Approaches and Techniques to Validate Internal Model Results

Abstract

The development of risk model for managing portfolio of financial institutions and insurance companies require both from the regulatory and management points of view a strong validation of the quality of the results provided by internal risk models. In Solvency II for instance, regulators ask for independent validation reports from companies who apply for the approval of their internal models. Unfortunately, the usual statistical techniques do not work for the validation of risk models as we lack enough data to significantly test the results of the models. We will certainly never have enough data to statistically estimate the significance of the VaR at a probability of 1 over 200 years, which is the risk measure required by Solvency II. Instead, we need to develop various strategies to test the reasonableness of the model. In this paper, we review various ways, management and regulators can gain confidence in the quality of models. It all starts by ensuring a good calibration of the risk models and the dependencies between the various risk drivers. Then applying stress tests to the model and various empirical analysis, in particular the probability integral transform, we build a full and credible framework to validate risk models.