MAPE greater than 100 % ?

[Former Member]

Hello Experts,

Can you tell me if the MAPE value can ever be greater than 100%, since it is a percentage value. After the forecasting run, we got these values way higher than 100%, like 20,000 and more....

Going by the formula, there might be a possibility for this to happen, if the expost forecast for is too large compared to history values,

Please anyone can tell me if the MAPE values generated can have values more than 100%, like 1000, 20,000 and so on........

Thanks

Susan

[Former Member · Accepted solution]

Hi Susan,

As you said , MAPE calculation is of mathematicl formula, any value is possible for that field. But the objective of forecasting is getting forecast results with minimal error values.

If you are selecting MAPE as one error measurement and if you are not sure of which is the forecast model can give you lowest MAPE, select auto model selection procedue strategies (51 to 56). Those should fetch you the best fit model with lowest possible MAPE for your historical data.

Best Regards,

Deepthi Naidu

[wilian_segatto]

Hi Susan!

Could you please elaborate more on your issue?

Give more details concerning transactions, scenarios, etc.

Thank you!

Will

[Former Member]

William,

We ran forecast on a selection ID that has all Product-Customer combinations using Auto Model 56. We ran in the bakground on Prod-cust level. The system has proposed a strategies like constant, croston, seasonal and so on.

We have set up a limit of 80 for the Error Metric MAPE in the Diagnosis group. If I am not wrong, MAPE is a percentage value and 80 should mean 80%. But we got alerts generated for almost all the Prod-cust combinations since the Actual MAPE values the system generated after using strategy 56 is abnormally high - in 10's of thousands like 2,956,009.830 and even 10,000,000.000 . These values are no where close to our limit 80. MAPE is the average of absolute percentage differences between Acutals and Expost forecast. So how could system generated MAPE be so high for all the combinations . Are we supposed to read it differently, because i thought auto model though may not propose the best fit model parameters, but atleast not give us a forecast which has such high MAPE values. I was expecting MAPE within 100 or worst case 200, 300 but not thousands and millions..

I am I missing something ?

Please help

Susan