Hello everyone,
As part of a robustness analysis I'm doing a GMM analysis.
Basically I used an OLS analysis with Fixed effect, here is my line of code :
Here are the VIF results:
VIFs are therefore low
Now for robustness, I use a GMM :
As you know, since this is a GMM, I'm obliged to do a VIF uncentered. Here are the results:
As you can see, I have a few variables with a high uncentered VIF: lnAGE and indep.
My questions are:
Should I use the VIF results from the classical regression or the uncentered VIF results from the GMM regression?
Lastly, is it correct to include indicator variables by year and by sector in a GMM (xtset company Year)?
Thank you for your answers
Loïc Dubois
As part of a robustness analysis I'm doing a GMM analysis.
Basically I used an OLS analysis with Fixed effect, here is my line of code :
Code:
regress Qtobin_w centBWOMEN centPERF c.centPERF#c.centBWOMEN centSIZE c.centSIZE#c.centBWOMEN c.centBWOMEN#i.quota lnAGE indep CEOChairmanDuality uncemployement popgrwoth corruption GDPgrowth Dummylegalsystem quota i.Year i.NAICS2digit, robust
Code:
Variable | VIF 1/VIF
-------------+----------------------
centBWOMEN | 1.59 0.627641
centPERF | 1.48 0.676269
c.centPERF#|
c.centBWOMEN | 1.18 0.847931
centSIZE | 1.19 0.838239
c.centSIZE#|
c.centBWOMEN | 1.07 0.932262
quota#|
c.centBWOMEN |
1 | 1.52 0.658244
lnAGE | 1.12 0.891564
indep | 1.75 0.570161
CEOChairma~y | 1.15 0.871665
uncemploye~t | 1.34 0.744624
popgrwoth | 1.83 0.545460
corruption | 1.63 0.611942
lnGDP | 1.86 0.537452
Dummylegal~m | 2.59 0.385997
quota | 1.44 0.692453
VIFs are therefore low
Now for robustness, I use a GMM :
Code:
ivreg2 Qtobin_w (centBWOMEN = L1.centBWOMEN L2.centBWOMEN) centPERF c.centPERF#c.centBWOMEN centSIZE c.centSIZE#c.centBWOMEN c.centBWOMEN#i.quota lnAGE indep CEOChairmanDuality uncemployement popgrwoth corruption GDPgrowth Dummylegalsystem quota i.Year i.NAICS2digit, gmm2s robust
As you know, since this is a GMM, I'm obliged to do a VIF uncentered. Here are the results:
Code:
vif, uncentered
Variable | VIF 1/VIF
-------------+----------------------
centBWOMEN | 1.62 0.616670
centPERF | 1.49 0.670903
c.centPERF#|
c.centBWOMEN | 1.22 0.820298
centSIZE | 1.21 0.829305
c.centSIZE#|
c.centBWOMEN | 1.07 0.933065
quota#|
c.centBWOMEN |
1 | 1.66 0.601833
lnAGE | 16.60 0.060248
indep | 11.92 0.083915
CEOChairma~y | 1.83 0.547125
uncemploye~t | 5.22 0.191751
popgrwoth | 3.66 0.273053
corruption | 6.16 0.162296
GDPgrowth | 4.55 0.219929
Dummylegal~m | 6.10 0.163808
quota | 1.62 0.615692
As you can see, I have a few variables with a high uncentered VIF: lnAGE and indep.
My questions are:
Should I use the VIF results from the classical regression or the uncentered VIF results from the GMM regression?
Lastly, is it correct to include indicator variables by year and by sector in a GMM (xtset company Year)?
Thank you for your answers
Loïc Dubois
