Abstract

This work addresses the feasibility of modeling the copper price through SARIMA approach. The period under study was 30 years (1991-2020), leaving the last year (2020) as the testing set, and the previous 29 years as the training set. The results with the nominal copper price by itself are of good quality in comparison with present literature (best MAPE for training set was 3.98%). As the exploratory data analysis (EDA) suggest a non-seasonal series, the best model obtained under this assumption was, following the notation (p, d, q) (P, D, Q)(S), the model (0, 1, 2)(0, 0, 0)(0). This model has 4.16% of MAPE in the training set, and it showed 4.31% MAPE in the testing period of one year (2020), using the rolling window technique. To improve forecast modeling, exogenous variables were included as a complement, using a SARIMAX approach. China's Leading Economic Index and Copper Consumption were the chosen alternatives. In the training data, adding any of them does not show an improvement. However, in the testing group, the economic activity becomes an interesting choice for performance, reducing the MAPE to 4.26%.