Statistical analysis and applications of financial network data in the era of digital intelligence

In the era of digital intelligence, financial systems generate vast volumes of financial network data characterized by high frequency, heterogeneity, and rich relational structure, rendering traditional models less effective at capturing the complex, time-varying dynamics of systemic risk. This review synthesized statistical methods and applications for these data along three strands: complex networks to study contagion and connectedness; higher-order and multilayer representations to capture group interactions and cross-market channels; and graph neural networks (GNNs) that fuse topology with rich node- and edge-level attributes for dynamic risk prediction. We outlined identification using network statistics, estimation in time-varying settings, and graph-based stress testing for macro-prudential applications. Open challenges include integrating multimodal data, improving causal interpretability and counterfactual evaluation, and scaling the construction of dynamic higher-order graphs. Advances in explainable GNNs aligned with structural contagion, together with models of spatiotemporal propagation that fuse textual and market-microstructure signals, can materially enhance real-time monitoring, forecasting, and macro-prudential policy design for systemic financial risk.