Probability of Default Model Validation

Abstract

In order for an institution to qualify for the Basel II internal ratings based (IRB) approach for measuring credit risk capital, a financial institution would need to have a robust validation framework in place. For that purpose, a model validation framework (together with two excel programs) had been developed by Oliver Wyman for OCBC Bank. In this project, we examine the applicability and implementation accuracy focusing on the two most important aspects of model validation: (1) measuring the discriminative power and (2) calibration accuracy. The Receiver Operator Characteristic (ROC) curve and its summary statistics, area under ROC (AUC) and accuracy ratio form the most important set of tools in evaluating the discriminative power of the models. The documentation as given in [1] and the implementation as provided by [4] has been found to be sound and accurate. The anchor point test and the curve calibration shape test form the two most important tests to evaluate the calibration accuracy of the models. Except for the failure to correctly compute the standard error of adjusted default rate as documented in section 4.1.6, the documentation and implementation as found in [1] and [5] respectively have been found to be sound and accurate.