|
Demand Forecasting Example There are a great many similarities when forecasting demand for all kinds of products and services. While you will probably use different model inputs, the principles are exactly the same. To forecast future sales, it’s necessary to understand what drives demand. This example shows how the neural net approach can be used to model a business which is at the mercy of macro-economic rather than market forces. This is not always the case, as often the factors affecting demand are under the business's control. This true-life example concerns a wholesaler supplying a fragmented customer base in the UK. He experienced declining sales between 1990 and 1992, owing to the economic recession during the period. In such strenuous times, an accurate picture of future sales was vital to predict optimum stock levels. One factor which was thought to drive demand was the general health of the economy—whether it was booming or receding. A suitable and readily available economic "thermometer" is the Coincident Indicator. The wholesaler believed that interest rates were a second factor affecting his business—because when interest rates were low, dealers tended to buy in bulk direct from manufacturers, rather than from him. Finally, the number of working days in the month was a third factor influencing sales. These three driving factors, and the sales figures themselves, were collected for each month from January 1988 to May 1992, and used to build a neural network model. Once built, the model could be dissected to quantify precisely how the business is affected by each of the different factors:
Note, in particular, how the neural network was able to deduce the unusual result that sales go up when interest rates go up. This is because wholesalers have to compete with their own suppliers—when interest rates are low, customers tend to buy in bulk from a manufacturer rather than in small quantities from a wholesaler. Although the wholesaler has no control over any of these factors, he can use the model to improve his business by forecasting future demand. This allows him to estimate how much stock he will require in future, and enable him to minimise working capital. To forecast future demand, future values of the input variables are needed. The number of working days per month is easily established; forecasts of the coincident indicator and interest rates are readily available. (Alternatively, 4Thought can forecast them for you!) Future demand can then be forecast.
While the forecast isn’t the best of news for the wholesaler, it will enable him to manage his business in the most cost-efficient manner possible during a very difficult period, through better control of stockholding, distribution vehicles and manpower. Additional Cases:
|
