最近遇到一些棘手的计量问题,因为没有吃透计量,只会运行模型,不明白个中原理,恳请高手不惜赐教,感谢万分!
模型中最重要的两个变量是lnstg lnstic
reg lninno lnstf lnstg lnstl lnstic lnci,r
Linear regression Number of obs = 420
F( 5, 414) = 608.44
Prob > F = 0.0000
R-squared = 0.9159
Root MSE = .43196
------------------------------------------------------------------------------
| Robust
lninno | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnstf | .5640627 .1184742 4.76 0.000 .3311767 .7969486
lnstg | .1531002 .0755189 2.03 0.043 .0046519 .3015486
lnstl | .0496971 .0959242 0.52 0.605 -.1388621 .2382562
lnstic | .1827958 .068943 2.65 0.008 .0472738 .3183178
lnci | .1333685 .0488801 2.73 0.007 .0372844 .2294526
_cons | -6.375354 .3941764 -16.17 0.000 -7.15019 -5.600517
------------------------------------------------------------------------------
. estat vif
Variable | VIF 1/VIF
-------------+----------------------
lnstf | 51.60 0.019379
lnstg | 25.56 0.039117
lnstic | 17.97 0.055649
lnstl | 17.74 0.056360
lnci | 6.57 0.152115
-------------+----------------------
Mean VIF | 23.89
. estat imtest,white
White's test for Ho: homoskedasticity
against Ha: unrestricted heteroskedasticity
chi2(20) = 86.95
Prob > chi2 = 0.0000
Cameron & Trivedi's decomposition of IM-test
---------------------------------------------------
Source | chi2 df p
---------------------+-----------------------------
Heteroskedasticity | 86.95 20 0.0000
Skewness | 17.57 5 0.0035
Kurtosis | 0.11 1 0.7457
---------------------+-----------------------------
Total | 104.62 26 0.0000
普通最小二乘的估计量系数很好,也通过了一定的显著性检验,但也存在很多问题。
参照别人的写法,大多进行固定效应与随机效应的检验。豪斯曼检验的结果如下:
---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+----------------------------------------------------------------
lnstf | .8429847 .9002719 -.0572873 .0471785
lnstg | -.2320736 -.0157891 -.2162846 .055165
lnstl | .0395976 -.2132562 .2528538 .0623562
lnstic | .8765391 .4017317 .4748074 .0716471
lnci | -.1389116 -.0493037 -.0896079 .006211
------------------------------------------------------------------------------
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(5) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 32.67
Prob>chi2 = 0.0000
(V_b-V_B is not positive definite)
p值为0,则应用固定效应模型。模型的计量结果如下:
Fixed-effects (within) regression Number of obs = 420
Group variable: number Number of groups = 30
R-sq: within = 0.8935 Obs per group: min = 14
between = 0.9121 avg = 14.0
overall = 0.8693 max = 14
F(5,385) = 645.91
corr(u_i, Xb) = -0.7727 Prob > F = 0.0000
------------------------------------------------------------------------------
lninno | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnstf | .8429847 .1280856 6.58 0.000 .5911499 1.094819
lnstg | -.2320736 .1140789 -2.03 0.043 -.4563693 -.007778
lnstl | .0395976 .1134273 0.35 0.727 -.1834169 .2626121
lnstic | .8765391 .1158131 7.57 0.000 .6488339 1.104244
lnci | -.1389116 .0443289 -3.13 0.002 -.2260686 -.0517547
_cons | -10.54612 .7819015 -13.49 0.000 -12.08346 -9.008792
-------------+----------------------------------------------------------------
sigma_u | .70377415
sigma_e | .31848193
rho | .83002205 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(29, 385) = 12.99 Prob > F = 0.0000
问题在于我最关注的那个变量lnstg的系数为负,还通过了5%的显著性检验,这个与我的预期恰恰相反。对于这个问题怎么解决。希望给予合理的解释
Instrumental variables (2SLS) regression Number of obs = 390
Wald chi2(4) = 4636.72
Prob > chi2 = 0.0000
R-squared = 0.9224
Root MSE = .4093
------------------------------------------------------------------------------
lninno | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnstic | .1253952 .0696047 1.80 0.072 -.0110276 .261818
lnstf | .7051549 .1027244 6.86 0.000 .5038187 .9064911
lnstg | .1255006 .0718182 1.75 0.081 -.0152604 .2662617
lnci | .1354696 .0282685 4.79 0.000 .0800643 .1908748
_cons | -6.896633 .3055345 -22.57 0.000 -7.49547 -6.297797
------------------------------------------------------------------------------
Instrumented: lnstic
Instruments: lnstf lnstg lnci L.lnstic lnstl
二阶段最小二乘的结果却是很好。而我同学的计量结果是,固定效应系数很好,但两阶段最小二乘的系数很不好。难道这二者相悖?我想知道出现这种情况的根源是什么。
在两阶段最小二乘的基础上加入相关的交互项,结果也都很理想。所以唯独这个固定效应模型的系数叫我很头疼。难道是内生性问题,可是我找的那个内生变量是lnstic(也是重要变量)而不是 lnstg(不是不找这个,而是找不到合适的工具变量,只能罢休)。知道其中原因的高手们,多多指导啊。谢谢啦