EBA’s principle for COVID impacted data for IRB models
1. EBA’s principle for COVID-19 impacted data for IRB models
The COVID-19 pandemic has impacted the global economies, including Europe, in big ways. The European Banking Authority (EBA) has presented a draft version of the principles using data from this period for IRB-models, and these principles will be included in the supervisory handbook at the end of 2022. The clarifications from EBA are not new guidelines (GL) or regulations but specification on EBA’s view on how current GL affect COVID-19 data and IRB models.
The reason for the need to specify their view regarding data from this period is the economic chock from the pandemic and the extraordinary effort from authorities to mitigate negative economic impact and consequently the data might be non-representative. These EBA clarifications will impact model validation, recalibration, and new model development.
Four core principles are presented from EBA; (1) GL on PD estimation, LGD estimation and the treatment of defaulted exposures (EBA/GL/2017/16) should apply also for the pandemic period, (2) significant decrease in risk weights and model parameters compared to pre-pandemic needs to be investigated in detail to ensure that data is representative before used in any types of models, (3) institutions should be cautious to use default data and loss rates observed during the period to recalibrate IRB models, especially if defaults and loss rates have improved significantly, not to lower long-run average default rates until it is certain that COVID-19 support measures don’t impact the performance. Finally, (4) recalibrations of downturn LGD models should not be done until there is a high level of certainty that the materialized losses from defaults during the pandemic has fully materialized.
An overall theme for all four principles is that institutions should be careful to draw statistical or economical conclusions from data during the pandemic at this time, and secondly to investigate the impact of COVID-19 support measures and details of the underlying drivers if there has been a significant change to data during the pandemic period compared to pre-pandemic. EBA emphasizes on recommendations regarding data that is non-representative and/or indicates significant decreases in risk weights. EBA GL on PD and LGD estimation apply for institution that have experienced significant increases in risk weights during the pandemic and/or default frequencies.
1.1 Guidelines regarding PD and LGD should apply
The first principles specify that all requirements from previous guidelines for data used in statistical models applies also for the pandemic period. If the data from the period is non-representative, then this data should not be excluded and appropriate adjustments to Margin of Conservatism (MoC) should be applied. The impacts of the pandemic and economic countering measures should be included in the review of IRB models. To assess non-representativeness in data, a selection of the topics needed to be assessed are:
- Application data – variable distributions of customers that applies for credit during different periods
- Definition of Default – have any changes in definition of default impacted the default rate?
- Detailed analysis of the change in risk parameters – If there has been a significant change in risk parameter distributions, is that due to COVID-19 or idiosyncratic risk changes?
- Lending standards and recovery policies – have COVID-19 affected lending standards and/or recovery policies that have changed the risk parameters or default rates, such as moratoria?
1.2 Significant decrease in risk weights and parameters for current models and model development
For institutions that have observed a significant increase in quality of IRB risk parameters (and a direct consequence also on risk weights) compared to 2019 should make detailed analysis if this change is a direct or indirect effect of the pandemic. If the input variables to the IRB models have changes significantly, it should be ensured that the differentiation power remains and reasonable economic reasons are maintained in the current model(s), if not, procedures defined by EBA needs to be followed. For cases where the reason for improvements in risk parameters are not clear and/or the differentiation power has significantly decreased, a detailed analysis should be carried out before any new model development is initiated. However, if institutions can determine that reasons for decreases in risk weights are due to changes in idiosyncratic risk, no further measures are needed regarding this topic.
1.3 Model recalibration
Before deciding to recalibrate IRB models, including recent data, should extra attention be paid to if data has change significantly in a similar way as for model development, and if changes may be impacted from economic measures including moratoria. To be able to decrease the Long-Run Average Default Rate (LRADR), EBA requires a significant analysis showing that the lower default rates are not a direct or an indirect consequence of COVID-19 support measures. The overall recommendation from EBA is to postpone any recalibrations using data from the pandemic period to a lower LRADR at this point. However, EBA provides clarifications in GL for institution that must update a model for some specific reason.
1.4 Postpone any downturn LGD recalibrations until there is a great certainty that losses related to COVID-19 has materialized
The main point of the fourth principle is to not update downturn LGD models and draw conclusion from realized loss data from defaults during the pandemic yet. Institutions should postpone updating downturn LGD models until there is a high level of certainty that these defaults have materialized, and all relevant loss data is available for analysis before making a recalibration of downturn LGD models.
2. Challenges for institutions
A pandemic is such an extraordinary event and the economic policies from central banks and governments have been unprecedented, therefore data from this period cannot be used without extra caution. These changes in the macro environment may have impacted the quality of the IRB models, risk parameter distributions may have changed significantly during the period, and some data may be non-representative.
A natural response to these problems might be to recalibrate the models with the most recent data and EBA has thus identified the need to provide restrictions and recommendation on this subject should the data be non-representative and/or indicate significant decreases in risk weights. Institutions should have patience and observe if the risk parameters trends back to the old averages or continue to show decreased risk according to historical calibrations.
FCG has in-depth experience and knowledge in all IRB areas and have worked with a majority of Nordic IRB banks. Please feel free to contact us for a discussion regarding the impact of data from the COVID-19 period.
Director & co-Head of Team Risk & Finance
Magnús Ólafur Sigurđsson