How to proceed with variables with Unit Root or non stationarity using fixed effects estimation??

[Sabrin Rahman]

I have a data set of N=280 and T =60. I did fisher test for each of my 9 variables. Only two have unit roots at levels and stationary at I(1), rest of the variables are stationary. My question is can I use first difference of the two variables and levels for the stationary variables in the same regression model using fixed effects? Or do I need to use first difference for all the 9 variables. Is Fixed effects estimation still valid with unit roots?

GII = FDI +FemaleFert+GOVTEXP+EDUCATION+POPUL+GDPGROWTH

MY GII and education variables are non stationary at level, so can I use in stata the below command???

xtreg d.GII FDI FEMFER d.EDUC Popul GDP, fe cluster (Country)

or do I need to difference all my variables?


Also what is the purpose of the Cointegration test? if cointegration exists does it mean I can use the non stationary variables in levels using normal fixed effects estimation???

[Andrew Musau]

You have \(N>T\), so you do not need to test for unit roots or cointegration. Proceed with FE.

[Sabrin Rahman]

Thank you Andrew for clarifying this. However I was wondering what if I use part of the sample, suppose low-income countries only, in that case, my T becomes 60 and countries (N) become 53. Also I have unbalanced data .

[Andrew Musau]

In this thread #5, you can look at the suggestion to use first-differences (FD) estimation instead of fixed effects in the case \(T>N\). Otherwise, if you go the dynamic panel route, take a look at this Stata Journal article that describes the command xtpmg. There is an application of the procedure there.