Abstract
We consider the assessment of mis-specification risk when forecasting Value-at-Risk on a daily horizon. In particular, we focus on Energy Markets (electricity,oil, gas), where the impact of model risk may be relevant. Within an AR-GARCH framework to capture known features of volatility, we consider nine competing distributions for the standardized innovations and we apply a recently proposed measure of model risk to quantify the amount of model uncertainty in the procedure. Our approach is made more robust by discarding, on a daily basis, the worst performing models by using a set of weights built upon the Bayesian Information Criterion. The analysis covers the period 2001-2015, allowing for an in-depth assessment of the dynamics of model risk.