Sunday, February 13, 2011

Outlier Correction with the Ex-Post Method

Entering content frame

Function documentation Outlier Correction with the Ex-Post Method Locate the document in its SAP Library structure


The ex-post method is one of two methods for outlier correction.


Touse the ex-post method you must have chosen a forecast model that generates an ex-post forecast. Forecast methods that do not generate an ex-post are for instance the weighted moving average method (method 14) and the adopt historical data method (60).
If you are using external forecasting, that is you have implemented user exit EXIT_/SAPAPO/SAPLMCPR_001, and want to use outlier correction, you must do so by implementing user exit EXIT_/SAPAPO/SAPLMCPF_003. See also: SAP Note 394076.


In this method the system uses the ex-post forecast to determine a tolerance lane. If a historical value lies outside this tolerance lane, the system views it as an outlier and corrects it.
This graphic is explained in the accompanying text
As opposed to earlier releases, after the system has replaced a value it recalculates the ex-post forecast and the MAD. It then uses the new values for the succeeding historical values.
You can decide whether the system corrects erroneous values to the ex-post value or to the relevant boundary of the tolerance lane.
Manual Outlier Correction with Historical Value Markings
This function is only available for the ex-post method.
You can assign a time series to a forecast profile either in the profile itself or in the forecast view of interactive planning, which is the recommended way. In the times series you can specify which periods are subject to outlier correction. If you enter a time series in this field, the system automatically sets the value of Sigma to zero. This means that the historical value in each period you select is changed to the ex-post forecast. This is in effect a manual outlier correction.


In Customizing you can change the settings for the ex-post method. If you do not, the system uses default settings. For more details, see below.
In the univariate forecast profile or directly in univariate forecasting (on the Settings tab page) select Ex-Post Method in the Outlier Correction field.
When you execute a forecast, the system:
       1.      Carries out a forecast with the chosen model using the ‘raw’ historical data. It calculates an ex-post forecast for all historical values.
       2.      Checks the periods in the initial phase (Customizing setting, default 3 periods) against the ex-post forecast. If the historical value is outside the tolerance lane the system corrects the value to either the ex-post value or the nearest boundary of the tolerance lane.
       3.      After the initial phase, if it has found and corrected an outlier, the system carries out another ex-post forecast and calculates MAD again. You can specify whether the system uses the local MAD (the value calculated up to and including the current period) or the global MAD. You can also specify in Customizing how many periods have to be free of outliers before the system calculates the ex-post forecast again. The default is 2.
You can save the above settings user-specifically in Customizing under Advanced Planning and Optimization ® Supply Chain Planning ® Demand Planning ® Basic Settings®Maintain User-Specific Settings for Outlier Correction. If a user does not have their own settings, those of DEFAULT_USER are used. It is possible to change these system defaults. For more information, see the IMG and the F1-help for the individual fields.

Leaving content frame

No comments: