Fusion Forecasting: An Integration of BiLSTM and Enhanced Transformers for Stock Prices

The corporate aim of seeking to attain the highest return for shareholders is still central to most financial papers. To improve the exactness and accuracy of the stock price forecasting, this research work propounds an advanced method known as BiLSTM-MTRAN-TCN. It improves the conventional transformer model with the novel Temporary Convolutional Network (TCN) to optimize it for stock forecasts and create the new MTRAN-TCN model. The BiLSTM-MTRAN-TCN model is a combined use of BiLSTM, MTRAN, and TCN each of which has unique features in the network architecture. As much as transformers are known to perform longer range dependencies they tend to underperform on sequential data. Such a problem is solved by BiLSTM that extracts bidirectional sequence information, and TCN also contributes to the model’s sequence dependency learning and generalization. This hybrid model was verified with the help of data of 5 of index stocks and 14 stocks from Shanghai and Shenzhen. The performance of BiLSTM-MTRAN-TCN was better than that of the existing methods, especially when it comes to the fit of index stocks. It showed the improvement of RMSE by the range of 24. 3% to 93. 5% was seen, and the R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values were found to improve by 0. 3% to 15. 6%. Further, this method demonstrated efficiency for different periods of the analysed data; there were no tendencies to untimeliness; thus, it could be concluded that the method is accurate and suitable for stock price prediction.