Globality in the metal markets: Leveraging cross-learning to forecast aluminum and copper prices

Over the last 25 years, aluminum and copper prices have fluctuated significantly, impacting mining companies, economies of mining countries and global financial markets. Accurate price forecasting can be useful for better budget planning, strategic decision making and portfolio management. To bridge recent developments in time series forecasting with commodity price forecasting, we investigate the performance of a recently prominent framework called global forecasting or cross-learning. Juxtaposed to the local approach, under which one model would be used for each time series requiring forecasts, this framework involves the training of a single (global) model using a set of relevant time series to create future predictions. In our analysis, we assume that incorporating stock prices of mining companies into global forecasting models can enhance the prediction accuracy of aluminum and copper prices compared to traditional forecasting methods. In this direction, we use a pool of auxiliary time series that includes the prices of the two commodities and the stock prices of major companies involved in their production and processing to train global feed-forward neural networks. The models differ in their approach to selecting relevant time series used in training and are either used as standalones or in combination with standard univariate models to forecast the two commodities’ prices 3, 6 and 12 months ahead. The performance of the models is then compared to benchmark univariate (local) models, namely the naïve method, ARIMA, and autoregressive feed-forward neural networks. The analysis suggests that the incorporation of information from different (albeit related) markets achieved through cross-learning is beneficial for forecasting the prices of the two commodities since, for all forecast horizons examined, the models achieving the lowest out-of-sample error were either standalone global models or local-global ensembles.