SES:
Single Exponential Smoothing
Exponential smoothing is the most common used and very effective method, it is surprisingly accurate in short-term forecasting, whereas in the SMA data are weighted equally, exponential smoothing give an exponentially decreasing weights over period of time.
where α is the smoothing constant.
This is the most widely used method, which valid for many situations when the most recent data is indicative of future, it is logical and easy to use than WMA. The main advantage of this method is that just a very small amount of historical data need to hold, to make the next forecast, all the need is the current forecast and a smoothing constant, in addition, a very simple calculation make the forecast time become extreme short, therefore, those advantages result in the widely used on short-term stocking and production work on computers.(Clark, 1969)
Against to the advantages, the most obvious issue is setting of the ‘α value’, this is usually be done by doing a trail and error test, such as MSE and MAPE, and find out the most low error ‘α value’. The other problem was pointed out by Chase (2009), where SES only forecast one period ahead, which lead to the long range forecasts are normally “flat”. In addition SES also require the most recent data must be indicative for future. Therefore, as long as the recent data is indicative for future, SES can be a very powerful tool for short-term forecasting demand.
Example:
The graph shows that as the alpha value getting larger, the forecast result will be more responsive to demand, the smaller alpha value, the more stable forecast result will get.
The value of alpha factor depends on the how responsiveness is needed which is depends on
- the cost of losing sales if no product/service is available
- the cost of holding inventory
- the cost of changing output levels to meet excess demand
Copyright © 2015 | University of Warwick, Coventry | [email protected]
POWERED BY KASPAR CHEN