[Raymond Zhang]

Wow,your answers are very helpful for me to understand GMM methods and your command xtdpdgmm.This command has been tweeted and speaded in China.

[Raymond Zhang]

sebastian Dear Pro Sebastian,
you said sys-GMM should holds the additional mean-stationarity assumption.I want to know how to test this mean-stationarity assumption.What Stata command can i use to test it?

Raymond

[Sebastian Kripfganz]

You can test the additional instruments for the level model with a difference-in-Hansen test (aka incremental Hansen test). See slides 48 to 52 of my 2019 London Stata Conference presentation.

Notice that the reliability of this test requires that the diff-GMM estimator is properly identified because this is essentially a test whether the sys-GMM estimator differs systematically from the diff-GMM estimator.

[Raymond Zhang]

So if the incremental overidentification test rejects the validity of the additional moment conditions for the level model, should I use diff-GMM? do you mean if Incremental overidentification test's p-value is larger than 5%,I should choose sys-GMM?

[Raymond Zhang]

Sebastian Kripfganz

[Sebastian Kripfganz]

In a nutshell, yes. Although you need to check the reliability of the diff-GMM estimator first, i.e. initially use the underidentification tests and the Hansen overidentification test for the diff-GMM estimator. If the Hansen test for diff-GMM fails, then the incremental Hansen test for the sys-GMM estimator is not valid.

[Raymond Zhang]

So can I do like this?
Step 1:do underidentification tests and the Hansen overidentification test for the diff-GMM estimator.
Step 2:if the Hansen overidentification test p-value>0.05,we can do the incremental Hansen test for the sys-GMM estimator .
Step 3:If the incremental Hansen test p-value<0.05,we should use Diff-GMM.If the incremental Hansen test p-value>0.05,we should use sys-GMM.

In Step 2,if the Hansen overidentification test p-value<0.05,does it mean both Diff-GMM and sys-GMM are invalid?

[Sebastian Kripfganz]

So can I do like this?
Step 1:do underidentification tests and the Hansen overidentification test for the diff-GMM estimator.
Step 2:if the Hansen overidentification test p-value>0.05,we can do the incremental Hansen test for the sys-GMM estimator .
Step 3:If the incremental Hansen test p-value<0.05,we should use Diff-GMM.If the incremental Hansen test p-value>0.05,we should use sys-GMM.

Basically yes.

In Step 2,if the Hansen overidentification test p-value<0.05,does it mean both Diff-GMM and sys-GMM are invalid?

Yes.

[Raymond Zhang]

Great!I have leraned so much!Thank you for your time!