cancel
Showing results for 
Search instead for 
Did you mean: 

MAPE Calculation as it relates to the ex post forecast

Former Member
0 Kudos

Hello,

When calculating the MAPE I understand the equation of acutual - forecasted/actual. What i am not fully understanding is how it relates to the ex post forecast. I understand the error total is calculated by using the ex post forecast results based on the system needs for initialization and the number of historical values. What i would like to know is how can i determine the number of historical periods that are aggregated into the MAPE error measurement?

Thank You,

Melissa

Accepted Solutions (0)

Answers (2)

Answers (2)

Former Member
0 Kudos

thank you

Former Member
0 Kudos

Hi Melissa,

MAPE: Is the mean absolute percentage error between the forecasted value and the historical value in the ex-post forecast.

Univariate Statistical forecasting is done based on corrected sales / demand history.

An outlier is an historical value that lies outside the tolerance lane. The outliers correction is an important activity.

There are two methods available for outlier correction:

1. Median Methof

2. Ex-post method.

In the forecasts profile maintenance, when we set Outlier Correction indicator as u2018Ex-post methodu2019, the system carries out an ex-post forecast twice: the first time when it performs outlier correction, the second time when it calculates the forecast based on the corrected history. The two ex-post values are not necessarily identical.

We can maintian customer-specific settings for Outlier Correction in IMG in APO:

Supply Chain Planning >> Demand Planning >> Basic Settings >> Maintain Customer Specific Settings for Outlier Correction

In this activity we can edit the settings for outlier correction with the ex-post method. These settings are saved user- specifically.

Number of Periods in Initial Phase:

In order to use the ex-post method for outlier control the system needs an initial period to create a smooth ex-post forecast. This ex-post forecast is the basis for detecting outliers later in the time series.

Number of Periods Without Outlier

In this field we define how many stable values the system must encounter before it recalculates the ex-post forecast again. For example, we leave the default value of 2 and the system encounters an outlier and corrects it. If the next two values are within the tolerance lane, it recalculates the ex-post forecast. If one of them is an outlier itself , it is corrected and if the next two values are not outliers, it recalculates the ex-post forecast then.

In the Master profile maintenace, we have setting for History horizon, we set how many historical periods we want to be used.

I hope this gives insight.

Please check and confirm.

Regards

Datta

Former Member
0 Kudos

I understand all of this . I understand outlier correction and it should have no bearing analytically on the matematical computation of MAPE. What i am trying to understand and maybe it is propriitary is what is the alogrithm for the MAPE calculation as it reltaes to an individual univariate profile? How is the aggrgation of the MAPE determind( how many months are used in the MAPE calculation)?How do i determine this? Is it based on the historical figure that i define in my model profile or is it some type of algorithm that is defined within the model or canned strategy? When you have a dynamic business that changes rapidly a MAPE over the entire data series sometimes is not relevent, however a MAPE of 3 or 6 months is relevent and can capture futre business. I would like to be able to understand what the models are aggregating for a MAPE.

Former Member
0 Kudos

Hello,

depending on how many months of history you provide to the forecast calculation:

- in the master forecast profile: history horizon

- in the univariate profile: option "without leading zeros", also model determines how many periods it needs to initialize itself

the system creates an expost with a given length

periods left = history horizon - zeros on the left - initialization period

MAPE will be computed for this periods, that is, for all periods where comparison expost / history is possible

The only way of controlling the precise number of periods used for the calculation I now of is to use a user-defined error calculation via BAdI and code it there.

hope this helps...