contradictory results in ivprobit and biprobit

[karamat khan]

Hi all,
i want to estimate the impact of export which is endogenous regressor in the model (lnexport (continuous) and exporter (0,1) on "y".. and i have three different proxies for "y" which are y1, y2 and y3 all three in binary form.. and "iv" is instrument variable in continuous

* Example generated by -dataex-. To install: ssc install dataex
(y1 lnexport exporter iv)
1         0 0   2.70805
0         0 0  2.995732
0         0 0 1.7917595
0  .6931472 1 4.6728287
0         0 0   4.65396
0         0 0 2.0794415
0         0 0 3.4011974
0 .26236427 1  4.787492
1         0 0  4.339467
1         0 0   3.64632
0         0 0  1.609438

because of binary nature of "y" and continuous and binary nature of export, i estimated two models ivprobit and biprobit

ivprobit y1 (lnexport=iv) lnfirm_age competition  lnlargest_own training_l10  ln_managexp i.isic, vce(cluster ccode)

Probit model with endogenous regressors         Number of obs     =      4,696
                                                Wald chi2(32)     =   2.92e+07
Log pseudolikelihood =  91.552004               Prob > chi2       =     0.0000

                                         (Std. Err. adjusted for 33 clusters in ccode)
--------------------------------------------------------------------------------------
                     |               Robust
                     |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------------+----------------------------------------------------------------
            lnexport |   2.567695   1.070549     2.40   0.016     .4694578    4.665933
    
--------------------------------------------------------------------------------------
Wald test of exogeneity (corr = 0): chi2(1) = 9.20        Prob > chi2 = 0.0024

for y1 in ivprobit the coefficients are positive and significant and Wald test suggest endogeniety, but biprobit suggest opposite results with insignificant coefficient for "exporter" and exogeniety... however i have consistent results for two other binary proxies of y2 and y3 in both ivprobit and biprobit regression using same "iv"

 biprobit (y1=exporter lnfirm_age competition  lnlargest_own training_l10  ln_managexp i.isic) (exporter=iv lnfirm_age com
> petition  lnlargest_own training_l10  ln_managexp i.isic), vce(cluster ccode)

Seemingly unrelated bivariate probit            Number of obs     =      4,700
                                                Wald chi2(30)     =          .
Log pseudolikelihood = -4117.8559               Prob > chi2       =          .

                                  (Std. Err. adjusted for 33 clusters in ccode)
-------------------------------------------------------------------------------
              |               Robust
              |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
y1            |
     exporter |   .2636794   .3614948     0.73   0.466    -.4448375    .9721962

--------------+----------------------------------------------------------------
exporter      |
           iv |   .3972677    .051231     7.75   0.000     .2968567    .4976787
  
--------------+----------------------------------------------------------------
          rho |  -.0661718   .1771233                     -.3926681    .2751372
-------------------------------------------------------------------------------
Wald test of rho=0: chi2(1) = .138756                     Prob > chi2 = 0.7095

.

1) can any one please explain why the results are are different for y1 when using same iv i have consistent results for y2 and y3 both in biprobit and ivprobit and how can i interpret these two different results or is there any way to correct this inconsistency?
2) please confirm if biprobit command is correct and im correctly using "exporter" in first equation
3) using biprobit can i use continuous iv when both endogenous regressor and outcome are binary?
4) if i outcome, endogenous regressor and iv all are in binary form... is it ok to use biprobit regression in that case too?

Thanks in Advance and sorry for long post, many questions ... i hope experts will help me to clear my doubts