Sargan-Hansen test

[Christopher Keegan]

Hi.

I am trying to obtain results for the Sargan-Hansen test. However, when I run a regression using xtabond2, I do not get the Sargan-Hansen test statistic. Is there a piece of code that I am missing??

Kind regards
Christopher Keegan

[Joro Kolev]

Seems to me that the user written command -tabond2- is returning the overidentifying restrictions test:

. use http://www.stata-press.com/data/r7/abdata.dta, clear

. xtabond2 n l.n l(0/1).(w k) yr1980-yr1984, gmm(l.n w k) iv(yr1980-yr1984, passthru) noleveleq small
Favoring space over speed. To switch, type or click on mata: mata set matafavor speed, perm.

Dynamic panel-data estimation, one-step difference GMM
------------------------------------------------------------------------------
Group variable: id                              Number of obs      =       751
Time variable : year                            Number of groups   =       140
Number of instruments = 103                     Obs per group: min =         5
F(10, 741)    =    207.78                                      avg =      5.36
Prob > F      =     0.000                                      max =         7
------------------------------------------------------------------------------
           n |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           n |
         L1. |   .5847306   .0526328    11.11   0.000     .4814035    .6880576
             |
           w |
         --. |  -.7204875   .0807712    -8.92   0.000    -.8790551     -.56192
         L1. |   .2072654   .0500393     4.14   0.000     .1090298    .3055011
             |
           k |
         --. |   .3839347    .052686     7.29   0.000      .280503    .4873663
         L1. |  -.0766793   .0343085    -2.23   0.026    -.1440327   -.0093259
             |
      yr1980 |  -.0308167   .0089215    -3.45   0.001    -.0483312   -.0133023
      yr1981 |  -.0649783   .0110817    -5.86   0.000    -.0867335   -.0432231
      yr1982 |  -.0375605   .0158624    -2.37   0.018     -.068701     -.00642
      yr1983 |   .0006808   .0209536     0.03   0.974    -.0404548    .0418164
      yr1984 |   .0269716   .0271112     0.99   0.320    -.0262523    .0801955
------------------------------------------------------------------------------
Instruments for first differences equation
  Standard
    yr1980 yr1981 yr1982 yr1983 yr1984
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    L(1/8).(L.n w k)
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z =  -7.33  Pr > z =  0.000
Arellano-Bond test for AR(2) in first differences: z =  -0.51  Pr > z =  0.608
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(93)   = 204.53  Prob > chi2 =  0.000
  (Not robust, but not weakened by many instruments.)

Difference-in-Sargan tests of exogeneity of instrument subsets:
  iv(yr1980 yr1981 yr1982 yr1983 yr1984, passthru)
    Sargan test excluding group:     chi2(88)   = 179.89  Prob > chi2 =  0.000
    Difference (null H = exogenous): chi2(5)    =  24.64  Prob > chi2 =  0.000

[Christopher Keegan]

My code is as follows:

xtabond rincome L.rincome

Thats it. But I don't understand why it does not display the Difference-in-Sargan tests.

The error is that the equation is not identified, regressors outnumber instruments.

[Sebastian Kripfganz]

You need to explicitly specify instrumental variables with the iv() or gmm() options. The way you specified it, you have 1 endogenous regressor but 0 instruments.

[Christopher Keegan]

Thank you this helps a lot

[Sandra Canizares]

I have a problem with Sargan-Hansen test in my panel data. I have more than 85,000 obervations and, applying xtabond2 command, I always get a p-value = 0.000 for the Hansen test. Can the problem be related to this large sample? I wonder if this test is valid in this case.
Thank you in advance for your answers.

[Sebastian Kripfganz]

With such a large data set, the Sargan-Hansen test will detect already small deviations from a correctly specified model. Those small deviations may not always have substantial implications for the coefficient estimates, but of course we cannot know how big the problem is unless we can find a specification which satisfies the test. If you cannot determine the source of a misspecification problem, you might simply have to interpret the results with caution.