Simulation based approach in evaluation of alternative VaR models in the presence of ARCH effects

In this paper, we examine the performance of 8 different models in VaR forecasting via simulation approach. Artificial returns are obtained from true data generating processes devoted to replicating realistically FTSE100 daily returns. Distributions of percentage errors of VaR estimates are considered in order to compare the performance of risk models. It is concluded that there is no one method that consistently outperforms others at all confidence levels according to all criteria. Nevertheless, it is inferred that unconditional methods work substantially worse than conditional. The second aim of this paper is to assess the performance of the backtesting procedure used in evaluation of alternative methods. It is found that making a judgment about the risk model only on the basis of backtesting results is dubious, since there is a high chance that it admits for VaR forecasting the model that substantially over- or underestimates true VaR