How big P-Value should we have in Sargan Test by Xtabond2 if we want to publish?

[Trang Phan]

Dear Statatist,

I get P-value for Sargan Test in Xtabond2 by 0.115. My econometrics teacher only told about the number of 0.05 during the class. However, when I asked my advisor on the result, he said that I can not plublish my paper if P-value of Sargan is only 0.115. I refered some articles to see in the line of Sargan Test, the numbers are around 3. For example, they release Sargan Test 3.38 (19). Then, Is 3.38 P-value?

Do anyone advise me on the safe P-value minimum in order to publish paper?

Thank you!

Trang,

[Sebastian Kripfganz]

The value of 3.38 is the value of the test statistic, not the p-value. The p-value is always between 0 and 1.

There is no general answer to your question about a "safe p-value". You might get an idea from an earlier discussion on Statalist:
P-values for Hansen and difference-in-Hansen tests!!!

[Marcos Almeida]

For example, they release Sargan Test 3.38 (19). Then, Is 3.38 P-value?

Assuming you perfomed the estimations in Stata, the output is straightforward on which is which.

Below, the example from the Stata Journal:

Sargan test of overid. restrictions: chi2(25) = 67.59 Prob > chi2 = 0.000

[Trang Phan]

Thank you Sebastion and Marcos for your replies.

1. Another question, can I break down my instrument like this (I mean that it is reansonable to seperate iv insead of one group iv(firmqual assetmat lnta cacl mvtobv absebit ratevol) to publish result)

xtabond2 debtmat l.debtmat tax firmqual levb assetmat lnta cacl mvtobv mabsebit ratevol, gmm(l.debtmat,laglimits(1 4) ) iv( firmqual,eq(both)) iv(levb,eq(both)) iv(l.assetmat,eq(lev)) iv(lnta,eq(lev) pass) iv(cacl,eq(dif) pass) iv(mvtobv,eq(lev) pass) iv(mvtobv,eq(dif) pass) iv(absebit,eq(lev) pass) iv(absebit,eq(dif) pass) iv(ratevol,eq(lev) pass) twostep robust

2. Why Hansen test is dot (.)

gmm(L.debtmat, lag(1 4))
Hansen test excluding group: chi2(0) = 0.00 Prob > chi2 = .
Difference (null H = exogenous): chi2(29) = 28.01 Prob > chi2 = 0.517

3. number of instruments (29) are too much for sample N=201 and T=9?

Thank you so much

Trang

[Marcos Almeida]

I'm not fully acquainted with this user-written program. That said, with regards to questions 1, 2 and 3, respectively:

You may wish to see the literature as well as the help files, but the best approach (mainly when we are starting to delve with a new program) is abiding by the standard recommendations.
P is a dot because the chi2 has no value, hence Stata cannot make further estimation.
Yes, I think so.

[Sebastian Kripfganz]

1. I strongly advise to never use the iv() option with suboption eq(both), which is unfortunately the default. This is not doing what you expect it to do. The following example illustrates that you obtain different results when you specify the instruments separately for both the first-differenced and the level equation:

webuse abdata
xtabond2 n L.n w, gmm(L.n, eq(diff)) iv(w, eq(both))
xtabond2 n L.n w, gmm(L.n, eq(diff)) iv(w, eq(diff)) iv(w, eq(level))

The second specification is what you actually have in mind. The first one should be avoided.
Irrespective of this point, iv(w k, eq(diff)) is the same as the separate specification iv(w, eq(diff)) iv(k, eq(diff)).

2. The difference-in-Hansen test is not reported because the model is underidentified after excluding your gmm() instruments (the number of instruments is smaller than the number of regressors).

3. In general, your overall instruments count should be fine.

[Marcos Almeida]

With regards to sample size under instrumental variables estimators, maybe you wish to take a look at these texts:

https://www.ncbi.nlm.nih.gov/pubmed/25124167

https://www.ncbi.nlm.nih.gov/pubmed/22152177

https://eml.berkeley.edu/~mcfadden/e240b_f01/ch4.pdf

[Trang Phan]

Dear Sebastian and Marcos

Much appreciate all your helpful comments. I will follow your suggests!

Regards,

Trang