Extended Geometric-VaR backtesting

Abstract This paper extends on the methods provided by Pelletier & Wei (2015). In the first part of the paper, we replicate the simulation-driven results where we extend with empirical research in the second part. Opposed to the Historical Simulation method used in Pelletier & Wei (2015), we create a set of risk specifications derived from Wong et al. (2016). These specifications are exploited using two distinct methods of innovations inspired by the findings of Bao et al. (2007). We empirically show that the Geometric-VaR test possesses high power against alternatives within the framework for our empirical specifications. Additionally, we show that parametrized GARCH specifications lead to better specified Value-at-Risk estimations than the Historical Simulation approach does. Lastly, we show that GARCH specifications with skewed innovations generally lead to better-specified Value-atRisk estimations than specifications that use a Filtered Historical Simulation instead.