Het voorspellen van een seizoengebonden vraag.
Report 91.3.TL.2885, Transport Technology, Logistic Engineering.
Forecasting the demand of a product is very important for every business
(industry) which is dependent on the behaviour of the market.
The seasonal variations of that demand can cause problems in the whole
organisation; in the field of production, inventory control and planning. This
makes forecasting of the periodic variation of the demand an imporant issue.
At first the potential savings ,which can be achieved by making a better
forecast, must be taken into account. The profits must exceed the costs and
Many mathematical models have been developed to describe time series for the
demand of a product. These time series can be used to make a forecast for the
future demand. The mathematical models make use of 'exponential smoothing'
techniques, causal relations and 'decomposition' techniques. It is important
to reduce the number of parameters used in a forecasting model. This will
minimize the calculation time and improve the transparency and the usage of
the forecast method.
To make a good forecast, the following aspects are important:
The seasonal and periodic variations of the demand of a product can have
several causes. These variations can be corrected in the forecasting model in
- sufficient data from the past to base the forecast on;
- experts with knowledge of the market should interpret the forecast;
- the demand of the product must not show too large variations and
unexpected changes, unless these changes can be corrected well and in time
by the forecasting model.
In general the corrections are made with help of one or more of the following
The following circumstances influence the choice of a method to forecast the
- additive seasonal components;
- multiplicative seasonal index;
- goniometric functions;
- seasonal adjusted 'dummy variables'.
It appears that there has to be a search for the best forecast method which
depends on the specific circumstances of the present situation.
- the demand forecast method used;
- the forecast horizon (number of periods ahead);
- the length and pattern of the available time series;
- the available time and budget;
- the accuracy desired;
- the knowledge of the causes of the periodic variations.
Practice and research shows that simple extrapolation performs often as well
as complex forecast methods. Besides this, combining different forecast
techniques can result in better forecasts.
Different factors play a different role in a specific forecast situation with
particular aspects. So the performance of a forecast method depends on the
specific situation. This explains the results of investigations in the
literature which often do not agree.
Reports on Logistic Engineering (in Dutch)
, TU Delft