Abstract
Abstract : This study investigates the performance of different backtesting techniques at evaluating market risk models with different conditional distributions. The study classifies existing backtesting techniques into two groups: Traditional backtesting techniques (Independence, unconditional Test, Conditional Coverage Test) that look only at the likelihood of the occurrence of a violation; and Modern backtesting techniques (Duration-Based Test, Dynamic Quantile Test, Dynamic Binary Test) that look either at the time elapsed between two consecutive violations and the relationship between violations and past violations. To achieve this, the study builds different types of conditional and unconditional market risk models. Unconditional market risk models include the historical simulation and variance covariance methods. The conditional market risk models are built by making use of eGARCH processes with different types of conditional distribution (asymmetric and extreme value distributions). The empirical analysis is based on daily return series of the following stock markets obtained from Bloomberg: - S&P500, FTSE 100, Africa All Share Index, and Nikkei 225. The sample period spans from 2006/01/02 to 2018/09/10. This sample period is then divided into two overlapping subsamples representing financial crisis, and tranquil period respectively. The results suggest that traditional backtesting techniques perform better at evaluating the different types of market risk models for both financial crisis and tranquil periods. However, the modern backtesting techniques perform well during a financial crisis and give misleading results during tranquil period. The finding of this study suggest that for an out-sample data of 250 days1, the best backtesting techniques to evaluate a model is the traditional backtesting techniques. Given the findings of this study, regulators and other decision makers can inform financial institutions operating in their respective jurisdictions to be cautious when using modern backtesting techniques in the process of evaluating market risk models for the computation of the regulatory minimum capital requirement.
M.Com. (Financial Economics)