When dealing with the case of two credit portfolios X and Y with the number of default NX(p) and NY(p), respectively, we develop a flexible dependence structure combining Archimedean copulas to model the dependency within each credit portfolio, with the multivariate generalised Pareto distribution introduced in Hendriks and Landsman (2017) to model the dependency between these credit portfolios. Special attention is paid to the copula generator functions h(·) possessing regularly varying and rapidly varying property, which allows us to devise important theoretical results for the conditional probability P(NX(p) ≥ nX|NY(p) ≥ nY ) and its limiting properties when default probability p is small.