Forecasting Chile’s mine copper production considering country risk and metal price: ARIMA, ARIMAX, and panel data models comparison
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Chile lidera la producción mundial de cobre de mina, un insumo estratégico para la industria global y la transición hacia una economía más sostenible. Anticipar con precisión la producción de cobre de mina en el corto, mediano y largo plazo es fundamental para la planificación presupuestaria, la formulación de políticas públicas y la toma de decisiones de inversión en el sector minero, particularmente en países con capacidades productivas significativas. Este estudio compara tres enfoques econométricos para estimar la producción de cobre de mina en Chile en un horizonte de corto a mediano plazo: ARIMA univariado; ARIMAX multivariado; y modelos multivariados de datos de panel. Los modelos multivariados incorporan el precio del cobre y el riesgo país como variables exógenas. Se emplean series históricas anuales y mensuales y periodos de entrenamiento variables para evaluar la capacidad de pronóstico de los modelos frente a datos reales observados entre 2021 y 2024 mediante cuatro métricas: MSE, RMSE, MAE y MAPE. Todos los modelos se estiman con y sin estrategias de estimación multi-paso. Los resultados muestran que la granularidad y la heterogeneidad de la información de producción, el uso de variables exógenas y la aplicación independiente de estrategias multi-paso solo contribuyen de forma marginal a obtener mejores estimaciones. Sin embargo, cuando se combinan, mejoran significativamente la precisión de las evaluaciones. El mejor modelo simple, ARIMA MM3 de producción total 2004–2020 (directo), y el mejor modelo integrador, datos de panel MM3 2004–2020 (iterativo), presentan un MAPE de 4,28% frente a 1,47% para todo el periodo de prueba. Este artículo propone un marco metodológico robusto y reproducible para la estimación de la producción minera de recursos estratégicos, destacando la importancia de considerar tanto la frecuencia de los datos como el entorno económico en los modelos de pronóstico de producción minera.
Chile leads the world’s mine copper production, a strategic input for the global industry and the transition to a more sustainable economy. Anticipating mine copper production in the short, medium and long term accurately is critical for budget planning, public policy formulation and investment decision-making in the mining sector, particularly in countries with significant production capacities.This study compares three econometric approaches to estimate the mine copper production in Chile for a short- to mid- term horizon: univariate ARIMA; multivariate ARIMAX; and multivariate panel data models. Multivariate models include copper price and country risk as exogenous variables. Annual and monthly historical series and variable training periods are applied to evaluate the forecasting capacity of models against real data observed between 2021 and 2024 through four measures: MSE, RMSE, MAE and MAPE. All models are estimated with and without multi-step estimation strategies.Results show that granularity and heterogeneity in production information, the use exogenous variables and the application of multi-step estimation incorporated independently only contribute marginally to obtain better estimates. However, when all are included together, they significantly improve assessments accuracy. The best simple model, total production ARIMA MM3 2004-2020 (direct), and the best integrative model, Panel data MM3 2004-2020 (iterative), present a MAPE of 4.28% vs 1.47% for the whole testing period.This paper proposes a robust and reproducible methodological framework for the estimation of mine production of strategic resources, highlighting the importance of considering both the frequency of data and the economic environment in mining production forecasting models.
Chile leads the world’s mine copper production, a strategic input for the global industry and the transition to a more sustainable economy. Anticipating mine copper production in the short, medium and long term accurately is critical for budget planning, public policy formulation and investment decision-making in the mining sector, particularly in countries with significant production capacities.This study compares three econometric approaches to estimate the mine copper production in Chile for a short- to mid- term horizon: univariate ARIMA; multivariate ARIMAX; and multivariate panel data models. Multivariate models include copper price and country risk as exogenous variables. Annual and monthly historical series and variable training periods are applied to evaluate the forecasting capacity of models against real data observed between 2021 and 2024 through four measures: MSE, RMSE, MAE and MAPE. All models are estimated with and without multi-step estimation strategies.Results show that granularity and heterogeneity in production information, the use exogenous variables and the application of multi-step estimation incorporated independently only contribute marginally to obtain better estimates. However, when all are included together, they significantly improve assessments accuracy. The best simple model, total production ARIMA MM3 2004-2020 (direct), and the best integrative model, Panel data MM3 2004-2020 (iterative), present a MAPE of 4.28% vs 1.47% for the whole testing period.This paper proposes a robust and reproducible methodological framework for the estimation of mine production of strategic resources, highlighting the importance of considering both the frequency of data and the economic environment in mining production forecasting models.
Description
Tesis (Master of Science in Engineering)--Pontificia Universidad Católica de Chile, 2025