Composite Model

[Former Member]

Can someone help,

I would like to run a composite model with differnt historical parmaters. When i attempt to pull individul univariate profiles into the composite profile(univariate profiles that have differnt historical parametrs) I get no results or inapproriate results. Is this possibl;e in SAP system? Or am i defining the composite model wrong?

Any help would be appreciated

[Former Member]

Thank you and i read some documentation in one of the SAP books which helped and gave a good answer to this. I still think it is beneficial to be able to bring different historical parameters(different historical parameters relevent to individual univariate profiles) into a single composite profile. A best fit would be nice not based on anything other than lowest MAPE. Let the composite model choose the lowest MAPE not based on optimization of parameters or historical data, but based on what the individual planner requires to analyze data. Is this functionality possible in any of the SAP versions? I believe my company has an older version of SAP?

[Former Member]

I believe you have created a master forecast profile (which calls the composite fcst profile) and a composite fcst profile with Univariate profiles assigned to it.

To start, you can use the modeu201D method with smallest MAD/MAPE in the composite fcst profile. Using this method SAP would choose the model with least MAPE error.

I would also suggest you to execute this in foreground first and then try in background.

Pls let me know the exact error that you would face after you have executed the above sequence.

Reg,Bopanna

[Former Member]

yes,

I ran the model in that seqence. The numbers are innappropriate. My question is Can you have different historical parameters within a composite model?

For example lets say i have one univariate profile A. The historical parameters that i have assigned to this profile are 24 periods.

Next i created univariate profile B, this profile i have assigned 36 periods to the historical parameters. The next step was to create a composite model and pull in Profile A and Profile B. Since the historical parameters are different, does this matter ?

Within the master composite profile which contains A & B what do you assign to the historical parameters field , because the 2 univariate profiles have different historical parameters?

[Former Member]

Hi Melissa,

In Composite profile we maintain weighing factor for different Univariate or other forecasts profiles.

System will execute forecasting seperately with Univariate profiles A and B will happen as per the settings in A and B.

However, the final forecast calculated and displayed will be based on the consolidation of the forecasts based on the weighing factors.

I hope this is clear.

Please check and confirm.

Regards

Datta

[Former Member]

No I am sorry i am still confused. I assigned each profile an equal weighting factor . I would like to know if i have diffrent historical periods for the univariate profiles that are contained in the master composite profile, what do i assign for the historical periods in the composite profile. This feild is required to perform the algorithm. What are the forecast results based on? Is the forecast based on the individul univariate profiles parameters ( whatever the system chooses as the pick best) or are they based on the historical parametrs defined in the master conposite? I need to know what historical paramtrs are being used?

[Former Member]

You can have only one historical period assigned for the master composite forecast profile.But what you can probably do is play around with the alpha factor by giving values between .1 to .9 .For eg. For univariate fct A you can have an alpha factor say 0.8 which would give more weightage to recent sales history and for univariate fcst B you may give 0.3 which shall give equal weightage across the sales history.