Abstract
In this paper we propose a new clustering algorithm based on copula functions. Copula functions (CF, hereafter) were introduced in Sklar (1959) in a probabilistic context and have become a powerful multivariate modeling tool in many fields. Among their advantages CFs allow to i) overcome the limitations of the linear correlation coefficient to measure dependence; ii) model the marginal distributions and the dependence structure separately: the former are linked to the shape of the distribution function, whereas the CF represents the kind of dependence; this allows to use the two–step estimation method called Inference for Margin (IFM) by Joe and Xu (1996); iii) fit any combinations of parametric and non parametric marginal distribution functions. In the next section we present briefly the procedure proposed together with some results obtained through a simulation study.