Dear Stata users,
By running xtabond2 I am using the System twostep GMM approach to estimate my dynamic panel data model. My dataset consists of 146 countries with observations from 1960-2010. The data set is unbalanced, hence certain gaps exist between years for some countries.
The point is to estimate the effect of a country being democratic (dummy = 1 if democracy, else 0) on Ln ofGDP pr. capita. Since the model is dynamic I also use three lags of Ln GDP pr. capita as independent variables. My linear regression model takes the form:
yc,t = β1Dc,t + β2yc,t-1 + β3yc,t-2 + β3yc,t-3 + αc + µt + ɛc,t
Dc,t is the dummy for democracy and my coeffecient of interest.
When I estimate the linear regression model by a fixed effect model with year dummies I run the following command:
xtreg LnGDP Democ L.LnGDP L2.LnGDP L3.LnGDP i.year, fe vce(robust)
I get a coefficient on the democracy dummy of β1 = 0,0056, hence 0,56 pct. growth following a democratic transition all else equal.
When I estimate the linear regression model by a two-step system GMM with year dummies I run the following code:
xtabond2 LnGDP Democ L.LnGDP L2.LnGDP L3.LnGDP i.year, ///
gmm(Democ, lag(1 2) collapse eq(diff)) ///
gmm(L.LnGDP L2.LnGDP L3.LnGDP, lag(1 4) collapse eq(diff)) ///
gmm(L.LnGDP L2.LnGDP L3.LnGDP, lag(1 4) collapse eq(level)) ///
iv(i.year) twostep robust
I hereby instrument by two lags of the democracy dummy as first-difference and four lags of each lagged dependent variable both as level and first-difference. I obtain the following results:
My question is: How can it be that the coefficient on Democ is 0.03710 and hereby almost ten times as large as the coefficient obtained by the fixed effect model above? I tried different lags of both Democ and LnGDP in the gmmstyle but the coefficient is not reduced significantly. Am I setting up the year dummies by Ivstyle in a wrong way?
Thank you in advance.
By running xtabond2 I am using the System twostep GMM approach to estimate my dynamic panel data model. My dataset consists of 146 countries with observations from 1960-2010. The data set is unbalanced, hence certain gaps exist between years for some countries.
The point is to estimate the effect of a country being democratic (dummy = 1 if democracy, else 0) on Ln ofGDP pr. capita. Since the model is dynamic I also use three lags of Ln GDP pr. capita as independent variables. My linear regression model takes the form:
yc,t = β1Dc,t + β2yc,t-1 + β3yc,t-2 + β3yc,t-3 + αc + µt + ɛc,t
Dc,t is the dummy for democracy and my coeffecient of interest.
When I estimate the linear regression model by a fixed effect model with year dummies I run the following command:
xtreg LnGDP Democ L.LnGDP L2.LnGDP L3.LnGDP i.year, fe vce(robust)
I get a coefficient on the democracy dummy of β1 = 0,0056, hence 0,56 pct. growth following a democratic transition all else equal.
When I estimate the linear regression model by a two-step system GMM with year dummies I run the following code:
xtabond2 LnGDP Democ L.LnGDP L2.LnGDP L3.LnGDP i.year, ///
gmm(Democ, lag(1 2) collapse eq(diff)) ///
gmm(L.LnGDP L2.LnGDP L3.LnGDP, lag(1 4) collapse eq(diff)) ///
gmm(L.LnGDP L2.LnGDP L3.LnGDP, lag(1 4) collapse eq(level)) ///
iv(i.year) twostep robust
I hereby instrument by two lags of the democracy dummy as first-difference and four lags of each lagged dependent variable both as level and first-difference. I obtain the following results:
| Reg. on: LnGDP | Coef. | Corrected std. Err. | z | P>|<| | [95% Conf. Interval] |
| Democ | 0.03710 | 0.0118498 | 3.13 | 0.002 | 0.0138786 to 0.060329 |
| LnGDP: | |||||
| L1. | 1.14025 | 0.0285374 | 39.96 | 0.000 | 1.084313 to 1.196177 |
| L2. | -0.10163 | 0.030588 | -3.32 | 0.001 | -0.1615781 to -0.0416753 |
| L3 | -0.01509 | 0.0250936 | -0.60 | 0.547 | -0.0642785 to 0..0340866 |
Thank you in advance.
Comment