Quantitative Risk

EBA: use of COVID-19 impacted data for internal credit risk models

Andreas Spyrides

EBA published, on the 21st of June 2022, draft principles in assessing representativeness of COVID-19-impacted data in re-calibration of IRB models. The Supervisory handbook that will be published later in 2022 will include these principles, aiming at ensuring a harmonised approach. COVID events and especially the use of moratoria and other public measures may have led to changes in default rates.

Principles for ensuring representativeness of the IRB-relevant data impacted by the crisis:

  1. The guidance laid down in the GL on PD and LGD should apply:
    • representativeness of data used for risk differentiation or risk quantification is to be analysed
    • non-comparability of the historical data underlying risk quantification should not lead to any data exclusions but should instead trigger an appropriate adjustment and increased margin of conservatism (MoC)
    • review of estimates should put specific emphasis on the potential impact of the crisis and the countering measures.
  2. Significantly decreased average risk weights or expected losses compared to end-2019 due to decreased average IRB risk parameter estimates should be duly analysed with respect to the representativeness of the sample used for model development to the current portfolio.
  3. Institutions should assess whether a re-calibration is necessary. Where there are indications of non-representativeness of those most recent realisations, institutions should postpone any recalibration to lower long-run averages including the most recent data until it is sufficiently certain that the trend of decreased realisations is sustainable and is not driven by the extraordinary COVID-19 support measures.
  4. Institutions should postpone any downward recalibration of downturn LGD including the most recent data until it is sufficiently certain that the effect of the COVID-19 pandemic has materialised in the observed loss rates. However, institutions should duly assess the loss data from defaults that occurred during the pandemic and apply appropriate margin of conservatism where there are indications that higher loss rates will be realised.