Short and sweet: Forecast Accuracy 

Andreas Kemmner

Forecast accuracy describes the deviation of a forecast from the later actual values. Forecast accuracy is expressed by a forecast error, which measures how much a forecast deviates from the later actual values. A forecast accuracy and a forecast error always refer to a specific forecast period, e.g. a three-month forecast. In practice, forecast errors are often measured for different forecast periods. Tracking the forecast accuracy is expected to lead to a continuous improvement of the forecast on the one hand and to the identification of a possible systematic forecast error (forecast bias) on the other hand. A forecast bias in the form of a continuously too high or too low forecast value can be used to correct future forecasts.

There are numerous different formulas for measuring the forecast error, whose respective advantages and disadvantages are hotly debated in research and practice.

Our advice:

 Forecast accuracy as an assessment criterion for the quality of a forecast falls short in many cases. No matter how much effort is put into forecasts, there is always a residual of chaotic market behaviour and thus a forecast error that cannot be further reduced. For most items, there is usually no systematic forecast error either. Such systematic forecast errors are more likely to be found in human forecasts. If powerful machine forecasts are used (statistical statements; AI procedures), the algorithms themselves reduce the error as much as possible. Forecast accuracy thus only has informative significance but does not represent a control criterion for making forecasts continuously more accurate. In practice, we even observe the danger that the focus on forecast accuracy leads to mathematical overengineering without any practical benefit in most cases; chaotic behaviour cannot be pressed into a single forecast value. On the other hand, because it is possible to measure the breadth of chaotic market behaviour, safety stocks can and must buffer the remaining forecast inaccuracy. For this reason, it is important to let the system calculate safety stocks and not set them manually.

Andreas Kemmner

Autor | Author

Prof. Dr Kemmner has carried out well over 150 national and international projects in over 25 years of consultancy work in supply chain management and reorganisation.

In 2012, he was appointed honorary professor for logistics and supply chain management by the WHZ.

The results of his projects have already received several awards.

Weitere Beiträge | READ MORE