Report and Webinar: Comparison of Traditional Modeling Techniques and Machine Learning for Prediction of LGD

olofNews, Reg&Tech Solutions, Risk & Analytics

 FCG’s modeling experts Thomas Aldheimer, Irina Cirmizi, Jonas Ljungqvist and Mats Ehnbom worked with Global Credit Data’s huge database of Basel compliant defaulted bank loans to explore different model building techniques.  The modeling work was advised by a group of expert banking practitioners forming a working group of GCD and used GCD standard definitions to develop a range of models, with a focus on Machine Learning.

The final report, available below, goes into detail on building 4 different LGD models, including a cure/non cure step.  There are detailed descriptions of the data selection, the drivers chosen and the methods used.

In addition, FCG’s Philip Winckle (currently senior advisor to FCG and former Executive Director of Global Credit data and formerly Head of Group Risk at SEB) and Thomas Aldheimer walked participants through the highlights of the report in a webinar on Tuesday 16th, see the session here. The webinar consisted of:

  1. Exploring use of GCD pooled data for building forward looking LGD models
  2. Comparing drivers observed in the data to those normally used in the industry
  3. Exploring the additional benefits of using Machine Learning to enhance model predictiveness

The report, the workshop and the slides of the webinar will benefit both hands on modelers and model managers as well as people working with credit risk data who want to see how it is used in practice.

Please do not hesitate to get in touch if you want to discuss the contents of the report or credit risk modeling in general. We also support in GCD related matters, for which you are welcome to contact Partner Jimmi Brink.

Download the report by submitting the form below: