Discussion Paper on machine learning for IRB models

In the recent years, Big Data has emerged as a result of the increase in data availability and storing capacity. Machine Learning (ML) models use this data as fuel that provides the necessary information for developing and improving features and pattern recognition capabilities. The European Banking Authority (EBA) published a report in 2020 of big data and advanced analytics, including ML. In addition, in June 2021, the EBA published a report on the current RegTech landscape in the EU. Furthermore, the EBA has published a Discussion Paper on ML for Internal Ratings-based (IRB) models.

Discussion Paper on Machine Learning for IRB models

Watch video

Executive summary

The EBA has published a Discussion Paper on the application of ML techniques in the construction and validation of models used in the IRB approach. The paper aims at analyzing the relevance of possible obstacles to the implementation of ML techniques in the IRB approach, based on some practical issues. The paper also includes the challenges and potential benefits of using ML, and sets out some principles and recommendations.

Main content

This Technical Note summarizes the main aspects of the Discussion Paper:

• Definition, learning paradigms and current use of ML in credit risk modelling. ML techniques are characterized by a high number of parameters and, therefore, require a large volume of data for their estimation, so that the models are able to reflect non-linear relations between the variables. Beyond this general definition, several learning paradigms may be used to train models.
• Challenges and potential benefits of ML techniques. The application of ML techniques poses a number of challenges and potential benefits. These depend on the context of their use, the complexity and interpretability of some ML models. The use of these techniques poses some challenges related to corporate governance, the implementation process, the categorization of model changes, or the use of unstructured data.
• Prudent use of ML techniques. ML techniques are more complex than traditional techniques and sometimes less transparent. Specific recommendations for these models include appropriate knowledge and application of model interpretability techniques, trying to keep the complexity for their use low, and applying appropriate model validation techniques.

Management Solutions has an expert working group that supports its clients in the development and implementation of their Machine Learning projects with focus on interpretability of outcomes and business integration - in each of the 6 defined lines of activity: i)data & IT Infrastructure; ii) modelling; iii) validation; iv) interpretability; v) regulation; and vi) others.

Download the technical note by clicking here.