Introducing the GED-Copula with an application to Financial Contagion in Latin America

While the Generalized Error Distribution (GED) has been used quite extensively in time series applications and has demonstrated a sound flexibility in the estimation process, there is so far no attempt to use this function in the construction of Copulas. Copulas are probability functions that link one multivariate distribution function to univariate distribution functions called marginals. These marginal functions are assumed to be continuous and to follow a uniform behaviour within [0,1]. In this paper we propose a new Copula function that, to our knowledge, has not been used in the literature of Copulas, until now: the bivariate GED-Copula. This function embeds other well-known distributions including the gaussian distribution. In order to assess the relative performance of this new Copula we investigate financial contagion in foreign exchange, stocks, bonds and sovereign debt markets in Latin America. Standard decision criteria provides strong evidence in favour of the GED-Copula against other Elliptical and Arquimidean alternatives.