Financial markets have become increasingly interconnected, amplifying the speed and magnitude of contagion effects during periods of uncertainty. This study proposes an integrated network-based and machine learning framework to examine financial contagion and systemic vulnerability in the Borsa Istanbul equity market. Using daily log-returns of 106 stocks over the period 2023–2025, a correlation-based financial graph network is constructed, and the structural backbone is extracted using the Minimum Spanning Tree approach. The resulting topology reveals a hub-dominated structure in which systemic importance is concentrated within a limited number of assets. A Graph Neural Network is employed to classify future volatility regimes. The model achieves an out-of-sample classification accuracy of 65%, indicating that network-aware learning captures predictive signals not available in isolated time-series models. The findings demonstrate that financial contagion is both structurally embedded and predictively exploitable when markets are modelled as complex networks. This study addresses a critical gap in the literature by extending predictive deep graph learning to the highly volatile dynamics of an emerging market, advancing beyond the static topologies and traditional econometric models prevalent in prior local research. The proposed framework offers practical implications for portfolio managers and macroprudential regulators by enabling the early identification of real-sector contagion hubs.
Cevher Özden (Wed,) studied this question.