Validation of default probability models: A stress testing approach

This study aims to evaluate the techniques used for the validation of default probability (DP) models. By generating simulated stress data, we build ideal conditions to assess the adequacy of the metrics in different stress scenarios. In addition, we empirically analyze the evaluation metrics using...

Full description

Saved in:
Bibliographic Details
Published in:International review of financial analysis 2016-10, Vol.47, p.70-85
Main Authors: Tsukahara, Fábio Yasuhiro, Kimura, Herbert, Sobreiro, Vinicius Amorim, Zambrano, Juan Carlos Arismendi
Format: Article
Language:English
Subjects:
Banking
Credit risk
Default probability
Entropy
Portfolio
Probability
Stock market delistings
Stress measurement
Validation studies
Validation techniques
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study aims to evaluate the techniques used for the validation of default probability (DP) models. By generating simulated stress data, we build ideal conditions to assess the adequacy of the metrics in different stress scenarios. In addition, we empirically analyze the evaluation metrics using the information on 30,686 delisted US public companies as a proxy of default. Using simulated data, we find that entropy based metrics such as measure M are more sensitive to changes in the characteristics of distributions of credit scores. The empirical sub-samples stress test data show that AUROC is the metric most sensitive to changes in market conditions, being followed by measure M. Our results can help risk managers to make rapid decisions regarding the validation of risk models in different scenarios. •We evaluate, using simulated data, stress tests aiming to support the decision-making of managers regarding the selection process of credit risk models.•We also empirically analyze the default probability validation metrics, using delisting information as a proxy for default, based on financial data of 30,686 public US firms.•We find that entropy based metrics are more sensitive to changes in the characteristics of the distributions of credit scores.
ISSN:1057-5219
1873-8079
DOI:10.1016/j.irfa.2016.06.007