[Joao Santos Silva]

Jaime,

Doing just one model is the sensible thing to do. From your description it looked as if you planed to do one model for each country.

Looking just at the imports from Chile is fine if that is what you are interested in. However, restricting the sample to the 20 main partners does not sound right as that is an endogenously selected sample. Suppose for example that all the 20 major partners are in Latin America; you may conclude that distance has a very small effect in your model when in fact it is because of distance that the main partners are in Latin America.

Best wishes,

Joao

[Jaime Giacomozzi]

Joao, thank you very much for your feedback, very useful! I will try to expand the data.

I wanted to bother you with one more question. If I use fixed effects, Stata eliminates the distance variable of the regression because 'it is constant within each group' (within each pair of countries: Chile-NZ; Chile-USA; Chile-Argentina, etc.... as I have just one destination country). Is there any way to get the coefficient for distance using fixed effects?

Thanks!

Jaime

[Joao Santos Silva]

No, that's not possible!

Best wishes,

Joao

[Jaime Giacomozzi]

Thanks Joao!!

[geoffroy Guepie]

Dear all,
I'm using the command ppml_panel_sg to estimate a structural gravity equation à la Anderson Yotov Larch (2015) where the nomalization of price indices is at the heart of the analysis to calculate the OMR and IMR, as a result I need to know which fixed effect is dropped when I launch the ppml_panel_sg estimator.

Thank you very much for your answer.

[Jaime Giacomozzi]

Hi again Joao! Hope you're well!

I do not know if this is the right forum for my question, but it's about The log of gravity. Hope you can answer it.

In the Results section of the paper, page 651, it says that 'OLS predicts that trade between two contiguous countries is 37% larger than trade between countries that do not share a border'. In footnote 27 says that the formula to compute this effect is (e^b -1) x 100%, where b is the estimated coefficient.

My question is: Why use that formula to compute the effect and not the coefficient itself of the regression (in this case, 0.314) What is the difference?

Thank you again!!

Kind Regards,
Jaime

[Joao Santos Silva]

Dear Jaime,

The coefficient is only an approximation that is valid when it is close to zero; otherwise we need the formula. To see why, just do the ratio between the multiplicative models with x₁ evaluated at x₁ = a and x₁ = a+1 and simplify it.

Best wishes,

Joao

[Metivier Jeanne]

Dear all,

I am estimating a gravity model in which there are 156 exporters and 20 years (the destination country is the same for all observations). So far, I have used OLS and PPML with time-fixed effects and exporter-fixed effects. I have 2 questions:
- should I use instead exporter-time FE?
- I also want to forecast trade values based on my estimation (I have data on the exogenous variables). How can I take into account the fixed effects in the forecast?

Thank you for your help.

Best regards.
Jeanne

[Joao Santos Silva]

Dear Jeanne,

1 - Yes, exporter-time FE is the standard.

2 - For forecasting you cannot use OLS because of the "retransformation problem"; you need yo use PPML and include the FE. How you do that depends on the command you use for the estimation.

Best wishes,

Joao

[Caroline arasavm Gao]

Hi Joao,
I am currently working on a paper using gravity model and I was wondering if my stata code is wrong.
my code is :
xtset country year
reg lnem lngdpjt lnpergdpj lndist lnfree contig comlang fta shock i.country i.year,r
xi:ppml em lngdpjt lnpergdpj lndist lnfree contig comlang fta shock i.country i.year,cluster(country)
dependent variable “em” means extensive margin,the variables "contig ,comlang ,fta and shock" are dummy variables.
Also,if I ran this code,some country dummy and year dummy will be omitted, So I am thinking if this situation will make bias?
Thank you very much!

[Joao Santos Silva]

Dear Caroline,

The code looks OK. Dropping those dummies will not not cause bias.

Best wishes,

Joao

[Caroline arasavm Gao]

Dear Joao,
Thanks a lot!
Also,I am confusing why the lndist variable wasn't omitted when I ran i.country and i.year in stata.Cause dist is not time varing.
Best wishes,
Caroline

[Joao Santos Silva]

Dear Caroline,

One of the dummies was dropped instead; in any case you cannot interpret the coefficient on such variables.

Best wishes,

Joao

[Caroline arasavm Gao]

Dear Joao,
Thanks a lot!
Oh I see,which means I can't interpret some trade policies like fta or comlang,cause dropped dummies (i.e. some countries or some years) maybe absord trade policy's effect on my dependent variable,right?
Best wishes,
Caroline

[Caroline arasavm Gao]

Dear Joao,
Dose this means I cant interpret the coefficient on lndist perfectly? If the coefficient on lndist is negtive ,which means it does fit gravity model ,but I cant claim it reduces extensive margin indeed.
This means I should focus on my key variables only. Right?
Best wishes,
Caroline