The Analytics of Risk Model Validation
A collection of papers in this new and developing area of research, this book focuses on the quantitative side of model validation and covers model validation in credit, market, and operational risk.
TABLE OF CONTENTS The Analytics of Risk Model Validation
Chapter 4: A Moments-Based Procedure for Evaluating Risk Forecasting Models
Kevin Dowd [*]
Abstract
This chapter examines the important problem of evaluating a risk forecasting model [e.g. a values-at-risk (VaR) model]. Its point of departure is the likelihood ratio (LR) test applied to data that have gone through Probability Integral Transform and Berkowitz transformations to become standard normal under the null hypothesis of model adequacy. However, the LR test is poor at detecting model inadequacy that manifests itself in the transformed data being skewed or fat-tailed. To remedy this problem, the chapter proposes a new procedure that combines tests of the predictions of the first four moments of the transformed data into a single omnibus test. Simulation results suggest that this omnibus test has considerable power and is much more robust than the LR test in the face of model misspecification. It is also easy to implement and does not require any sophistication on the part of the model evaluator. The chapter also includes a table giving the test bounds for various sample sizes, which enables a user to implement the test without having to simulate the bounds themselves.
[*] Nottingham University Business School, Jubilee Campus, Nottingham, UK
1. Introduction
There has been a lot of interest in the last few years in risk forecasting (or probability-density forecasting) models. Such models are widely used by financial institutions to forecast their trading, investment and other financial risks, and their risk forecasts are often a critical factor in firms risk management decisions. Given the scale of their reliance on them, it...
Copyright Elsevier Ltd. 2008 under license agreement with Books24x7
