A novel mixed-frequency deep learning forecasting model for natural resource prices: A case study of copper futures price

Accurate prediction of natural resource prices plays a significant role in national economic development. However, existing research often focuses solely on same-frequency forecasting, neglecting the rich information contained in high-frequency data. To bridge this gap and explore whether mixed-frequency prediction improves the forecasting performance, this study develops an innovative mixed-frequency deep learning forecasting model grounded in Pearson correlation coefficient analysis, long-short-term memory, particle swarm optimization, and mixed-frequency data sampling regression. Taking copper price as an example, this study first applies Pearson correlation analysis to select the most relevant influencing factors from mixed-frequency variables. These factors include policy uncertainty, macroeconomic conditions, energy costs, and other non-ferrous metal prices. Subsequently, the proposed mixed-frequency deep learning model is used for predicting copper price. Experiments include comparisons with the benchmark model, multi-step prediction, statistical hypothesis testing, in-depth evaluation of forecasting effectiveness, and robustness analysis. The final experimental results demonstrate that the proposed mixed-frequency deep learning model significantly outperforms the comparison models, effectively improving prediction accuracy. This study not only expands the scope of futures price prediction research, but also provides a new perspective for time series prediction work in other fields.