two-step system GMM - Query!

[Abdullah Ijaz]

Hello everyone,

I have got a query related to two-step system GMM while using xtbond2 command. I am posting my code and results here and would like to clarify certain doubts to make sure that I have applied this model in the right manner. I understand that right specification of model depends on theoretical backing, I just want to know whether these results make sense econometrically.

I am trying to understand the link between conflict and business. There are 120 countries in sample covering period 2006-2018. Hence, my GMM code is:

xtabond2 BR BRlagged conflict GDP credit procedures trade Gov i.Year, gmm(BRlagged conflict GDP credit procedures trade Gov, collapse) iv( i.Year, equation(level)) nodiffsargan twostep robust orthogonal small

Where BR is business rate (my dependent var), conflict (independent var) and rest are my controls.

- For gmm() style, I assume my independent controls including the laggedBR can be endogenous so I have used all of my controls + independent var.
- For IV() style, I am only using i.Year as I assume that i.Year wont be endogenous in first difference.

This is the output I get:

Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: countrynum                      Number of obs      =      1238
Time variable : Year                            Number of groups   =       120
Number of instruments = 104                     Obs per group: min =         1
F(20, 119)    =     14.05                                      avg =     10.32
Prob > F      =     0.000                                      max =        13
------------------------------------------------------------------------------
             |              Corrected
          BR |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    BRlagged |    .365565   .0780428     4.68   0.000     .2110324    .5200976
    conflict |  -.0003031   .0001801    -1.68   0.095    -.0006596    .0000535
         GDP |    .358728   .2343814     1.53   0.129    -.1053706    .8228266
      credit |   .1314865   .2481946     0.53   0.597    -.3599635    .6229365
  procedures |  -.0049294   .0119675    -0.41   0.681    -.0286263    .0187675
       trade |   .0094576   .0043159     2.19   0.030     .0009116    .0180036
         Gov |  -.0216178   .5368458    -0.04   0.968    -1.084626    1.041391
             |
        Year |
       2006  |          0  (empty)
       2007  |   .2016201   .1241664     1.62   0.107    -.0442419    .4474821
       2008  |    .117198   .1279461     0.92   0.362     -.136148     .370544
       2009  |   .1438569   .1492302     0.96   0.337    -.1516339    .4393476
       2010  |   .1786523   .1530077     1.17   0.245    -.1243182    .4816227
       2011  |   .2129109   .1419142     1.50   0.136    -.0680934    .4939152
       2012  |    .226956   .1463006     1.55   0.123    -.0627337    .5166458
       2013  |   .2539669   .1463097     1.74   0.085    -.0357409    .5436747
       2014  |   .2644678   .1424316     1.86   0.066     -.017561    .5464965
       2015  |    .297177   .1476913     2.01   0.046     .0047335    .5896205
       2016  |   .3377692   .1450907     2.33   0.022     .0504751    .6250633
       2017  |   .3124527    .148403     2.11   0.037        .0186    .6063055
       2018  |   .3404902   .1484727     2.29   0.024     .0464994    .6344811
             |
       _cons |  -4.485134   2.544286    -1.76   0.080    -9.523074    .5528061
------------------------------------------------------------------------------
Instruments for orthogonal deviations equation
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    L(1/12).(BRlagged conflict GDP credit procedures trade Gov) collapsed
Instruments for levels equation
  Standard
    2006b.Year 2007.Year 2008.Year 2009.Year 2010.Year 2011.Year 2012.Year
    2013.Year 2014.Year 2015.Year 2016.Year 2017.Year 2018.Year
    _cons
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    D.(BRlagged conflict GDP credit procedures trade Gov) collapsed
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z =  -2.24  Pr > z =  0.025
Arellano-Bond test for AR(2) in first differences: z =   0.22  Pr > z =  0.829
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(83)   = 519.95  Prob > chi2 =  0.000
  (Not robust, but not weakened by many instruments.)
Hansen test of overid. restrictions: chi2(83)   =  92.93  Prob > chi2 =  0.214
  (Robust, but weakened by many instruments.)

As the results suggest, and also looking at AR and Hansen tets, it seems my model is right econometrically. However, do you see any major issues with the code and this model in econometric sense?

Looking forward to your replies and feedback!

[Jared Greathouse]

The man to speak to about this would likely be Sebastian Kripfganz. However to really know if it makes sense, you'd need to give a little more detail on the specifics of the question.

[Abdullah Ijaz]

The man to speak to about this would likely be Sebastian Kripfganz. However to really know if it makes sense, you'd need to give a little more detail on the specifics of the question.

Thanks for your quick feedback.

My primary question is that for iv() option, is it fine to take years as exogeneous. Hopefully, there is nothing wrong with that. Also, wanted to know if the code is right or I am just using it in the wrong way.

[Sebastian Kripfganz]

First of all, it seems that you are using an outdated version of xtabond2 which still contained a couple of bugs. Please update to the latest version!

I would specify time dummies in the same way as you did.

The default settings for the gmm() option are appropriate for predetermined variables (which normally includes the lagged dependent variable) but not for endogenous variables. For the latter, only instruments from lag 2 onwards are valid.

You might find the following presentation slides useful:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

[Abdullah Ijaz]

Sebastian Kripfganz Many thanks for your answer and explaining this to me. Just a quick question on this: The default settings for the gmm() option are appropriate for predetermined variables (which normally includes the lagged dependent variable) but not for endogenous variables.

You said normally includes lagged dep vars. So can it be completely wrong to include (independent vars or controls vars who are treated as endogeneous)? I have seen the similar method in one of papers. So i got confused.

[Sebastian Kripfganz]

With the default settings, you cannot use endogenous variables in gmm(). You need to add the lags() suboption with a minimum lag of 2.