Fixed and Random Effects with Unit root test

[Mohammad Qabaja]

Hello,

I want to ask if I performing the Fixed Effects and Random Effects for the models of the study, should I check the stationarity of variables by unit root test and then apply the differences for the variables not stationary in level then I perform the Fixed Effects and Random Effects?

or there is no need to apply the stationarity for the variables in Fixed Effects and Random Effects models?

for my study N=11 and T=32 and variables stationary in I(0) and I(1)

Thank you

[Maxence Morlet]

You should most certainly run a panel unit root test on each of your variables, and transform the variables displaying at least one unit root until they do not display any unit roots in order to avoid any spurious results and guarantee internal validity. For most panel unit root tests, you will need strongly balanced data (e.g. for the Levin-Lin-Chu or Harris Tzavalis). If you do not have strongly balanced panel data, I would suggest that you employ the Fisher type test.

[Mohammad Qabaja]

You should most certainly run a panel unit root test on each of your variables, and transform the variables displaying at least one unit root until they do not display any unit roots in order to avoid any spurious results and guarantee internal validity. For most panel unit root tests, you will need strongly balanced data (e.g. for the Levin-Lin-Chu or Harris Tzavalis). If you do not have strongly balanced panel data, I would suggest that you employ the Fisher type test.

Thank you Maxence for your reply

I did the panel unit root test for each variable and it shows they are stationary at the level and at first difference. So, when I perform the Fixed and random effect command can I add (d) as the difference in it like the following:-

xtreg d.y x1 d.x2 d.x3, fe

is it possible?

[Maxence Morlet]

If the level of your variables is stationary, then for interpretation purposes, I would suggest using the level in your regression unless theory specifically dictates that you should use the difference.

You can of course use the difference, although this might be difficult to interpret.

[Mohammad Qabaja]

If the level of your variables is stationary, then for interpretation purposes, I would suggest using the level in your regression unless theory specifically dictates that you should use the difference.

You can of course use the difference, although this might be difficult to interpret.

Thank you again for your reply

The problem I cant use the level in most of the variables because they are not stationary. I have seven variables that will be used in four models and each model has four variables. But there something that is not clear in most of researches that using fixed and random effects didn't show they check the stationary of they did the unit root test.

[Eric de Souza]

In #3 you state "I did the panel unit root test for each variable and it shows they are stationary at the level and at first difference."
Then in #5 you say "The problem I cant use the level in most of the variables because they are not stationary."
Which is the case?

[Mohammad Qabaja]

In #3 you state "I did the panel unit root test for each variable and it shows they are stationary at the level and at first difference."
Then in #5 you say "The problem I cant use the level in most of the variables because they are not stationary."
Which is the case?

sorry, I didn't understood your question. The variables of the study mixed stationary in level and first difference and that why I wrote I cant use the level for all variables