Dear Statalist users.
I am using the xtabond2 to draw advantage of the twostep System GMM approach in order to estimate my dynamic panel data model. I am considering two groups of countries: The European Union (28 countries) and the Central and Eastern Europe (10 countries) both groups from 1990 - 2014 with an unbalanced dataset. Hence, N1 = 700, N2 = 250 and T = 25.
The point is to estimate the effect of investments in renewable energy for the two groups of countries on (ln) of gross domestic product pr. capita. Since I assume that the model is dynamic I also use the first three lags of ln GDP pr. capita as independent variables. My linear regression model takes the form:
LnGDPc,t = β1RECc,t + β2LnGDPc,t-1 + β3LnGDPc,t-2 + β3LnGDPc,t-3 + αc + µt + ɛc,t,
RECc,t is gross inland renewable energy consumption and my coefficient of interest.
To estimate this model I run the following code:
where tdum are yearly time dummies for each of the years from 1993 to 2014.
I hereby instrument by two lags of REC as first-difference and four lags of the first lagged dependent variable as first-difference. This works well when I estimate the model for the European Union countries (N=28). However, when I run the same regression for the Central and Eastern European countries (N=10) I obtain the following output:
The problem is that many of the coefficients are omitted. My best guess is that this relates to problem of instrument proliferation when I have too few degrees of freedom when considering a lower number of countries. If I run the regression without year dummies the coefficient on the lagged dependent variable is no longer omitted. However, according to Roodman (2009) simply removing the time dummies never seems as the best practise.
My question: How do I come around this problem in Stata? I have read threads like https://www.statalist.org/forums/for...m-time-dummies[\URL] but I cannot find the answer here.
Any help is appreciated. Thank you in advance.
Best regards,
Emil F. Mahler.
I am using the xtabond2 to draw advantage of the twostep System GMM approach in order to estimate my dynamic panel data model. I am considering two groups of countries: The European Union (28 countries) and the Central and Eastern Europe (10 countries) both groups from 1990 - 2014 with an unbalanced dataset. Hence, N1 = 700, N2 = 250 and T = 25.
The point is to estimate the effect of investments in renewable energy for the two groups of countries on (ln) of gross domestic product pr. capita. Since I assume that the model is dynamic I also use the first three lags of ln GDP pr. capita as independent variables. My linear regression model takes the form:
LnGDPc,t = β1RECc,t + β2LnGDPc,t-1 + β3LnGDPc,t-2 + β3LnGDPc,t-3 + αc + µt + ɛc,t,
RECc,t is gross inland renewable energy consumption and my coefficient of interest.
To estimate this model I run the following code:
Code:
xtabond2 L(0/3).LnGDP REC tdum4-tdum25, ///
gmm(L.LnGDP, lag(1 4) collapse equation(diff)) ///
gmm(REC, lag (1 2) collapse equation(diff)) ///
ivstyle(tdum4-tdum25, equation(level)) twostep robust
I hereby instrument by two lags of REC as first-difference and four lags of the first lagged dependent variable as first-difference. This works well when I estimate the model for the European Union countries (N=28). However, when I run the same regression for the Central and Eastern European countries (N=10) I obtain the following output:
Code:
Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: country1 Number of obs = 220
Time variable : year Number of groups = 10
Number of instruments = 28 Obs per group: min = 22
Wald chi2(25) = 2.01e+07 avg = 22.00
Prob > chi2 = 0.000 max = 22
------------------------------------------------------------------------------
| Corrected
LnGDP | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
LnGDP |
L1. | 1.06558 .0589094 18.09 0.000 .9501197 1.18104
L2. | 0 (omitted)
L3. | -.0614096 .059251 -1.04 0.300 -.1775394 .0547201
|
REC | -2.52e-06 1.69e-06 -1.49 0.135 -5.82e-06 7.87e-07
tdum4 | -.044087 .010539 -4.18 0.000 -.0647431 -.0234308
tdum6 | .0105556 .0108727 0.97 0.332 -.0107544 .0318657
tdum7 | -.008127 .0136221 -0.60 0.551 -.0348258 .0185717
tdum8 | 0 (omitted)
tdum9 | 0 (omitted)
tdum10 | -.028836 .0148624 -1.94 0.052 -.0579658 .0002938
tdum11 | 0 (omitted)
tdum12 | 0 (omitted)
tdum13 | 0 (omitted)
tdum14 | 0 (omitted)
tdum15 | 0 (omitted)
tdum16 | 0 (omitted)
tdum17 | 0 (omitted)
tdum18 | .0298336 .0065639 4.55 0.000 .0169685 .0426987
tdum19 | -.0188079 .0147789 -1.27 0.203 -.0477741 .0101583
tdum20 | 0 (omitted)
tdum21 | 0 (omitted)
tdum22 | -.057359 .0440904 -1.30 0.193 -.1437746 .0290566
tdum23 | 0 (omitted)
tdum24 | 0 (omitted)
tdum25 | 0 (omitted)
_cons | 0 (omitted)
------------------------------------------------------------------------------
Instruments for first differences equation
GMM-type (missing=0, separate instruments for each period unless collapsed)
L(1/2).REC collapsed
L(1/4).L.LnGDP collapsed
Instruments for levels equation
Standard
tdum4 tdum5 tdum6 tdum7 tdum8 tdum9 tdum10 tdum11 tdum12 tdum13 tdum14
tdum15 tdum16 tdum17 tdum18 tdum19 tdum20 tdum21 tdum22 tdum23 tdum24
tdum25
_cons
My question: How do I come around this problem in Stata? I have read threads like https://www.statalist.org/forums/for...m-time-dummies[\URL] but I cannot find the answer here.
Any help is appreciated. Thank you in advance.
Best regards,
Emil F. Mahler.
Comment