Groupe d’études et de recherche en analyse des décisions

Testing for bivariate extreme-value dependence

Jean-François Quessy

In this talk, I will describe new tests for the hypothesis of bivariate extreme-value dependence. All these test statistics are functionals of either Kendall's process or its version with estimated parameters. The procedures considered are based on linear combinations of moments and on Cramér-von Mises distances. It will be shown how a suitably adapted version of the Multiplier Central Limit Theorem enables the computation of asymptotically valid p-values. Some results about the power of the tests in small samples and asymptotically under local alternatives will be presented. The methods will be illustrated on the Cook & Johnson data set.