[Anthony Macedo]

Thank you very much João.

My sample is small due to some data limitation and we underestimated the non-randomness effect, but I will discuss it with my supervisor and try to expand the sample.
Again, thank you for helping so many people like me.

[Anthony Macedo]

Thank you very much João.

My sample is small due to some data limitation and we underestimated the non-randomness effect, but I will discuss it with my supervisor and try to expand the sample.
Again, thank you for helping so many people like me.

Regarding the time-invariant variables like distance or language, if I should forget RE it is then preferable to drop them, right?

[Joao Santos Silva]

Anthony,

If you manage to include multiple exporters in your dataset and if you include dummies for restination (as you have now) and also for origin, then you can keep those variables (but you would not be able to include country characteristics). Otherwise, those variables will be dropped so you might as well exclude them.

Best wishes,

Joao

[Anthony Macedo]

Thank you very much for your attention João,

Including multiple exporters is not very feasible because I am focusing in a specific product from one country. Although I can try to increase the number of destination countries.
In this circumstance, would it make sense to estimate my model (with or without time-invariant variables) using xtpoisson or xtpqml?

[Joao Santos Silva]

Indeed, those commands will be suitable for what you want to do. If you get problems with convergence, just use -ppml- with dummies.

Best wishes,

Joao

[Anthony Macedo]

Thank you very much Joao.
Your answers and advices are very useful.

Best regards,
Anthony

[Cesar Huaman]

Dear Mr Silva,

I have a doubt about the PPML estimator. I read your paper "The Log of Gravity" and i understood this estimator help to reduce de heterocedasticity and selection bias but if my database don´t have zero values in the dependent variable i also can use this estimator or I just have to limit my regressions to use fixed effect and random effect of linear regressions.

Thank you very much,

César Huamán

[Joao Santos Silva]

Dear Cesar,

If you are estimating a gravity equation, you should use PPML even if there are no zeros in your data.

Best wishes,

Joao

[JFelipe PinedaG]

Hi Joao, could you help me with these questions please


Im estimating a gravity equation with the PPML estimator as suggested in "The Log of Gravity".

I have the following cases

1 - Regress the Log(trade) as my dependent variable and observations > 0 and cluster , the RESET test is NOT Rejected .

2 - Regress the Log(trade) as my dependent variable and all observations and cluster , the RESET test is Rejected .

3 - Regress the trade as my dependent variable and all observations and cluster , the RESET test is Rejected .

4- Regress the trade as my dependent variable and observations > 0 and cluster , the RESET test is Rejected .


The coefficients between 1 and 2, are pretty similar.
The coefficients between 3 and 4, are pretty similar.


The result given by the case 1 is appropriate? , even when i have the Log(trade) and observations > 0, my P>|z|´s RESET test is 0.511.

the command is a u suggest .

predict u, xb

gen u2 = u^2

and include u2 in the original regression




Thank u very much.


Regards,

Felipe

[Joao Santos Silva]

Dear Felipe,

In 2, how can you estimate the regressions with all observations if the dependent variable is logged?

Anyway, models where the dependent variable is logged (1 and 2) are unlikely to be helpful, so I would focus no 3 (because there is no reason to exclude the zeros as in 4). If you are worried about the result of the RESET, you may try to add variables to the model, or change the way the regressors enter the model.

Best wishes,

Joao

[JFelipe PinedaG]

Hi Joao,

In 2 i have trade + 1. Im going to focus on 3 then.

Thank u

[Cesar Huaman]

Thank you Joao,

Best wishes,

César

[Cesar Huaman]

Dear Joao,

I posted before and I would like your advise once again if you can. I am estimating a gravity model of trade using extensive and intensive margins, I am using panel data and i would like to use the ppml estimator but I have some questions:

- I don´t know if it can use the PPML estimator with this methodology.
- How can I adapt the PPML estimator to a panel data. Just with dummies over time and over countries?

Thank you,

César

.

[Joao Santos Silva]

Dear Cesar,

Using PPML with panel data is fine, but you may not want to use the -ppml- command. If you search this forum you will see that there are other commands that are more suitable for models with many fixed effects, although the results will be exactly the same that you obtain using -ppml- and dummies.

Best wishes,

Joao

[Hashim Akin-Olagunju]

Dear Joao,

I'm working on the effect of SPS measure on cocoa trade with 5 exporters and 19 importers for a 12 year period. I ran the gravity model using -xtpoisson .....robust fe- command. I understand that the dependent variable for a poisson regression should be raw trade value (preferably in '000 units). However, using count data for the dependent variable brought spurious results. I tried it with transformed dependent variable [ln(1+trade)] and the result seemed okay. I conducted Reset test and the probability was not significant. Please, I need your direction.

Regards,
Hashim