[Gretl-users] Cointegration Engle-Granger ?

[Sven Schreiber]

john w schrieb:
> Hi guys,
> I tried the Engle-Granger cointegration test but I noticed something.
> Here is my procedure:
> 
> 1.I estimated a equatation in cointegration dialogue with the following
> variables l_CPI, l_REER, MMR and l_Y using 12 lags and got these results
> down.
> 2. These ADF tests for these variables are wright and the results are
> the same if you do ADF tests through main Gretl window using the same
> number of lags.
> 3.In step 5 we have the cointegration results and they are the same if
> you estimate a OLS model through main Gretl window.
> 4.In step 6 we have ADF tests for the residuals. Tests are without a
> constant. I tried to check the same test through main Gretl window using
> the same specification ie. I saved residuals from OLS regression (which
> I done through the main Gretl window) and tested with the ADF test
> without a constant. Here are the results:
> 
>> From Engle-Granger cointegration:
> Step 6: Dickey-Fuller test on residuals
> Augmented Dickey-Fuller tests, order 12, for uhat
> sample size 86
> unit-root null hypothesis: a = 1
> 
>   test without constant
>   estimated value of (a - 1): -0,333671
>   test statistic: t = -2,46579
>   asymptotic p-value 0,4903
> 
> 
>> From maing Gretl window:
> Augmented Dickey-Fuller tests, order 12, for uhat
> sample size 86
> unit-root null hypothesis: a = 1
> 
>   test without constant
>   model: (1 - L)y = (a-1)*y(-1) + ... + e
>   estimated value of (a - 1): -0,333671
>   test statistic: t = -2,46579
>   asymptotic p-value 0,01324
> 
> Test statistics and estimated values, sample size are the same, but
> p-values are different?
>> From the first test ie. E-G one could conclude that there is no 
> cointegration and from the second (from OLS saved residuals) that there
> is a cointegration relationship.
> Why is that?
> Did I make same mistaque in comparing?
> 
>

Yes, EG has a different distribution than ADF. gretl is right.
cheers,
Sven