cancel
Showing results for 
Search instead for 
Did you mean: 

Performing Outlier correction and not using it for forecast

Former Member
0 Kudos

Hello Experts,

Can we allow the system to perform the outlier correction(using Expost method) and still use the History that we have corrected manually in planning books?

For example: We have

KF1: Base History

KF2: Corrected History KF-9AVCORHIST

KF3: Corrections(+/-)

KF4: Final history

Final History = Base history + Corrections(+/-)

Is it possible to save the system Corrected history (KF2) which holds the values after outlier correction while still always using KF4 to generate the Forecast. The idea is that if we can first highlight where the outliers in the history , and then let the user have the control to decide whether to consider that as a outlier, if so they can make a manual correction in the corresponding bucket in KF3: Corrections(+/-). But always have the forecast based of KF4 with an outlier correction being done on KF1 and store the corrected values in KF2.

What I have observed is that, if there is an outlier correction made to the history, then the system always uses this corrected history to generate the forecast. In which case, we cannot use the manually corrected history. But if you use the manually corrected history(KF4), then the outlier correction is not performed at all.

Appreciate your response.

Thanks

Aston

Accepted Solutions (1)

Accepted Solutions (1)

Former Member
0 Kudos

Hi Alicia,

We are running APO 5.0, SP13, so my responses are based on that.

I don't think you can have the system natively do what you describe. When the system executes the statistical forecast, it reads the history from the key figure you specify in the forecast profile, performs the operations outlier correction, average number of days, lifecycle planning, promotion adjustment (if they are specified in the univariate profile) and writes the result to the Corrected History KF. It then uses the Corrected History KF as the basis of the statistical forecast calculation. So you can't keep 9AVCORHIST separated from KF4. The system will always use KF4 to calculate 9AVCORHIST.

But there may be a way to get the info you are looking for. You will need another KF. Call it KF5. Run the stat forecast with a generic constant model for your CVCs, using outlier correction (but no lifecycle planning, promotion adjustment or average number of days correction). Then run a background job to save (HISTORY - 9AVCORHIST) to KF5. The difference is your outlier correction. Now run your regular stat forecast process. Alternatively, you could save the difference directly to KF3??

Probably this is not the elegant solution you were looking for, but it's quick and dirty.

We do not set the flag "Read corrected history values from the planning version" for any forecast profile (transaction MC96B). If you do, I think this chucks my idea out the window.

Dan

Answers (0)