The application of machine learning and challenger models in IRB Credit Risk modelling


The application of machine learning and challenger models in IRB Credit Risk modelling

The use in model estimation and risk driver selection

Machine Learning techniques show great potential to improve the IRB Credit Risk model landscape. Despite its many advantages, there is currently still limited use of machine learning in IRB modelling. This blog describes how machine learning based challenger models can be used to support and improve IRB models.

Challenger models for model estimation

The recent surge in data availability and storing capacity, combined with increased computing power, creates the opportunity for machine learning to be applied in credit risk modelling.

The main challenge is that machine learning (hereafter “ML”) models are more complex, making their results less transparent to interpret, justify and explain to management functions and supervisors. Therefore, the incorporation of ML models in the internal ratings-based (hereafter: “IRB”) model landscape has been limited.

As identified in the EBA ML discussion paper, one of the use cases for ML in IRB modelling are challenger models. Challenger models are models applied in parallel to traditional models, to benchmark model performance, explore alternative modelling assumptions and identify data patterns that may not be captured by traditional models.
This blog series aims to provide insights on how machine learning can be incorporated as challenger models in the context of IRB modelling and model estimation.

Challenger Models – Model Estimation

Download: Challenger Models - Risk Driver Selection

This blog is part of a blog series around the use of challenger models in Credit Risk modelling. In our previous blog we described the use of Machine Learning based challenger models in the risk driver selection process. Download the report to find out more!

Challenger Models – Risk Driver Selection
Did you find this useful?