A Preliminary Investigation of the Impact of Gaussian Versus t-Copula for Data Perturbation

Abstract

In this paper, we provide a preliminary investigation of t-copulas for perturbing numerical confidential variables. A perturbation approach using Gaussian copulas has been proposed earlier. However, one of the problems with the Gaussian copulas is that it does not preserve tail dependence. In this investigation, we show that the t-copula can be used effectively to provide all the benefits that a Gaussian copula provides and, in addition, maintain tail dependence as well. We illustrate this approach using two examples. We hope to perform a comprehensive investigation of this approach in the future.