A novel perspective on forecasting non-ferrous metals' volatility: Integrating deep learning techniques with econometric models

This study puts forward a new perspective on non-ferrous metals' volatility prediction in the futures market. Two hybrid deep learning architectures are constructed by embedding assorted convolutional neural networks (CNN) into long short-term memory (LSTM) models, and combining the LSTM networks with various generalized autoregressive conditional heteroscedasticity (GARCH)-type models. We illustrate the numerical implementation of all proposed models on four non-ferrous metal indices. Our findings suggest that the GARCH-LSTM model outperforms other alternatives by examining diverse error metrics. This study marks a significant advancement in the application of integrated deep learning models to enhance the prediction performance of commodity volatility.

Automated summary

This study proposes a new perspective on predicting the volatility of non-ferrous metals in the futures market. Two hybrid deep learning architectures are constructed by combining convolutional neural networks (CNN) and long short-term memory (LSTM) models, as well as LSTM networks and various generalized autoregressive conditional heteroscedasticity (GARCH) models. The findings show that the GARCH-LSTM model outperforms other alternatives in predicting commodity volatility. This study marks a significant advancement in enhancing the prediction performance of commodity volatility using integrated deep learning models.