Clustering of upcrossings of high values (Q2655060)

Let \(\{X_n\}_{n \geq 1}\) be a stationary sequence satisfying some local dependence condition. Let \(\{u_n\}_{n \geq 1}\) be a sequence of real levels. For an integer \(T \geq 1\), the sequence \(\{X_n\}\) has a \(T\)-upcrossing of \(u_n\) at \(i\) if the event \(A_{i,n}=\{X_i \leq u_n <X_{i+T}\}\) occurs. The point process of \(T\)-upcrossings is defined by \[ N_n^{(T)}(B)= \sum_{i=1}^n \mathbb I_{A_{i,n}}\delta_{i/n}(B),\: B\subset [0,1],n \geq 1, \] where \(\mathbb I_A\) denote the indicator of event \(A\) and \(\delta_a\) is the unit mass at \(a\). The intensity of the point process limit of \(N_n^{(T)}\) is linked to a parameter \(\eta^{(T)}\in [0,1]\) called the \(T\)-upcrossings index. The authors give explicit formulas for the \(T\)-upcrossings index using the joint distributions of finite number of terms of \(\{X_n\}\) and analyse the effect that systematic sub-sampling has on the value of the upcrossings index. Finally, they apply the results to the extremal Markovian sequence and to the negatively correlated uniform \(AR(1)\) process.