GMM 3SLS Underidentification Problem

[Selin Ozdamar]

Dear Statalisters,

I estimated my 3 equations model using 3SLS estimator (used reg3 command). I got the results and commented on them. however, I decided to do further diagnostics (i.e. overidentification) and I have to use gmm command to do this. I have panel dataset, that consists of 117 countries for 1990-2021 period. I had an issue running gmm command brings me underidentification error:


reg3 (ln_gdppc = ln_inf invt hc saving trade tec telp ln_fcons gini_disp) ///
(gini_disp = ln_inf invt hc unemp telp rural cre ln_fcons ln_gdppc) ///
(ln_fcons = ln_inf invt hc rural agedepend ln_lexp ln_gdppc gini_disp), 3sls
//Turning it to GMM:
gmm (eq1: ln_gdppc- {b0} - {xb: ln_inf invt hc saving trade tec telp ln_fcons gini_disp})
(eq2: gini_disp- {c0} - {xc: ln_inf invt hc unemp telp rural cre ln_fcons ln_gdp})
(eq3: ln_fcons- {d0} - {xd:ln_inf invt hc rural agedepend ln_lexp ln_gdppc gini_disp}),
instruments(eq1: ln_inf invt hc saving trade tec telp unemp rural cre agedepend
ln_lexp)
instruments(eq2: ln_inf invt hc saving trade tec telp unemp rural cre agedepend
ln_lexp)
instruments(eq3: ln_inf invt hc saving trade tec telp unemp rural cre agedepend
ln_lexp)
winitial(unadjusted, independent) wmatrix(unadjusted) twostep

This is the error message that I got: note: 1338 missing values returned for equation 1 at initial values.
model is not identified; there are more parameters, 10, than instruments, 1
r(481);

So is the problem about missing values or the number of parameters?

How can I fix this issue? I would appreciate any answer/help. Thanks in advance.

[George Ford]

Should ln_gdp be ln_gdppc in eq2 to match other equations and reg3 specification?

Missings may be the issue. Figure out what's causing that, drop the variable, and try again. If reg3 runs, then it may not be the issue.

[George Ford]

this post deleted due to incorrectness.