A Multivariate Approach to Time Series Forecasting of Copper Prices with the Help of Multiple Imputation by Chained Equations and Multivariate Adaptive Regression Splines
-
Fernando Sánchez Lasheras [1]
;
Javier Gracia Rodríguez
[1]
;
Paulino José García Nieto
[1]
;
Esperanza García-Gonzalo
[1]
;
Gregorio Fidalgo Valverde
[1]
-
[1] Universidad de Oviedo
Universidad de Oviedo
Oviedo, España
-
- Localización: 15th International Conference on Soft Computing Models in Industrial and Environmental Applications (SOCO 2020): Burgos, Spain ; September 2020 / coord. por Álvaro Herrero Cosío , Carlos Cambra Baseca , Daniel Urda Muñoz , Javier Sedano Franco , Héctor Quintián Pardo , Emilio Santiago Corchado Rodríguez , 2021, ISBN 978-3-030-57802-2, págs. 691-701
- Idioma: inglés
- Texto completo no disponible (Saber más ...)
-
Resumen
This research presents a novel methodology for the forecasting of copper prices using as input information the values of this non-ferrous material and the prices of other raw materials. The proposed methodology is based on the use of multiple imputation with chained equations (MICE) in order to forecast the values of the missing data and then to train multivariate adaptive regression splines models capable of predicting the price of copper in advance. The performance of the method was tested with the help of a database of the monthly prices of 72 different raw materials, including copper. The information available starts on January 1960. The prediction of prices from September 2018 to August 2019 showed a root mean squared error (RMSE) value of 318.7996, a mean absolute percentage error (MAPE) of 0.0418 and a mean absolute error (MAE) of 252.8567. The main strengths of the proposed algorithm are two-fold. On the one hand, it can be applied in a systematic way and the results are obtained without any human with expert knowledge having to take any decision; on the other hand, all the trained models areMARS. This means that the models are equations that can be read and understood, and not black box models like artificial neural networks.
