Xtabond2 : Instruments and Groups

[Darnell Grant]

My instruments significantly outweigh my groups, using the collapse option does bring this number downward but still not feasible. Any recommendations on what I can change to my the analysis better

. xtabond2 logRGDP l.logRGDP  l.logHEX l.logEFI d.logFDI l.logITR l.logTOT l.logGFC d.logSSE, gmm (l2.logRGDP) iv(logHEX logEFI logFDI logITR logTOT logGFC logSSE)
>  
Favoring speed over space. To switch, type or click on mata: mata set matafavor space, perm.
Warning: Number of instruments may be large relative to number of observations.

Dynamic panel-data estimation, one-step system GMM
------------------------------------------------------------------------------
Group variable: country1                        Number of obs      =        46
Time variable : year                            Number of groups   =         5
Number of instruments = 46                      Obs per group: min =         6
Wald chi2(8)  =  30618.59                                      avg =      9.20
Prob > chi2   =     0.000                                      max =        12
------------------------------------------------------------------------------
     logRGDP |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     logRGDP |
         L1. |   .9834148    .025527    38.52   0.000     .9333829    1.033447
             |
      logHEX |
         L1. |  -.0051808   .0085478    -0.61   0.544    -.0219342    .0115726
             |
      logEFI |
         L1. |  -.2870912   .1323939    -2.17   0.030    -.5465784    -.027604
             |
      logFDI |
         D1. |   .0264561   .0081658     3.24   0.001     .0104514    .0424607
             |
      logITR |
         L1. |   .0081671   .0231965     0.35   0.725    -.0372972    .0536313
             |
      logTOT |
         L1. |   -.083688    .037738    -2.22   0.027    -.1576532   -.0097228
             |
      logGFC |
         L1. |  -.0198461   .0102797    -1.93   0.054     -.039994    .0003019
             |
      logSSE |
         D1. |  -.0790158   .0843087    -0.94   0.349    -.2442577    .0862262
             |
       _cons |   .8132458   .3243634     2.51   0.012     .1775053    1.448986
------------------------------------------------------------------------------
Instruments for first differences equation
  Standard
    D.(logHEX logEFI logFDI logITR logTOT logGFC logSSE)
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    L(1/19).L2.logRGDP
Instruments for levels equation
  Standard
    logHEX logEFI logFDI logITR logTOT logGFC logSSE
    _cons
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    D.L2.logRGDP
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z =  -1.39  Pr > z =  0.165
Arellano-Bond test for AR(2) in first differences: z =  -1.38  Pr > z =  0.168
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(37)   =  57.51  Prob > chi2 =  0.017
  (Not robust, but not weakened by many instruments.)

Difference-in-Sargan tests of exogeneity of instrument subsets:
  GMM instruments for levels
    Sargan test excluding group:     chi2(33)   =  51.99  Prob > chi2 =  0.019
    Difference (null H = exogenous): chi2(4)    =   5.52  Prob > chi2 =  0.238
  iv(logHEX logEFI logFDI logITR logTOT logGFC logSSE)
    Sargan test excluding group:     chi2(30)   =  57.35  Prob > chi2 =  0.002
    Difference (null H = exogenous): chi2(7)    =   0.15  Prob > chi2 =  1.000

[Beri Parfait]

I am not very good at GMM models, but I would suggest that you should reduce the number of explanatory variables. Also, consider adding "robust" and "orthogonal" to your stata code to control for heteroscedasticity. As noted in Roodman(2009), you should also not be excited with the P-value of 1.000 as shown on your Sargan Test for exogeneity. It is preferable to have it at less than or equal to 0.25. Finally, you may wish to consider reviewing the paper "A note on the theme of too many instruments" by David Roodman

[Sebastian Kripfganz]

These GMM estimators are designed for small-T, large-N models. Your N is only 5 groups. There is no way to estimate such a model with such few groups.