System GMM and Fixed effects generate contradictory results

[David Shin]

Dear all,

I am working on my thesis and I have some problem with result which I don't quiet understand.

My dependent variable is (Heal), I.Vs are (Mil, Demo, Interaction); control varibles are (lnpop, lngdpp, pop65, gov_exp_gdp DE)

I have few questions.

1) I got contradictory result for I.Vs (Mil, Demo, Interaction) signs are just opposite in SGMM and FE estimation

Variables	Sign SGMM	Sign FE	Note
Mil	Positive	Negative
Demo	Positive	Negative
Interaction	Negative	Positive

2) This is my first time using SGMM. Is my command for SGMM correct? I am confused with lags, GMM and IV options. I input collapse in GMM option.

3) Finally, the AR(1), AR(2) and Sargan & Hansen Test results interpretation.

As I understoond,

AR(1) Pr > Z = 0.000 (good sign)
AR(2) Pr > Z = 0.340 (good sign)
Sargan Test = 0.000 (good sign)
Hansen Test = 0.982 (good sign)

Am I right?



I would be very grateful to receive any comments.

Thank you very much in advance!

Kind regards,
David



For SGMM,

I coded

"xtabond2 HEAL l.HEAL Mil Demo Demo_interaction DE LnGDPP LnPOP POP65 Govt_Exp_GDP, gmm(l.HEAL Mil Demo Demo_interaction DE Govt_Exp_GDP, collapse) iv(LnGDPP LnPOP POP65) twostep"

and I got following result.

For FE estimation,

xtreg HEAL c.Mil##c.Demo DE LnGDPP LnPOP POP65 Govt_Exp_GDP, fe

[Sebastian Kripfganz]

The results are not directly comparable. Your system GMM estimates are for a dynamic model while the fixed-effects estimates are for a static model.

Your system GMM estimates are likely to suffer from instrument proliferation. Despite the collapse suboption, the number of instruments is still too large. You could restrict the number of lags used to form these instruments with the suboption laglimits() or re-think your assumptions. Do you really need to assume that so many variables are endogenous?

xtabond2 shows you a warning message that uncorrected two-step standard errors are unreliable. This should be taken serious. Windmeijer-corrected robust standard errors should be computed with the robust option.

Too many instruments together with non-robust two-step standard errors turn the Hansen test result unreliable. The "too high" p-value is a clear indication in that regard. The Sargan test is not really useful in the context of system GMM estimation because the one-step weighting matrix is not optimal. (In any case, a p-value of 0.000 is not a good sign as you would reject the null hypothesis of correct model specification.)

As it is the first time you are using the system GMM estimator, I strongly recommend that you read the related literature. You might want to start with Roodman (2009). How to do xtabond2: An introduction to difference and system GMM in Stata. Stata Journal 9 (1), pp. 86-136.

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

To expand on Sebastian's useful response, models with lagged dv's are very different than models without. You're holding constant a lot of stuff with the lagged dv which changes the effects of the other ivs. They often give extremely different results. This holds whether you use system GMM, 2SLS or whatever to estimate the model with the lagged dv. There are literatures on this - I think in Political Science or Sociology.