xtabond2 or xtdpdsys? which one should be preferred

[Panos Tzouvanas]

Hello Stata users,

I am relatively new in the dynamic panel estimations. I am facing a problem while I run some DPD regressions. I run the same model with two different codes i) xtabond2 and ii)xtdpdsys. I have posted the code and the results below in order to illustrate my point.
Model 1, xtabond2:

  xtabond2 L(0/1).roa index gg Lvr logbv loge tdum3-tdum12, gmm(l.roa, lag(2 4) collapse) iv(index gg Lvr logbv loge tdum3-tdum12, eq(diff)) iv(index gg Lvr logbv loge tdum3-tdum12, eq(level)) twostep robust
Favoring space over speed. To switch, type or click on mata: mata set matafavor speed, perm.

Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: id                              Number of obs      =       694
Time variable : time                            Number of groups   =       104
Number of instruments = 35                      Obs per group: min =         1
Wald chi2(16) =    230.39                                      avg =      6.67
Prob > chi2   =     0.000                                      max =        11
------------------------------------------------------------------------------
             |              Corrected
         roa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         roa |
         L1. |   .2720502   .2867352     0.95   0.343    -.2899404    .8340408
             |
       index |    .264512   1.555302     0.17   0.865    -2.783825    3.312849
          gg |   3.430854   1.396779     2.46   0.014     .6932173     6.16849
         Lvr |   -.532701   .5594641    -0.95   0.341     -1.62923    .5638285
       logbv |    .152656   .6946184     0.22   0.826    -1.208771    1.514083
        loge |   .3537036   .5354133     0.66   0.509    -.6956871    1.403094
       tdum3 |   -1.22701   .6092102    -2.01   0.044     -2.42104   -.0329798
       tdum4 |  -1.362138   .9582666    -1.42   0.155    -3.240306      .51603
       tdum5 |  -2.954374   .7888081    -3.75   0.000     -4.50041   -1.408339
       tdum6 |  -1.112729   1.077059    -1.03   0.302    -3.223727    .9982686
       tdum7 |  -.8603397   .7190427    -1.20   0.231    -2.269638    .5489581
       tdum8 |  -1.664296    .716404    -2.32   0.020    -3.068422   -.2601695
       tdum9 |  -1.929806   .9292252    -2.08   0.038    -3.751054   -.1085586
      tdum10 |  -2.183652   1.052817    -2.07   0.038    -4.247135   -.1201689
      tdum11 |   -3.10171   1.233736    -2.51   0.012    -5.519789   -.6836317
      tdum12 |  -4.162872   1.499625    -2.78   0.006    -7.102082   -1.223662
       _cons |   2.063816    4.16028     0.50   0.620    -6.090183    10.21781
------------------------------------------------------------------------------
Instruments for first differences equation
  Standard
    D.(index gg Lvr logbv loge tdum3 tdum4 tdum5 tdum6 tdum7 tdum8 tdum9
    tdum10 tdum11 tdum12)
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    L(2/4).L.roa collapsed
Instruments for levels equation
  Standard
    index gg Lvr logbv loge tdum3 tdum4 tdum5 tdum6 tdum7 tdum8 tdum9 tdum10
    tdum11 tdum12
    _cons
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    DL.L.roa collapsed
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z =  -1.54  Pr > z =  0.124
Arellano-Bond test for AR(2) in first differences: z =   0.57  Pr > z =  0.568
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(18)   = 107.54  Prob > chi2 =  0.000
  (Not robust, but not weakened by many instruments.)
Hansen test of overid. restrictions: chi2(18)   =  25.29  Prob > chi2 =  0.117
  (Robust, but weakened by many instruments.)

Difference-in-Hansen tests of exogeneity of instrument subsets:
  GMM instruments for levels
    Hansen test excluding group:     chi2(17)   =  25.08  Prob > chi2 =  0.093
    Difference (null H = exogenous): chi2(1)    =   0.21  Prob > chi2 =  0.647
  gmm(L.roa, collapse lag(2 4))
    Hansen test excluding group:     chi2(14)   =  22.06  Prob > chi2 =  0.077
    Difference (null H = exogenous): chi2(4)    =   3.23  Prob > chi2 =  0.520
  iv(index gg Lvr logbv loge tdum3 tdum4 tdum5 tdum6 tdum7 tdum8 tdum9 tdum10 tdum11 tdum12, eq(diff))
    Hansen test excluding group:     chi2(3)    =   4.55  Prob > chi2 =  0.208
    Difference (null H = exogenous): chi2(15)   =  20.75  Prob > chi2 =  0.145
  iv(index gg Lvr logbv loge tdum3 tdum4 tdum5 tdum6 tdum7 tdum8 tdum9 tdum10 tdum11 tdum12, eq(level))
    Hansen test excluding group:     chi2(3)    =   4.26  Prob > chi2 =  0.235
    Difference (null H = exogenous): chi2(15)   =  21.03  Prob > chi2 =  0.136

Model 2, xtdpdsys:

 xtdpdsys roa index gg Lvr logbv loge tdum3-tdum12, lags(1) maxldep(1) maxlags(2) twostep vce(robust) artests(2)

System dynamic panel-data estimation         Number of obs         =       694
Group variable: id                           Number of groups      =       104
Time variable: time
                                             Obs per group:    min =         1
                                                               avg =  6.673077
                                                               max =        11

Number of instruments =     36               Wald chi2(16)         =     53.03
                                             Prob > chi2           =    0.0000
Two-step results
------------------------------------------------------------------------------
             |              WC-Robust
         roa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         roa |
         L1. |   .0533535    .063961     0.83   0.404    -.0720077    .1787148
             |
       index |   1.046043   2.768901     0.38   0.706    -4.380903    6.472988
          gg |   .7522948   .7417949     1.01   0.311    -.7015964    2.206186
         Lvr |  -2.592315   1.174064    -2.21   0.027    -4.893438   -.2911906
       logbv |   1.001701   4.457059     0.22   0.822    -7.733975    9.737376
        loge |   1.074724   2.146775     0.50   0.617    -3.132878    5.282325
       tdum3 |   -.973988   .7854983    -1.24   0.215    -2.513536    .5655603
       tdum4 |  -.8094584   1.893513    -0.43   0.669    -4.520677     2.90176
       tdum5 |  -2.361437   1.745995    -1.35   0.176    -5.783525    1.060651
       tdum6 |  -.9949049   1.913985    -0.52   0.603    -4.746247    2.756437
       tdum7 |  -.5033503   2.010152    -0.25   0.802    -4.443176    3.436475
       tdum8 |  -1.055596   2.125202    -0.50   0.619    -5.220915    3.109722
       tdum9 |  -1.022225   2.129166    -0.48   0.631    -5.195313    3.150863
      tdum10 |  -.6484834   2.229763    -0.29   0.771    -5.018738    3.721771
      tdum11 |   -1.10356   2.381747    -0.46   0.643    -5.771698    3.564577
      tdum12 |  -2.655637   2.464671    -1.08   0.281    -7.486303    2.175028
       _cons |   4.468879   26.89985     0.17   0.868    -48.25386    57.19162
------------------------------------------------------------------------------
Instruments for differenced equation
        GMM-type: L(2/2).roa
        Standard: D.index D.gg D.Lvr D.logbv D.loge D.tdum3 D.tdum4 D.tdum5
                  D.tdum6 D.tdum7 D.tdum8 D.tdum9 D.tdum10 D.tdum11 D.tdum12
Instruments for level equation
        GMM-type: LD.roa
        Standard: _cons

I know that I have done something wrong because I do not have the same number of instruments and my results are quite contradicting.
Therefore, 1) I would like to know if you can detect any mistakes in the two codes.
2) How can I perform identification tests for txdpdsys? There are some propestimation tests but I would like to check some extra features [e.g. Difference in Hansen (J) test].

Thank you in advance. Any help is highly appreciated!

[Sebastian Kripfganz]

1) xtdpdsys does not collapse instruments. Also, compare the lists of instruments below the two regression tables for further differences in the specifications.
2) The difference-in-Hansen test is not implemented as a postestimation command for xtdpdsys.

In principle, you can do everything with xtabond2 that can be done with xtdpdsys plus many other things.

As an aside: It is not advised to specify the time dummies as instruments for both the first-differenced and the level equation. In general, they should be specified as level instruments only.