How Grant Thornton can help you understand the challenges surrounding the implementation and validation of machine learning techniques in IRB models.

This publication will provide an overview of the current and future uses of Machine Learning in the world of Internal Rating Based (IRB) Models which allow banks to model their own inputs for calculating Risk Weighted Assets.

What are Internal Rating Based (IRB) Models?

Risk Weighted Assets (and as a result capital requirements) for credit risk can be calculated either using the IRB or Standardised Approach.  IRB models represent a more advanced and risk sensitive approach, breaking down into three component parts (PD, LGD, EAD):

  1. Probability of Default (PD): What is the chance that the borrower will default on a 1 year time horizon?
  2. Loss Given Default (LGD): How much will I get back (and therefore how much will I lose) in the event of default?
  3. Exposure at Default (EAD): If the borrower defaults, what will be the size of my exposure?

A Bank’s IRB approach can be classified either as Foundation or Advanced:

(1) Foundation IRB (FIRB) Approach:  Banks estimate PD and use regulatory prescribed estimates of LGD and EAD.

(2) Advanced IRB (AIRB) Approach:  Banks use their own estimates of PD, LGD, and EAD.