fiscal response to the pandemic – JobKeeper and other payments to households, as well as cash grants, tax relief and other stimulus measures to support firms – which contributed to a lower level of defaults relative to what historical economic relationships would suggest. This means that ECL models calibrated using pre-COVID information may not be fit-for-purpose today.

Due to changing customer behaviour, uncertain sectoral risks and economic forecasts and upcoming regulatory changes such as APS 220, there will be a downstream impact on AASB 9 modelling in the next few years. Components of an organisation’s ECL models may also become important for other purposes, such as consideration of Climate Vulnerability Assessments, as required by APRA for some banks.

We expect to see continued focus on alignment and integration between expected, unexpected and stressed loss regimens to reflect data, definitional and model interdependencies.

In addition to the model considerations mentioned, it is also important that organisations understand how these feed into the financial statement disclosure. Good practice dictates that when uncertainty increases so does the need for transparent disclosures. To that end, it is important that the Credit function and the Finance function work closely together to ensure accurate and transparent disclosures.

Questions organisations should be asking include:

• How can we ensure that our model roadmaps and landscape is holistic across expected loss, unexpected loss and stressed loss regimens?
• Are the disclosures sufficiently transparent to inform users of financial statements?

Organisations have had to deal with several challenges and significant uncertainty in estimating the effect of COVID-19 on their loan portfolios. In addition to AASB 9, adjacent regulatory changes have stretched organisations' capacity across both first and second line. Going forward, organisations should focus on their strategic credit model journey over the medium term. Roadmaps should be holistic across expected loss, unexpected loss and stress loss regimens and reflect data, definitional and model interdependencies.

Authors:

Jonathan Sykes

Stephen Smith

Will Chan

Acknowledgements:

Richard Tedder

Henri Venter

Jasmine Armini

1 AASB 9 incorporates the requirements of IFRS 9.

2 APS 220 Attachment E – COVID-19 Adjustments.