Gaussian process models applied for monthly forecast coal price mineral : a case study of Mozambique

Resumo

The search for models capable of predicting price movements and economic variables is recurrent in the
finance literature. The understanding of these prices’ behavior is essential for proper inflation control and planning
of production in exporting countries. Although the research in this regard is relatively vast, presenting studies
on statistical or econometric learning models of time series. Some approaches have deserved greater prominence
in modeling and their prediction, such as moving averages methods, exponential smoothing, seasonal ARIMA,
Autoregressive Vector (VAR), and ARCH or GARCH, among others. However, due to the peculiar characteristics
of the commodities market, such series often describe movements such as randomness, changing levels, explained
by various market factors and exogenous aspects. Linear models, as previously mentioned, may not be entirely
appropriate for capturing this nonlinear behavior and therefore requiring other modeling forms. This work is
part of this context since it proposes the application of Gaussian process models as an alternative for modeling
and forecasting the behavior of the monthly prices of mineral coal in Mozambique. The time series refers to
the monthly sales of coal mineral between 2011 to March 2020. The data were collected on the website of the
bank of Mozambique, in the external sector database. The proposed model is compared to the other models using
proper metrics. Our findings show that the Gaussian processes model presented promising results in the forecast
one month ahead. The results obtained of this research can provide useful predictions for the coal prices that
can assist the treasury managers in previewing the economic performance and potentially improve Mozambique’s
performance on the global commodities market.