This study presents improved methods for long-term forecasting of mineral demands. Intensity of use, both in its simple, original form and as described by richer economic relations is one such method, particularly when intensity of use is estimated using rigorous statistical methods. Additionally, this dissertation explores the implications of the learning curve for long term forecasting of mineral demands. This study begins by considering the empirical evidence which applies when a learning curve is present. Then, if a learning pattern is present, the learning model is used to examine an economic measure for specified levels of economic activity. This dissertation also provides some empirical results on the learning curve in mineral industries and demonstrates how the learning model can be used as an economic forecasting tool. As an alternative to the intensity of use and learning models, there is a vector model, either using time varying coefficients or expressed as a transcendental function, to capture dynamics. This model estimates the time varying parameters from the vector space instead of the variable space. The major advantage of this model is that it honors correlations between variables. This is especially important in ex ante forecasting in which explanatory variables themselves must be forecast to obtain a forecast of the dependent variable.