Multi-step-ahead copper price forecasting using a two-phase architecture based on an improved LSTM with novel input strategy and error correction

Abstract

Accurate copper price forecasting plays a vital role in many aspects of economics. However, the complicated fluctuations of copper price make it a challenging job. This study develops a novel architecture based on long short-term memory model optimized by genetic algorithm (GA-LSTM) and error correction strategy for multi-step-ahead copper price forecasting. The proposed architecture includes the following two sub-phases: (1) initial forecasting of copper price; (2) error correction. In the first phase, to improve the forecast accuracy, a number of recent copper price data and some selected influencing factors are combined to construct a hybrid input of GA-LSTM model. The hybrid input strategy simultaneously considers the evolution rule of historical price data and the causal relationship between the influencing factors and copper price data, which provides more information for GA-LSTM model. One real copper price data series with a time span of 30 years is used for validating the forecasting ability of the proposed architecture and a same length of iron ore price data series is also adopted for testifying the generalization ability and robustness of the proposed architecture. The experiment results illustrate that the proposed architecture outperforms the benchmark models considered in this study.