A copula-based data augmentation strategy for the sensitivity analysis of extreme operational losses

In this work, we aim to assess the importance of macroeconomic and financial variables for operational losses of UniCredit Bank. To achieve this, we consider the Shapley effects as a variance-based measure of importance. However, the small number of observations of extreme losses makes the estimation of the Shapley effects challenging. To address this issue, we proposed augmenting the sample of extreme observations using vine copulas and calculating the Shapley effects on the augmented sample. The effectiveness of this procedure is supported by a numerical simulation. Findings obtained with our methodology applied to the UniCredit Bank data show its usefulness for the risk management of operational losses.