Abstract
This paper outlines the principles involved in the Box-Jenkins approach to forecasting. Attention is centred chiefly on the situation where future values of a time series are to be forecast exclusively in terms of its past record, although an extension of the method to allow inclusion of information provided by a leading indicator is briefly described. The Box-Jenkins methodology is outlined and illustrated by its application to a seasonal sales forecasting problem. Some practical experience in application is summarized, difficulties which may be encountered outlined and methods of overcoming them discussed.