COVID-19 & IFRS 9 Expected Credit Loss | Practical implications of the PRA’s guidance on loss estimation

COVID-19 & IFRS 9 Expected Credit Loss | Practical implications of the PRA’s guidance on loss estimation

Lenders are in uncharted territory, especially when it comes to calculating IFRS 9 ECL. In order to provide direction to firms, the Prudential Regulatory Authority (PRA) released two statements in March (20 March and 26 March). In both releases there are two main themes with respect to IFRS 9 impairment:

• Allocation to Stage 2 and identification of defaults within the current environment; and
• Use of reliable forward-looking information that takes account the unprecedented financial support measures whilst ensuring that the effect is the crisis short lived.

We explore these two themes in more detail below and consider the practical implication for firms performing expected credit loss calculations under IFRS 9 for loans. We also highlight the short-term and long-term effects that the pandemic could have on expected credit loss estimation.

Identifying defaults and exposures that trigger significant increase in credit risk (SICR)

Firms need to reassess default and SICR (or Stage 2) triggers in the context of the COVID-19 pandemic. The PRA stated that, all other things being equal, payment holidays, other support measures and schemes or covenant breaches should not automatically lead to firms moving assets to Stage 2 or Stage 3. The practical implementation of this guidance may be challenging for many firms as they need to be able to identify assets that have moved to Stage 2 or Stage 3 solely due to the effects of the pandemic, and apply an override where it is appropriate to reverse this.  For smaller portfolios, firms could perform an assessment at an exposure level to review and override stage allocation. However, for large portfolios, firms will need to find a systematic way of performing this review that is both practical and supportable.

For large portfolios, firms will need to identify segments that are high risk and that need to be reviewed. One of the ways of identifying high risk segments is using machine learning or decision trees. However, unless firms have existing tools they can leverage, sophisticated solutions like these are not viable in the current environment. A simpler analysis would be to identify the reason for the change in stage allocation. If the reason for the change is exclusively due to the effect of the pandemic, an override needs to be applied. This could for example be done by analysing internal customer behaviour in the few months leading up to the crisis, perhaps supported by qualitative rationale at a customer/industry segment level. External data sources, which are available in the market, can also be leveraged to inform this assessment, especially those that collect customer behaviour and spending across banks.

Another point of consideration is whether the data collected over this time forms a complete picture of a customer’s circumstances. For example, many customer may give up trying to get through to a banks call centre to ask for a payment holiday, and some may not even try. These customers may take enforced payment holidays which would therefore normally trigger SICR as their account hits the 30 days in arrears backstop. Therefore, in the absence of further analysis to assess the customer’s circumstance prior to the pandemic, the exposure would be deemed as SICR which may not be appropriate.

Lastly, the PRA’s guidance on stage allocation is likely to result in an increased use of post model adjustments (PMAs) and overrides. Quantification of PMAs or overrides could be made using alternative modelling techniques. We discuss below the use of a simulation-based model to provide reasonable benchmarks quickly for non-retail portfolios. Nonetheless, in order to be able to monitor the use of overrides as the pandemic plays out, firms will need to ensure that they have appropriate model performance monitoring frameworks in place. Firm’s model risk management frameworks should also provide for an appropriate level of oversight and independent challenge of the PMA calculation process, in a similar manner to that applied to the core IFRS 9 models.

Modelling ECL: taking account of monetary stimulus, using reliable forward-looking information and ensuring the shock is short lived

In both its statements, the PRA highlights that forward-looking information should take account of the unprecedented economic support measures already announced. It also advises firms to only use reliable and supportable forward-looking information, which it concedes is very challenging to prepare currently. Finally, it expects firms to use forward-looking information that reflects the anticipated short-term nature of the shock. This is easier said than done however, and many firms will find it challenging to produce ECL numbers that adhere to all these requirements.

One way of overcoming the challenge of incorporating “all current information” in loss estimation particularly for non-retail portfolios is to use a market implied default risk measure, such as expected default frequency (EDF). By their nature, EDFs will already capture all the relevant information that is available at a specific point in time. For this reason they are attractive at the moment because they will already have all expectations of the negative effect of the pandemic on business cash flow together with the positive impacts of central bank and government related stimulus priced in.

There are various sources of EDF data in the market. One example is the Credit Research Initiative (CRI) established by the National University of Singapore, which provides publically available EDFs. As an example, Figure 1 below shows average the EDF for a basket of companies that operate in cyclical industries in the UK. It is clear that the default risk for these obligors has spiked in the recent weeks, highlighting the effect of the crisis, but crucially this should also take into account the financial support measures announced recently by the Bank of England.

Figure 1: Credit Research Initiative (National University of Singapore) EDF for the consumer cyclical sector in the UK

The next challenge for firms is forecasting losses using reliable economic forecasts. When using EDF data, a credit cycle index (CCI) can be constructed that can be used to calculate a PD term structure (a forecast of the path of PD over a specified time horizon). A Monte Carlo simulation-based approach can be used to simulate the various paths of CCI over a forecast horizon. When applied within an ECL framework, the result is a loss distribution from which the unbiased expectation can be determined. Whilst many firms will have adopted a discrete scenario approach using three to five defined scenarios, a simulation-based approach can also be used to inform ECL estimates. This approach has the advantage that is it very quick to implement and does not require the development of new economic scenarios or associated probability weights. Our view is that a simulation approach is therefore a powerful way to triangulate the output of a scenario based ECL estimate, and potentially offer guidance as to the scenario weightings or level of manual overlay that should be applied due to COVID-19.

Lastly, it can be challenging to build a simulation-based model that is both accurate and reflects the short-term nature of the shock. This is due to the fact the pandemic itself and its economic impact is unprecedented. Economic systems usually observe protracted recovery periods following times of stress and therefore accurate models will reflect this trend. As discussed in a separate article, we have done our own analysis using the EDF time series in Figure 1. Using an autoregressive model with parameters to account for seasonality leads to the best overall fit. However, this model predicts that default risk and hence ECL will take approximately 3 years to stabilise to average historic levels and does not reflect the short-term impact and anticipated rapid rebound of the economy. Where a time series forecast does not intuitively reflect the guidance provided by the PRA about the timing and pace of the recover, it would be sensible to apply manual adjustment to the model parameters before the output can be used for impairment benchmarking purposes.

Modelling IFRS 9 ECL in a post-COVID-19 pandemic world

Once the pandemic has passed, the landscape within which IFRS 9 ECL is calculated may have permanently changed and there are short-term effects that firms need to consider.

Firms will need to think about how their models will be recalibrated once the new loss and default data is available in the aftermath of the pandemic. For example, there will be a period of months in which clients with more than 30 days in arrears do not move to Stage 2 and clients with more than 90 days in arrears are not considered a default. In addition, we are likely to see an unusual impact on asset prices from the rare combination of severely reduced cash flow for many businesses alongside government support.

Modelling choices will need to be made based on which of these effects to include and which to ignore. Whatever choices are made, new types of data will need to be captured. For example, changes to the way forbearance or loan covenants are treated will need to be captured accurately. This data will allow the effects to be either included or stripped out in the future, based on the objectives of the model being developed.

The future trends that will be observed in the loss and default data will not be directly caused by poor economic conditions in a conventional sense. Rather, the economy will have reacted to an external shock event (primarily a drop in consumption and the forced closure of many businesses for a period of time) that caused an increase in defaults and losses. Therefore firms will need to be cautious when recalibrating impairment models using the new data to avoid models producing spurious results. Expert judgement is likely to play an important role in the recalibration process, while firms will need to have a robust model performance monitoring framework in place to fully understand the impacts.

In summary, lenders are required to manage the impact of COVID-19 across their business, including profitability and balance sheet volatility. Firms need to carefully consider the guidance provided by the PRA in the context of IFRS9 ECL estimation. In the short term, they need to think of practical ways to override stage allocation to reflect as accurately as possible borrowers’ individual circumstances. Firms that use a discrete scenario-based approach may benefit from considering alternative approaches to informing loss estimates, such as a simulation-based model, as this may have a number of advantages now as well as in any future period of economic uncertainty. Finally, as new types of data become necessary to meet future modelling objectives, it is important that systems and processes are modified to ensure all relevant data is captured and stored.