Forecast errors

[Former Member]

Hi experts:

I have some urgent questions:

a. How does one decide as to the values to put in the forecast error upper limit in a diagnosis group? Is it the business users' or the DP consultant's responsibility?

b. Do all forecasting models use MAD to best-fit a model / parameters in the ex-post horizon or it varies by the FC strategy used?.

c. I know I can search DP tables using DP*  but can you recommend some DP related tables to become familiar with based on your experiences?

d. Any practical tips, best practices, gotchas to consider when using transports / TSOBJ in DP?

[rajkj · Accepted solution]

Hi,

Please find below my answers to your queries.

a. How does one decide as to the values to put in the forecast error upper limit in a diagnosis group? Is it the business users' or the DP consultant's responsibility?

• Typically, business should own this responsibility. DP consultant can help them to set the value (or define diagnosis group) and track the forecast error.

b. Do all forecasting models use MAD to best-fit a model / parameters in the ex-post horizon or it varies by the FC strategy used?.

• Forecast error method depends on your settings defined in forecast profile (t.code /sapapo/mc96b)

c. I know I can search DP tables using DP*  but can you recommend some DP related tables to become familiar with based on your experiences?

• If your DP environment is heavily customized, then it makes sense to know the DP tables. Otherwise, it is not required.

d. Any practical tips, best practices, gotchas to consider when using transports / TSOBJ in DP?

• This transaction helps you to collect all the related objects together and transfer collectively to the destination system. It is very much useful and kind of mandatory for transferring your completed development to upper landscapes.
• However, the subsequent transfers can also be handled individually. For instance, macro adjustments that can be managed through macro book interface.

Thanks,

Rajesh

[Former Member · Accepted solution]

Rajesh,

Thanks for the answers. My follow up questions are:

b. Are there guidelines as to which error measure (MAD OR MAPE) to be used for which F/C strategy?

d. Where can I learn about how to  use the TSOBJ transaction, meaning properly filling in the fields to collect / send subsequent transfers?

Regards,

[Former Member · Accepted solution]

b. for a single product, minimizing MAD or MAPE is almost equivalent. The main advantage of MAPE is that, being an dimensionless quantity, allows a comparison across different products: you can talk about a 100% error in product X and product Y - this would be MAPE - but an error of 35 units for product X and product Y is misleading, as you would require for a meaningful comparison to know what is ie. the average sales of each product. Both have roughly the same sensitivity to outliers. The SAPPRESS book (Inventory Planning with SAP SCM) recommends (more or less) to use MAD, MAPE... etc... to choose the best forecast, the use the standard deviation to optimize parameters. The discussion there on forecasting errors may help you to make sense of all these.

regards,

J.

[rajkj · Accepted solution]

Hi John,

I could not agree more with James's response. DP practitioners recommend using MAPE over MAD. Then, usage of t.code /sapapo/tsobj is fairly simple. It just collects the required objects and appends to the transport request. Even, if you make mistake, you can correct the transport request and reexecute the t.code. Probably, with a couple of executions you can master it.

Thanks,

Rajesh

[Former Member · Accepted solution]

James,

Thanks for your response. I have awarded the points to you and Rajesh.

[Former Member · Accepted solution]

Hi Rajesh and other experts,

The environment to be supported  is customized and please do advise of a handful of key tables and badi to look at in advance.

I know you have deep expertise in DP and your urgent help will be greatly appreciated.