IBP Demand Planning Statistical Forecasting - How to select Alpha,Beta,Gama?

[Former Member]

Hi,

We are currently on Release 4.0 Support Pack 2 and attempting to get the most out of the statistical forecasting tools that are avaialble. Considering that the only outcomes are Linear, Trend or Seasonal generating consistant reliable forecasts appears to be a hit-miss approach.  All Alpha, Beta and Gama entries need to be manually entered for IBP to then select the optimum result.  Does anyone have a solution that can use historical data to generate an IBP starting point for the number of historical months / A-B-G parameters?

Thanks

John

[Irmi_Kuntze]

Dear John

first before I come to your real question: I will not be able to provide you with a satisfying answer.

Second the remark that you should make sure that your input keyfigure must not have any UoM or Currency conversion. If it does have, create a calculated KF that just copies from your input

Third remark, statistical FC will be executed on the Base Planning level. If you want to have execution on a more aggregated level, than again: Provide a calculated one that copies on aggregated level the real input

And fourth remark: In statistical FC you have some law of big numbers. Don't try to execute the FC on the level of your product-customer-location, if not really necessary. Usually (depended on your kind of business / industry), the data input to run on Customer level is not sufficient for any data model. Run on aggregated level of e.g. Product-Location  or Product - Customer Region.

Without these basics, you will not get a proper result, no matter how good the models are.

_________________________

Now to your question: Answer for you is "No, not yet".

In an On-Premise-Solution you would probably be able to see some of the results in HANA, but you are on release 4, not on 3, so I assume you don't have access via HANA studio

As Product Development Team is aware of the difficulties, they are working on the Demand module which will come with IBP 5 / 6.

There you will have much better possibilities, not only in the restricted number of models but as well in regards to usability and evaluation.

You will have e.g. Pre-Processing- and Post-Processing Steps possible (e.g. substitute missing values by median or mean), Ex-Post Forecast, error measures such as MAPE, MSE, MASE or WMAPE etc, and you'll have a bunch of Forecast models .

Until than, experience does help a lot... In good old APO-DP-times, default values for alpha, beta, gamma had been 0.3, so that in any case is a good starting point. Elsewise, it is indeed quite some trial & error.

If anyone know some tricks, please post!

[Former Member]

Hi Irmhild, Your response provides me with a very good update and understanding of the current Statistical Forecasting status.  Our SAP contractor is using HANA studio to access the results and has made some progress at establishing alpha Beta Gama parameters, but more importantly whether a product should use 12-24-36 months of history  .  Experience is important.  FYI we are forecasting at the aggregate level and then using a 3-month proportional disaggregation to generate a demand at the customer/product/location level

[Alecsandra]

Hi John,

When you run the statistical forecast, the system picks the best set of values for these constant parameters. So you can define several values and the optimal one that returns the least error for the given data set is chosen.

Rgds,

Alecsandra

[Former Member]

Hi Alecsandra,

Technically you are correct. The issue centers around the fact that the Demand Planner needs to firstly manually establish the actual value parameters that will be pick-up for evaluations.  We have  approx 70 entries covering the SIngle, Double and Triple parameters. There does not appear to be any statistical forecast analysis tool to assist in determining whether these 70 entries are adequate or if the number should be 100, 140 etc.  Additionally, should the profiles use 12, 24, 36 or 48 months history to provide the least error result.  It all seems to be trial and error when it comes to establishing parameters and profiles - are you aware of any tools to assist in this area?

Rgds

John