1: \begin{abstract} 

2: % Multivariate extreme value models are widely used in fields like

3: % finance, insurance, telecommunication or environmental sciences. The

4: % structure of the tail dependence is characterized by the 

5: % so-called \emph{stable tail dependence function} (\textsc{stdf}). Estimating the latter is  the keystone of statistical

6: % inference for  multivariate extremes.

7: % Although extensive studies have proved consistency and the asymptotical normality of empirical version of the \textsc{stdf},

8: % it still lacks non-asymptotic bounds. The main purpose of this

9: % paper is  to fill this gap. By taking advantage of VC-type concentration

10: % inequalities for  relative deviations, upper bounds

11: % are derived with the expected rate of convergence in $O(k^{-1/2})$.

12: % %by firstly stating VC-type concentration tools %needed to derive such a non-asymptotic bound. 

13: % %This Vapnik-Chervonenkis theory allowed us to derive an upper bound with expected speed convergence in $O(k^{-1/2})$.

14: % %The point will be made that

15: % The concentration tools involved in this analysis rely on the more general study of maximal deviations in low probability regions, and thus directly apply to a particular pattern recognition setting, the classification of extremal observations.\\

16: 

17: 

18: % {\sc Key Words:} Standard Tail Dependence Function, Multivariate extreme value theory, Concentration inequalities, VC theory, Extremal Prediction

19: 

20: % \end{abstract}