The aim of the discussion paper is to set supervisory expectations on how new sophisticated machine learning models can coexist with and adhere to the Capital Requirements Regulation (CRR) when used in the context of IRB models. The discussion paper seeks stakeholders’ feedback on many practical aspects related to the use of machine learning in the context of IRB with the aim of providing clarity on supervisory expectations on their use. The consultation runs until 11 February 2022.

Machine learning techniques have been increasingly used in banks for business purposes. However their use remains limited the context of IRB models to calculate regulatory capital for credit risk. This is due to the complex nature of such models, which leads to challenges in interpreting and justifying their results to supervisors.

This discussion paper is a first step to engage the industry and the supervisory community to investigate the possible use of machine learning in the context of IRB models and to build up a common understanding of their general aspects, and the related challenges and opportunities in complying with the prudential requirements.

The paper identifies the challenges and the opportunities which institutions may face in using machine learning to develop CRR-compliant IRB models. The paper discusses the main concerns stemming from this analysis of the CRR requirements and provides a set of principle-based recommendations that would ensure a prudent use of machine learning models in the context of the IRB framework.
Consultation process

Comments to this consultation can be sent to the EBA by clicking on the "send your comments" button on the consultation page. Please note that the deadline for the submission of comments is 11 February 2022. All contributions received will be published following the end of the consultation, unless requested otherwise.

A public hearing will take place virtually in the form of a webinar on 16 December 2021 from 14:00 to 16:00 CET. The EBA invites the stakeholders to register using this link by 14 December 16:00 CET.