Thermal coal price forecasting via the neural network

• Published in: Intelligent systems with applications Vol. 14; p. 200084
• Main Authors: Xu, Xiaojie, Zhang, Yun
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.05.2022; Elsevier
• Subjects: algorithms; coal; commodity exchange; data collection; energy; Machine learning technique; Neural networks; policy analysis; Price forecasts; prices; Thermal coal; Time series data; Price forecasts; Neural networks; Machine learning technique; Thermal coal; Time series data
• ISSN: 2667-3053; 2667-3053
• DOI: 10.1016/j.iswa.2022.200084

•We build a non-linear auto-regressive neural network model for forecasts of thermal coal prices.•Closing prices recorded on a daily basis of thermal coal traded in China Zhengzhou Commodity Exchange are used for analysis.•The model is simple and leads to performance of good accuracy and stabilities.•The model could serve as technical forecast tools and for policy analysis. Thermal coal price forecasts represent an essential issue to investors and policy makers, given its importance as a strategic energy source. The current work aims at exploring usefulness of non-linear auto-regressive neural networks for this forecast problem based upon a data-set of closing prices recorded on a daily basis of thermal coal traded in China Zhengzhou Commodity Exchange during January 4, 2016 – December 31, 2020, which is an important financial index not sufficiently explored in the literature in terms of its price forecasts. Through testing a variety of model settings over algorithms, delays, hidden neurons, and data splitting ratios, the model that produces performance of good accuracy and stabilities is reached. Particularly, the model has five delays and ten hidden neurons and is constructed with the Levenberg-Marquardt algorithm based on the ratio of 80%–10%–10% of the data for training–validation–testing. It leads to relative root mean square errors of 1.48%, 1.49%, and 1.47% for the training, validation, and testing phases, respectively. Usefulness of neural networks for the price forecast issue of thermal coal is demonstrated. Forecast results here could serve as standalone technical forecasts and be combined with other forecasts when conducting policy analysis that involves forming perspectives of trends in prices.