ywh19860616 发表于 2013-12-30 11:31 
在没有做其他处理情况下,你是指什么处理?
还有,你开始用的是什么命令?就是除了xtabond这个命令。
总 ...
数据在这里,用差分GMM是这样
. xtabond lntrade lngdp1 lnpgdp1 lngdp2 lnpgdp2 lndis colo lang adj , twostep
note: lndis dropped from div() because of collinearity
note: colo dropped from div() because of collinearity
note: lang dropped from div() because of collinearity
note: adj dropped from div() because of collinearity
Two-step results
colo | (omitted)
lang | (omitted)
adj | (omitted)
_cons | (omitted)
------------------------------------------------------------------------------
Warning:
gmm two-step standard errors are biased; robust standard
errors are recommended.
Instruments for differenced equation
GMM-type: L(2/.).lntrade
Standard: D.lngdp1 D.lnpgdp1 D.lngdp2 D.lnpgdp2
Instruments for level equation
Standard: _cons
drop的变量是没有系数显示的,然后检验也没有值显示
. estat abond
cannot calculate AR tests with dropped variables
Arellano-Bond test for zero autocorrelation in first-differenced errors
cannot calculate test with dropped variables
+-----------------------+
|Order | z Prob > z|
|------+----------------|
| 1 | . . |
| 2 | . . |
+-----------------------+
H0: no autocorrelation
. estat sargan
Sargan test of overidentifying restrictions
H0: overidentifying restrictions are valid
cannot calculate Sargan test with dropped variables
chi2(20) = .
Prob > chi2 = .
系统GMM用了.
xtdpdsys lntrade lngdp1 lnpgdp1 lngdp2 lnpgdp2 lndis colo lang adj
note: lndis dropped from div() because of collinearity
note: colo dropped from div() because of collinearity
note: lang dropped from div() because of collinearity
note: adj dropped from div() because of collinearity
colo | -9.057994 4.891049 -1.85 0.064 -18.64427 .528287
lang | 2.834002 .5084988 5.57 0.000 1.837362 3.830641
adj | 3.74681 2.595319 1.44 0.149 -1.339921 8.833542
这里变量虽然被drop了,但回归结果仍然显示了它们的系数
检验有一个有值显示
. estat abond
artests not computed for one-step system estimator with vce(gmm)
Arellano-Bond test for zero autocorrelation in first-differenced errors
+-----------------------+
|Order | z Prob > z|
|------+----------------|
+-----------------------+
H0: no autocorrelation
. estat sargan
Sargan test of overidentifying restrictions
H0: overidentifying restrictions are valid
chi2(22) = 63.88466
Prob > chi2 = 0.0000
为什么有这样的区别呢?差分GMM时Warning:
gmm two-step standard errors are biased; robust standard
errors are recommended.是说这个方法不适用吗。
我说的没做处理是说回归时我没有设定内生变量前定变量滞后项之类的,都是用的默认。这样会有影响吧?麻烦你了:)