Cointegration - exogenous variables

[EViews Gareth]

You're right that we don't have a function to do that. You might be best off seeing if there is an R routine that will perform the cointegration you want, and calling that from within EViews, or, if there isn't, using R or Matlab within EViews to do the generalised eigenvalues for you.

[jjcale5]

Hello,

I am studying on an article about environmental kuznets curve.

The model is logco2 = logGDP + logGDP^2 + logGDP^3
The man who prepared the paper had found the variables in his cointegration analysis like
-3.96 for logGDP(-1) 1.20 for logGDP^2(-1) and -0.09 for loggdp^3

and at last, he wrote his model like ;

logco2 = 3.96logGDP - 1.20logGDP^2 + 0.0923logGDP^3

is it normal ?
if it is the case, why ?

thanks very much

[Pandelis]

Dear Thomas von Brasch, EViews Glenn and EViews Gareth

I have exactly the same problem. I do not have access to PcGive that apparently gives the choice to test the existence of co-integration in a VECM with exogenous stationary variables (and then compare the Trace Statistic with the Harbo et al. 1998 critical values), thus I am trying to perform this task in E-Views.

My latest effort was:

i) I estimated a "standard" VEC version of my model (i.e. exogenous variables included in the equations but not in the co-integrating vector)
ii) then I went to "Proc", "Make a System" and I edited the equations as for the co-integrated vector to include the additional variables (it does not allow the parameters in the co-integrating vector to be estimated, so I used fixed values, as a working exercise).
iii) then using Maximum Likelihood as an estimation method, I had an output which, however, did not include standard deviations and z-ratio test results. Just the estimated coefficients and "standard" diagnostics such as R-square etc.
iv) the"view options" on the estimation output, as you have already indicated, does not include a co-integration test, which was the main reason for performing this exercise to begin with.

The way I see it, unless if I find a way to include a "c.i. test option", even if I manage to have full estimation output (which, by the way is possible, if I use the OLS option instead of ML) I would still have to manually calculate the Trace Statistic.

So:

a) Is Is it possible to have a formula for the Trace Statistic as an E-views program so that I can use it to derive the test value in the modified VECM?

b) If, instead, I plainly estimate the VECM with the exogenous variables treated as endogenous (similarly to lexie) how correct would that be? I know is not the same and that would involve the estimation of a larger number of variables (also degrees of freedom concerns due to a small sample). However, the alternative for time being is to estimate the model with the exogenous variables NOT in the co-integrating vector, thus the critical values presented in the related E-Views option to be inapplicable. Also, since the model is not in the Rahbek and Mosconi 1999 form, the Harbo et al 1998 critical values are also inapplicable.

I would appreciate your feedback. I also strongly recommend the next version of E-Views to include the Rahbek and Mosconi 1999 test option.

P.M.