The checkerboard copula and dependence concepts

We study the problem of choosing the copula when the marginal distributions of a random vector are not all continuous. Inspired by three motivating examples including simulation from copulas, stress scenarios, and co-risk measures, we propose to use the checkerboard copula, that is, intuitively, the unique copula with a distribution that is as uniform as possible within regions of flexibility. We show that the checkerboard copula has the largest Shannon entropy, which means that it carries the least information among all possible copulas for a given random vector. Furthermore, the checkerboard copula preserves the dependence information of the original random vector, leading to two applications in the context of diversification penalty and impact portfolios.