Outlier Correction in the DMA Model For the forecast with the dynamic moving average (DMA) model, the system corrects the outlier before it calculates the forecast. The system does so in various ways, depending on whether it is correcting the outlier for the order item forecast or for the average demand size per order item (demand/item) forecast.
If the largest order is greater than the
Outlier Correction Parameter for Item in DMA Model
, and greater than the average of the forecast weekly order items plus k times the order item deviation, the system restricts the order to the average of the forecast weekly order items plus k times the order item deviation. In doing so it excludes the outlier both from the average of the forecast weekly order items and from the outlier deviation.
You can specify the
Outlier Correction Parameter for Item in DMA Model
in the forecast profile on the
Model Parameter
tab page. You get to the forecast profile on the
SAP Easy Access
screen under
.
You can enter
k
in the
Outlier Correction Parameter for Item
on the
General
tab page.
If the largest demand/item is at least the
Param1: Outlier Correction for Demand/Item in DMA Model
parameter on the
Model Parameter
tab page, the system limits the demand/item to two times the demand/item average, or to the demand/item average plus
k
times the demand/item deviation (not including the outlier).
You can enter
k
in the
Outlier Correction Parameter for Item
on the
General
tab page.
If the four conditions under Defining Optimal Amount of Historical Data in DMA Model are not met, the system proceeds as follows:
If both of the last two periods over 13 weeks are at least the
Param4: Relevant Periods for Demand/Item in DMA Model
parameter in the forecast profile on the
Model Parameter
tab page, the system carries out a t-test for this time span to determine whether there is a significant difference in the demand/item between the two periods.
The significance depends on the
Significance Level T-Test 13:13 for Demand/Item in DMA Model
parameter in the forecast profile on the
Model Parameter
tab page.
If the difference is relevant, the system uses the last 13 weeks. If not, it does the same check for two periods over 26 weeks. For this to happen, each of the last 26 week periods must be at least the
Param5: Relevant Periods for Demand /Item in DMA Model
parameter in the forecast profile on the
Model Parameter
tab page. The system now carries out a t-test for this time span, to determine whether there is a significant difference in the demand/item between the two periods.
The significance depends on the
Significance Level T-Test 26:26 for Demand/Item in DMA Model
in the forecast profile on the
Model Parameter
tab page.
If there is a relevant difference, the system uses the last 26 weeks. If not, it uses the whole 52 weeks for the calculation.
The following is true for each of the above calculations:
If the largest demand/item is at least as large as the
Param2: Outlier Correction for Demand/Item in DMA Model
in the forecast profile on the
Model Parameter
tab page, the system limits the largest demand/item as follows:
Average demand/item plus k times the demand/item deviation