Is "robust" option in "xtreg, fe" and two- step system GMM important?

[Huong Nguyen]

Hello everyone,

I am dealing with a panel dataset at country level covering 153 countries from 2005 to 2019.

My results are quite consistent (I mean the sign of key variables) when I use fixed effect regression and two- step system GMM WITHOUT robust option.

In fixed effects regression, my code is following:

xtreg dependent variable independent variable i.year i.countrycode, fe

Similarly, with two- step system GMM, I use xtabond2. I only get good results when I exclude "robust" option.

I know that if I can get good results with "robust" option, it is perfect. But when I add robust option, my results are changed significantly.

My question is: Is there any chance to accept my regression results without robust option?

Thank you for spending your time with my concern!

[Carlo Lazzaro]

Huong:
limiting my reply to -xtreg- only, with 153 panels cluster-robust standard errors are almost mandatory.
In addition, how did you -xtset- your dataset? Thanks.

[Huong Nguyen]

Dear Prof.Carlo Lazzaro,

Thank you very much for your support.

I used "xtset countrycode year". So, in your opinion, my results would be unreliable without robust option, right?

My two-step system GMM code is:

xtabond2 l(0/2).Y X1 X2 X3 X4 yr*, gmm(Y X1, lag (2 4) collapse) gmm(X2 X3 X4 , collapse) iv(yr*) two

yr* is year dummy variables.

Best regards,

Huong

[Huong Nguyen]

Dear Prof.Carlo Lazzaro,

I would like to explain more about my fixed effect code.

I already control for time- and country-specific effects by running the code as follows:

xtreg Y X1 X2 X3 X4 i.year i.countrycode, fe (1)

I also get good results when If I exclude "i.year i.countrycode" and add "vce(cluster countrycode)" by using the following code:

xtreg Y X1 X2 X3 X4 , fe vce(cluster countrycode) (2)

In your opinion, which one ((1) or (2)) would be better?

Best regards,

Huong

[Carlo Lazzaro]

Huong:
my question was intended to understand how you -xtset- your data to rule out that -countrycode- was actually your -panelid-.
Without -xtset- details from your side, I would go:

xtreg Y X1 X2 X3 X4 i.year , fe vce(cluster countrycode)

[Huong Nguyen]

Dear Prof.Carlo Lazzaro,

Thank you very much for your useful answer.

Best regards,

Huong

[Carlo Lazzaro]

Houng:
Carlo is enough. Thanks.

[Tasfaye Fikadu]

Hello everyone,

I am dealing with a panel dataset at country level covering 153 countries from 2005 to 2019.

My results are quite consistent (I mean the sign of key variables) when I use fixed effect regression and two- step system GMM WITHOUT robust option.

In fixed effects regression, my code is following:

xtreg dependent variable independent variable i.year i.countrycode, fe

Similarly, with two- step system GMM, I use xtabond2. I only get good results when I exclude "robust" option.

I know that if I can get good results with "robust" option, it is perfect. But when I add robust option, my results are changed significantly.

My question is: Is there any chance to accept my regression results without robust option?

Thank you for spending your time with my concern!

Dear Nguyen,

did you use the "robust" option after two step sytem GMM?. Mine is also fine with exclusion of "robust"


Have you seen the work of David Roodman on page 123

[ The Stata Journal (2009) 9, Number 1, pp. 86–136] stated as ( I did not understand whether he is suggesting or not )]


“The level(), robust, cluster(), noconstant, and small options are also mostly standard. level() controls the size of the reported confidence intervals, the default being 95%. In one-step GMM, xtabond2’s robust option is equivalent to cluster(id) in most other estimation commands, where id is the panel identifier variable, requesting standard errors that are robust to heteroskedasticity and arbitrary patterns of autocorrelation within individuals; in two-step estimation, where the errors are already robust, robust triggers the Windmeijer correction. cluster() overrides the default use of the panel identifier (as set by tsset) as the basis for defining groups. It implies robust in the senses just described. Changing the clustering variable with this option affects onestep “robust” standard errors, all two-step results, the Hansen and difference-in-Hansen tests, and the Arellano–Bond serial correlation tests. cluster() is available only in the Mata version of xtabond2, which requires Stata 9 or later. noconstant excludes the constant term from X and Z; however, it has no effect in difference GMM because differencing eliminates the constant anyway. small requests small-sample corrections to the covariance matrix estimate, resulting in t-test instead of z-test statistics for the coefficients and an F test instead of a Wald χ2 test for overall fit”

[Sebastian Kripfganz]

With two-step system GMM, you should almost always use "robust" standard errors. While the name of this option is potentially misleading, it applies a correction to the standard errors which can otherwise be severely biased in small samples; see e.g. slide 16 in my 2019 London Stata Conference presentation:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

Especially if your standard errors differ substantially, this indicates that the "robust"/bias-corrected version is needed.

[Tasfaye Fikadu]

With two-step system GMM, you should almost always use "robust" standard errors. While the name of this option is potentially misleading, it applies a correction to the standard errors which can otherwise be severely biased in small samples; see e.g. slide 16 in my 2019 London Stata Conference presentation:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

Especially if your standard errors differ substantially, this indicates that the "robust"/bias-corrected version is needed.

Dear Sebastian,
Thank you very much for your answer and the link.