. gen y= .
. replacey=ui+4*x+0.5*z1+0.9*z2+e if x<0
. replace y=ui+(-3.5*x)+0.5*z1+0.9*z2+e if x>=0拐点为0的倒U型曲线
1、不考虑内生性问题,半参数回归的拐点为0.29704595
2、考虑内生性问题,半参数回归的拐点为0.2846916,明显更准确一些
3、动态面板门槛一般模型
------------------------------------------------------------------------------
y | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
z1_b | .5240811 .022296 23.51 0.000 .4803817 .5677806
z2_b | .8778075 .0143141 61.32 0.000 .8497524 .9058627
x_b | 3.889159 .0412546 94.27 0.000 3.808302 3.970017
cons_d | -.0907999 1.120286 -0.08 0.935 -2.286521 2.104921
z1_d | -.0548423 .0431717 -1.27 0.204 -.1394572 .0297726
z2_d | .0427477 .029696 1.44 0.150 -.0154553 .1009508
x_d | -7.281174 .0611229 -119.12 0.000 -7.400972 -7.161375
r | .0636142 .2980837 0.21 0.831 -.5206191 .6478475
------------------------------------------------------------------------------
拐点0..0636,比半参数准确很多。拐点左侧斜率较准确,右侧不准确。
4、动态面板门槛扭结模型
------------------------------------------------------------------------------
y | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
z1_b | .4901399 .0052905 92.64 0.000 .4797707 .5005092
z2_b | .8977091 .0033338 269.28 0.000 .891175 .9042431
x_b | 3.908986 .0379181 103.09 0.000 3.834668 3.983304
kink_slope | -7.357911 .0615421 -119.56 0.000 -7.478531 -7.23729
r | .0636142 .1010927 0.63 0.529 -.1345238 .2617523
------------------------------------------------------------------------------
拐点位置以及左右两侧斜率,都超级准确。