Forecast Errors

[Former Member]

Hi Experts

1)How MSE , RMSE , MAPE ,MPE works in generation of the forecast.

As there is nothing mentioned in SAP HELP except Formula.

How these errors influence in generating accurate Forecast.

2)Wat error is given much importance to consider the accuracy of the forecast and model fit.Because when i am generating forecast for one forecast model , MAD is less and Other model MSE is less and some other model MAPE is less and Mad is high.Here in this context what model should be considered.

with regards

sai

[Former Member · Accepted solution]

Hi Sai,

1) Refer to this [Univariate Forecasting|http://help.sap.com/saphelp_scm50/helpdata/en/ac/216b6b337b11d398290000e8a49608/frameset.htm]

The system looks at the history and plots some dots into future as in the figures shown in the above link. The system then trys to join this dots and get a kind of trend like constants, trend or season etc. But if you want to draw line or curve, not all dots can be linked. Some of them lie up or down the average line as shown in the sap help link.

It is the variation of these points from the trend line is variance and every other values are derived from this.

Lets say the system found upward trend as the possible model,

Here is how the calculations are made:

Actualpoint -

---

averagelinepoint-variance-%variance-Absolute error(AE)---%AE

12_____________10_____________2______28.57______2_______________18.18

15_____________12_____________3______42.86______3_______________27.27

16_____________14_____________2______28.57______2_______________18.18

17_____________16_____________1______14.29______1________________9.0

16_____________18_____________-2______28.57_____2________________18.18

21_____________20_____________1______14.29______1________________9.0

___________________________________________________________________________

SUM--

---

7

---

11--

---

99.81

___________________________________________________________________________

Total of variance(Error): 2321(-2)+1 = 7

% variance or % error = (2/7)*100 = 28.57

Average Erorr = sum(variance)/6 = 7/6 = 1.16

Mean Absolute Error = avg(MAE) = 11/6 = 1.83

Mean absolute Percentage Error(MAPE) = 99.81/6=16.636 and so on for the other errors like MSE etc.

There is no direct explanation like the above in SAP HELP.

2) MAD or MAPE is usually the benchmark. How ever it is up to the business that decides which error value they want to use. Comparing their business pattern they may find that when the MAD is low the forecastign profile used in more appropriate for our business.

[Former Member · Accepted solution]

Hi Venkat

Really appreciate for your immediate reply.

1)Can you please elaborate more on MAPE and MAD.

MAD or MAPE is usually the benchmark. How ever it is up to the business that decides which error value they want to use. Comparing their business pattern they may find that when the MAD is low the forecastign profile used in more appropriate for our business.

2)Please throw some light on MSE,RMSE?

[Former Member · Accepted solution]

Hi Sai,

1)I hope you read the whole explanation and understood it. MAD or MAPE can be used as a forecasting error measure. Lower the error better the model. The business users can decide which error to use. If they have no clue just ask them to use MAD.

2)

Actualpoint -

---

averagelinepoint-variance-%variance-Absolute error(AE)-%AE----SE

12_____________10_____________2______28.57______2_______________18.18___4

15_____________12_____________3______42.86______3_______________27.27___9

16_____________14_____________2______28.57______2_______________18.18___4

17_____________16_____________1______14.29______1________________9.0____1

16_____________18_____________-2______28.57_____2________________18.18__4

21_____________20_____________1______14.29______1________________9.0____1

___________________________________________________________________________

SUM--

---

7

---

11

---

99.81-

___________________________________________________________________________

SE- Square of Error (variance)

MSE= mean Square error = (49414+1)/6 = 3.833

RMSE = Root Mean square error = sqrt(3.83) =1.96

In my experience, I haven't seen a user even looking at this error value for forecast analysis...theya re usually happy with either MAD or MAPE.

For more detailed explanation of these errors try reading any statistics book.

[Former Member · Accepted solution]

Hi Venkat

Thanks for your detailed explanation.

Now i am clear wrt forecasting errors.

with regards

Sai