EViews.com

Hi guys,

First time poster here. I need your help with a VAR that I'm trying to estimate using Eviews.

Basically, I have 4 series that I'm trying to estimate a VAR with. 3 out of my 4 series have seasonality though, and I'm not sure whether I should adjust seasonality before estimating my VAR or not, and what would be the benefits of doing that.

I tried both ways. When I adjust seasonality using the Census X12 program, VAR estimation is much smoother and seems to make more sense. But the problem is the Granger causality test isn't conclusive at all and none of my series are causing any other, which as I understand it, makes it pointless to estimate a VAR with such series?

When I don't adjust seasonality, the Granger causality test is conclusive and results actually make sense. But VAR estimation is a real pain in the ***, can't find the optimal lag due to the number of observations (24 quarterly observations), hence VAR stationarity becomes impaired and impulse responses to a Cholesky 1% shock are all over the place as you could imagine. Also, when the series are seasonally adjusted, unit root tests show all series are I(1), whereas when they are not seasonally adjusted, some are I(1) and others I(2).

What do you guys think? Any input would be greatly appreciated.

Thanks,

B.

Come on guys, any idea?

Although trend, seasonality and etc. might lead to spurious relations, adjusting or controlling such effects is quite difficult. The risk (of any filtering) is that you may change the behaviour of the actual data. As you see, there is an important trade-off here. Common practice is to make necessary adjustments during the preliminary data analysis stage. The correct approach, however, would be to build a general model that accounts for such factors as well (if possible).

In your case, you can also try using seasonal dummies as exogenous variables in the VAR equation. However, it seems you do not have enough observations to do a proper VAR analysis to begin with. Or a seasonal adjustment for that matter...

Thanks for the reply mate. And you're totally right, the number of observation is definitely what messing me up here.