Our objective is to develop an econometric theory applicable to macroeconomic forecasting. Such a theory must, therefore, allow for non-stationary processes which are subject to intermittent structural breaks, using models that may differ from the mechanism which generated the data, and that are selected from data evidence. In this volume, we primarily focus on the more tractable case without structural breaks, but note any important differences which result when there are breaks. We begin by setting the scene, and then we review the checkered history of our subject. We develop a framework within which to analyse economic forecasting, then consider alternative forecasting methods. The final section examines forecasting in an artificial (computer-generated) process as a precursor to the remaining analysis.