# st: Different Sargan test results in xtabond Stata9 and xtabond Stata10

 From W.L.Zang@Bradford.ac.uk To statalist@hsphsun2.harvard.edu Subject st: Different Sargan test results in xtabond Stata9 and xtabond Stata10 Date Fri, 14 Aug 2009 12:57:14 +0100

```Dear all

I understand that xtabond in GMM model reports the coefficients of the first
differenced model in Stata9 and reports coefficients of the level model in
Stata10.  However, I am not sure why Stata9 and Stata10 produce different
Sargan test results.  For example,

In stata 9
. xtabond y dummy01-dummy06, pre(x1, lag(0,.)) pre(x2, lag(0,.)) pre(x3,
lag(0,.)) pre(x4, lag(0,.)) noconstant
Arellano-Bond dynamic panel-data estimation     Number of obs      =       157
Group variable (i): country                     Number of groups   =        23

Wald chi2(.)       =         .
Time variable (t): year                         Obs per group: min = 6
avg =  6.826087
max =         7
One-step results
D.y      Coef.   Std. Err.      z    P>z     [95% Conf. Interval]
y
LD.    .4557214   .0657212     6.93   0.000     .3269103    .5845325
x1
D1.     27.0095   15.70867     1.72   0.086    -3.778926    57.79792
x2
D1.    1.282466   .5059736     2.53   0.011     .2907759    2.274156
x3
D1.   -.4017822   .3392372    -1.18   0.236    -1.066675    .2631106
x4
D1.   -.0032462   .4207785    -0.01   0.994    -.8279568    .8214645
dummy01
D1.   -5.386602   1.854018    -2.91   0.004    -9.020411   -1.752793
dummy02
D1.   -3.725648   1.957494    -1.90   0.057    -7.562265    .1109692
dummy03
D1.    -4.19972   2.157489    -1.95   0.052    -8.428322    .0288815
dummy04
D1.   -8.418777    2.30855    -3.65   0.000    -12.94345   -3.894103
dummy05
D1.    -6.44409   2.577985    -2.50   0.012    -11.49685   -1.391331
dummy06
D1.   -7.722238   2.969409    -2.60   0.009    -13.54217   -1.902302
Sargan test of over-identifying restrictions:
chi2(163) =   171.09     Prob > chi2 = 0.3164
Arellano-Bond test that average autocovariance in residuals of order 1 is 0:
H0: no autocorrelation   z =  -3.99   Pr > z = 0.0001
Arellano-Bond test that average autocovariance in residuals of order 2 is 0:
H0: no autocorrelation   z =  -1.12   Pr > z = 0.2628

In stata 10
. xtabond y dummy01-dummy06, pre(x1, lag(0,.)) pre(x2, lag(0,.))
pre(x3,lag(0,.)) pre(x4, lag(0,.)) noconstant
Arellano-Bond dynamic panel-data estimation
Number of obs         =       157
Group variable: country
Number of groups      =        23
Time variable: year                      Obs per group:    min =         6
avg =  6.826087
max =         7
Number of instruments =    127           Wald chi2(11)         =    104.34
Prob > chi2           =    0.0000
One-step results
y      Coef.   Std. Err.      z    P>z     [95% Conf. Interval]
y
L1.    .4557214   .0657212     6.93   0.000     .3269103    .5845325
x1      27.0095   15.70867     1.72   0.086    -3.778925    57.79792
x2     1.282466   .5059736     2.53   0.011      .290776    2.274156
x3    -.4017822   .3392372    -1.18   0.236    -1.066675    .2631106
x4    -.0032462   .4207785    -0.01   0.994    -.8279568    .8214645
dummy01   -5.386602   1.854018    -2.91   0.004    -9.020411   -1.752794
dummy02   -3.725647   1.957494    -1.90   0.057    -7.562264    .1109696
dummy03    -4.19972   2.157489    -1.95   0.052    -8.428322    .0288813
dummy04   -8.418778    2.30855    -3.65   0.000    -12.94345   -3.894103
dummy05   -6.444089   2.577985    -2.50   0.012    -11.49685   -1.391331
dummy06   -7.722238   2.969409    -2.60   0.009    -13.54217   -1.902302
Instruments for differenced equation
GMM-type: L(2/.).y L(1/.).x1 L(1/.).x2 L(1/.).x3 L(1/.).x4
Standard: D.dummy01 D.dummy02 D.dummy03 D.dummy04 D.dummy05 D.dummy06
. estat sargan
Sargan test of overidentifying restrictions
H0: overidentifying restrictions are valid
chi2(116)    =  171.0927
Prob > chi2  =    0.0007

The coefficients in stata9 and stata10 are the same.  My question is why Sargan
test results are different. In Stata9, chi2(163) =171.09 with p value 0.3164
and in Stata10,  chi2(116) =171.0927 with p value 0.0007? Which Sargan test
result should I follow, Stata9 model or Stata10 model?

Thank you very much for your help.

Best Wishes,

Linda

------------------------------------------------------------
This mail sent through IMP: http://webmail.brad.ac.uk
To report misuse from this email address forward the message