Whatever the specific definition, the common denominator of any kind of forecast is a reference to the future. This implies that all sources of uncertainty associated with describing present and past must also be associated with forecasting-and one more: the specification error inherent in the future dimension. In particular, this error should be associated with the distinction between causality and correlation, i.e, the understanding of behaviour, the necessary prerequisite for prediction. Thus, the key representational problem, the gap between model and reality, and the conditions for controlling that gap, becomes particularly evident in forecasting. This article is an attempt at clarifying some of the uncertainties most prevalent in forecasting, as a means to show with what respect they must be treated. The discussion relates to forecasting in general, but with some special references to air traffic. (C) 1999 Elsevier Science Ltd. All rights reserved.
