[Joao Santos Silva]

Isabel,

PPML does not account for endogeneity; for thatyou need to look at ivpoisson. The PPML results are always preferable to those obtained buy OLS because PPML is much more robust.

Best wishes,

Joao

[Isabel Cour]

Prof. Joao thank you very much for your help, I will try ivpoisson. Regards,

[Isabel Cour]

Joao,

I tried ivpoisson and I lag my regressor one and two period but only for the variable were significative in the ppml specification model, is this correct? Thank you very much for all your answer, you help a lot already. Kind regards,

 ivpoisson gmm lnimpo lngdp_2 lntariff (lnintus = LLlnintus Llngdp)

Step 1
Iteration 0:   GMM criterion Q(b) =  .00251777  
Iteration 1:   GMM criterion Q(b) =  .00181401  
Iteration 2:   GMM criterion Q(b) =  .00181401  

Step 2
Iteration 0:   GMM criterion Q(b) =  .00252415  
Iteration 1:   GMM criterion Q(b) =  .00214509  
Iteration 2:   GMM criterion Q(b) =  .00214509  

Exponential mean model with endogenous regressors

Number of parameters =   4                         Number of obs  =         78
Number of moments    =   5
Initial weight matrix: Unadjusted
GMM weight matrix:     Robust

------------------------------------------------------------------------------
             |               Robust
      lnimpo |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
  lnintus    |   .0238666   .0070002     3.41   0.001     .0101466    .0375867
     lngdp_2 |   .0371496   .0016081    23.10   0.000     .0339977    .0403015
    lntariff |  -.0085163   .0044145    -1.93   0.054    -.0171684    .0001359
       _cons |   2.572713   .0319768    80.46   0.000     2.510039    2.635386
------------------------------------------------------------------------------
Instrumented:  lnintus
Instruments:   lngdp_2 lntariff LLlnintus Llngdp

[Joao Santos Silva]

Dear Isabel,

I cannot really help you because I do not know exactly what you want to do and why; I suggest that you discuss this with your colleagues or with your adviser (if you have one). Anyway, your sample is really small, so do not push your data too much.

Best wishes,

Joao

[Isabel Cour]

Dear Joao,

I am analysing one importer and 6 exporter for the period 2000-2015, I examine the effect of techno at total flow and sector flows of export. My control variables are gdp exporter and importer , tariff, distance. I tried a pooled ols, ppml, and I want to check endogeneity, because the cross country correlation suggest a causality between techno variable and the volume of export(both variables may be determinated simultaneously). Thank you very much for any comments. Kind regards,

[Isabel Cour]

Dear Joao,

Please may you explain your suggestion about "do not push data too much". I need to control by endogeneity because trade can impact on technology and technology can impart on trade so that is why. However as you say the sample is small. Thank you very much and any comment is welcome, kind regards.

[Isabel Cour]

Dear Joao,

When analyse endogeneity we should take into account those variables that have significant effect on the dependent variables?. For instance, after estimate ppml model if some of the variables like it happens in my model were dropped because colineality issues or just have not significant effect on the dependent variable, should not be include in the ivpoisson because you merely check whether the result of the ppml estimation, are or not affect by endogeneity. In other words, if some independent variables are dropped, they does not affect the result of the estimation so it does not count, however I am not sure whether include or not does variables that have not significant impact on the dependent variables and are not dropped. Any advice is very welcome!! Thank you prof Joao!!!

[Joao Santos Silva]

For that you need ivpoisson.

Best wishes,

Joao

[Isabel Cour]

thank you very much prof. Joao!!!! last question regarding to the specification of ivpoisson I copy last post " For instance, after estimate ppml model if some of the variables like it happens in my model were dropped because colineality issues or just have not significant effect on the dependent variable, should not be include in the ivpoisson because you merely check whether the result of the ppml estimation, are or not affect by endogeneity. In other words, if some independent variables are dropped, they does not affect the result of the estimation so it does not count, however I am not sure whether include or not does variables that have not significant impact on the dependent variables and are not dropped" Sorry for the question but I saw sam paper where the include or the variables even those one that were dropped in the model and I am not sure whether that is right or not. Thank you again prof. Joao!!! kind regards,

[Joao Santos Silva]

Indeed, when using ivpoisson it is a good idea not to include the variables and the observations dropped by ppml. However, you should include all other variables, even if not significant.

Best wishes,

Joao

[Isabel Cour]

Thank you very much prof. Joao for!!!! kind regards,

[Isabel Cour]

Dear Joao,

PPML estimation requires that the dependent variable should be specified in level rather than in logaritms? I just notes by comparison here in this post. if i do not specify in level, the estimation is unstable?. Thank you!!! I would appreciate an answer. Regards,

[Joao Santos Silva]

In PPML the dependent variable is indeed in levels rather that in logs, otherwise the results are meaningless.

Best wishes,

Joao

[Isabel Cour]

Prof. Joao,

Thank you very much for you help, I really appreciate it. It seems I have some technical issues with stata becuase my resutl still are not good otherwise the database is special because I can not get what it is wrong. Is it possible to forward you my database in stata to your mail for checking? I am using pooled ols and ppml, then cheking endogeneity via ivpoisson. thank you very much, regards!

[Joao Santos Silva]

Dear Isabel,

Please email me explaining why you need to do this and why you cannot get help from someone at your institution and I'll see if it is appropriate for me to help you in that way, OK?

Best wishes,

Joao