Short Term Forecasting: Horses for Courses

Within the British Gas Corporation the Headquarters Operational Research Department has assisted various clients with a range of short term forecasting problems. As a result, over the years considerable experience has been gained with the use of alternative forecasting procedures. The authors felt that it would be interesting to compare the usefulness of these alternative approaches, not as usual by their forecasting performance measured in terms of some statistical parameter, but rather in terms of some qualitative factors. Amongst these we identify the objectives of the forecaster, nature of the environment, ease of use and nature of the time series itself. Such factors we believe may well be more important than the forecasting performance as determined by strictly statistical measures. This paper discusses, in these terms, our experience of using alternative forecasting models for appliance sales and gas sendout forecasting problems.