Forecasting crude oil prices using a convolutional neural network with time-delay embedding

This paper proposes a novel hybrid forecasting model, TDE-CNN, to model the complex dynamics of crude oil price movements. The model integrates Time-Delay Embedding (TDE) Method with a Convolutional Neural Network (CNN) to leverage both spatial and temporal information. The TDE-CNN model uses the TDE method to transform raw crude oil data into higher-dimensional space to reveal underlying spatio-temporal patterns, while the CNN effectively models these patterns for improved predictive accuracy. The TDE-CNN model is applied to forecast major crude oil spot price movements, and its forecasting performance has been comprehensively and rigorously evaluated. Empirical results demonstrate that the TDE-CNN model achieves lower forecasting errors compared to benchmark models, as measured by Mean Squared Error (MSE). Additionally, the Diebold-Mariano test confirms that the improvement in forecasting accuracy is statistically significant.