Thank you very much for your kind help. I will follow your instructions.

I agree with Omer in using Hansen test.

The best that you have to do is to use both of them tests and try to find instruments that will give you better Pvalues. However, if u want to goose only one of them then I agree with the previous answers and thus Hansen Jstat is preferable. 

I am using STATA command xtabond (gmm) and I am also struggling with tests.

My syntax in Stata:

xtabond depvar indep1 indep2 indep3 year, twostep vce(gmm)

estat abond

estat sargan

Also, I did AB tests (Arellano-Bond test for zero autocorrelation in first-differenced errors: results: in first differences z=-1.5628, p>0.1181; in second differences z=1.3832, p> 0.1666 ), and Sargan test of overidentifying restrictions (chi2(43)=12.97, p>chi2=1). Is it still necessary to do Hansen test?

Can you help me with Stata command for Hansen?

Thanx!

doubt me too!

hello. Can you help me with Stata command for Hansen test after two-step SYS-Gmm vce (robust) estimation?Everery body answers, i will be grateful.

To put it simply, both tests are useful to test instrument validity. Personally, I run both tests when running systemt-GMM models. To the best of my knowledge, results from running Hansen or Sargan tests are in consensus. However, assuming that these two tests provide contradictory results, I would prefer to choose different instruments in order to satisfy both tests.

On the other hand, take special attention to instrument proliferation issue (i.e. the use of too many instruments). To avoid this - since you are applying Roodman's xtabond2 - use the "collapse option" in STATA.

Good luck!

Zohreh Asadi and @ Danijela Sokolic

You can do like this

xtabond2 y x1 x2 ....

eststo result_1

esttab result_1 ,label se stats(N j ar1p ar2p hansenp, labels("Observarsions" "No. of instruments" "AR1 (p-value)" "AR2 (p-value)" "Hansen-J (p-value)"))

Cheers,

Duc-Nguyen.

Hi Nguyen, how to report Difference-in-Hansen tests in your commands? Thanks so much

Hansen test - strongly recommended.

@ Tien Thong Nguyen,

Use this command to obtain Difference-in-Hansen stat:

xtabond2 y l.y x1 x2 x3 y*, gmm(l.y) iv(x1 x2 x3 y*) twostep robust orthogonal small

Ngozi, your videos are really great. Thank you for sharing those.

The P-value of Sargan test is very sensitive to the proliferation of instruments... for more details see (Roodman, 2009) .

Thanks for the positive feedback @Elma Satrovic, deeply appreciated! 💕 I hope I have the honor of getting your subscription?😃

@Zohreh Asadi, I have detailed videos on GMM estimation techniques on my YouTube Channel https://www.youtube.com/c/CrunchEconometrix. Kindly watch them. You'll be glad you did! 😊

Hi Nguyen, have you use 'outreg2' to exported the result of the regression? If you have done so, could you please tell me how to add AR(1), AR(2) and Sargan test in the outreg2 formulation? Thank you so much!

Dear Churun Liu : I use esttab command to export the results (see earlier comment)

both

Dear Duc Nguyen Nguyen

I would like to know how I do to include the Sargan test statistics on the command line you published (esttab). I appreciate it if I can help.

Bosede Ngozi Adeleye , i've watched your videos and they are amazing. Thank you for sharing

@Ngozi ADELEY, your video is really helpful for the whole economic research community. It really helped me a lot. Thank you so much.

What should be the p values for Hansen test?

Dear researchers i'm using Stata command xtabond2 and my pvalues for sargan is 0.08 and for hansen its 1. i'm using GDP as my dependent variable. i used many instruments but still hansen comes as 1. what should i do to make hansen less than 1?

thanks in advance

Churun Liu Once you have written the command for GMM. Export your results with the following command

outreg2 using results.doc, se bdec(3) e(ar1p ar2p sarganp hansenp) word replace

If the p value of sargan test is1; it means there are so many instruments in the model. Collapse option is required according to Roodman(2009)

but still appeared same problem with collapse option is TRUE in PVARGMM in R.

Just add the “collapse” to the gmm command, because your instruments are overmuch.

try the collapse option and play around with your model

Careful, your null hypothesis is wrong. The test of overidentifying assumptions has the null hypothesis as follows.

First notation: assume a model with k endogenous regressors, w exogenous regressors and g excluded instruments. Then, in order to identify the equation, you need to have g=k excluded instruments. In this case the Sargan-Hansen test cannot be used.

If g>k then you have overidentification. Then the null hypothesis is that g-k excluded instruments are uncorrelated with the error-term (weakly exogenous) provided that at least k excluded instruments are valid.

So our of the g excluded instruments, you can test the validity of g-k instruments but you can never test validity of all instruments (as a set).

You can pick any from Sargan and Hansen.

Collapse and Hansen its better

Sargan

If you use both; one for robustness then it will be good.

report Sargen , if you use one-step.. report Hansen, if you use two-step

Nariman Sayd , why?