Conditional versus Unconditional Models for VaR Measurement

Abstract

Accurate risk prediction plays an increasing role in effective risk management process. Nowadays, the most common measure of risk is the Value at Risk (VaR) but there are many ways of its measurement. In this paper unconditional and conditional models are applied and compared using backtesting procedures. We examine several models like Gaussian, stable, NIG and generalized Pareto distributions as an unconditional models and GARCH, EWMA and GARCH-EVT models as a conditional approach. Results are presented for long and short position applying 5 world indexes, 4 exchange rates and 2 metals. Conditional models perform better than unconditional models. Especially the GARCH-EVT model enables to estimate the VaR correctly regardless of considered assets.