Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • xtabond2 with missing ar(2) test


    Dear all,

    I am estimating system gmm on a panel with a gap. In fact I have a panel 2006-2014, but 2011 is a missing year.
    The results I get after applying xtabond 2 are the following.


    Code:
    command:
    xtabond2 y l.y  l2.y  x y z i.year if ser==0, /*
    */ robust twostep cluster(id) small orthog /*
    */ gmm(L.y,  lag(2 5)) gmm(L2.y, lag(3 .)) /*
    */ iv(yr1-yr8 x y z)
    
    ------------------------------------------------------------------------------
    Arellano-Bond test for AR(1) in first differences: z =  -6.05  Pr > z =  0.000
    Arellano-Bond test for AR(2) in first differences: z =      .  Pr > z =      .
    ------------------------------------------------------------------------------
    Sargan test of overid. restrictions: chi2(2)    =  13.86  Prob > chi2 =  0.001
      (Not robust, but not weakened by many instruments.)
    Hansen test of overid. restrictions: chi2(2)    =   2.55  Prob > chi2 =  0.280
      (Robust, but weakened by many instruments.)
    
    Difference-in-Hansen tests of exogeneity of instrument subsets:
      gmm(L2.y, lag(3 .))
        Hansen test excluding group:     chi2(0)    =   2.42  Prob > chi2 =      .
        Difference (null H = exogenous): chi2(2)    =   0.13  Prob > chi2 =  0.939
    My question is: why is AR(2) test missing?

  • #2
    The ar tests are missing because you have year fixed effects dummy variables in the model

    Comment


    • #3
      Ammari, thank you for the response. I think that time dummies are necessary in the system gmm model and suggested in the paper by Roodman too. there are other models i have estimated, with time dummies, but the ar2 test is reported. .i was wondering if the problem lies in the missing year in the panel? or possibly second lag of dependent variable which somehow distorts the model?

      Comment


      • #4
        If the 2011 data is missing, that would mean that you cannot estimate an AR(2) model in first differences because you never observe the 3rd lag in levels; hence the missing value for the AR(2) test.

        Regarding the time dummies, I recommend to have a look at the following Statalist discussion:
        System GMM - Time dummies
        https://twitter.com/Kripfganz

        Comment


        • #5
          Hi all. I am facing the same situation as Mina (to be precise, I only have 4 periods, but because of the lagged dependent variable in system GMM, I am left with only three periods, and at least 4 periods are necessary for the AR(2) test). When this is the case, which is the best way moving forward? Is there an alternative test for serial correlation that we can use in the context of system GMM? or should we just state in the paper the impossibility of conducting that test? Thank you!

          Comment


          • #6
            After the latest version of xtdpdgmm, you can use the new estat serialpm postestimation command, which uses the Jochmans portmanteau test for serial correlation. This test already works for 3 time periods.

            See: https://www.statalist.org/forums/for...45#post1676045

            More on GMM estimation of linear dynamic panel data models:
            Last edited by Sebastian Kripfganz; 16 Oct 2023, 05:27.
            https://twitter.com/Kripfganz

            Comment

            Working...
            X