Re: VECM estimation: How to determine significances

[SnakeCharmerII]

Hi,

I was wondering how to determine significance in the Cointegrating Eq of the VECM model, because only Standard errors in ( ) & t-statistics in [ ] appear in the output of Eviews.

Searching in the forums, I read the suggestion of doing it by hand with,

scalar tstat = @qtdist(1-(alpha/2),n-1)

But I'm not sure if this is correct (what makes me doubt is that Stata reports Z p-values, from a Standard Normal distribution rather than a t-distribution).

Can anyone help me?

[SnakeCharmerII]

Ok, I've done the following experiment:

1) I've estimated VECM models in STATA and in EVIEWS until I obtained nearly the same results in terms of s.e. and t-statistics
2) With these values, I've tried different statistical ways of calculating the p-values, by hand, until I got the same p-values reported by Stata.
3) It seems that the formula to determine the significance (the p-value) is (express with eviews code):
scalar zstat = (1-@cnorm(@abs(tau)))*2
Where tau is the value of the t-statistic associated with the estimator of the non-deterministic parameter in the cointegrating equation of the VECM model (it is specially accurate with few degrees of freedom, you can try it with Stata).

This is an interesting result, as it is a two-tailed test of Ho: tau = 0 based on the assumption of a standard gaussian distribution, NOT a Student t-distribution.

Opinions? Suggestions?