panel unit root test fixed effect random effect

[Morshadul Hasan]

1. In what bases of Unit root test result I can go for fixed and random effect modelling?
2. Can I go for Fixed effect or Random effect modelling if there is unit root (non-stationary variables) in my variables?
3. What if some of my variables have unit root and some there is no unit root (stationary variables)?

NB: I have five countries 27 years data

[Maxence Morlet]

Tough to allow for a unit root for only 27 years of data, but I do not have any empirical evidence to support this to be honest...

Since you have T>N, you could go for Driscoll-Kraay (1998) standard errors, although in your case T/N does not really tend to infinity...

Concerning the estimation method of your coefficients (and not the variance-covariance matrix of the errors), definitely resort to two-way fixed effects, just by default. Assumptions are more plausible than random effects' assumptions.

[Dr. Iqra Yaseen]

hello Maxence Morlet .....my query is that do i need to go for stationary test with PPMLHDFE model of PPML. my panel data set has 24 years and 44 countries i.e. N>T....thanks in advance...keenly seeking your reply

[Maxence Morlet]

You cannot really allow for a unit root with 24 years of data, just go for Poisson pseudo max likelihood model, use cluster-robust standard errors, and include at least two-way fixed effects (if you only have country and time variation).

[Dr. Iqra Yaseen]

thank you Maxence Morlet for your help. Good wishes for your future.