An overview and framework for PD backtesting and benchmarking

In order to manage model risk, financial institutions need to set up validation processes so as to monitor the quality of the models on an ongoing basis. Validation can be considered from both a quantitative and qualitative point of view. Backtesting and benchmarking are key quantitative validation tools, and the focus of this paper. In backtesting, the predicted risk measurements (PD (Probability of Default), LGD, EAD) will be contrasted with observed measurements using a workbench of available test statistics to evaluate the calibration, discrimination and stability of the model. A timely detection of reduced performance is crucial since it directly impacts profitability and risk management strategies. The aim of benchmarking is to compare internal risk measurements with external risk measurements so as to better gauge the quality of the internal rating system. This paper will focus on the quantitative PD validation process within a Basel II context. We will set forth a traffic light indicator approach that employs all relevant statistical tests to quantitatively validate the used PD model, and document this approach with a real-life case study. The set forth methodology and tests are the summary of the authors' statistical expertise and experience of world-wide observed business practices.