The impact of Covid-19 on expected credit loss of retail portfolios — the case of moratoriums

Most countries have taken multiple tiers of measures to relieve their citizens from the sudden and sharp fall in economic activity. Among them was the implementation of a moratorium on payments for loans held by individuals and businesses.

The European Banking Authority, to its credit, demonstrated quick reflexes, issuing a statement and then guidance on default recognition eligibility and on the specification of Significant Increase in Credit Risk (SICR) in the case of a payments moratorium.

The payments moratorium will nonetheless affect all the components of the expected credit loss (ECL) function, in a variety of ways — and therefore the IFRS 9 risk component models that produce it.

Most retail PD and lifetime PD (LPD) estimates are based on behaviour scores or behaviour-related attributes. If a business-as-usual classification of exposures continues during the moratorium, the following dynamics are expected:

• Delinquency bucket based predictors will present an artificially improved picture, with no delays recorded in the behaviour period.
• Payment counters and balance-change related predictors, on the other hand, will suggest an increase in risk levels.

Given that bucket-based criteria are usually the predominant drivers in risk classification, the overall expectation is a concentration of exposures — because their characteristics become static — and a shift towards the best PD pools, because of the higher importance of bucket-related predictors. The higher the participation rate in the moratorium, the more intense these dynamics are expected to be. In fact, these effects are expected to continue in the first months following the end of the moratorium, until payment behaviour history is restored.

Beyond doubt, those patterns do not represent the true status of the portfolio. The overall risk level of the portfolio is not reducing, the classification is misleading, and the relationship between PD pools and PD is not yet known.

Though each situation may be different, the general objective is to perform a classification that considers the credit risk profile of the client and the level of exposure to the risks of the current economic environment over the medium term — that is, when the moratorium is expected to end. This could be achieved by combining:

• A frozen version of the PD pools, or behaviour scores, from before the impact of Covid-19, representing the credit risk profile;
• With the forecast impact of Covid-19 by sector or type of activity.

This approach introduces a forward-looking aspect into the PD estimates and may have various levels of complexity: in the way the combination takes place (score level, overlay and so on), in the level of judgemental decision, and in the use of relevant historical data, fresh data as it becomes available, or relationships identified in economic sectors in other countries and generalised.

The cure rate component of the loss given default (LGD) estimate is where the main interest needs to be placed, as it is more sensitive to economic conditions, and most remedial strategies focus on increasing cure rates.

Cure rates are expected to be affected in the following ways:

• Exposures in default for less than 12 months (new defaults), for which a higher cure rate is usually expected, will be affected the most. For these exposures the ability to make payments and become current is greatly impaired by distressed economic conditions, so the 12-month cure rate should be readjusted downwards. If a macro layer exists for this component, it could be employed considering the macroeconomic projections for GDP and other macroeconomic variables; if not, stressing the estimate is recommended. However, depending on the shape of the economic recovery, cure rates in the subsequent period may be significantly higher than any past observation. As such a pattern may not be available in any training data used to produce the estimate, the cure rate in 2021 for months in default 13–24 may exceed any pre-existing macro-based model estimate.
• For exposures currently 13 months or more in default, expected cure rates should also be adjusted downwards, while expectations of the cure rate bouncing back in 2021 (that is, at 25+ months in default by then) are low.

Exposures not in default (stage 1 and stage 2) may be classified in two categories:

• For exposures eligible for or already in the moratorium, any default event will occur after the moratorium, with the exception of default triggered by unlikeliness to pay (UTP). The macroeconomic projections for this period (2021) are in most cases favourable, so to a degree the existing or similar cure estimates may be close to reality.
• For exposures not eligible for the moratorium, the expected pattern in the event of default should be similar to that of already defaulted exposures.

On expected recoveries from the realisation of property, property indices should be adjusted according to scenarios for the type of recession that will affect the economy. Cash recoveries usually refer to a longer-term recovery period, so less impact is expected. It is also worth noting the need for an extension of probation for already forborne exposures that are eligible for the moratorium.

Exposure at default (EAD) estimates are expected to be affected only mildly by Covid-19 and the moratorium measures. The main dynamics that can be expected are:

• For fixed-term exposures, the prolongation of the term with capitalisation of interest should be reflected in the estimated EAD in the term structure.
• For revolving exposures, as long as limit monitoring, cash advances and authorisation tracking are in place, no impact is expected. It is worth noting, though, that in recessionary periods inactive revolving facilities may become active, with high default rates and high limit usage.

The first step in responding to the new reality that Covid-19 has introduced, and to the economic recession expected for 2020, is to understand and predict how portfolios will be affected. Only then can decisions be informed and appropriate actions taken to manage portfolios through these difficult circumstances.

To that end, adjustment or recalibration of the ECL component estimates will be required, probably together with a further level of granularity that better reflects expectations of a differently sized effect on particular segments. Any such change is a natural trigger for independent model validation.