MAPE Values after APO Stat Run --- Seasonality not picking the system

[Former Member]

Dear Gurus,

In our project we are using Forecast Strategy 56 for the APO Stat Run.

Due to System is not picking the seasonality, we tried to test with other Strategies with 54.

For that purpose we have taken large Parent SKUs and for those 10 parent SKU have created the selection and ran the Stat Run using 54 strategy.

Now my concern is after completion of APO Stat Run when I check spool checks I was not able to see for those whole 10 SKUs what is the overall MAPE value.

How to check the overall MAPE value in spool list job log or else in planning book.

And also pls provide some bright suggetions to pick the system seasonality.

Regards,

SN

[former_member254329]

Hi Subhan,

You cannot check overall MAPE from the logs table. The logs will only show you a single MAPE value for a given selection ID.You either have to build your own weighted MAPE formula through a DP macro, or a BEx query (if data is in infocubes) or else to get an overall result.

In order to help you picking up the seasonality can you answer the following questions?

Do you work in weekly / monthly buckets? What's a typical season for your industry? (ex: December to March) What are your settings for Periods per season in the forecast profile? Do you sometimes have lags before your season actually starts? How big is your history horizon?

Regards,

MZ

[Former Member]

Hi Subhan,

MAPE error is available in stat fcst log spool if it is ticked as forecast errors block in Stat Fcst profile(Univariate tab).

-Anup

[Former Member]

Subhan,

I am providing a generic answer such that it helps every body, esp. first timers.

Strategy 56, which in one of the automatic forecasting strategies in DP, runs a battery of tests on historical data. The result of these tests is a model whose forecast results in least MAD. While this may NOT ALWAYS be the BEST FIT model, you are not expected to adopt it without inspection. The eventual model proposed by the systems could be any of the constant, trend, seasonal or seasonal trend models. This is by far the MOST exhaustive test for arriving at a forecast model that system thinks is "best fit" but you may not think so, which is fine.

If strategy 56 does not propose seasonal model and YOU THINK it is not right, you can try forecasting with strategy 52 that tests purely for presence of seasonality or 53 that tests for both trend and seasonality or 55 that also tests seasonality.

The first place you should check is the messages tab of interactive forecast screen to SEE the messages you get and the forecast error calculated by the system for each of these auto tests. You will find MAPE in the forecast error tab of interactive forecast screen

If you STILL THINK your history is seasonal, and none of the automatic strategies (tests) reveal positive seasonality, you can directly run forecast using seasonal models 30 or 31 or 35 or 40 or 41 and DO note the forecast error calculated by the system in each case and also the messages. If there is no seasonality, you will see a message like  "seasonal test is negative"

The next place you should check is the table of forecast version comparison to compare the results of multiple forecasts ran on your selection.

Now, dump them all to excel and compare what model was "ideal" according to YOU.

Now tabulate all of these errors to know which one indeed offered lowest MAD or MAPE or whatever the measure you selected as forecast error.

You are still not bound by lowest MAD or MAPE as a basis to accept the forecast proposed by the system. You can PICK/ADOPT what you YOU THINK is right.

I hope this gives some directions.

More here

http://help.sap.com/saphelp_apo/helpdata/en/ac/216b6e337b11d398290000e8a49608/content.htm

Thanks

BS