A Hybrid GRU-Transformer Model for Financial Forecasting and Risk Management

Financial sector demands precise forecasting models to mitigate risk and optimize asset management. This study presents a novel Hybrid GRU-Transformer Model that integrates Gated Recurrent Units (GRUs) for capturing local temporal patterns and the Transformer T5 architecture for modeling global dependencies. Financial datasets sourced from Kaggle, including stock prices and trading volumes, were preprocessed through normalization and categorical encoding, followed by an 80:20 dataset split for training and testing. The proposed model outperformed existing methods such as ARFIMA-LSTM and EMD-CNN-LSTM, achieving superior metrics, including a Mean Squared Error of 0.08 and an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R^{2}}$</tex> of 0.95. Results underline the model's accuracy and robustness in predicting financial risks and trends. By enhancing predictive performance and risk management strategies, this model addresses the dynamic complexities of financial markets. Future research will explore its adaptability across diverse instruments and markets to further its application in real-time forecasting scenarios. The system is implemented using Python integrated with MATLAB.