non-stationary dependent variable in dynamic panel data model

[Veronika Label]

Dear Experts
I would greatly appreciate if you could let me know how can I import data if I have non stationary dependent variable when I using GMM method. This method using first difference (Eviews you do to me automatically recalculates at first difference), but when i have non stationary dependent variable (economic growth) although it will be also the first differences, but its instruments are at levels. Please let me know if I must make import data as first difference in case dependent variable or not.

Thank you for answering me

[Sebastian Kripfganz]

There is no need to transform your variables to account for non-stationarity in dynamic panel models.

[Veronika Label]

Dear Mr. Kripfganz. Also when I have non-stationary independent variable and unit root test tell me is stationary in second difference.

[Tobias Wendler]

Hi together,

am I correct by assuming that for difference GMM stationarity in first-differences is sufficient, or is unit root in levels a problem due to the instruments ?

Thanks for any hints and best
Tobi

[Cati Sacerdote]

There is no need to transform your variables to account for non-stationarity in dynamic panel models.

Hello, I hope this is not too old to start up again. Why is it that there is no need to transform your variables in this case? And would this require AH IV/GMM or would it work as OLS too? Wouldn't having a non-stationary dependent (and lagged dependent) variable - for example real GDP - be an issue due to the weak instruments problem?

[Sebastian Kripfganz]

What I meant in the quoted sentence was that nonstationarity in dynamic models (with a lagged dependent variable) does not lead to the familiar spurious-regression problem. This is only a problem in static models. Moreover, nonstationarity is generally less of a problem in fixed-T models, because the distributions are derived under large-N asymptotics, while unit roots cause a problem under large-T asymptotics.

Having said that, a nonstationary variable could indeed lead to a weak-instruments problem.

Whether you are using IV/GMM or OLS is a totally different story. It depends on your assumptions about the exogeneity/endogeneity of your regressors and the presence of unobserved unit-specific effects.