Are GARCH models necessary for Expected Shortfall?

Abstract: Following the Basel Committee on Banking Supervision’s decision to move from Value at Risk to Expected Shortfall, risk managers will have to alter their methods for reporting risk. This paper sheds light on the question of which volatility models and distributional assumptions that works best for this new method of risk measurement by evaluating forecasts for the Swedish index OMXS30. The empirical results indicate that the choice of model becomes less important for Expected Shortfall than for Value at Risk, and that the Student’s t distributed Value at Risk model improves accuracy compared to a normally distributed model. The EWMA model, proposed by the RiskMetrics Group, as well as the IGARCH and GJR-GARCH models generates adequate Value at Risk estimates at the 97.5 percent confidence level.