先认真阅读一些相关文献。面板数据集的截面纬度应大于时间纬度（N〉T）。估计后可从残差自相关检验（要求差分残差服从一阶负相关，无二阶和更高阶的相关）和sargen/hansen检验来检验工具变量的有效性，sargan是用于同方差，hansen检验更为一般可适用于异方差。gauss程序dpd98或stata程序xtabond2是估计GMM的现成程序，但gauss程序容易出错，还是xtabond2好用一些。

 xi: xtabond2 d.x dl(1/1).x  dl(1/1).y, gmm( x y, lag(2 8)c )

>   iv(i.year,equation(level)) noc  small robust h(1)

i.year            _Iyear_1980-2006    (naturally coded; _Iyear_1980 omitted)

Favoring space over speed. To switch, type or click on mata: mata set matafav

> or speed, perm.

Warning: Number of instruments may be large relative to number of observation

> s.

Warning: Two-step estimated covariance matrix of moments is singular.

  Using a generalized inverse to calculate robust weighting matrix for Hansen

>  test.

  Difference-in-Sargan statistics may be negative.

Dynamic panel-data estimation, one-step system GMM

-----------------------------------------------------------------------------

> -

Group variable: id                              Number of obs      =       95

> 0

Time variable : year                            Number of groups   =        3

> 8

Number of instruments = 41                      Obs per group: min =        2

> 5

F(2, 38)      =    412.71                                      avg =     25.0

> 0

Prob > F      =     0.000                                      max =        2

> 5

-----------------------------------------------------------------------------

> -

             |               Robust

       D.x |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval

> ]

-------------+---------------------------------------------------------------

> -

         x |

         LD. |   .1555062   .0066141    23.51   0.000     .1421166    .168895

> 8

       y |

         LD. |    .034008   .0368939     0.92   0.362    -.0406797    .108695

> 8

-----------------------------------------------------------------------------

> -

Instruments for first differences equation

  GMM-type (missing=0, separate instruments for each period unless collapsed)

    L(2/8).(x y) collapsed

Instruments for levels equation

  Standard

    _Iyear_1981 _Iyear_1982 _Iyear_1983 _Iyear_1984 _Iyear_1985 _Iyear_1986

    _Iyear_1987 _Iyear_1988 _Iyear_1989 _Iyear_1990 _Iyear_1991 _Iyear_1992

    _Iyear_1993 _Iyear_1994 _Iyear_1995 _Iyear_1996 _Iyear_1997 _Iyear_1998

    _Iyear_1999 _Iyear_2000 _Iyear_2001 _Iyear_2002 _Iyear_2003 _Iyear_2004

    _Iyear_2005 _Iyear_2006

  GMM-type (missing=0, separate instruments for each period unless collapsed)

    DL.(x y) collapsed

-----------------------------------------------------------------------------

> -

Arellano-Bond test for AR(1) in first differences: z =  -1.08  Pr > z =  0.28

> 1

Arellano-Bond test for AR(2) in first differences: z =   0.91  Pr > z =  0.36

> 1

-----------------------------------------------------------------------------

> -

Sargan test of overid. restrictions: chi2(39)   = 310.42  Prob > chi2 =  0.00

> 0

  (Not robust, but not weakened by many instruments.)

Hansen test of overid. restrictions: chi2(39)   =  37.82  Prob > chi2 =  0.52

> 4

  (Robust, but can be weakened by many instruments.)

Difference-in-Hansen tests of exogeneity of instrument subsets:

  GMM instruments for levels

    Hansen test excluding group:     chi2(37)   =  37.64  Prob > chi2 =  0.44

> 0

    Difference (null H = exogenous): chi2(2)    =   0.18  Prob > chi2 =  0.91

> 2

  gmm(x y, collapse lag(2 8))

    Hansen test excluding group:     chi2(23)   =  22.73  Prob > chi2 =  0.47

> 6

    Difference (null H = exogenous): chi2(16)   =  15.09  Prob > chi2 =  0.51

> 8

  iv(_Iyear_1981 _Iyear_1982 _Iyear_1983 _Iyear_1984 _Iyear_1985 _Iyear_1986

> _Iyear_1987 _Iyear_1988 _Iyear_1989 _Iyear_1990 _Iyear_1991 _Iyear_1992 _Iy

> ear_1993 _Iyear_1994 _Iyear_1995 _Iyear_1996 _Iyear_1997 _Iyear_1998 _Iyear

> _1999 _Iyear_2000 _Iyear_2001 _Iyear_2002 _Iyear_2003 _Iyear_2004 _Iyear_20

> 05 _Iyear_2006, eq(level))

    Hansen test excluding group:     chi2(14)   =  17.87  Prob > chi2 =  0.21

> 3

    Difference (null H = exogenous): chi2(25)   =  19.95  Prob > chi2 =  0.74

> 9