Abstract:

Our main objective has been to build a model which takes into account both the schedule of the different channels and the competition between them, forecasts reasonably well and with less error than extrapolation, is easy to use, can be used on different targets, can make short term or medium term forecasts, is relatively inexpensive, and is, therefore, useful to those who need forecasts. To this end, the method uses recent audience data in which the television programs have been coded. We shall focus here on the model built in France, but the method has also been applied to data from other countries. The forecast of each TV program is obtained from the combination of two forecasts: a total audience forecast and a share forecast. While the total audience is forecasted by extrapolation of the recent past, the share forecast is obtained from the combination of three basic models: an "intelligent" extrapolation" model, a "share model" (both of which take into account the program schedule of all channels), and a simple extrapolation model (which does not take into account the schedule). The model has been installed at France Television in the spring of 1993. We have found that it reduces to a fair extent the average absolute error obtained by simple extrapolation. Furthermore, we are working on improvements of the model. The interaction between the user and the forecasting model is done under Access. As a result, the whole system is quite user-friendly because Access is used both to prepare a schedule and to retrieve the forecasts.