Forecasting Sales Data with Neural Nets: A Case Study. (with C. Chatfield) (1996)

Using the Chatfield-Prothero sales data as the main example, a variety of neural network (NN) models are fitted and the resulting forecasts are compared with those obtained from Box-Jenkins modelling and from Holt-Winters forecasting. The results suggest that there is plenty of scope for going badly wrong with NN models and that it is unwise to apply them blindly in `black-box' mode. Rather the wise analyst needs to use traditional modelling skills to select a good NN model, for example in making a careful choice of input variables and of an appropriate architecture. The BIC criterion is recommended for comparing different models. Great care is also needed when fitting a NN model and using it to produce forecasts. Unfortunately, even with such care, the NN forecasts were found to have disappointing accuracy in this case study, perhaps because of the shortness of the series.
Last modified on 07/30/96