Accurately predicting stock prices is crucial for investment and risk management, but the non-stationarity of the financial market and the complex correlations among stocks pose challenges to traditional models (ARIMA, LSTM, XGBoost), resulting in difficulties in effectively capturing dynamic patterns and limited prediction accuracy. To this end, this paper proposes the Financial Spatio-Temporal Graph Attention Network (FSTGAT), with the following core innovations: temporal modelling through gated causal convolution to avoid future information leakage and capture long- and short-term fluctuations; enhanced spatial correlation learning by adopting the Dynamic Graph Attention Mechanism (GATv2) that incorporates industry information; designing the Multiple-Input-Multiple-Output (MIMO) architecture of industry grouping for the simultaneous learning of intra-group synergistic and inter-group influence; symmetrically fusing spatio-temporal modules to construct a hierarchical feature extraction framework. Experiments in the commercial banking and metals sectors of the New York Stock Exchange (NYSE) show that FSTGAT significantly outperforms the benchmark model, especially in high-volatility scenarios, where the prediction error is reduced by 45–69%, and can accurately capture price turning points. This study confirms the potential of graph neural networks to model the structure of financial interconnections, providing an effective tool for stock forecasting in non-stationary markets, and its forecasting accuracy and industry correlation capturing ability can support portfolio optimization, risk management improvement and supply chain decision guidance.

This study introduces a Spatio-Temporal Graph Attention Network (STGAT) for non-stationary financial systems to address challenges in stock price prediction, such as non-stationarity and complex dependencies, by integrating a gated causal temporal convolution module and a novel graph attention module. STGAT employs causal time-modeling to prevent future information leakage, a dynamic industry-correlation graph attention mechanism for adaptive weight learning, and a multi-scale MIMO framework for industry-grouped feature extraction. Its symmetric integration of temporal and spatial modeling balances time-series dynamics with inter-stock correlations. Experiments on the New York Stock Exchange’s commercial banking and metals sectors demonstrate that STGAT substantially outperforms XGBoost, LSTM, and SARIMA in predictive accuracy, particularly in high-volatility scenarios. This research underscores the efficacy of graph neural networks in modeling interconnected financial systems and offers insights for advancing portfolio optimization and risk management.

Crude oil prices exhibit complex dynamics---such as volatility clustering, asymmetric temporal dependencies, and nonlinear responses to geopolitical and economic shocks---that pose challenges to accurate forecasting. To address this, we propose a hybrid model, GBK-Net, which integrates three complementary components: GARCH for modeling time-varying volatility, BiLSTM for capturing bidirectional temporal dependencies, and Kolmogorov-Arnold Network (KAN) for refining nonlinear relationships. Empirical validation is conducted on 39 years of daily WTI crude oil prices (1986--2025), covering major events such as the 2008 financial crisis and the COVID--19 pandemic. Comparative experiments against benchmark models---including GARCH, EGARCH, LSTM, and CNN--LSTM--KAN---demonstrate that GBK-Net achieves the lowest prediction error (RMSE = 2.49; R² = 0.981), with statistical tests confirming its superiority. We also benchmark nested specifications using approximately unbiased tests [1]. This work contributes a robust forecasting tool for energy economics, with implications for strategic planning, risk hedging, and financial decision-making in volatile energy markets. Unless otherwise noted, we model and evaluate on shifted log-returns to accommodate the 2020 negative-price episode; results using arithmetic returns are reported as robustness checks in the Appendix.