Pedroni cointegration and GM-FMOLS

[tcfoon]

Hello,

I attempt to perform the Pedroni cointegration test and the GM-FMOLS using RATS version 8. However, I faced several problems and I need your help.

1. My estimation model consists of at least 8 explanatory variables, but 3 of them are dummy variables. I am not sure whether I should include all 8 variables to test for the panel cointegration with Pedroni approach or I should run the test without the dummy variables. If I am not mistaken, Pedroni (1999) allow us to test for maximum 7 explanatory variables.

2. From my reading, I understand that if the panel dataset are cointegrated. I should estimate the relationship between the variables of interest using GM-FMOLS. However, some of the studies include dummy variables in their GM-FMOLS model, while some other studies exclude the dummy variable and they include the dummy variables only in the short run. I’m a bit confused about this.

Here, I hope to get a feedback from you all as soon as plausible.

[TomDoan]

To quote the assumptions from Pedroni's paper:

The variables yit and Xit are assumed to be integrated of order one, denoted I(1), for each member i of the panel, and under the null of no cointegration the residual eit will also be I(1).

Dummy variable X's aren't I(1) so they aren't included in the number of tested variables. I would think that most types of dummies would alter the asymptotics, so you couldn't use the tabled values anyway.

[tcfoon]

Hello TomDoan,

Thanks for your valuable advice. If I don't tests the with dummy, then can I include dummy when estimate the long run coefficient with group mean FMOLS (if the variables are cointegrated)?

Thanks,

[TomDoan]

In general, no. The FM correction would depend upon how the dummies entered into all the endogenous variables, since they would induce different types of correlation depending upon the assumptions.