Hausman

[Guest]

HI all!
Can anyone tell me what is the difference between simple hausman and hausman with sigmamore option?

[Marcos Almeida]

Hi, Maham,

I guess that, ideally, we should take the difference between the variances as positive definite. The sigmamore reinforces that.

Added here, a comment from the Stata manual (http://www.stata.com/manuals13/rhausman.pdf)

sigmamore specifies that the covariance matrices be based on the estimated disturbance variance
from the efficient estimator. This option provides a proper estimate of the contrast variance for
so-called tests of exogeneity and overidentification in instrumental-variables regression.

This is the furthest I can go. Hopefully you'll get further explanation.

Best,

Marcos

[Marcos Almeida]

Hi, Maham,

I guess that, ideally, we should take the difference between the variances as positive definite. The sigmamore reinforces that.

Added here, a comment from the Stata manual (http://www.stata.com/manuals13/rhausman.pdf)

sigmamore specifies that the covariance matrices be based on the estimated disturbance variance
from the efficient estimator. This option provides a proper estimate of the contrast variance for
so-called tests of exogeneity and overidentification in instrumental-variables regression.

This is the furthest I can go. Hopefully you'll get further explanation.

Best,

Marcos

[Guest]

Thanks a lot Marcos.
One more question. Can sigmamore be used as an alternative for simple hausman?
Regards

[Carlo Lazzaro]

Yes,
but take a look at -help hausman- and related entry in Stata 13.1 .pdf manual.

[Marcos Almeida]

Hi Maham,

Carlo already gave the appropriate answer ("yes") and pointed to the right way to catch up with the issue: "help" in Stata, and the manual.

Just to make things absolutely clear to every level of the audience: sigmamore is an option you add to the "hausman" commands, like that:

. hausman fe re, sigmamore

Best,

Marcos

[Guest]

Thanks a lot guys. Really helpful.

[Carlo Lazzaro]

Maham:
just an aside to what Marcos has already comprehensively explained, obtained from -hausman- entry in Stata 13.1 pdf manual (pag 768):
-sigmamore- option for -hausman- specification test allows the covariance matrices of both estimators to be based on the estimated disturbance variance of the efficient estimator;
-sigmaless- option for the same test allows the covariance matrices of both estimators to be based on the estimated disturbance variance of the consisten estimator;
- sigmamore - option for -hausman- has a more econometric flavour than -sigmaless- , in that, quoting Stata 13.1 -pdf manual related entry (same page),

...provides a proper estimate of the contrast variance for so-called tests of exogeneity and overidentification in instrumental-variables regression

[Guest]

Maham:
just an aside to what Marcos has already comprehensively explained, obtained from -hausman- entry in Stata 13.1 pdf manual (pag 768):
-sigmamore- option for -hausman- specification test allows the covariance matrices of both estimators to be based on the estimated disturbance variance of the efficient estimator;
-sigmaless- option for the same test allows the covariance matrices of both estimators to be based on the estimated disturbance variance of the consisten estimator;
- sigmamore - option for -hausman- has a more econometric flavour than -sigmaless- , in that, quoting Stata 13.1 -pdf manual related entry (same page),

Thanks a lot Carlo!

[sonia kaur]

Hi all!! I am sonia kaur. I am currently working on my thesis ( determinants of bank profitability) in which i have to analyse approximately 17000 banks over 64 quarters. The determinants i have chosen are lag roa, size , credit risk ratio, a few more ratios as well and inflation , gdp and interest rates.

I have a few issues that i need to clarify and i really hope i can get some insights from you guys.

1) firstly i did pooled regression and used the xtset command, my data is identified as unbalanced panel snd over 64 quarters. ( this includes the lagroa variable).

Then i go on to run my fe and re using the same variables as the pooled but the problem i face is in the hausman test. My test is giving me a positive definite error and i did some research and tried to change the commands to fe_all, re_all , store both of them and then use the fommand " hausman fe_all re_all, sigmamore. And i do not get the error anymore. Now my question is can i actually do this to solve the problem or am i forcing the data to work my way by using this commands? Is there an underlying problem i am not seeing? Im really confused. Please do give me your insights.

[Carlo Lazzaro]

Sonia:
welcome to this forum.
Although exceptions do exist, usually -xtreg- outperforms pooled OLS when it comes to panel data.
It is pretty frequent that -hausman- gives back the message you mention.
The -sigmamore- option often fixes the issue.
Eventually, your code for -hausman- is correct.
As an aside, please note that Stata can handle both balanced and unbalanced panel datasets, the latter with no extra effort/details from your side.

[sonia kaur]

Carlo Lazzaro thats great news!! Thankyou so so much! Can i just check with you , if my intial data is auto set as quarter for this regression is there any issues i might face as i am more familiar with running regressions using yearly data. So i was wondering if the error i was receiving could be because of that.

[Carlo Lazzaro]

Sonia:
I do not think that quarters are an issue.
-hausman- works asyntotically: hence, stumbling upon some methodological cabblestone along the way is the rule rather than the exception.
On a different note, I would be more concerned about: endogeneity; model misspecification (eg, non-linearity) and (less relevant but sometimes annoying) heteroskedasticity.

[sonia kaur]

Carlo Lazzaro okay carlo!! Thankyou so much! Are there any tests you suggest i run to check for those problems mentioned above?? Cause ive done a fe re and hausman test now ( thankgod it was correct) . My next test is gmm. Do you think gmm will solve those issues ( if present) ?

[Carlo Lazzaro]

Sonia:
see https://www.statalist.org/forums/for...ity-panel-data.
I do not think that

...gmm will solve those issues ( if present)...