Abstract

Empirical beta copula is defined by using ranks of data and the beta distribution. It can be considered as a non-parametric method for describing dependence structure of multivariate data. Since no parameter is needed in the estimation of the copula, this method is very easy to use. However, when the sample size is large, the computation of beta functions in this method shows difficulty. To enhance utility of this copula, we consider its parameterization by separating the data into cells of a K by K grid, and investigating the method of selecting K. This would not only make it easier in computation but also provide a smoother copula density.