xtabond2 or xtivreg2

[Enza Testa]

Hi guys,
i'm using STATA 13 to estimate my dyamic panel model. The aim of my thesis is to understand if globalization affects economic growth. I start from a Fixed effects model and I want to spve the endogeneity problem. Since i use a growth model, someone suggests me to use a dynamic one. I use a 5-years panel and the variables are the average over the 5 years. My dependent variable is the GDP growth rate and the regressors are different globalization indices that I used one by one, plus several macroeconomic variables used like controls, I show you an example with one of them. I add time dummies in order to take into account time effects (I add them one by one as suggested in this Forum).

GDPgrowth_5 lnFertility_5 newinvest newGen_exp newoglob are the endogenous variables
newlsc lnLife_5 are the predetermined variables

 xtabond2 GDPgrowth_5 l.GDPgrowth_5 newoglob iGDP newlsc lnLife_5 lnFertility_5 newinvest newGen_exp newInfl2 trade newdbagdp dem_5 y70 y75 y80 y85 y90 y95 y00 y05, gmm (GDPgrowth_5 lnFertility_5 newinvest newGen_exp  newoglob, lag (2 3) eq(diff)) gmm (newlsc lnLife_5, lag (1 2) eq(diff)) iv (newInfl2 dem_5 trade newdbagdp iGDP, eq(diff)) iv(y70 y75 
> y80 y85 y90 y95 y00 y05, eq(level)) twostep nolevel robust small ar(3)
Favoring space over speed. To switch, type or click on mata: mata set matafavor speed, perm.
Instruments for levels equations only ignored since noleveleq specified.
Warning: Two-step estimated covariance matrix of moments is singular.
  Using a generalized inverse to calculate optimal weighting matrix for two-step estimation.
  Difference-in-Sargan/Hansen statistics may be negative.

Dynamic panel-data estimation, two-step difference GMM
------------------------------------------------------------------------------
Group variable: id2                             Number of obs      =       224
Time variable : period                          Number of groups   =        69
Number of instruments = 56                      Obs per group: min =         0
F(20, 69)     =     18.47                                      avg =      3.25
Prob > F      =     0.000                                      max =         4
-------------------------------------------------------------------------------
              |              Corrected
  GDPgrowth_5 |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
  GDPgrowth_5 |
          L1. |  -.0932418   .1432099    -0.65   0.517    -.3789377    .1924541
              |
     newoglob |   .2020836   .0897642     2.25   0.028     .0230089    .3811584
         iGDP |   -.104824   .0246037    -4.26   0.000    -.1539071   -.0557409
       newlsc |   .0821822   .0827541     0.99   0.324    -.0829077     .247272
     lnLife_5 |   .1667886   .0705606     2.36   0.021      .026024    .3075532
lnFertility_5 |   .0314123    .023665     1.33   0.189     -.015798    .0786226
    newinvest |   .2726478   .0819279     3.33   0.001     .1092061    .4360894
   newGen_exp |  -.0661485   .0816824    -0.81   0.421    -.2291005    .0968035
     newInfl2 |   .0081002   .0026686     3.04   0.003     .0027764     .013424
        trade |   .0010489    .000336     3.12   0.003     .0003785    .0017193
    newdbagdp |    .005614   .0203926     0.28   0.784    -.0350681     .046296
        dem_5 |  -.0376581   .0283473    -1.33   0.188    -.0942093    .0188931
          y70 |          0  (omitted)
          y75 |   .0095499   .0068016     1.40   0.165     -.004019    .0231188
          y80 |          0  (omitted)
          y85 |   .0015786   .0043232     0.37   0.716    -.0070458    .0102031
          y90 |    -.00236   .0079309    -0.30   0.767    -.0181817    .0134617
          y95 |   .0027976   .0132303     0.21   0.833    -.0235961    .0291912
          y00 |          0  (omitted)
          y05 |          0  (omitted)
-------------------------------------------------------------------------------
Instruments for first differences equation
  Standard
    D.(newInfl2 dem_5 trade newdbagdp iGDP)
  GMM-type (missing=0, separate instruments for each period unless collapsed)
    L(1/2).(newlsc lnLife_5)
    L(2/3).(GDPgrowth_5 lnFertility_5 newinvest newGen_exp newoglob)
------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z =  -2.08  Pr > z =  0.038
Arellano-Bond test for AR(2) in first differences: z =  -1.12  Pr > z =  0.264
Arellano-Bond test for AR(3) in first differences: z =  -0.09  Pr > z =  0.931
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(36)   =  78.74  Prob > chi2 =  0.000
  (Not robust, but not weakened by many instruments.)
Hansen test of overid. restrictions: chi2(36)   =  43.12  Prob > chi2 =  0.193
  (Robust, but weakened by many instruments.)

Difference-in-Hansen tests of exogeneity of instrument subsets:
  gmm(GDPgrowth_5 lnFertility_5 newinvest newGen_exp newoglob, eq(diff) lag(2 3))
    Hansen test excluding group:     chi2(1)    =   4.41  Prob > chi2 =  0.036
    Difference (null H = exogenous): chi2(35)   =  38.71  Prob > chi2 =  0.306
  gmm(newlsc lnLife_5, eq(diff) lag(1 2))
    Hansen test excluding group:     chi2(20)   =  26.00  Prob > chi2 =  0.166
    Difference (null H = exogenous): chi2(16)   =  17.12  Prob > chi2 =  0.378
  iv(newInfl2 dem_5 trade newdbagdp iGDP, eq(diff))
    Hansen test excluding group:     chi2(31)   =  35.96  Prob > chi2 =  0.247
    Difference (null H = exogenous): chi2(5)    =   7.16  Prob > chi2 =  0.209

I have the following questions:

1 - I tried to avoid the proliferation of instruments using only two lags but I obtain a warning and I don't know if it is related to this problem.
2 - I tried to change globalization indices but the GDP lagged is never significant. This may suggest that a static model using instrumental variables like xtivreg2 (but in this case I have to fins instruments) may better fit to my data to solve the endogeneity problem? Or alternatively, it might make sense to run an AB model without the dependent lagged variable to make it a static model?

I would appreciate some feedbacks (also to my AB model, I'm not 100% sure I've used the AB options well)
Thank you!

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions.

I'm not an AB user. David Roodman has written a nice explanation of the issues.
I'm not sure exactly how you've done this, but if your growth rate is over 5 years and you've only lagged it one year, then there is an undesirable level of overlap in construction of the two variables. You might also want to think about whether growth or level is a better way to represent your model.

[Enza Testa]

Hi Phil thank you for your answer, I read Roodman but I have still some doubts.
If I have gdp at t as dependent variable, the lagged variable is the gdp growth rate at t-1 so in the 5 years before, I think that there is not overlapping.
I know that maybe in a variable in level there is more impact from the lagged variable with respect to a growth rate, but l I chose the growth rate since I want to have the same measure as the static model and compare them.
Do you have any insight regarding the use of the AB in a static way? Someone suggest me this solution but I'm not pretty confident with that. I prefer to use a IV model in this case.

Thank you!