Comparing Expected Shortfall forecasts by using Filtered Historical Simulation combined with a HEAVY or GARCH model

We use Filtered Historical Simulation with a time-varying volatility model to forecast Expected Shortfall (ES) forecasts. Either the HEAVY or GARCH model is used as the volatility model. Shephard and Sheppard (2010) introduced the HEAVY model which has quicker adjustments than the conventional GARCH model to structural breaks in the volatility process. The aim of this paper is to compare the specification of the ES forecasts and their out-of-sample predictive ability. 34 assets including trades assets, indexes computed by MSCI and exchange rates are used in our analysis. We show that the HEAVY model is favoured for particular assets in terms of the correct specification of the model and the predictive performance.