how to get same result with xtdpdsys and xtabond2?

[Joseph Danquah]

Hello, i am currently working on my thesis and i am stuck with overidentification test. I used the xtdpdsys command and i got the expected coefficient and significant values for my variable of interest. However when i use the sargan test, i get a p-value of 0.0000 which makes me feel uncomfortable. i believe the sargan test has flaws and i want to use the hansen j test rather but there seem to be no command for it. The only way to get it is to use the xtabond2 command. I am not too good with that command due to the endogenous and exogenous options, it confuses me. i would be glad if someone can assist me to get the right xtabond2 command from my results which will give me the exact results as the xtdpdsys. I have posted the results here

xtdpdsys lngreenfield lngdppc angdppc inf open cit lnpop cr lngreenfieldL1, nocons lags(1) artests(2) vce(robust)
note: L.lngreenfield dropped because of collinearity

System dynamic panel-data estimation Number of obs = 165
Group variable: id Number of groups = 19
Time variable: year
Obs per group:
min = 8
avg = 8.684211
max = 9

Number of instruments = 52 Wald chi2(8) = 4224.77
Prob > chi2 = 0.0000
One-step results
--------------------------------------------------------------------------------
| Robust
lngreenfield | Coef. Std. Err. z P>|z| [95% Conf. Interval]
---------------+----------------------------------------------------------------
lngdppc | .5713249 .4883243 1.17 0.242 -.3857731 1.528423
angdppc | .0017639 .0391657 0.05 0.964 -.0749995 .0785273
inf | -.0350448 .021586 -1.62 0.104 -.0773525 .007263
open | .0013623 .0107646 0.13 0.899 -.0197359 .0224605
cit | -.1055809 .0387232 -2.73 0.006 -.181477 -.0296848
lnpop | 1.435553 .2608094 5.50 0.000 .9243757 1.94673
cr | .8368848 .534084 1.57 0.117 -.2099007 1.88367
lngreenfieldL1 | -.1874725 .1027013 -1.83 0.068 -.3887632 .0138183
--------------------------------------------------------------------------------
Instruments for differenced equation
GMM-type: L(2/.).lngreenfield
Standard: D.lngdppc D.angdppc D.inf D.open D.cit D.lnpop D.cr
D.lngreenfieldL1
Instruments for level equation
GMM-type: LD.lngreenfield

[Sebastian Kripfganz]

Please see the following topic on Statalist: how to make xtdpdsys and xtabond2 equivalent?

As an additional comment: You do not want to include the lagged dependent variable lngreenfieldL1 manually in xtdpdsys. This is included automatically. A manual inclusion might result in unwanted effects.

Moreover, it seems strange to exclude the constant from a system GMM estimation.

[Joseph Danquah]

thank you very much, if i include the constant term, i get difference result which is not favourable, but when i exclude it, i get the expected results as i need. is there a reason to justify the exclusion of the constant term. Also the hansen J test gives a p-value of 1.0000, is it extreme or it ok since i need to justify my model to be overidentified.

[Sebastian Kripfganz]

Chosing the model based on the desired results is not a scientific approach. If you have such a strong prior opinion about your results, then there is no point of even estimating any model. Moreover, if the desired result is only obtained with a specific model specification, this lack of robustness should raise a lot of question marks.

The Hansen p-value of 1.0000 is a clear indication that your model suffers from severe problems, in particular a too-many-instruments problem. Your number of instruments (52) is far too large relative to the number of groups (19). With this small sample size, there is not much hope to get robust GMM estimates in a dynamic model.

[Joseph Danquah]

hello, thanks for the enlightenment, does that mean i need to estimate without using robust standard errors, how does that correct heteroscedasticity if i do not use robust, if there a way or an estimator which would seem better