Greetings,
I'm currently estimating a system of simultaneous equations in which some of the dependent variables are also in the explanatory part of other equations. Hence, there is an endogenous system with some assumed exogenous variables (educational and health public expenditure change and its square value). The issue is about the implementation of the individual fixed effects and the time fixed effects. My panel is n=100 and t=20. and i am using the following code:
This delivers an optimal result for the Hansen/Sargan tests of overid, but it is a pooled form of the model. In which I have the description of endogenous and exogenous variables assumed in the estimation.
When I introduce the classical dummy variables by using i.country into the equations the program crash:
Providing the error message as
I assume that reg3 is taking as endogenous the individual fixed effects introduced by the dummy variables. Hence, making the computations unfeasible.
The solution I came about is that to avoid such problem of induced endogeneity within the program is to use as exogenous the fixed effects at a country level. Using the command:
My experience indicate to me that this approach is making a strong assumption of exogeneity of the time-invariante effects in the countries, which they may be correlated with the explanatory variables. but I can't figure out another way to estimate the model with fixed effects inclusion. So, I am wondering if this could be a good approach to introduce the fixed effects in a 3sls panel system. I heard that some studies uses the type of two stage regressions where during the first stage the time invariant effects are captured and used in the second stage. but I wanted further references for this approach.
With regards,
I'm currently estimating a system of simultaneous equations in which some of the dependent variables are also in the explanatory part of other equations. Hence, there is an endogenous system with some assumed exogenous variables (educational and health public expenditure change and its square value). The issue is about the implementation of the individual fixed effects and the time fixed effects. My panel is n=100 and t=20. and i am using the following code:
Code:
reg3 (y ln_gdp ln_gdp_sq ln_gini)(ln_gdp y ln_gini)(ln_gini ln_gdp ln_gdp_sq y), inst( d.educ_gvmnt_exp d.educ_sq d.health_exp d.health_sq) first
Code:
Endogenous variables: y ln_gdp ln_gdp_sq ln_gini Exogenous variables: D.educ_gvmnt_exp D.educ_sq D.health_exp D.health_sq
Code:
reg3 (y ln_gdp ln_gdp_sq ln_gini i.country)(ln_gdp y ln_gini i.country)(ln_gini ln_gdp ln_gdp_sq y i.country), inst( d.educ_gvmnt_exp d.educ_sq d.health_exp d.health_sq)
Code:
Equation is not identified -- does not meet order conditions
The solution I came about is that to avoid such problem of induced endogeneity within the program is to use as exogenous the fixed effects at a country level. Using the command:
Code:
reg3 (y ln_gdp ln_gdp_sq ln_gini)(ln_gdp y ln_gini)(ln_gini ln_gdp ln_gdp_sq y), inst(i.country d.educ_gvmnt_exp d.educ_sq d.health_exp d.health_sq)
With regards,
