Factors of predictive power for mineral commodities

Abstract

In our study, we individually forecast 26 metal prices one-month ahead and outperform the predefined benchmark model, a random-walk (with drift) in 18 (18) cases. These forecasts are based on an overview over a large set of potential predictors for mineral commodities, originating from studies which only consider a selection of attributes and apply them to predict specific commodities or commodity indices. We pre-select the relevant, commodity-specific, individual factors through a correlation analysis, followed by a BIC based regression selection.

The results of our out-of-sample, one-month ahead forecasts show a significant outperformance for 18 of the 26 commodities considered, especially those in the minor metals sector. The differences in predictability between the metal groups are remarkable, as we are able to forecast 13 of 17 minor metals, 5 of 6 industrial metals, but no precious metal, highlighting the heterogeneity in metal commodity markets. Focusing on the influential factors, the value factor has a dominating, highly significant, negative effect in the prediction and determination of prices.