How is corrected history calculated?

[Former Member]

Hello Experts,

I have a situation where the sales data for all calendar month/year in a horizon may not be available at the lowest level of CVC. Say for example, I am working with customer, location, product, region, branch, sector as characteristics, then for a particular product for a definite history horizon, the sales history may not be available for each month of the year at the customer + location + product + region + branch + sector level.

Hence in order to get a decent forecast generated out of system, I need to fill the gaps and hence rely on 'corrected history' functionality of APO.

Just wanted to know how does the system fill the gaps as corrected history where it does not find a value? Can we influence the behavior?

How to account the corrected history to generate forecast rather than actual history to generated forecast?

As I know, in the forecast profile creation for univariate forecast profile, under section 'read historical data' if we expand the button for 'additional settings' we get the option to read corrected history from planning version and also the option to correct the history. However, when the first forecasting run happens, the planning version does not have any corrected history and hence forecasting does not give any result. In that case we run the first iteration without the check box selected, i.e. on actual history. Later we move to run forecast on corrected history.

Is my understanding correct here?

Regards

Manotosh

[rajkj · Accepted solution]

Hi Manotosh,

In your key fig settings (Planning area maintenance menu option Extras -> Forecast Settings), you can assign key fig for 'Corrected History'. This will be used by your forecast profile. Based on your settings (Adjust Corrected History), the system substitutes the gaps in corrected history from original history.

Then, you have another feature called model reinitialization (applicable for exponential forecast models). This allows you to define a strategy to treat the gaps in your original history flexibly. You have 3 options-

1. Copy history from previous periods

2. Use constant model - replicate one constant number in all the periods where gap exists

3. Allow use of older seasonal indices - calculate seasonal value based previously calculated seasonal index and average value.

Thanks,
Rajesh

[Former Member · Accepted solution]

Hi All,

just a further extension to the above situation. Consider that I have a history horizon as 12 M. History is available for first 6 M and not available for next 6 M. Found that system does not give any corrected history and hence does not give any forecast.

On the other hand, if there are intermittent gaps in history (and not a constant gap) system suggests corrected history and hence a forecast.

Is there a range/limit/cut off of blank history periods after which system does not suggest any corrected history and hence a forecast?

Also, We are using two KFs as history and forecasting based on them . One history as sales order qty and another history as invoiced qty. Can I have two KFs for corrected history corresponding to each history KF? As I understand, we can assign only one KF as corrected history (9AVCORHIST or a custom KF) in Planning area definition for forecast settings. Is there a way that I can use two history KFs and corresponding two KFs for corrected history and base my forecast on these corrected history KFs?

Any pointer would be of great help.

Regards

Manotosh

[rajkj · Accepted solution]

Hi Manotosh,

To generate forecast, system needs history and number of periods required depends on the forecast model/strategy you selected as well as the available historical data. There is no property or config setting available to control this behavior.

You can have multiple history and corrected history key figures. To have this flexibility, don't assign any forecast related key figures to your planning area (Extras -> forecast settings).

Define different forecast profiles and assign a corrected history key figure to the profile as shown in the following screenshot.

Thanks,
Rajesh

[Former Member · Accepted solution]

Hi Rajesh,

Thank you so much for your revert. Makes the concept clear.

Thanks again.

Regards

Manotosh