LSTM-augmented vine copula modelling for energy-finance contagion analysis

Amid mounting geopolitical tensions and rapid transformations in global energy dynamics, the transmission of risk between energy and financial systems has become a pressing concern in safeguarding financial stability. This study introduces a unified modelling framework that fuses artificial intelligence with wavelet decomposition, volatility modelling and copula theory to uncover evolving tail dependencies and contagion pathways across crude oil, renewable energy, and financial markets. We enhance the conventional three stage methodology, which consists of the Maximum Overlap Discrete Wavelet Transform, ARMA EGARCH filtering, and Vine Copula modelling, by integrating a Long Short Term Memory neural network. Our empirical investigation, leveraging daily observations from global oil benchmarks, clean energy indices, and financial sector metrics, uncovers pronounced shifts in tail dependencies and contagion intensities during turbulent periods. The prospective volatility signal generated by the LSTM strengthens the model’s ability to capture time varying tail behavior and nonlinear contagion. Using daily data from 2015 to 2025, the framework reveals strong short run asymmetry, with downside contagion dominating, while medium term dynamics show gradual structural adjustments. Out sample tests indicate that the enhanced model outperforms DCC, rolling copulas, GRU and attention-based networks in forecasting tail dependence. Event driven spillover analysis further shows that major shocks reshape transmission routes and shift contagion hubs across markets. By combining deep learning signals with interpretable dependence and network analysis, the framework offers a concise and effective tool for monitoring systemic risks and supporting stress testing and macroprudential supervision.