Abstract/Summary

A new model risk measure and estimation methodology based on loss functions is proposed in order to evaluate the accuracy of volatility models. The reliability of the proposed estimation has been verified via simulations and the estimates provide a reasonable fit to the true model risk measure. An empirical analysis based on several assets is undertaken to identify the models most affected by model risk, and concludes that the accuracy of volatility models can be improved by adjusting variance forecasts for model risk. The results indicate that after crisis situations, model risk increases especially for badly fitting volatility models.