BiMT-TCN: A Cutting-Edge Hybrid Model for Enhanced Stock Price Prediction

In the face of the rapid evolution and escalating complexity of financial markets, precise stock price prediction has become a critical area of research for scholars and practitioners alike. Stock markets are subject to a vast array of influencing factors, both internal and external, which complicates prediction efforts. This study proposes BiMT-TCN, a novel model combining Bidirectional Long Short-Term Memory (BiLSTM), a modified Transformer, and Temporal Convolutional Network (TCN), aimed at enhancing the accuracy and stability in stock price prediction. BiLSTM facilitates the capture of bidirectional dependencies, which aids in decoding the intricate patterns within time-series data. The modified Transformer integrates global information, enhancing the model’s capacity to manage long-range dependencies effectively. TCN, known for its parallel processing and proficiency in capturing deep historical patterns, further bolsters model stability and generalizability. Empirical evaluations on major indices such as SSE, HSI, and NASDAQ demonstrate that BiMT-TCN consistently outperforms state-of-the-art models, achieving R2 scores of 0.9779, 0.9776, and 0.9969 respectively, along with significantly lower RMSE, MAE, and MAPE values. The implications of this work extend to practical investment decision-making, where improved forecast precision can enhance risk management, optimize trading strategies, and inform financial planning in volatile markets.