1: \begin{abstract}

2: We propose a new class of estimators for Pickands dependence function

3: which is based on the concept of minimum distance estimation. An explicit

4: integral representation of the function $A^*(t)$, which minimizes a weighted

5: $L^2$-distance between the logarithm of the copula $C(y^{1-t},y^t)$ and

6: functions of the form $A(t) \log(y)$ is derived. If the unknown copula

7: is an

8: extreme-value copula, the function $A^*(t)$ coincides with Pickands

9: dependence function. Moreover, even if this is not the case, the function

10: $A^*(t)$ always satisfies the boundary conditions of a Pickands dependence

11: function. The estimators are obtained by replacing the unknown copula

12: by its

13: empirical counterpart and weak convergence of the corresponding process

14: is shown. A comparison with the commonly used estimators is performed

15: from a theoretical point of view and by means of a simulation study. Our

16: asymptotic and numerical results indicate that some of the new estimators

17: outperform the estimators, which were recently proposed by Genest and Segers

18: [\textit{Ann. Statist.} \textbf{37} (2009) 2990--3022].

19: As a by-product of our results, we obtain a simple test for the

20: hypothesis of

21: an extreme-value copula, which is consistent against all positive quadrant

22: dependent alternatives satisfying weak differentiability assumptions of

23: first \mbox{order}.

24: \end{abstract}