Hi there,

I am estimating the relationship between growth and inequality using dynamic panel data.

Originally, I set out to do a pooled ols, random effects and fixed effects regression. Within this I was using GDP per capita at the beginning of the period as an endogenous variable.

However I am now estimating using GMM xtabond2 command. The literature suggests to use the lagged dependent variable as an instrument. I want to know whether I can still use GDP per capita at the beginning of the period as an endogenous instrument as opposed to growth in the last period.

I am now thinking there are 3 possible regression I could run:

(1) xi: xtabond2 GDPgrowth laggedGDPpc initialgini initialginisq educ pppi i.year, gmm(laggedGDPpc) iv(initialgini initialginisq educ pppi i.year) noleveleq nodiffsargan robust small

(2) xi: xtabond2 GDPgrowth L1_GDPgrowth initialgini initialginisq educ pppi i.year, gmm(L1_GDPgrowth) iv(initialgini initialginisq educ pppi i.year) noleveleq nodiffsargan robust small

(3) xi: xtabond2 GDPgrowth L1_GDPgrowth laggedGDPpc initialgini initialginisq educ pppi i.year, gmm(L1_GDPgrowth laggedGDPpc) iv(initialgini initialginisq educ pppi i.year) noleveleq nodiffsargan robust small


(1) uses initial GDP p/capita. (2) uses lagged GDP growth (the lagged dependent variable). (3) uses both as instruments. They all give fairly similar results...

Which would be best regression to run?

MAIN QUESTION: can I use lagged GDP p/capita as an endogenous instrument or do I have to directly use the lagged dependent variable, in this case lagged GDP growth?