Machine learning in the context of IRB internal models – upcoming EBA guidance

On November 11, the European Banking Authority (EBA) issued for consultation a paper on the use of machine learning (ML) in the context of IRB internal models. These models are used by banks to determine capital requirements for credit risk.

The EBA study aims to identify the main challenges and possible benefits of new ML-based IRB models. These models have an internal function in banks to maintain credit risk precautions. The second objective of the study is to provide essential recommendations that should ensure the proper use of ML models by banks in the future. Then their use would be in line with the regulatory requirements of the Capital Requirements Regulation (CRR).

Thus, the paper discusses the general scope of ML usage in the context of internal models and the limitations that currently exist. It also highlights the challenges and advantages of using this technology to develop IRB models. The most important part of the paper, however, are the recommendations, in which the EBA points out the need to:

• ensure an adequate level of knowledge of the model’s operation for all relevant individuals,
• avoid unnecessary complexity of the model,
• ensure an appropriate level of understanding of the model,
• monitoring, documenting, and validating ML models and keeping them up to date, in particular for ML models with limited “explainability” or that are frequently updated.

The above development will now be subject to appropriate consultation. Ultimately, these guidelines will have an overwhelming impact on the application of machine learning techniques in the context of IRB internal models. They will probably also influence the actions and position of local market regulators, including the Financial Supervision Authority.

Do you want to be up-to-date? Subscribe to our newsletter!