Page 1 of 1

Incorrect J-statistic in GMM output??

Posted: Tue May 01, 2018 6:06 am
by oksanakim
Dear Moderators and users.

I am estimating the model using GMM and several instrumental variables in Eviews 10. The output shows the J-statistic and also the probability of J-statistic. My understanding was that this J-statistic is a Sargan statistics and that output is ready to test the Null hypothesis: instruments are valid. However, in the literature, I came across a claim that the GMM-reported J-statistic is actually incorrect and that it should be amended by multiplying the reported one by a number of observations, the p-value has to be recalculated to get the actual J-statistic that can be used to test the null. How true is this?? I am attaching the document where I found this. I am confused. Thank you.

Re: Incorrect J-statistic in GMM output??

Posted: Tue May 01, 2018 7:31 am
by EViews Gareth
That document is extremely old and out of date. If you're using EViews 7, 8, 9, 10 or later, it is not applicable.

Re: Incorrect J-statistic in GMM output??

Posted: Tue May 01, 2018 7:37 am
by oksanakim
I see. I am using an unstructured panel and GMM estimation. So the J-stat with a related probability is basically ready to test for the Null: instruments are valid, correct? I am attaching the file with an example of my estimation, marking the J(Prob of J) that I am inquiring about. Thanks a lot.