VALUE AT RISK ESTIMATION USING GARCH-TYPE MODELS

This paper examines five GARCH-type models, including RiskMetrics in the Value at Risk estimation. These models are applied to an optimized internally diversified portfolio, comprised of three benchmark indexes from three dfferent countries (Romania, UK and USA) in order to asses the overall performance of the daily VaR estimates at various probability levels (1%, 2.5% and 5%). Study results indicate that all symmetric models outperform the asymmetric ones, both for normal and Student's t distributions. We also find that GARCH(1,1) underestimates 1% VaR in comparison to RiskMetrics and GARCH-t(1,1) that perform very well. Moreover, GARCH-t(1,1) gives better 2.5% VaR estimates and RiskMetrics outperforms GARCH(1,1) and GARCH-t(1,1) for 5% VaR estimates.