In recent years, Big Data has emerged as a result of increased data availability and data storage capacity. Machine Learning (ML) models use this data as fuel that provides the necessary information for developing and improving features and pattern recognition capabilities. In June 2021, the European Banking Authority (EBA) published a report on the current RegTech landscape in the European Union (EU). In November of the same year, the EBA published a Discussion Paper on ML for Internal Ratings-based (IRB) models in order to gather feedback from the public.
Follow-up report on ML for IRB models
Executive summary
The EBA has published a follow-up Report summarising the main findings of the consultation on its ML discussion paper and providing an overview of current use cases of ML techniques for IRB models.
Main Content
This technical note provides an overview of the current use cases of machine learning techniques and analyzes how the use of these techniques in credit risk models interacts with two other legal frameworks: the General Data Protection Regulation (GDPR) and the future Artificial Intelligence Act (AI Act).
- Selective use for IRB models. The consultation findings show that, where applied, ML techniques are used for some steps of the IRB Approach only. In particular, results show that institutions predominantly focus on applying ML techniques to PD model development.
- Challenges in developing and validating IRB models using ML techniques. Developing and validating IRB models using ML techniques poses specific challenges that can be summarized into three categories:
- statistical issues
- human skill-related issues
- interpretability issues
- Compliance with GDPR and Artificial Intelligence Act regulatory frameworks. Decision-making on the use of ML techniques in credit risk models should include ethical and legal as well as consumer and data protection considerations in alignment with these two frameworks.
Download the technical note on Follow-up report on Machine Learning for IRB models.