It has to be borne in mind that it is demand that is being forecast. Averaging past sales is all very well, but it says little about the marketplace at the moment. If sales tend to be fairly static, and the averaging period fairly short, this may not matter much. Even so, the fact that sales have been static in the past is no guarantee that they will continue like that. However, it is relatively easy to amend the averaging process.
The most basic method is to attach different weightings to the various months in the averaging calculation. Instead of each month being awarded an equal say in the calculated average, "older" months are given lower weightings and more recent months progressively higher ones. The level of sales last month thus figures more prominently in the forecast than that of sales six months ago. Selecting the weightings which give the best forecast may require a little trial and error, but the basic guideline is simple: the stronger the trend, the higher the weight that should be given to recent months. The forecast can then be seasonally adjusted, if appropriate.
A far more commonly used technique is exponential smoothing. This does the same thing but in a rather neater and more efficient manner. Rather than allowing weightings to be chosen by management - usually in a fairly arbitrary way - exponential smoothing uses evenly (geometrically) decaying weights. All management has to do is select the "alpha" factor that determines the "height" of the decay curve. An alpha factor of 0.15, for example, means that 15% of the new forecast value comes from last month's sales, with the other 85% coming from previous months. Because particular months do not have individual weightings there is no need to hold them in the computer, which not only saves on storage but also speeds the calculation. It also means that all of the sales history is utilised.
But this method, too, can be improved upon. Double exponential smoothing is a variant which is usually employed when growth is high; triple exponential smoothing caters for any cyclical factors. There are also techniques incorporating tracking signals that help in detecting errors, not to mention still more sophisticated procedures, such as Box-Jenkins analyses, which replace exponential smoothing itself. However, walking comes before running. For the companies still at the Bloggs and Company stage, exponential smoothing might be a realistic aim for the near future.
Up to now the intensely theoretical and mathematical nature of forecasting techniques has generally ensured that their use was stifled at birth. These days some companies are lucky enough to have integrated computer-based sales order processing and production control systems with an in-built forecasting module. Even these are not the whole answer: there is also a need for desktop-accessible tools which might do for forecasting what WordPerfect and Lotus 1-2-3 have done for word processing and spreadsheets.
But software companies respond to the market. Consider depreciation and discounted cash flow calculations, for example. These are no less mathematically complicated than exponentially smoothed forecasts. Nevertheless, spreadsheets contain a whole panoply of built-in depreciation and DCF tools, and no built-in forecasting functions. The irony is that the construction of forecasts and projections is among the most common uses of a spreadsheet.
Companies need to wean themselves off of their old-fashioned methods of forecasting. They simply do not deliver sufficiently accurate results. It is time that managements woke up to this, and started demanding the tools with which to do the job.
(Malcolm Wheatley is a freelance writer.)