Automatic ARIMA forecasting

[ESVAL2016]

Good evening, for all
I want to use "Automatic ARIMA forecasting" in EViews9, and I am dealing with two-time series, one of them is non-stationary its represent the endogenous variable(Y), and another one its stationary series and its represent the exogenous variable(X), OK.

My problem is:
When I applied "Automatic ARIMA forecasting", to perform this model, I observed the non-stationary series not transformed to be stationary series for example(DY)!!

How to solve this problem in EViews programme?

Best Regards,

[ESVAL2016]

Hello, for all

Why are there a big difference in the result in Automatic ARIMA forecasting AND estimate equation !!? Which method is the right way??

Thanks for your help.

[EViews Gareth]

There should not be big differences.

[ESVAL2016]

For example,

My model (1,1,0)(1,0,1)12 by estimate equation, I have AIC = 7.347764 AND R^2 =38%
The same model by Automatic ARIMA forecasting, I have AIC= 7.500708 AND R^2=94%

I think its big difference !!

[EViews Gareth]

Was the same sample used? Automatic forecasting will drop more observations, usually.

[ESVAL2016]

Yes, I used the same sample in both methods !!
Also, Automatic ARIMA doesn't transform my non-stationary series !!
Therefore,

I think estimate equation box is the best way to find the better model
What is your opinion?

[EViews Gareth]

I doubt the same sample was used for estimation of the chosen model. Check the output carefully.

[ESVAL2016]

Only, one difference in the work, I take the first difference of my original series. Because the automatic way is not transformed the non-stationary series !!

[ESVAL2016]

What is your opinion?
Which way is the right way to find the best model??

[EViews Gareth]

Not really sure what the issue is.

If you believe the series should be differenced, difference it and then use the automatic arima selection.

If you believe the series should be differenced and follows an exact arima specification, then use that specification and don't use the automatic selection.