Weak instriments after ivreg2

[Dario Maimone Ansaldo Patti]

Dear All,

I am running an IV estimation using the ivreg2 command. When looking at the tests reported at the end of the table I have some doubts. Below the results of the tests:

Underidentification test (Kleibergen-Paap rk LM statistic):             11.832
                                                   Chi-sq(5) P-val =    0.0372
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):               19.474
                         (Kleibergen-Paap rk Wald F statistic):         23.098

I use clusterd standard errors. The Underidentification test checls if the instruments are correlated with the endogenous variable. This is clear to me. The Kleibergen-Paap test in my understanding checks if that correlation is weak. Reading this post I read "Strictly speaking, the Kleibergen-Paap statistic can be used for tests of underidentification, but not for weak identification". Therefore it is suggested that the command weakivtest, implementing the test developed by Montiel-Olea and Pflueger (2013) should be used instead.

I cannot understand this point. When reading the aforementioned test, it is said that Monterl-Olea and Pflugger test is meant to check weak instruments in linear instrumental variables regression that is robust to heteroscedasticity, autocorrelation, and clustering. My questions:

1) Which is the difference between Montiel-Olea and Pflueger and Kleibergen-Paap tests? Are they testing different things? Is the difference due to the fact that Montiel-Olea and Pflueger test is robust to heteroskedasticity, autocorrelation and clustering?

2) May I rely on Kleibergen-Paap test to confirm that my instruments are not weak?

Before incidently "discovering" the above post, I was used to rely on Kleibergen-Paap test to confirm that my instruments are strong. But now I am a little bit confused.

Thanks in advance.

Dario

Thanks for your attention

[George Ford]

Rejecting the Underid test is not good. The weakid test is provided so use that one.

[Dario Maimone Ansaldo Patti]

George Ford thanks!