One statistical technique for short-range sales forecasting, exponential smoothing, is a type of moving average that represents a weighted sum of all past numbers in a time series, with the heaviest weight placed on the most recent data. To illustrate, consider this simple but widely used form of exponential smoothing —— a weighted average of this year’s sales is combined with the forecast of this year’s sales to arrive at the forecast for next year’s sales. The forecasting equation, in other words, is next year’s sales = a (this year’s sales) + (1-a)(this year’s forecast) The a in the equation is called the “smoothing constant” and is set between 0.0 and 1.0. If, for example, actual sales for this year came to 320 units of product, the sales forecast for this year was 350 units, and the smoothing constant was 0.3, the forecast for next year’s sales is
(0.30)(320) + (0.7)(350) = 341 units of products
Determining the value of a is the main problem. If the series of sales data changes slowly, a should be small to retain the effect of earlier observations. If the series changes rapidly, a should be large so that the forecasts respond to these changes. In practice, a is estimated by trying several values and making retrospective tests of the associated forecast error is then chosen for future smoothing.
Evaluation of past sales projection methods The key limitation of all past sales projection methods lies in the assumption that past sales history is the sole factor influencing future sales. No allowance is made for significant changes made by the company in its marketing program or by its competitors in theirs. Nor is allowance made for sharp and rapid upswings or downturns in business activity, nor is it usual to correct for poor sales performance extending over previous periods.
The accuracy of the forecast arrived at through projecting past sales depend largely upon how close the company is to the market saturation point. If the market is nearly 100 percent saturated, some argue that it is defensible to predict sales by applying certain percentage figure to “cumulative past sales of the product still in the hands of users” to determine annual replacement demand. However, most often the company whose product has achieved nearly 100 percent market saturation finds, since most companies of this sort market durables or semi durables, that its prospective customers can postpone or accelerate their purchases to a considerable degree.
Past sales projection methods are most appropriately used for obtaining “check” forecasts against which forecasts secured through other means are compared. Most companies make some use of past sales projections in their sales forecasting procedures. The availability of numerous computer programs for time-series analysis and exponential smoothing has accelerated this practice.
