On extremal dependence: some contributions (Q1936535): Difference between revisions

Let \(X,Y\) be random variables with common continuous distribution function \(F\). One well-known tail dependence coefficient is defined by \[ \lambda=\lim_{x\uparrow \omega(F)} P(Y>x\mid X> x), \] where \(\omega(F)\) is the upper limit of the support of \(F\), provided this limit exists. Let \((X_1,Y_1)\), \((X_2,Y_2)\) be two random pairs identically distributed as \((X,Y)\). The upcrossings-tail dependence coefficients investigated in this paper are defined by \[ \mu_{Y\mid X}= \lim_{x\uparrow \omega(F)} P(Y_1\leq x <Y_2\mid X_1\leq x <X_2), \] \[ \mu_{X\mid Y}= \lim_{x\uparrow \omega(F)} P(X_1\leq x <X_2\mid Y_1\leq x <Y_2). \] \((X_1,Y_1), (X_2,Y_2)\) are called asymptotic upcrossings-tail independent if \(\mu_{Y\mid X}=0\) or \(\mu_{X\mid Y}=0\). It is shown that these coefficients are related to multivariate tail dependence coefficients already known in the literature. The coefficients of tail dependence are applied to stationary sequences and, hence, measure tail dependence in time. Connections with the extremal index, the upcrossing index as well as with local dependence conditions are established.