How can we do data scrubbing?

I mean the sales data of a consumer electronic industry, which includes promotion, seasonality, trends, etc, and this needs to be seperated.

Any ideas are welcomed.
 

When you say data. What is the source of this data. You would need to use statistical models to estimate the seasonality, trend, etc.
 

When I say history data, it is meant that it is actual sales data from the execution system. I can't use this data directly as base for forecasting. I need to first decompose it, and then use it.
 

Thanks for clearing that up. I pressume the transactional data has mixed in it sales that where a result of promtional activity
etc. Unless those sales in the execution system have some indicator of the type of sales, you will have to use statistical methods to estimate promotional sales, etc.

Unless you need to have separated data for trends and seasonality for reporting purposes, very specific forecasting methods or similar, you don't need to separate it. Depending on the statistical model used, it needs to have the actual sales with the trend and seasonality included to be able to estimate the forecast. In some way the model is able to extract the trend
and the seasonality from the sales, you don't have to separate them in different key figures (in general).
 

For promotions it's different. In this case you need to clean the history. 
General rule to use about what type of information you need to clean from the actual sales is all the effects that you not expect that will be reproduced and that the statistical model is not able to consider. This includes promotions but can also be an unexpected hot summer that has increased sales too much, canibalization from other products, ...

Some common ways to clean the history are:
- use the outlier: it will clean all the sales history that will be outside some tolerance values. I personally don't like too much because you loose too much information about the actual sales.
- for promotions: if you have been able to estimate the effect of a promotion and you have this information you can clean directly the history.
If you don't know the effect of a promotion but you know the periods on which you had promotions, you can use statistical methods to estimate them.
- manual correction

This was a very helpful information. Infact, we intend to use an advanced forecasting method, that shall decompose the data into trend, seasonality. But, for promotion we will have to devise a way.

The trend and seasonality data gets seperated internally through the forecasting heuristic (no link with the SNP heuristic obviously). 

Firstly, Model initialization happens where the system determines the necessary model parameters for the chosen forecast strategy. 

Based on the model, the basic value, trend value and seasonal value is determined for each historical value during the model
initialization phase. The system is able to do this because it is able to identify patterns in the historical data eg., is there a peak
or trough in historic sales repetitively for a particular data set - this can give it the seasonal index, and similarly for trend.

Following this the indices are used for the future periods to forecast. So you really have to do little to seperate the seasonal
and trend values. 

Promotion planning is a differend ballgame though.In order to analyse promotion patterns in historic data there is a procedure to be followed.

On the Promo Planning desktop - Choose the object view Historical analysis from the workspace toolbar and enter here the historic period for which you wish to decompose promotion - The system applies linear regression to the historical data, thus creating an ex-post forecast, and displays the results in the row Ex-post estimate. The difference between the historical data and the ex-post estimate is displayed in the Promotion pattern row. 

That was really what I was looking out for. You are correct about decomposing data. 
May be, I will require your help on promtion segregation as I have never used it. I will try this out and come back to you. In the mean time, if you could provide more information/steps to seperate out promotion effect, that would be very helpful.
But, all the same, thanks a lot.

Get help for your SAP APO problems
SAP APO Forum - Do you have a SAP APO Question?

SAP Books
SAP APO Books  - Certification, Interview Questions and Configuration

SAP APO Tips
SAP APO Tips and SAP Advanced Planner/Optimizer Discussion Forum

Best regards,
SAP Basis, ABAP Programming and Other IMG Stuff
http://www.erpgreat.com

All the site contents are Copyright © www.erpgreat.com and the content authors. All rights reserved.
All product names are trademarks of their respective companies.  The site www.erpgreat.com is in no way affiliated with SAP AG.
Every effort is made to ensure the content integrity.  Information used on this site is at your own risk.
 The content on this site may not be reproduced or redistributed without the express written permission of
www.erpgreat.com or the content authors.