Systemic risk in banks, as one of the most significant threats to financial stability, arises from the complex interaction between banks’ balance-sheet characteristics and their network structure. Nevertheless, much of the existing literature has predominantly examined the determinants of systemic risk using linear and independent frameworks, devoting limited attention to the potential nonlinearity of these relationships and their variation across different market conditions.
This study aims to investigate the nature of relationships (linear, semi-linear, and nonlinear), as well as the individual and interaction effects of financial variables and network-based indicators on banks’ systemic risk. Systemic risk is measured using the ΔCoVaR metric, while the XGBoost algorithm, combined with the SHAP explainability framework, is employed to capture nonlinearities and interaction effects among explanatory variables. The analysis is conducted separately for the bubble formation and bubble collapse phases.
The findings reveal that during the bubble formation phase, the impact of financial and network variables on systemic risk is predominantly nonlinear and driven by interaction effects. Financial variables such as loan growth and leverage exhibit a substantial contribution to interaction effects, while network indicators—including closeness centrality and betweenness centrality—also play a prominent interactive role. During the bubble collapse phase, the relative importance of interaction effects increases further, and relationships become more linear and direct, reflecting the materialization of systemic risk within the banking network.
These results highlight the importance of dynamic and forward-looking systemic risk management and underscore the need for supervisory and macroprudential policies that explicitly account for both banks’ financial characteristics and their network interactions.