Long-range metal consumption forecasts using innovative methods

Brazil's aluminium consumption in the year 2000 is forecast to be in the range of 800 000 to 1 300 000 tonnes; these bounds correspond to annual rates of growth in GDP of 3.6% and 6.1% respectively. This range of consumption conforms closely to that produced by an extensive translog consumption model which includes GDP, price of aluminium and its chief substitute, copper, and time as a proxy for technical change in aluminium using products and product technology. Analysis of various models by ex post forecast errors on a test period showed the most accurate models to be a simple learning model for cumulative IU and a time varying coefficient consumption model, followed by the extensive translog consumption model. The models with the largest ex post errors included the linear IU, lognormal IU, and extensive learning IU models.