Page 1 of 1

Using Cointreg in a VEC model

Posted: Mon Dec 29, 2014 4:52 pm
by stoddj
I wish to use variables in the cointegration portion of a VEC model, some of which will NOT be included in the VAR portion. Therefore the COINTREG routine naturally suggests itself. My difficulty is in a first step, trying to get 'up to speed.' That is, I'm unable to replicate the cointegration results of a simple VEC model estimate using COINTREG.

I do realize that the form of presentation is different in COINTREG and VEC, in that in a VEC model the cointegrating regression is represented by its error term, so that the dependent variable is given with a coefficient of 1.000. And of course the COINTREG must be estimated with a one period lag. But after allowing for these minor issues, the output of the COINTREG and VEC (cointegration portion) look very different indeed -- quite different degrees of significance, for example. Is there a simple way to do what I'm after, or am I missing something quite fundamental? Thanks, JS

Re: Using Cointreg in a VEC model

Posted: Tue Jan 20, 2015 12:56 pm
by stoddj
This is by way of a partial 'reply' to my own post. I'd still like an answer to those questions posed, about differeneces in the form of the cointegrating equation when estimated as 'stand alone,' as opposed to when it is embedded within a VEC model. As I said before, this may be a beginner-level error/confusion on my part. But if so, I seek correction. :oops:

Mitigating these issues, however, I now see that EViews 9 (Beta Version), in its new ARDL (Auto-Regressive Distributed Lag) routine, provides a means of posing variables in the cointegrating equation set at a different numbers of lags than those shown in the VAR portion of the VEC. The unit root requirements on the cointegrated variables within the ARDL form are much loser -- i.e., as long as none of them is I(2), one may have a mix of I(1) and I(0) variables.