A Novel Zinc Price Forecasting Method Based on Multi-Factor Selection and LSTM Network

Zinc is an indispensable base material for the development of national economy and the construction of national defense industry, and the price forecasting is of great significance for investors, policy makers and researchers. Considering the complexity, dynamic and strong nonlinearity of zinc price changes, and it is usually affected by a variety of external factors, it is difficult to obtain a satisfactory forecasting effect only by analyzing the underlying pattern of historical price data changing. To solve the aforementioned problem, a novel zinc price forecasting method based on factor selection and long short-term memory network is proposed. First, a number of factors that may be related to price changes are collected and Granger causality test is employed to remove the non-causal factors. Then, XGBoost is used to analyze and sort the remaining factors by importance, and the factors with high importance for forecasting are selected. Finally, utilizing the historical price of zinc and the selected external factors, a multivariable forecasting model based on long short-term memory network is established to forecast the future price of zinc. Compared with other state-of-the-art methods from value and direction prediction accuracy and fitting ability, the proposed model has superior performance for zinc price forecasting.