On the uncertainty of VaR of individual risk (Q2332768)

The authors study the effect of uncertainty on VaR by considering a set of uncertainty measures. Using the relationship between VaR and distribution function, they transform the non-convex VaR maximization problem to an associated convex optimization problem. The robust VaR and an associated worst scenario can then be identified. The robust VaR is found to be the VaR under the physical probability measure with a higher confidence level, and the adjustment on the confidence level is independent of the distribution of the loss. The worst scenarios obtained are concave distortions of the physical distribution. It is shown that the choice of the loss model is still important when there is an uncertainty model.