How to find lowest percentage error in DP Forecast to bsiness?

[former_member309586]

Hi Gurus,

I just wanted to know how to get lowest percentage errors for my forecast and make business happy ?

I am using "Auto Model selection-2" and i wanted to have lowest percentage errors and what should i do for it?

What alpha , Beta and Gamma values need to keep in forecast profile?

Please let me know it will be really great

Thanks for the answers

Regards,

raj

[srinivas_krishnamoorthy · Accepted solution]

In the automodel 2, you can specify that MAPE be used a selection criteria for alpha beta and gamma. You can set this in Univariate Profile tab under Model Parameters section and Error Measure field. This will ensure that an appropriate forecast model is selected which you can check in selection assignment to forecast profile screen. This does not guarantee business happiness though, but gives a more quantitative route for a solution.

[former_member309586 · Accepted solution]

Hi Gurus,

Thanks for the email. I agree with you . I had run the forecast and see that MAPE is arround 98 to 1000% and i am not sure why this error is coming up and what are the factors which are effecting these percentage error.

I tryed to give value alpha=0.3 and Beta=0.3 and error is around 70-98 %

please let me know how to reduced my error percentage during forecast?

Thanks for your instant responce

Regards,

Raj

[srinivas_krishnamoorthy · Accepted solution]

Automodel is trying to fit forecast profiles with varying values of alpha, beta and gamma. If we have a large enough range of the three as part of automodel, it picks up the combination with least MAPE. I think the model is optimized with the proposed value of alpha beta and gamma.

If you need to reduce error further, you may want to do appropriate outlier correction or correct history for any one-time distortions or try to check other parameters like seasonality period etc. One may have to do lots of trial and error to come up with the parameters with least error. It may work for present period but not for next if the dynamics of underlying history numbers have changed.

In the sdn website, try to reach out for Articles section where you may want to read through focused articles on error measurements during forecasting and how to make the process more efficient. I do not think you may get any more info than that.

[Former Member · Accepted solution]

Auto model Selection procedure 2 (56) - Tests for Constant, Seasonal and/or Trend models while attempting to

minimize Error by adjusting Alpha, Beta and Gamma..

This model adjusts Alpha, Beta and Gamma between the values of 0.1 to 0.5 in 0.1 increments while calculating the Mean Absolute Deviation (MAD) for each Iteration. The iteration with the Lowest MAD is the Model chosen with the appropriate smoothing factors.

MAD is calculated by the absolute difference between the "Ex-post forecast" and the "Actual value". The MAD is the error calculation used to determine the best fit in the 'Auto model 2' forecast model. Must use Outlier Correction as 'Ex-Post Method'.

Though this model gives more precise results, not recommended to run in background as it carries detailed tests with longer time.