Assessing Variable Importance via Shapley Values for Extreme Operational Losses: A Study on Banking Dataset

We present a comprehensive analysis of extreme operational losses in UniCredit Bank over a ten-year period, with a primary emphasis on evaluating variable importance using Shapley values. Our investigation begins by fitting advanced models to capture the severity distribution of operational losses exceeding the 95th percentile threshold. Specifically, we employ a VGAM model with a generalized Pareto distribution (GPD), a Generalized Beta 2 model, and an Extended Generalized Pareto Distribution (EGPD) to accurately characterize the nature of extreme losses. In addition, we explore the interdependencies among explanatory variables by utilizing Vine copulas, enabling the generation of simulated values. These simulated values play a vital role in data augmentation, facilitating the estimation of variable importance through the application of Shapley values. By leveraging the power of Shapley values, our study provides a comprehensive assessment of the influence of each variable on the occurrence of losses across various event types, revealing key drivers of operational losses. The findings of this research highlight the critical role played by Shapley values in assessing variable importance. Through our analysis, we identify the variables that significantly impact operational losses and gain deeper insights into their individual contributions. This knowledge is invaluable for designing effective risk management strategies and equipping financial institutions with actionable insights to mitigate operational risks more efficiently. By integrating advanced statistical modeling techniques, including extreme value theory, Vine copulas, and Shapley values, our thesis presents a comprehensive framework for assessing variable importance in operational risk management. The research outcomes contribute a robust methodology and practical insights to the field, empowering financial institutions to make informed decisions and implement effective risk mitigation strategies.