Abstract
The paper discusses the practical aspects of modeling joint distribution of pairs of national stock indices via copula functions. Parameters of marginal distributions and the association parameter describing the dependence structure are estimated using empirical Bayes method numerically implemented with the help of random walk Metropolis algorithm. A comparison of parametric and semiparametric approaches to copula model construction is performed. The problem of selection of a class of pair copula functions approximating such empirical characteristics of stock indices dependence as Kendall’s concordance, joint empirical cumulative distribution function, and tail behavior.