Journal of Credit Risk

Risk.net

Calibration and mapping of credit scores by riding the cumulative accuracy profile

Marco van der Burgt

  • This paper introduces a new algorithm for mapping credit scores to ratings, based on step-wise partitioning of the cumulative accuracy profile.
  • Testing the algorithm by simulating two different scoring models reveals that the algorithm generates significantly different rating grades and a monotonous default probability scale.
  • The algorithm also shows that rating calibration is a trade-off between optimizing discriminatory power and avoiding concentrations in rating grades.
  • The algorithm produces a number of rating classes, which behaves as a power-law of the Accuracy Ratio of the credit scores and the unconditional probability of default.
     

A lot of literature on credit risk scoring techniques exists, but less research is available regarding the mapping of credit scores to ratings and the calibration of ratings. This paper introduces an algorithm for mapping credit scores to credit ratings and estimating a probability of default (PD) per rating grade. The algorithm is based on stepwise partitioning of the cumulative accuracy profile, such that requirements like stable ratings and a monotonous PD scale, as stated by the European Banking Association’s regulatory technical standards, are fulfilled. We test the algorithm by simulating different PD models and score distributions. These tests reveal that the algorithm maps credit scores to significantly different rating grades. Each rating cor- responds to a PD, which is a monotonous function of the rating grade. The tests also show that the total number of rating grades, which result from the mapping algorithm, strongly depends on the ability of the scoring model to discriminate between defaulting and nondefaulting counterparties.

To continue reading...

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here: