CausalGraphX Explains Systemic Financial Risk with GNNs.

Global financial systems are inherently interconnected, making them susceptible to systemic risks where a single failure can trigger widespread defaults. Traditional risk models often fail to capture these complex, non-linear dynamics. While Graph Neural Networks (GNNs) show promise for relational data, they typically learn correlations and lack the explainability crucial for regulators. Researchers have introduced CausalGraphX, a new framework that combines GNNs with counterfactual reasoning to offer explainable systemic risk assessments. CausalGraphX uses a Graph Attention mechanism to model institutional vulnerability and employs adversarial regularization to ensure it identifies causal drivers, not just spurious correlations. A key feature is its optimization-based approach for generating counterfactual explanations, allowing regulators to ask "what if" questions, such as the minimum capital injection needed to prevent a bank's default. Validated on large synthetic financial networks, CausalGraphX significantly outperforms existing models in predicting cascading defaults while providing sparse, plausible, and actionable insights.