Periods per season field in auto model 1

[Former Member]

WE have decided to use auto model 1 , strategy 53, for a selection ID.Historical horizon is 52 weeks. What value should be used in periods per season field. The data pattern is purely random. Sometimes it shows trend, sometimes linear regression, and other times constant, and so on...so we decided to use auto model 1, as clear grouping of material/location combinations was not possible. I am confused between the values of 1 and 52, what should i use.

[satish_waghmare3]

Hello Shweta,

Please refer these two threads on similar topic - "Seasons Per Period"

   and

As all the Univariate models are basically Time Series models.  Time series models use patterns from the past to try and predict the future(extrapolation). These models required number of periods for forecast model initialization and to identify the data pattern, and so it has an impact on history horizon.

Following is how I have understand model initialization period required by forecast models -

Constant Model requires ...1 Period...because it has to just use that one period and then extrapolate into future.

Trend Model requires - 3 Periods...to know whether trend is increasing(upward) or decreasing(downward) minimum 3 periods are essential.

Seasonal - 1 complete season...as this will tell the system how the demand is shaping up, which all months it rises or contracts and which all months it flattens..

With same logic for Trend + Seasonal model requires....1 complete Season and 3 Periods.

Hence when you will use Automatic model selection(strategy#53) you will likely to receive an error message not enough history data.

Model Initialization - Forecasting - SAP Library

How strategy 53 works, please refer below link :

Automatic Model Selection Procedure 1 - Forecasting - SAP Library

Hope this will help.

Thank you

Satish Waghmare

[Former Member]

Shweta,

Your confusion is well placed.

Since you have only 52 weeks of history, the periods per season can only be 52!!. SAP should have made this a bit more transparent by giving an option to configure seasonal cycles per n history periods. e.g. Nov to Jan (13 weeks) can be a highest demand season but May to July mid  (16 weeks ) may be lowest demand season. Again July mid to Aug (7 weeks) may be high demand season but not as high as Nov to Jan.! If this is the case (which is pretty much true in most geos if the product is say Hot Chocolate), I do not know how SAP explains the seasonality with only 52 weeks history with 1 period (week) in number of periods per season. When using 3 years of history in months in my system, the number of periods per season used is 12. So you can juxtapose. We use Auto models too for some combinations.

Another alternate is something called Seasonal Planning in DP but I haven't laid my hands on it. As the name suggests there is probably something more useful in there to explain seasonality. I suggest you explore that option.

Thanks

BS