On the use of t-copulas for economic capital calculations

David G. Maher

Abstract

ABSTRACT

We examine three methods for constructing correlated Student's t random variables. Our motivation arises from simulations that utilize heavy-tailed distributions for the purposes of stress testing and economic capital calculations for financial institutions, and we make several observations regarding the suitability of the three methods for this purpose. Moreover, we show that one of the methods produces undesirable values, despite possessing apparent theoretical soundness. Onthe other hand, another construction produces desirable values, despite having a theoretical shortcoming. This highlights the need for model validation through both theoretical formulation and aspects of its ultimate output.