Detecting Financial Contagion Through Higher-Order Networks: A Deep Learning Approach to Emerging Market Risk

Abstract How can we detect and quantify multi-dimensional financial contagion in emerging markets before crises manifest? Systemic risk assessment in emerging financial markets requires capturing complex contagion mechanisms that extend beyond traditional pairwise relationships, yet existing frameworks fail to model the higher-order dependencies that characterize crisis propagation. This study develops an integrated framework combining Transfer Entropy networks, hypergraph-based risk analysis, and deep learning stress testing to quantify multi-dimensional systemic vulnerabilities in Turkish financial markets from 2015-2025. The methodology introduces three novel metrics: Risk Virality Score using PageRank on directional information flow networks, Hyperedge Anomaly Detection for early warning capabilities, and Financial Immunity Score synthesizing contagion resistance, recovery speed, and structural risk exposure. Empirical analysis of ten sectoral indices and seven macroeconomic variables reveals severe vulnerability in the banking sector contrasting with exceptional resilience in eight other sectors, identifies 25,319 higher-order risk relationships invisible to pairwise analysis, and demonstrates early warning capabilities with 2-4 week lead times before major stress episodes. The BiLSTM-Attention stress testing reveals that domestic political uncertainty generates more persistent impacts than external shocks across most sectors. These findings challenge conventional risk assessment approaches and support enhanced macroprudential frameworks incorporating network effects and higher-order relationships for emerging market financial stability.