Ignoring partial demand while forecasting

[Former Member]

HI all,

We are in SCM 4.0. We consider past 36 months of demand to foreacst for next 24 months. In the last 6 months we had a huge spikes in the demand due to unexpected orders and now the demand is stable. I am looking for the ways to ignore the laast 6 months of huge demand for the future forecasting as it will have great impact on our future numbers.

Couple of options we are looking at to reduce the effect are:

1. Using Outlier functionality with various sigma values

2. Trying out with Trend Damping Profile and/or History damping profiles

However, we are looking for the possible ways to ignore the last 6 monhs demand completely

Your thoughts are well appreciated.

Thanks,

Sai

[Former Member · Accepted solution]

Hello Sai,

There are multiple ways to tackle your issue:

1.Maintain fixed dates in History horizon in master forecast profile

2.Maintain an offset of 6 months in history horizon in master fcst profile

3.Perform manual Sales history adjustments for last 6 months

I would have preferred option 3 since at any point of time you will consider 36 months of past history for stat fcst generation.

Thanks,Bopanna

[Former Member · Accepted solution]

Thanks Bopanna for your prompt reply.With the first two options we have to change the settings eveytime which we would like to avoid. However, we thought about the third option but we want to keep that as our last option since it is a tiresome process making the adjustments manually.

[somnath_manna]

One radical approach can be using Promotions.

See /people/somnath.manna/blog/2009/05/29/demand-forecasting-in-tough-times for further details.

Hope this helps.

Somnath

[ZoltanMBiro]

Hi,

One other option would be to use outlier correction as you suggested, but in conjunction with a Historical Value Markings profile. In that profile you would specify only those 6 months where your data is problematic. Thus the outlier correction would correct only these 6 months. Use a low sigma value, so that outlier correction is sensitive.

Kind regards,

Zoltan Biro.