For you to apply VAR model, all variables must be stationary at first difference. If variables are stationary at first and second difference, apply the autoregressive model. If variables are stationary purely at levels, first difference or a combination of both, apply the ARDL

Hope my opinion is useful. Please upvote me if it is

Thanks very much Ebele, that is clear.

In order to apply VAR all variables have to be of the same order of integration.

If all the variables are I(0) you have a VAR model in level. If all the variables are I(d)  with some d > 0 you have two distinct cases:

a)if  the variables are cointegrated an error correction term has to be included in the VAR, the model becoming a VECM (Vector error correction model);

b) if the variables are not cointegrated they have to be differenced d times and you will have a VAR model in difference.

A good reference for VAR models is Walter Endres, "Applied Econometric Times Series". You can find a pdf of the book on Internet, just google it.

regards,

Bogdan

Thanks Bogdan:

Your comments are so useful, I'm just confused about some papers which applied VAR model using variables that are co-integrating in I(1) and I(2). 

Dear Abdelatif,

Pay attention to this fact that any cointegration in variables makes you apply The Vector Error Correction instead of VAR Model.

Yours,

Ehsan

Thanks Dr. Ehsan: 

I appreciate your clarifying.