A continuous updating weighted least squares estimator of tail dependence in high dimensions (Q125412)

This paper deals with a new estimator for the tail dependence function in high dimensions. Given random vectors \(X_i = \left(X_{i1},\dots,X_{id}\right) \in \mathbb R^d\) for \(1 \leq i \leq n\), their tail dependence function \(l\) is estimated by employing an initial estimator \(\widehat{l}_{n,k}\) based on the same data and then minimizing a weighted least square loss over all possible functions \(l_\theta\) in a parametric family \(\left\{l_\theta: \theta \in\Theta\right\}\). The weighting matrix is allowed to depend on \(\theta\) as well and is hence jointly estimated with \(\theta\). The authors prove consistency and asymptotic normality of their estimator under reasonable assumptions and address goodness-of-fit testing as well. The choice of the initial estimator \(\widehat{l}_{n,k}\) is discussed in detail with several common examples. Afterwards extensive simulation studies are performed showing a competitive behavior of the investigated method. An application to European stock markets is also shown.