A Markov product for tail dependence functions

Abstract

We introduce a Markov product structure for multivariate tail dependence functions, building upon the well-known Markov product for copulas. We investigate algebraic and monotonicity properties of this new product as well as its role in describing the tail behaviour of the Markov product of copulas. For the bivariate case, we show additional smoothing properties and derive a characterization of idempotents together with the limiting behaviour of n-fold iterations. Finally, we establish a one-to-one correspondence between bivariate tail dependence functions and a class of positive, substochastic operators. These operators are contractions both on $L^1(\mathbb{R}_+)$ and $L^\infty(\mathbb{R}_+)$ and constitute a natural generalization of Markov operators.