Hello All,
I have a model where the endogenous variable is instrumented using two IV's say z1 and z2 (this comes from theoretical as well as empirical literature). We also know that while doing first stage regression, the endogenous variable is run as a function of IV's and other control variables. Therefater the predicted values from this regression are used for your further second stage regression. I use xtivreg2 to perform IV-2sls/gmm2step method. Having gone through Econometric Analysis of Cross Section and Panel Data I found that while doing IV-2sls/GMM, one should incorporate the quadratic terms as well as cross product of instruments. Since I have control variables in first stage, can I incorporate the cross product as well as higher order tems of the control variables too in my xtivreg2 command along with that of instruments? In fact while I do that,I find that the strength of the instruments increase as reflected by Cragg-Donald as well as KPW F stat. I understand that this somewhere hints that any of the control variables (a or b) are somewhere potential instruments of the endogenous variable. But then these control variables are also in the second stage regression, so is that appropriate?
I follow two codes below:
In the second code, I incorporate various control variables as well as their cross product and quadratic terms.
Let me know if any further information is required. I tried to keep it short to convey exact doubt.
Eagerly waiting for some reply.
Please help,
Mohina
I have a model where the endogenous variable is instrumented using two IV's say z1 and z2 (this comes from theoretical as well as empirical literature). We also know that while doing first stage regression, the endogenous variable is run as a function of IV's and other control variables. Therefater the predicted values from this regression are used for your further second stage regression. I use xtivreg2 to perform IV-2sls/gmm2step method. Having gone through Econometric Analysis of Cross Section and Panel Data I found that while doing IV-2sls/GMM, one should incorporate the quadratic terms as well as cross product of instruments. Since I have control variables in first stage, can I incorporate the cross product as well as higher order tems of the control variables too in my xtivreg2 command along with that of instruments? In fact while I do that,I find that the strength of the instruments increase as reflected by Cragg-Donald as well as KPW F stat. I understand that this somewhere hints that any of the control variables (a or b) are somewhere potential instruments of the endogenous variable. But then these control variables are also in the second stage regression, so is that appropriate?
I follow two codes below:
Code:
I: xtivreg2 ROTA (l.Leverage l.Leverage_Square l.Lev*HHI l.Lev*Size = l.z1 l.z1_Square l.z1*HHI l.z1*Size l.z2 l.z2_Square l.z2*HHI l.z2*Size L.z1*z2) l.HHI l.Size l.Growth, fe cluster(firm) endog (l.Leverage l.Leverage_Square l.Lev*HHI l.Lev*Size)
Code:
I: xtivreg2 ROTA (l.Leverage l.Leverage_Square l.Lev*HHI l.Lev*Size = l.z1 l.z1_Square l.z1*HHI l.z1*Size l.z2 l.z2_Square l.z2*HHI l.z2*Size L.z1*z2 l.a l.b l.ab l.a_square l.b_square l.z1*a l.z1*b l.z2*a l.z2*b) l.a l.b l.HHI l.Size l.Growth, fe cluster(firm) endog (l.Leverage l.Leverage_Square l.Lev*HHI l.Lev*Size)
Eagerly waiting for some reply.
Please help,
Mohina
