cancel
Showing results for 
Search instead for 
Did you mean: 

Automatic Procedure 2 generating negative values

lalit_behere
Explorer
0 Kudos

Hi all,

We are using Automatic Procedure 2 (Strategy 56) for Forecast generation.

Forecast Profile Settings are :

Forecast Horizon: 24 P (Months)

History Horizon: 36 P (Months)

According to the Planning Area Settings, for Forecast KF-Negative values are not allowed.

For few codes when I tried to generate forecast interactively its giving an error message:

“Negative values are not allowed for the forecast key figure”

While for other codes even if they have a decline trend, system is not generating negative values instead populating zeros.

Note:

There are 4 weeks in past for which History is not available. But these weeks do not fall into the History Horizon of 36 P.

Analysis


I performed no. of tests and found different results:

  1. 1. When I deleted all the values prior to those 4 missing weeks, system is generating the forecast Properly.

Message:  Automatic Procedure 2 called

                 Seasonal test negative

                 Trend Test Positive

                 Constant model selected.

  1. 2. When I randomly put some approximate values in those 4 missing weeks again system is generating the Forecast properly.

Message:  Automatic Procedure 2 called

                 Seasonal test negative

                 Trend Test Positive

                 Trend model selected.

  1. 3. When I reduces the history horizon from 36 P to 35 P, system is generating the Forecast.

Message:  Automatic Procedure 2 called

                 Seasonal test negative

                 Trend Test Positive

                 Constant model selected.

Q1. Why the system is generating negative values for few codes while populating zeros for others ?

Q2. Why those 4 weeks are impactig the forecast, even when they are outside the History Horizon ?

Q3. Why constant model is selected in some cases even if “Trend test” is positive ?



Please let me know if you need any further information. TIA.

           

Regards

Lalit

Accepted Solutions (0)

Answers (3)

Answers (3)

Former Member
0 Kudos

Hello,

For the benefits of the others consultants, please indicate the good or the helpful answer!

Thanks, Marius

former_member187488
Active Contributor
0 Kudos

Hi,

For Q1, Are the two sample codes have exactly the same historical data? I think the result depends on historical data only, instead of depending on codes (characteristic value).

For Q2, What is number of periods per season? It may require more historical data than the historical periods to initialize the model.

For Q3, You should understand that a test on a certain model is positive does not mean it will select this model. It only means it will go on to do further test on this model (iteration on different alpha/beta/gamma values and check forecast errors.) If a test is negative, no further test will be made on this model. Final model selection is based on forecast error comparison.

Best Regards,
Ada

Former Member
0 Kudos

Hi Lalit,

The automatic model selection give you a excellent result if everything is perfect. Please check in the activity, t.code  /SAPAPO/MC8T if the message 402 - Forecast values will be negative is active or no. Should not be active if you don't want to have negative values. 

For your issue, in the forecast profile try to work with Trend Dampening Using Phi Parameter. With this parameter you can reduce the trend, so you will not have negative values. You can choose Phi's value from 0.01 to 0.5. With 0.01 you will actually cut the trend. 

Please check this note too: 2088665

Hope that can help you.

Thanks.

Regards, Marius

lalit_behere
Explorer
0 Kudos

Hi Marius,

Thanks for the reply.

I would like to highlight some points here:

1. I am getting the message "Negative values are not allowed for the forecast key figure" while generating the forecast interactively not through background job.

2. The message 402 - "Forecast values will be negative" is already unchecked in activity.

3. When I plotted the graph for the values in excel, its an upward trend.

I don't think there are any negative trend values except the missing weeks values.

Q. Is there anything related to those 4 missing weeks in the past ?

Regards

Former Member
0 Kudos

Hi Lalit,

Try with "0" allowed. If is not working with "0" try to add 1 everywhere. For this you can use macro to populate the missing weeks in the past with "0" or "1".

Thanks.

Regards, Marius