Rules of thumb for "acceptable" and "good" p-values of Hansen test and AR(2) test for sys-GMM / diff-GMM?

[Yu Wang]

Dear Stata experts,

I'm working on a growth regression (dependent variable is logGDPpc) using xtabond2.

I found that sys-GMM or diff-GMM is sensitive to model specifications.

Though I "tried" some specifications,

the A-B AR(2) test p-value is around 0.1 (cannot reach 0.2); and

the Hansen test p-value ranges from 0.2 to 0.6.

Thus I wonder, are there any rules of thumb for "acceptable" and "good" p-values of Hansen test and AR(2) test?

Is a p-value of a AR(2)/Hansen test merely exceeding 0.1 acceptable, or any suggestions to remedy this?

Thanks for any advice!

[Yu Wang]

Sorry to bump this thread, any tips would be appreciated !

[Kevin Grier]

well you are actually worried about the power of the test here, as failing to reject the null is validating what you want to do. Power is tricky, but reducing the size (making it easier to reject the null) helps with power. I am not aware of rules of thumb for what is "good enough". Since the Sargan/Hansen test is merely for overidentification and doesn't validate identification, it's really not that important.

[Yu Wang]

well you are actually worried about the power of the test here, as failing to reject the null is validating what you want to do. Power is tricky, but reducing the size (making it easier to reject the null) helps with power. I am not aware of rules of thumb for what is "good enough". Since the Sargan/Hansen test is merely for overidentification and doesn't validate identification, it's really not that important.

Thanks for your reply!

Can I understand in such a way: AR(2) test is more important (than Hansen test, if there is a tradeoff).