hausman检验选择re,还是fe?

33604

收藏 2018-05-11

看了网上的帖子，知道怎么做hausman检验（如下），但是结果我不知道该如何选择

xtreg ROE Rate LnSize Lev Owner1 Rel indeRate State,fe
note: State omitted because of collinearity

Fixed-effects (within) regression             Number of obs    =    1,451
Group variable: Stkcd_int                      Number of groups  =       570

R-sq:                                        Obs per group:
   within  = 0.0340                                        min =       1
   between = 0.0234                                        avg =       2.5
   overall = 0.0202                                        max =       9

                                             F(6,875)       =    5.14
corr(u_i, Xb)  = -0.2482                      Prob > F       =    0.0000

------------------------------------------------------------------------------
      ROE |    Coef. Std. Err.    t P>|t|    [95% Conf. Interval]
-------------+----------------------------------------------------------------
      Rate | 1.264523 .5174366    2.44 0.015    .2489611 2.280085
   LnSize | .0057314 .0065719    0.87 0.383 -.0071671    .01863
      Lev | -.094453 .0324405 -2.91 0.004 -.1581234 -.0307827
   Owner1 | .2114925 .0600829    3.52 0.000    .0935691 .3294159
      Rel |  -.0092853 .0101952 -0.91 0.363 -.0292952 .0107246
indeRate | -.106998 .0863222 -1.24 0.215 -.2764209 .0624248
   State |       0  (omitted)
   _cons |  -.0095531 .1516732 -0.06 0.950 -.3072389 .2881328
-------------+----------------------------------------------------------------
   sigma_u |  .07853109
   sigma_e | .0664082
      rho |  .58306009 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(569, 875) = 2.72                   Prob > F = 0.0000

. est store fe

. xtreg ROE Rate LnSize Lev Owner1 Rel indeRate State,re

Random-effects GLS regression                Number of obs    =    1,451
Group variable: Stkcd_int                      Number of groups  =       570

R-sq:                                        Obs per group:
   within  = 0.0154                                        min =       1
   between = 0.0838                                        avg =       2.5
   overall = 0.0795                                        max =       9

                                             Wald chi2(7)    =    72.94
corr(u_i, X) = 0 (assumed)                   Prob > chi2    =    0.0000

------------------------------------------------------------------------------
      ROE |    Coef. Std. Err.    z P>|z|    [95% Conf. Interval]
-------------+----------------------------------------------------------------
      Rate | 1.391345 .4053853    3.43 0.001    .5968044 2.185885
   LnSize | .0200519 .0030728    6.53 0.000    .0140294 .0260744
      Lev |  -.0977056 .0179754 -5.44 0.000 -.1329367 -.0624745
   Owner1 | .0821221 .0212242    3.87 0.000    .0405234 .1237208
      Rel |  -.0011336 .0057371 -0.20 0.843 -.0123781 .0101108
indeRate |  -.0620039 .0467672 -1.33 0.185    -.153666 .0296582
   State | .0087122 .0105634    0.82 0.410 -.0119917 .0294161
   _cons |  -.3104223 .0669143 -4.64 0.000 -.4415719 -.1792728
-------------+----------------------------------------------------------------
   sigma_u | .0562195
   sigma_e | .0664082
      rho |  .41748321 (fraction of variance due to u_i)
------------------------------------------------------------------------------

. est store re

. hausman fe

               ---- Coefficients ----
         |    (b)       (B)          (b-B)    sqrt(diag(V_b-V_B))
         |    fe          re       Difference       S.E.
-------------+----------------------------------------------------------------
      Rate | 1.264523    1.391345    -.1268219       .3215641
   LnSize | .0057314    .0200519    -.0143205       .0058093
      Lev | -.094453 -.0977056       .0032526       .0270051
   Owner1 | .2114925    .0821221       .1293704       .0562093
      Rel | -.0092853 -.0011336    -.0081517       .0084278
indeRate | -.106998 -.0620039    -.0449941       .0725559
------------------------------------------------------------------------------
                        b = consistent under Ho and Ha; obtained from xtreg
         B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test:  Ho:  difference in coefficients not systematic

               chi2(6) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                        =    24.76
            Prob>chi2 =    0.0004

扫码加我拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

全部回复

leeyooyun

2018-5-11 11:35:51

接上一楼
. hausman fe re,constant sigmamore

               ---- Coefficients ----
         |    (b)       (B)          (b-B)    sqrt(diag(V_b-V_B))
         |    fe          re       Difference       S.E.
-------------+----------------------------------------------------------------
      Rate | 1.264523    1.391345    -.1268219       .3214884
   LnSize | .0057314    .0200519    -.0143205       .0058087
      Lev | -.094453 -.0977056       .0032526       .0270015
   Owner1 | .2114925    .0821221       .1293704       .0562035
      Rel | -.0092853 -.0011336    -.0081517       .0084267
indeRate | -.106998 -.0620039    -.0449941       .0725465
   _cons | -.0095531 -.3104223       .3008693       .1360995
------------------------------------------------------------------------------
                        b = consistent under Ho and Ha; obtained from xtreg
         B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test:  Ho:  difference in coefficients not systematic

               chi2(7) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                        =    23.02
            Prob>chi2 =    0.0017
            (V_b-V_B is not positive definite)
. local model "fe re"

. esttab `model',mtitle(`model')

--------------------------------------------
                  (1)          (2)
                     fe             re
--------------------------------------------
Rate             1.265*       1.391***
               (2.44)       (3.43)

LnSize          0.00573       0.0201***
               (0.87)       (6.53)

Lev             -0.0945**    -0.0977***
               (-2.91)       (-5.44)

Owner1             0.211***    0.0821***
               (3.52)       (3.87)

Rel             -0.00929       -0.00113
               (-0.91)       (-0.20)

indeRate          -0.107       -0.0620
               (-1.24)       (-1.33)

State                0       0.00871
                  (.)       (0.82)

_cons          -0.00955       -0.310***
               (-0.06)       (-4.64)
--------------------------------------------
N                   1451          1451
--------------------------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

以上是我的跑出来的结果，不明白hausman fe和hausman fe re,constant sigmamore有什么区别？
我一开始用hausman fe，根据Prob>chi2 =    0.0004选择fe
后来发现hausman fe re,constant sigmamore会出现 (V_b-V_B is not positive definite)，并且从输出的结果看，re比fe要更显著一些
请问我应该选择re,还是fe？

另外，这个可以当做稳健性检验么？还是需要另外再做稳健性检验？

扫码加我拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

Splendid_Yeung

2018-5-16 17:29:50

hausman fe re,constant sigmamore执行命令后，Prob>chi2 = 0.0017，P值小于0.05，则选择固定效应模型。
应该不能作为稳健性检验，看到的论文Hausman检验更多是判断选择固定效应还是随机效应模型。
hausman fe re,constant sigmamore这个命令，论坛中有人说，在进行Hausman 检验时，有可能会遇到检验统计量为负的情形，这时候要求添加 sigmamore 或者 sigmaless 选项，作为修正。