Extreme VaR scenarios in higher dimensions (Q2463674)

Let \(X_1,\dots,X_n\) be a sequence of random variables with given marginal distributions \(F_i\). The lower bound for the quantile (Value at risk, VaR) of \(X=X_1+\dots+X_n\) at a given level \(\alpha\) is investigated. Using a copula technique, the authors describe the dependence structure of \(X_j\) which yields the worst VaR. In the case where \(F_i\) are uniform on \([0,1]\), the obtained result coincides with the known sharp bound \(n(1+\alpha)/2\). Examples of numerical computations for Pareto margins are given.