Forecasting Volatility of New Energy Stock Prices: An Integrated CEEMDAN and BiLSTM-Transformer Hybrid Model

Stock volatility exhibits nonlinear and noisy characteristics, making its accurate prediction crucial for stock market analysis. This paper proposes a hybrid model integrating CEEMDAN with BiLSTM-Transformer, applied to the task of forecasting the volatility of new energy stocks. Additionally, the model combines sentiment indices with volatility as multi-dimensional input data, enhancing its applicability. Evaluation metrics include MAE, MSE, RMSE, and R², with comparisons against CEEMDAN-BiLSTM, CEEMDAN-Transformer, and BiLSTM-Transformer models. The experimental results demonstrate that CEEMDAN-BiLSTM-Transformer outperforms its counterparts, exhibiting strong robustness and generalizability.