reg3 (Three-stage estimation for systems of simultaneous equations) equation is not identified

[Suna Marasli]

Dear Statalist,

Using panel data of firms, I am trying to estimate a model where

i. firms' past productivity (lagged as t-3) predicts firms' management quality (t-1),
ii. management quality (t-1) predicts same period's productivity (t-1)
iii. productivty at (t-1) predicts export volume at (t)

The data is set as a panel using xtset, with unique firm identifiers and years. The data is not perfectly balanced but there are not so many missing observations. The number of firms (balanced observations) is 330.

I thought the model I want to investigate can be estimated using the Three-stage estimation method (the reg3 command) for systems of simultaneous equations. Accordingly, I designed the model as below:

global mng " L.management L3.productivity L.capital L.size" // firm size and physical capital as control variables in this equation
global prd "L.productivity L.management i.industry" // industry dummies as control variables in this equation (numerous dummies of industry)
global xvo "export L.productiviry i.ownership i.age i.location" // ownership, age category and location factor variables as control variables (numerous dummies of location)
global enlist "L.management L.productivity export" // endogenous variables in the system
global exlist "L3.productivity L.capital L.size i.industry i.ownership i.age i.location" // exogenous variables in the system
reg3 $mng $prd $xvo, endog($enlist) exog($exlist)

But whatever changes in the model I tried, I always got the r(481) error of "Equation is not identified -- does not meet order conditions"

I think my model satisfies the requirement of having as many exogenous variables in the remaining system as there are endogenous right-hand-side variables in an equation, but maybe I am wrong. Why do you think I keep receiving this error?

Maybe reg3 is not suitable for my purpose, because I have panel data with likely heteroscedasticity. I saw that using bootstrap option can be useful, or that Wooldridge recommends using GMM-3SLS estimation, but unfortunately I cannot understand those methods enough and I don't feel comfortable using the methods which I don't understand well.

How can I estimate my model, I'd greatly appreciate if you could give me ideas.

Best regards,
Suna

[Amare Terefe Gashaye]

Dear all, would you forward me your suggestion:

If at least one of the dependent variables in a system of simultaneous regression equations is dummy, how could we use the rreg3 STATA function which inherently use linear probability fitting irrespective of the nature of the dependent variable? Is it possible to use logistic or probit models with rreg3 in STATA?