How to perfrom a residual-based test for cointegration

[yunxiangcao]

On papge 236 of EViews 7 Users Guid II, It is shown that "To perfrom the Engle-Granger test, open an estimated equation and select View/Cointegration and"...However, I failed to find out the option in an equation object.

The option could be shown when I select two series (denoted as X and Y, respectively), which I want to perform the Engle-Granger test for cointegration between, and open as a GROUP, then select in the GROUP object "View/Cointegration test/Single-Equation Cointegration Test".

(1) Why it is different from what the Users Guide illustrates for residual-based tests?

When I perform the Engle-Granger cointegrationt est in a GROUP object, the outputs offer two groups of tau-statistics and z-statitics, one for the regression with variable X as a dependent variable, and the other for the regression with variable Y as a dependent variable.

My following questions are, (1) it is necessary to do both regressions? (2) the Probs obtained are differents for these two regressions, why? based on which regression I should make statistical reference? (3) Moreover, from the same regression, the Prob based on tau-statistic and that based on z-statistic are different too, one is statistically significant while the other not. Which one I should take?

The Version I used is EViews 7.2

Thanks!

[EViews Glenn]

To get the EG as a view off of an estimated equation, you must estimate using cointegrating regression method.

You get two different test statistics when computed off of a group since there is no dependent variable in a group (note that this is not the case for the EG off of an estimated equation, which is why there is additional output in the former). As is well known, you can run the cointegration test using each variable in the cointegrating relation as the dependent variable. There is no right answer as to which is the best result. There is discussion of this issue in the literature.