Hello everyone,
I have questions about GMM regressions.
In fact, after having obtained my results with fractional logit regression, I would like to do a robustness analysis by checking for endogeneity. However, I'm a beginner and I have a few questions.
First of all, I need to use the xtset command. My first problem is that I don't have any unique company data by year, so I get this message:
« . xtset entFE year
repeated time values within panel
r(451); »
To solve this problem, I created a time variable: gen time = _n and then the command . "xtset entFE time" command:
Panel variable: entFE (unbalanced)
Time variable: time, 1 to 2126
Delta: 1 unit
Then, here is my basic regression:
fracreg logit forpourcent GEND centBoarddiversity i.GEND#c.centBoarddiversity BoardSize Independentboard CEOChairmanDuality debttoasset_w lnageyear_w lncompanysizemarketcap_w roa_w reelage independent busy master phd charteredaccountant law i.Year i.industryFE , vce(robust)
forpourcent: is my dependent variable (voting proportion) which takes a continuous value between 0 and 1
GEND: is my first variable of interest and is equal to 1 if a woman is nominated and 0 if it is a man
centBoarddiversity: is my second variable of interest and is a mean-centred variable representing the proportion of women on the board (mean-centered because there is an interaction term).
i.GEND#c.centBoarddiversity: is my interaction variable between my two variables of interest.
the other variables are control variables.
I therefore potentially suspect endogeneity on my two variables of interest GEND and centBoarddiversity. I therefore identified instrumental variables (FEMMEFRACTION roa_w and NBADMIN) that influence the endogenous variables (GEND and centBoarddiversity) but not on the dependent variable (forpourcent). (
Here is my GMM regression:
ivreg2 forpourcent (centBoarddiversity GEND = L1.FEMMEFRACTION L2.roa_w L3.NBADMIN) i.GEND#c.centBoarddiversity BoardSize Independentboard debttoasset_w lnageyear_w lncompanysizemarketcap_w reelage independent busy master phd charteredaccountant law i.year i.industryFE, gmm2s robust
The results screen is in attachments.
My model is therefore significant, with an R2 that I think is OK, a significant underidentification test at 5% and a non-significant Hansen J statistic. Based on these results, I think we can say that my GMM is correct with the relevant instruments.
Here are my questions:
1)Is it OK to create a time variable to solve my first problem?
2) Is my GMM model correct?
3) Is it possible to do a GMM model with two endogenous variables and an interaction?
4) Is having 3 instrumental variables coherent? Should we only have two?
5) Is it necessary for the three instrumental variables (all of them) to influence all the endogenous variables (GEND and centBoarddiversity)? (With standard regressions, I find that FEMMEFRACTION and roa_w are significant on GEND but not the NBADMIN variable; and that NBADMIN and roa_w are significant on centBoarddiversity but not the FEMMEFRACTION variable).
6) Are there any other tests that need to be carried out to corroborate these results?
7) Do I need to take any additional precautions (bearing in mind that the VIF and correlations are acceptable)?
8) Initially the roa_w variable was a control variable (fracreg logit) but I decided to put it as an instrument in the GMM. Is this problematic or acceptable?
Sorry for the long message and thank you in advance for your help.
I have questions about GMM regressions.
In fact, after having obtained my results with fractional logit regression, I would like to do a robustness analysis by checking for endogeneity. However, I'm a beginner and I have a few questions.
First of all, I need to use the xtset command. My first problem is that I don't have any unique company data by year, so I get this message:
« . xtset entFE year
repeated time values within panel
r(451); »
To solve this problem, I created a time variable: gen time = _n and then the command . "xtset entFE time" command:
Panel variable: entFE (unbalanced)
Time variable: time, 1 to 2126
Delta: 1 unit
Then, here is my basic regression:
fracreg logit forpourcent GEND centBoarddiversity i.GEND#c.centBoarddiversity BoardSize Independentboard CEOChairmanDuality debttoasset_w lnageyear_w lncompanysizemarketcap_w roa_w reelage independent busy master phd charteredaccountant law i.Year i.industryFE , vce(robust)
forpourcent: is my dependent variable (voting proportion) which takes a continuous value between 0 and 1
GEND: is my first variable of interest and is equal to 1 if a woman is nominated and 0 if it is a man
centBoarddiversity: is my second variable of interest and is a mean-centred variable representing the proportion of women on the board (mean-centered because there is an interaction term).
i.GEND#c.centBoarddiversity: is my interaction variable between my two variables of interest.
the other variables are control variables.
I therefore potentially suspect endogeneity on my two variables of interest GEND and centBoarddiversity. I therefore identified instrumental variables (FEMMEFRACTION roa_w and NBADMIN) that influence the endogenous variables (GEND and centBoarddiversity) but not on the dependent variable (forpourcent). (
Here is my GMM regression:
ivreg2 forpourcent (centBoarddiversity GEND = L1.FEMMEFRACTION L2.roa_w L3.NBADMIN) i.GEND#c.centBoarddiversity BoardSize Independentboard debttoasset_w lnageyear_w lncompanysizemarketcap_w reelage independent busy master phd charteredaccountant law i.year i.industryFE, gmm2s robust
The results screen is in attachments.
My model is therefore significant, with an R2 that I think is OK, a significant underidentification test at 5% and a non-significant Hansen J statistic. Based on these results, I think we can say that my GMM is correct with the relevant instruments.
Here are my questions:
1)Is it OK to create a time variable to solve my first problem?
2) Is my GMM model correct?
3) Is it possible to do a GMM model with two endogenous variables and an interaction?
4) Is having 3 instrumental variables coherent? Should we only have two?
5) Is it necessary for the three instrumental variables (all of them) to influence all the endogenous variables (GEND and centBoarddiversity)? (With standard regressions, I find that FEMMEFRACTION and roa_w are significant on GEND but not the NBADMIN variable; and that NBADMIN and roa_w are significant on centBoarddiversity but not the FEMMEFRACTION variable).
6) Are there any other tests that need to be carried out to corroborate these results?
7) Do I need to take any additional precautions (bearing in mind that the VIF and correlations are acceptable)?
8) Initially the roa_w variable was a control variable (fracreg logit) but I decided to put it as an instrument in the GMM. Is this problematic or acceptable?
Sorry for the long message and thank you in advance for your help.
Comment