Stable Probabilistic Graphical Models for Systemic Risk Estimation

The interdependencies within the global financial system can cause ripple effects especially during crisis in a process called contagion. We study contagion because the transmission of shocks during a crisis can have a significant impact on society and the global economy. We apply Stable Graphical Models (SGM), a class of multivariate alpha-stable densities that can be represented as Bayesian networks whose edges encode linear dependencies between random variables. We are motivated by the lack of a generalized and sufficiently flexible model that can capture leptokurtic features exhibited in financial time series. Using data from 24 developed and emerging countries between 2000 and 2023, we study the process of contagion across 6 crisis and 7 tranquil periods. Our results show that the incidence of contagion is more expressed during crisis periods, demonstrating the model's ability to identify and characterize the structural relationship between random variables.