EViews.com

Hi, I'm testing for a relationship between Government Spending and GDP, I've confirmed cointegration for some countries, and a lack for others.
For the cointegrated one's, I estimate a VEC, and then perform a causality test via "view - lag structure - granger causality/block exogeneity test"

Is this correct, also, if so do I just do the same but with a VAR for the non-cointegrated countries.

Any help would be appreciated.

Dear Bobby,

Yes, you could do that for the cointegrated variables that is using Vector Error-Correction Model (VECM) readily in Eviews. For your information, you should use first difference VAR instead of VAR for the variables / countries are not cointegrated.

Alternatively, you could run the Error-Correction Model (ECM) with the unequal lag structure, where this is more realistic for the real world study. You could refer to Hsiao (1981) - Journal of Monetary Economics or other relevant studies for the non-equal lag structure causality tests..

Hope the above comments and suggestions helped you.

Thank you,

Warmest regards,
tcfoon

Hello!

I acutally posted a related question somewhere else because I didn't see that this kind of stuff is discussed here. If anybody wants to answer my question, feel free here: http://forums.eviews.com/viewtopic.php?f=4&t=1464

However, bobby, you can estimate the VAR in levels as well, as long as you overfit the model for the highest number of integration that could be in your data. Wald-tests conducted for such a VAR are asymptotically valid as long as you do not include the overfitted lags in your test. The idea goes back to Toda & Yamamoto, 1995, "Statistical Inference in possibly integrated systems" (or something like this).

Best wishes
Anna