The Use of Machine Learning Techniques in Internal Models

Introduction

On 28th July 2025, The European Central Bank (ECB) has published a revised version of its Guide to Internal Models¹, incorporating significant updates to reflect changes in the regulatory framework under CRR3 and the revised Basel standards. Building on the experience gained since the guide’s initial publication in 2019, this updated edition provides enhanced transparency which also allows for harmonisation in the supervision of internal models used for credit risk, market risk, and counterparty credit risk. 

A key enhancement in this revision is the dedicated section on The Use of Machine Learning (ML) Techniques in Internal Models. Recognising the increasing adoption of ML in banking, the ECB sets out clear supervisory expectations to ensure that such models are:

• Developed using robust, replicable methodologies;
• Governed by well-defined internal policies specifying scope, limitations, and alignment with risk management and governance frameworks;
• Subject to appropriate levels of human oversight, explainability, and documentation, especially in relation to model outputs and override practices.

These principles aim to ensure that the use of ML does not compromise transparency, model reliability, or regulatory compliance. The guide also clarifies the need for institutions to justify the complexity of ML-based models with demonstrable performance improvements and to mitigate risks associated with their integration into credit decisioning, capital allocation, and stress testing processes.

The updated guide supports the simplification of model landscapes while reinforcing prudent model risk management practices. It has been informed by industry feedback and collaborative input from national competent authorities, ensuring that it remains a practical and forward-looking reference for both banks and supervisors.

Access the full publication: The Use of Machine Learning Techniques in Internal Models [ 2661 kb ]

Authors:

Andreas Spyrides, Quantitative Risk Services Leader

Kimia Mirsalehi, Consultant, Quantitative Risk