Dear Statalist,

I am replicating the paper "Country or leader? Political change and UNGA voting" by Dreher and Jensen. To understand how leader change in one country affects their voting in line with the US, they regress multiple lagged explanatory variables such as US imports as a share of GDP (t-1) and US aid/GDP(t-1) as well as leader change. You can see the fixed effects estimator regression below.

xtset id year
xtreg inlinekey_usa change_leader colour_l prs_l lgdp_pc_l gg_l imports_gdp_l odacomUSA_gdp_l, fe cluster(id)

To solve the potential simultaneity problem with some of the explanatory variables, they employ a system GMM estimator with xtabond2 command. Below you can see the regression command.

xtabond2 inlinekey_usa colour_l prs_l lgdp_pc_l gg_l imports_gdp_l odacomUSA_gdp_l change_leader time* ///
, gmm(imports_gdp_l odacomUSA_gdp_l, c) ivstyle(change_leader prs_l lgdp_pc_l gg_l) small two robust

I need to test the validity of the instruments; exogeneity and relevance. But I keep getting error messages saying all instrumented variables=instrumental variables.

My questions are as follows:
1. Does it make sense to test validity of the instruments because they are internal instruments, exactly the same lagged variables used in xtreg regression?
2. If I should test it, how can I make it work in stata? Should I create the instruments with gen x= x-1 [_n+1]
3. Are there any additional tests designed for lagged explanatory variables as IV that I can make use of in this example?

Thank you!