Estimation window strategies for value-at-risk and expected shortfall forecasting

Tobias Berens, Gregor Weiß and Daniel Ziggel

Abstract

Compared with the large number of value-at-risk (VaR) and expected shortfall (ES) forecasting models proposed in the literature, few contributions have been made to address the question of which estimation window strategy is preferable for forecasting these risk measures. To fill this gap, we apply different estimation window strategies to a set of simple parametric, semiparametric and nonparametric industry-standard risk models. Analyzing daily return data on constituents of the German Deutscher Aktienindex (DAX), we evaluate forecasts by backtesting the unconditional coverage and independent and identically distributed properties of VaR violations, the ES forecasting accuracy and the conditional predictive ability. We thereby demonstrate that the selection of the estimation window strategy leads to significant performance differences. The results indicate that forecast combinations are the preferable estimation window strategy.