Extreme value theory with operator norming (Q2443883)

Let \(X_{1},\dots,X_{n}\) be iid random vectors in \(\mathbb{R}^{d}\) whose distribution \(\mu\) is operator regularly varying at infinity. Consider an arbitrary direction \(\theta\in\mathbb{R}^{d}\backslash\left\{ \mathbf{0} \right\} \) and define the directorial maximum \[ M_{n}\left( \theta\right) =\max_{i=1,\dots,n}\left\langle X_{i},\theta \right\rangle . \] The authors view \(M_{n}\left( \theta\right) \) as a stochastic process indexed by \(\theta\). They develop a general theory for such directional extremes, where the tail index is allowed to vary with the direction.