INTRODUCTION

At the present time, a good deal of effort and expenditure is being spent on the production of economic forecasts. There exists a wide range of forecasting techniques, from sophisticated regression procedures to naïver models or intuitive guesses, and it would seem essential for forecasters to attempt to assess the success or otherwise of such exercises. In this context, the two relevant questions are how good (in some sense) is a particular set of forecasts and can the forecasting procedure be modified to yield an improved performance. One element of the first question might be an enquiry as to whether or not better forecasts are being achieved by some other forecaster. Most of the available techniques for evaluating forecasts are almost entirely concerned with discovering the “best” forecasting methods from some set or, equivalently, in ranking methods. We shall argue that much of this work has little or no operational significance and that a wider viewpoint is required.

An analysis of forecast performance can be carried out at two levels. At the subjective level, one might look particularly closely at any very large forecast errors which have been made, and attempt to assess in retrospect what underlying conditions led to such errors and whether or not such conditions could have been foreseen at the time the forecasts were made. If they could, then allowance for these possibilities should be made in future forecasting exercises.