Page 1 of 1

VECM Causality

Posted: Wed Nov 07, 2012 3:26 am
by CharlieEVIEWS
Hi there!

Apologies for cross-posting, this may have previously been on an incorrect board.

I was wondering if there was a simple way of jointly testing long and short-run causality in VECM models in Eviews - that is (obviously where r=1!): assume Y=(a,b), \deltaY(t)=\alpha\beta(Y(-1)) +\gamma\delta Y(-1)+\gamma\delta(-2)+...+epsilon.

Under accepted definitions of VECM causality testing (e.g. mosconi and giannini 1992), we want to jointly test - for example, a causing b: the short-run causality by a standard wald test on \gamma_{a,1}=...=\gamma_{a,k-1}=0 in the second equation of the system, and \alpha_2\beta_1=0 (or, alternatively, \alpha1.\alpha2, for variable 2 causing variable 1, with all of the coefficients on the difference terms of variable 2 in the first equation being jointly equal to zero and \alpha2.\beta1 for variable 1 causing variable2, with all the coefficients on the difference terms of variable 1 in the second equation being jointly equal to zero)?

Sorry if I have misinterpretted the issue, but it is how i understand the issue from the literature such as:

http://web.uvic.ca/~jaclarke/clarkemirza_2006.pdf



What I am essentially asking is if there is a way to jointly test the (non-linear) restriction (for example) such as:

alpha_1*beta_2 = 0 (in the cointegrating vector and the error correction mechanism respectively)

AND

gamma_{2,t-1}=...=gamma_{2,t-k}=0, where gammas are the coefficients on the lagged (endogenous) variables.

Thanks

Re: VECM Causality

Posted: Mon Nov 12, 2012 10:54 am
by CharlieEVIEWS
What is essentially being asked here is: is it possible to undertake a test of the variety of Section 4.5 in the following paper:

http://dare.uva.nl/document/305

Despite the fact that it says in Section 4.6:

"Finally, it may be noted that none of these packages have a built-in option to test non-causality hypotheses of the type discussed in Section 4.5."

However, this paper was written in 2003, and I was wondering if indeed any commercial software packages had begun to integrate such tests?

Best wishes, thanks again



p.s. - an earlier, unsuccessful attempt to answer this question was made in post:
http://forums.eviews.com/viewtopic.php?t=4259&f=4