Improved forecasting techniques

Exponential smoothing is the basic model, which can, of course, be improved. It splits demand into two components:

• average, which is recalculated each period

• random fluctuations about the averages.

The next stage of complexity is to add either:

• a constant increase or decrease in demand, using double exponential smoothing, or

• a seasonal variation of demand, using base series forecasting.

Alternatively, entirely different approaches to intrinsic forecasts can be used giving:

• unweighted average trends (regression analysis)

• profitability of changes in demand (Baysian forecasting)

• curve fitting (Fourier analysis)

• or more complex methods. (These are noted but not discussed as exponential methods are a better option.)

The mathematics can become quite complex and are usually left to the computer. In general, the more complex the model the more history is required which is why simple forecasting has been acceptable until recently. Intrinsic forecasts are best for data collected under similar conditions, even when the demand is erratic. Then it is quite possible that exponential forecasting with ? = 0.1, or even long moving averages are best. If a change from single to double exponential smoothing gives a significant improvement in the forecast for an item, then perhaps more sophisticated methods would work better, but there are fewer and fewer items benefiting from the better forecasting techniques. As a general rule forecasting can benefit more by improving the customer communications to detect changes...