A forecast should never introduce more variability than is actually in the sales.
A noisy forecast used to calculate stock levels will result in excessive stock builds, driving up stock handling costs.
Noise is often introduced into the forecast when forecasters simply chase the noise in the sales and try to forecast it, rather than forecasting the signals within the sales pattern.
Comparing the forecast and the sales
Comparing the variability of the sales with the variability of the forecast error gives a good indication of the accuracy of the forecast and verifies that the forecast is not introducing more variability than is in the sales.
- Where there are strong signals in the demand data you would expect the forecast to predict these well enough to make the variability of the forecast error very small.
- Where there are few or no signals in the demand data you would expect the forecast to just simulate the general noise, with variability similar to that of the demand.
|
A simple graph can quickly highlight when forecasts are introducing excessive noise:
- SKUs should fall within the green area, showing that their forecast error variability is less than their demand variability.
- If the opposite is true, and they consequently fall within the red area, the forecast is introducing more noise than is present in the demand.
- When variabilities are equal, SKUs fall within the gold area.
|
|
|