I have firm-year panel data (cluster by firm id). My main treatment variable is HARD_FREEZE (dummy):
- firms that never hard-freeze: HARD_FREEZE=0 for all years
- firms that hard-freeze: HARD_FREEZE=0 in pre-freeze years and HARD_FREEZE=1 in the freeze year
- I drop post-freeze years (I do not study years after the freeze)
Outcome: CSR_DISC (continuous). I also include CSO_PRESENCE and the interaction HARD_FREEZEXCSO_PRESENCE,
plus standard controls and year and industry (ff_12) fixed effects.
I am concerned about self-selection/endogeneity in HARD_FREEZE and attempted a Heckman-style correction by:
1) Probit model for HARD_FREEZE
2) Generate inverse Mills ratio (mills) from the probit linear prediction
3) Include mills in the CSR regression (estimated on all observations)
My code (without asdoc formatting) is:
probit HARD_FREEZE SIZE LEV MB OCF SD_OCF BOARD_SIZE GENDER_RATIO BOARD_IND FUND_STATUS ///
FUND_RATIO PLAN_SIZE i.year i.ff_12, vce(cluster id)
predict lefthat if e(sample), xb
gen mills = normalden(lefthat) / normal(lefthat) if e(sample)
reg CSR_DISC HARD_FREEZE CSO_PRESENCE HARD_FREEZEXCSO_PRESENCE SIZE LEV MB OCF SD_OCF ///
BOARD_SIZE GENDER_RATIO BOARD_IND SUS_SCORE SUS_COMM FUND_STATUS FUND_RATIO PLAN_SIZE ///
mills i.year i.ff_12, vce(cluster id)
Questions:
1) Is this a correct way to implement a Heckman correction for “self-selection into HARD_FREEZE”
when CSR_DISC is observed for all observations in the analysis sample?
(i.e., is adding mills from probit(HARD_FREEZE) to an OLS on all obs valid?)
2) If this is not correct, what is the recommended Stata approach for a continuous outcome with an
endogenous binary regressor/treatment (HARD_FREEZE)?
- firms that never hard-freeze: HARD_FREEZE=0 for all years
- firms that hard-freeze: HARD_FREEZE=0 in pre-freeze years and HARD_FREEZE=1 in the freeze year
- I drop post-freeze years (I do not study years after the freeze)
Outcome: CSR_DISC (continuous). I also include CSO_PRESENCE and the interaction HARD_FREEZEXCSO_PRESENCE,
plus standard controls and year and industry (ff_12) fixed effects.
I am concerned about self-selection/endogeneity in HARD_FREEZE and attempted a Heckman-style correction by:
1) Probit model for HARD_FREEZE
2) Generate inverse Mills ratio (mills) from the probit linear prediction
3) Include mills in the CSR regression (estimated on all observations)
My code (without asdoc formatting) is:
probit HARD_FREEZE SIZE LEV MB OCF SD_OCF BOARD_SIZE GENDER_RATIO BOARD_IND FUND_STATUS ///
FUND_RATIO PLAN_SIZE i.year i.ff_12, vce(cluster id)
predict lefthat if e(sample), xb
gen mills = normalden(lefthat) / normal(lefthat) if e(sample)
reg CSR_DISC HARD_FREEZE CSO_PRESENCE HARD_FREEZEXCSO_PRESENCE SIZE LEV MB OCF SD_OCF ///
BOARD_SIZE GENDER_RATIO BOARD_IND SUS_SCORE SUS_COMM FUND_STATUS FUND_RATIO PLAN_SIZE ///
mills i.year i.ff_12, vce(cluster id)
Questions:
1) Is this a correct way to implement a Heckman correction for “self-selection into HARD_FREEZE”
when CSR_DISC is observed for all observations in the analysis sample?
(i.e., is adding mills from probit(HARD_FREEZE) to an OLS on all obs valid?)
2) If this is not correct, what is the recommended Stata approach for a continuous outcome with an
endogenous binary regressor/treatment (HARD_FREEZE)?
HTML Code:
. do "C:\Users\lenovo\AppData\Local\Temp\STDa04_000000.tmp"
. asdoc probit HARD_FREEZE SIZE LEV MB OCF SD_OCF BOARD_SIZE GENDER_RATIO BOARD
> _IND FUND_STATUS FUND_RATIO PLAN_SIZE i.year i.ff_12 , robust cluster (id) nest repl
> ace drop(i.year i.ff_12 ) dec(4) tzok save(qqqqa)
Iteration 0: Log pseudolikelihood = -358.75999
Iteration 1: Log pseudolikelihood = -327.72275
Iteration 2: Log pseudolikelihood = -324.57221
Iteration 3: Log pseudolikelihood = -324.53113
Iteration 4: Log pseudolikelihood = -324.53103
Iteration 5: Log pseudolikelihood = -324.53103
Probit regression Number of obs = 3,900
Wald chi2(40) = 94.12
Prob > chi2 = 0.0000
Log pseudolikelihood = -324.53103 Pseudo R2 = 0.0954
(Std. err. adjusted for 282 clusters in id)
------------------------------------------------------------------------------
| Robust
HARD_FREEZE | Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
SIZE | -.1925695 .0740725 -2.60 0.009 -.3377488 -.0473902
LEV | -.0087306 .4015024 -0.02 0.983 -.7956607 .7781996
MB | .0002073 .0073871 0.03 0.978 -.0142711 .0146857
OCF | -3.458301 1.245414 -2.78 0.005 -5.899267 -1.017335
SD_OCF | 3.875817 2.853238 1.36 0.174 -1.716427 9.46806
BOARD_SIZE | -.0004025 .0297384 -0.01 0.989 -.0586887 .0578837
GENDER_RATIO | .3799744 .6574852 0.58 0.563 -.9086729 1.668622
BOARD_IND | .2277899 .6650473 0.34 0.732 -1.075679 1.531259
FUND_STATUS | -4.173805 2.029593 -2.06 0.040 -8.151735 -.1958759
FUND_RATIO | -.0761674 .3856172 -0.20 0.843 -.8319632 .6796285
PLAN_SIZE | .0687444 .0637423 1.08 0.281 -.0561882 .1936771
|
year |
2005 | -.2409666 .4322112 -0.56 0.577 -1.088085 .6061518
2006 | .0026222 .381609 0.01 0.995 -.7453177 .7505622
2007 | -.220559 .4408242 -0.50 0.617 -1.084559 .6434405
2008 | .4790183 .321579 1.49 0.136 -.1512649 1.109302
2009 | .6453316 .312803 2.06 0.039 .032249 1.258414
2010 | .4765423 .325337 1.46 0.143 -.1611064 1.114191
2011 | .2260907 .3614312 0.63 0.532 -.4823015 .934483
2012 | .361 .3465188 1.04 0.298 -.3181643 1.040164
2013 | .4685907 .3371055 1.39 0.165 -.192124 1.129305
2014 | .2758176 .3544123 0.78 0.436 -.4188177 .9704528
2015 | .438984 .3475913 1.26 0.207 -.2422823 1.12025
2016 | .3682024 .3422007 1.08 0.282 -.3024986 1.038903
2017 | .3257874 .3579784 0.91 0.363 -.3758373 1.027412
2018 | .7495029 .3326064 2.25 0.024 .0976063 1.401399
2019 | .5081082 .3590654 1.42 0.157 -.1956471 1.211864
2020 | .7474426 .3516579 2.13 0.034 .0582058 1.436679
2021 | .3653477 .4157055 0.88 0.379 -.4494201 1.180115
2022 | .0776221 .4568847 0.17 0.865 -.8178554 .9730996
|
ff_12 |
2 | .0701796 .2884724 0.24 0.808 -.495216 .6355752
3 | -.2088294 .2146219 -0.97 0.331 -.6294806 .2118218
4 | -.3456759 .3760136 -0.92 0.358 -1.082649 .3912973
5 | -.3439714 .2487806 -1.38 0.167 -.8315724 .1436295
6 | -.1551492 .2504892 -0.62 0.536 -.6460991 .3358007
7 | .1184574 .3654636 0.32 0.746 -.5978381 .8347529
8 | -.406023 .2438484 -1.67 0.096 -.883957 .0719111
9 | .4415868 .2073734 2.13 0.033 .0351424 .8480312
10 | -.0921214 .2570592 -0.36 0.720 -.5959483 .4117054
11 | .4643857 .2316409 2.00 0.045 .0103778 .9183935
12 | -.3317513 .2383857 -1.39 0.164 -.7989788 .1354762
|
_cons | -1.089957 .7413996 -1.47 0.142 -2.543073 .3631599
------------------------------------------------------------------------------
Click to Open File: qqqqa.doc
.
. predict lefthat, xb // Get the linear prediction
(4681 missing values generated)
. gen mills = normalden(lefthat) / normal(lefthat) // Generate the inverse Mill's ratio
(4,681 missing values generated)
.
.
. asdoc reg CSR_DISC HARD_FREEZE CSO_PRESENCE HARD_FREEZEXCSO_PRESENCE SIZE LEV MB OC
> F SD_OCF BOARD_SIZE GENDER_RATIO BOARD_IND SUS_SCORE SUS_COMM FUND_STATUS FUND_RATIO PLA
> N_SIZE mills i.year i.ff_12 , robust cluster (id) nest replace drop(i.year i.ff_12 ) dec
> (4) tzok save(qqqkkq)
Linear regression Number of obs = 3,210
F(45, 265) = 82.25
Prob > F = 0.0000
R-squared = 0.7327
Root MSE = 9.339
(Std. err. adjusted for 266 clusters in id)
------------------------------------------------------------------------------------------
| Robust
CSR_DISC | Coefficient std. err. t P>|t| [95% conf. interval]
-------------------------+----------------------------------------------------------------
HARD_FREEZE | 6.264576 2.157665 2.90 0.004 2.016228 10.51292
CSO_PRESENCE | 4.940515 1.003364 4.92 0.000 2.964935 6.916096
HARD_FREEZEXCSO_PRESENCE | 11.74723 3.839324 3.06 0.002 4.187765 19.30669
SIZE | 4.118176 5.715745 0.72 0.472 -7.135875 15.37223
LEV | 2.109673 3.31258 0.64 0.525 -4.412651 8.631998
MB | -.0698221 .0285052 -2.45 0.015 -.1259476 -.0136966
OCF | 97.2196 104.8254 0.93 0.355 -109.1771 303.6163
SD_OCF | -79.62506 118.0351 -0.67 0.501 -312.031 152.7809
BOARD_SIZE | .3070938 .1880339 1.63 0.104 -.0631367 .6773242
GENDER_RATIO | -7.023461 12.48731 -0.56 0.574 -31.61043 17.56351
BOARD_IND | -7.603016 8.288169 -0.92 0.360 -23.92206 8.716026
SUS_SCORE | .4259821 .0275958 15.44 0.000 .3716473 .480317
SUS_COMM | 2.088117 .948821 2.20 0.029 .21993 3.956304
FUND_STATUS | 91.92552 127.6094 0.72 0.472 -159.3319 343.1829
FUND_RATIO | .0192403 3.532275 0.01 0.996 -6.935654 6.974135
PLAN_SIZE | -1.082472 2.037977 -0.53 0.596 -5.095159 2.930215
mills | -25.47822 34.18466 -0.75 0.457 -92.78633 41.82989
|
year |
2006 | -4.373616 7.821997 -0.56 0.577 -19.77479 11.02755
2007 | 1.747812 1.767486 0.99 0.324 -1.73229 5.227915
2008 | -15.13568 22.58608 -0.67 0.503 -59.60667 29.33532
2009 | -18.25729 27.63952 -0.66 0.509 -72.67829 36.16372
2010 | -13.55586 22.55711 -0.60 0.548 -57.96983 30.85811
2011 | -7.140786 14.95481 -0.48 0.633 -36.58615 22.30457
2012 | -8.83993 19.07473 -0.46 0.643 -46.39723 28.71737
2013 | -9.791559 22.35421 -0.44 0.662 -53.80603 34.22291
2014 | -3.596534 16.41043 -0.22 0.827 -35.90795 28.71488
2015 | -3.999571 21.40539 -0.19 0.852 -46.14585 38.14671
2016 | -1.341657 19.25826 -0.07 0.945 -39.26032 36.57701
2017 | 1.654498 17.95445 0.09 0.927 -33.69703 37.00602
2018 | -7.953823 30.78313 -0.26 0.796 -68.56445 52.65681
2019 | -.9603528 23.44141 -0.04 0.967 -47.11547 45.19477
2020 | -5.065944 30.69163 -0.17 0.869 -65.49642 55.36453
2021 | 4.119329 19.18789 0.21 0.830 -33.6608 41.89945
2022 | 11.0519 10.25829 1.08 0.282 -9.146224 31.25002
|
ff_12 |
2 | -5.528289 3.408065 -1.62 0.106 -12.23862 1.182041
3 | 4.454642 6.694695 0.67 0.506 -8.72692 17.6362
4 | 16.19785 11.59032 1.40 0.163 -6.622988 39.01868
5 | 14.32454 10.57837 1.35 0.177 -6.503813 35.15289
6 | 4.73878 5.014423 0.95 0.346 -5.1344 14.61196
7 | -.2542357 5.377176 -0.05 0.962 -10.84166 10.33319
8 | 17.30009 12.58141 1.38 0.170 -7.472164 42.07234
9 | -20.45107 13.75656 -1.49 0.138 -47.53713 6.634986
10 | 4.658465 3.76444 1.24 0.217 -2.753552 12.07048
11 | -16.90209 13.98326 -1.21 0.228 -44.43452 10.63034
12 | 9.345867 10.47308 0.89 0.373 -11.27517 29.9669
|
_cons | 32.26755 62.6324 0.52 0.607 -91.0529 155.588
------------------------------------------------------------------------------------------
Click to Open File: qqqkkq.doc
.
end of do-file
Comment