I'm currently engaged in a panel data analysis project in R, focusing on multiple economic variables (such as GDP, Gross Fixed Capital Formation, etc.) for EU27 countries spanning the period 1995-2015. I plan to break these variables into five-year time spans. My question is about the appropriate timing for conducting a unit root test on these variables. Should I perform the unit root test before breaking the variable into five-year spans or after?

Thank you!


1 Answer 1


You already do not have enough data to run unit root tests (it is argued in various sources that minimum T for unit root test should be 25-30 e.g. see Baltagi (2005), Econometric Analysis of Panel Data chapter 12). Hence if you want to perform unit root test despite of having too low T you should not cut the T even more before you run the test.

Moreover, if you do the test only to check for stationarity before you run your regression, and if you plan to run regressions on 5 year sub samples, it is waste of time to check for unit roots.

With such small T you do not need to worry about stochastic trends, simply add time fixed dummies to allow every year having some specific effect. This would not be efficient with large T, and in such cases you should test for unit root in panel data, but you simply do not have the required amount of observations to run credible unit root tests. Only do them if this is some sort of school project where its required.

  • $\begingroup$ Thank you very much, you helped me a lot $\endgroup$
    – kostas2323
    Jan 13 at 1:18
  • $\begingroup$ @RichardHardy yes thanks for catching it $\endgroup$
    – 1muflon1
    Jan 13 at 11:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.