0
$\begingroup$

I have my data for 10 years and I used ARIMA model to forecast data for 30 years.

Now, I want to measure the forecasting accuracy for the next 30 years.

How do I calculate each one of MAD, SMAPE, MAPE and etc....?

Each one of these formulas needs actual and forecast data. How could I do with 30 years forecast, where there is no actual data?

$\endgroup$
  • $\begingroup$ If there is no actual data to validate the forecasts against, then there is no way to assess forecast accuracy. You have to forecast data that you actually have (i.e. pseudo out-of-sample forecasting) to be able to assess forecast accuracy out of sample. $\endgroup$ – Richard Hardy Aug 24 '17 at 16:01
  • $\begingroup$ @Richard Hardy. Huge and Big THANKS. I will not use pseudo out-of-sample forecasting but from your comment I got the answer. $\endgroup$ – YAcaCsh Aug 25 '17 at 11:04
  • $\begingroup$ You are welcome! Good luck with forecasting! $\endgroup$ – Richard Hardy Aug 25 '17 at 11:19
  • $\begingroup$ Easy, just wait 30 years and you will see ;) Seriously though, the right approach has been mentioned in an answer, but I still wouldn't forecast 30 years into the future based on 10 years data. $\endgroup$ – David Ernst Aug 25 '17 at 13:24
  • $\begingroup$ @user7019377 :) heh. It is industrial problematic. There is no data enough and they need to forecast. I cannot say it will be precise but can be good enough. $\endgroup$ – YAcaCsh Aug 25 '17 at 13:30
0
$\begingroup$

To answer this question.

You need to train your model on the data that you have first.

Like train your forecasting model on 5 over 10 years data.

Then select the best state that you can use to evaluate the rest of data and forecasting accuracy.

Later, forecast data for 30 years.

In this case, you validate your model and you have the worthy case to forecast by.

$\endgroup$
  • $\begingroup$ Just be aware that you cannot train and evaluate the model on the same data. You should train the model on the training subsample and then evaluate on another subsample. Otherwise you will get overly optimistic evaluations. $\endgroup$ – Richard Hardy Aug 25 '17 at 11:20
  • $\begingroup$ Yes totally agree $\endgroup$ – YAcaCsh Aug 25 '17 at 11:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.