October 2022

The EBA’s new Principles aim to protect IRB risk estimates, which rely on historical data, from distortion by pandemic-era economic support measures. The new rules are already in force, and supervisors are probing compliance. Banks need to address the implementation challenge now.

The COVID-19 pandemic saw European governments launch unprecedented economic rescue packages comprising direct financial relief such as furlough schemes, credit payment moratoria, and public guarantee schemes. These programmes were in place for much of 2020 and most or all of 2021. Some continue to operate.

As a result, observed default and loss rates give little indication of a downturn in credit quality during the pandemic. Now however, as noted by the European Supervisory Authorities,1 a likely recession is anticipated - due to slowing growth, higher inflation and rising energy prices - and exposures with COVID support measures could be especially vulnerable2. For example, applying repayment moratoria does not automatically mean that forbearance is needed or that a loan is non-performing, but it could signal an increased credit risk that may not be reflected in NPL ratios.

In short, supervisors are concerned that COVID-19 responses may have artificially lowered default and loss rates, or rendered risk estimates too optimistic. Pandemic measures could have distorted IRB-relevant data:

  • Directly, by altering credit processes - such as not treating forbearance as a trigger of default
  • Indirectly, by distorting IRB model inputs - such as when government grants enhance borrowers’ liquidity, leading to a potential temporary improvement in their credit score

In response, June 2022 saw the EBA publish four new Principles aimed at ensuring that IRB model data remains representative. This six-page document, issued without consultation, has important implications for European banks’ prudential IRB risk modelling and is also relevant for IFRS9 credit modelling.

The principles will be included in the EBA’s soon to be finalised supervisory handbook (expected at the end of 2022), but they apply immediately. We observe that supervisory authorities are already asking Significant Institutions and smaller banks how they are applying these principles to their risk estimate review frameworks – including the annual monitoring and validation of IRB models.

What do the new principles say?

The first principle focuses on applicable guidance. It states that banks are expected to follow the EBA’s Guidelines on PD and LGD while implementing the principles for IRB modelling data. Among other things, that includes:

  • Analysing the representativeness of data
  • Factoring all defaults into risk quantification
  • Applying Appropriate Adjustment (AA) and Margin of Conservatism (MoC) to non-representative data
  • Conducting the annual review of risk estimates

The second principle is concerned with IRB model-based risk estimates. It requires banks to identify and analyse any significant decreases in average risk weights or expected losses, compared to end-2019, arising from lower average IRB risk parameter estimates including PD, LGD, ELBE and CCF3. Depending on the extent of any decrease and the materiality of affected portfolios, banks may need to:

  • Perform enhanced analysis of representativeness
  • Recalibrate or redevelop models
  • Adjust abnormal values to make them more conservative, or apply treatment for missing values

The third principle focuses on observed default and loss rates affected by the pandemic. Where there are indications of non-representativeness, banks should verify whether the cause of any decrease arises from COVID-related measures. Where the decrease in long-run average default rates (LRADR) is driven with COVID-related measures, banks should avoid recalibrating models to these lower values until more recent default information is available and they can be confident that lower default rates are sustainable. Once again, they may also need to recalibrate model assumptions or apply a MoC.

The fourth principle is concerned with downturn LGD. As a period of crisis, the pandemic will probably need to be reflected in downturn LGD calculations. The EBA recommends that recalibration of downturn LGD should be postponed until the effects of the pandemic have fully materialized in observed loss rates. Until then, the MoC should be used to address any data shortfall, making up for the absence of a correctly calculated downturn LGD.

How are these principles relevant for IRB models lifecycle?

Banks need to include the new COVID-specific principles into their ongoing review of estimates, including model monitoring and validation.

That includes conducting an ad-hoc impact analysis of all IRB models, to identify those with a significant drop in risk parameters and which are also material for the institution. Periodic model validation exercises will need to change too, ensuring that model monitoring and development reflect the new principles. Lastly, the new principles will require banks to redevelop or recalibrate impacted models, introducing adequate AAs and MoCs where needed.

What challenges are expected?

For banks, the greatest practical challenges are likely to centre around data. Locating and verifying all relevant data on COVID-19 response measures is not trivial. Separating the impact of COVID-19 measures from changes arising from other factors – such as reconstructed Definition of Default triggers – will be complex and subjective.

Judgemental estimates about the impact of economic relief measures will also be needed, and ad-hoc changes such as MoC made where data is judged unrepresentative. In addition to altering systematic monitoring and validation processes, a further challenge may arise from the need to plan and execute ad-hoc analyses focusing on the unique features of the COVID-19 environment.

What does this mean for banks?

Most larger banks have made a start on data identification and the adjustment of their model monitoring processes, but there is a long way to go. Smaller institutions face less complexity, but typically have fewer in-house capabilities to call on. Every institution will need to ensure it has sufficient resources and skills available across teams ranging from data to model validation, covering both IRB and IFRS9, and backed up by appropriate management and governance.

It will take a substantial effort for banks to fully integrate the new principles into their credit risk modelling activities. This, and the need for immediate implementation, means that work should start now.

Observed default rates


1. Link: ESAs warn of rising risks amid a deteriorating economic outlook | European Banking Authority (europa.eu) 

2. As also earlier in 2020 noted by EBA (link: Earlier assessment by EBA: For publication - Thematic note on moratoria and public guarantees.pdf (europa.eu)), the disruption caused in some sectors by the COVID-19 pandemic and the confinement measures will inevitably drive the default rates higher [after 2020].

3. Probability of Default, Loss Given Default, Expected Loss Best Estimate, Credit Conversion Factor