Evaluation of Selected Models for Value at Risk Calculation

We compared different newer models (e.g. CAViaR and one of the most recent approaches HAR-QREG) to the more traditional approaches (e.g. RiskMetrics and GARCH(1,1)) for value at risk calculation. As samples for different asset classes we chose MDAX and CDAX as representatives for the German capital market, gold, Brent crude oil, wheat, and corn for alternative investments, and the EUR/USD exchange rate representing the currency market. The prediction quality of each model was tested using back testing methods like the conditional coverage and dynamic quantile test. It turned out that the newer models are able to outperform the traditional approaches, but all fail to model corn return due to an extreme price drop