PPML-gravity with country-specific fixed effect- distance and other regressors?

[Bushra Faizi]

Dear Zylkin and Silva,
Thank you very much for responding to the previous questions.
I am estimating the PPML gravity model, using bilateral trade values in the services sector (cross-section data) as a dependant variable, and country-specific fixed effects as regressors. Other regressors include distance, contiguity, common colony, and importer's GDP constrained to 0.8.
The distance, common colony, and contiguity, all are negative and significant. If I skip the contiguity and common colony, that does not affect R and distance is still negative.
My question is:
1. How do you interpret that contiguity and common colony reduce trade in services?
2. Do you advise me to run the regression excluding common colony and contiguity? I have a problem in explaining the theory behind the negative co-efficient of contiguity and common colony.
Thank you.
NOTE: ppmlhdfe did not work in my case,

[Joao Santos Silva]

Dear Bushra Faizi,

Before replying to your question, what do you mean when you say that ppmlhdfe did not work for you?

Best wishes,

Joao

[Bushra Faizi]

Dear Joao,
When I use ppmlhdfe, it omits a number of fixed effect coefficients and for different countries in different sectoral regression.
My objective to run the ppml/ppmlhdfe is to obtain the fixed effect co-efficient from gravity model, to be used in the next stage of the analysis, i.e to calculate the tariff equivalent.
When I run the ppml command I get the importers' fixed effect coefficients and omits only 1 country's FE, same country in each sectoral regression. This is how PPML works for me and PPMLHDFE doesn't.
I hope I put it clearly.
Thank you

[Bushra Faizi]

Dear Joao,
I am a PhD student and I am new to ppml method. I want to make sure that I am doing the estimation correctly. I am using the following commands to estimate gravity model with ppml:

1. Estimate the model: Command 1: ppml TRD comcol ldist COMLANG_OFF lgdp_imp CONTIG imp_fe_* exp_fe_1-exp_fe_`=`NoC'-1', iter(30) noconst
2. Constraint the co-efficeint of GDP (importer) Command 2: test _b[lgdp_imp ]=0.8,coef
I get the following results (Along with the fixed effect co efficient for all the countries)
Kindly advise if I am running the correct commands.
My question at the start of this thread is related to these results.
Thank you.
Bushra,

[Joao Santos Silva]

Dear Bushra Faizi,

The problem you have is that if you include importer fixed effects the coefficient on importer GDP is not identified; alternatively, one of the fixed effects is not estimated. That is why the different commands have to drop either one of the FE, or the GDP. You need to think carefully about what you are doing, in particular about which coefficients can be identified, and adjust your model accordingly. Anyway, the obvious thing to do is to drop GDP from the model as its effect is picked up by the fixed effects.

Best wishes,

Joao

[Bushra Faizi]

Dear Joao,
Let me try that and get back to you.
Thank you ,

[Bushra Faizi]

Dear Joao,
I run the ppml and ppmlhdfe, both give almost the same results and I am getting all the fixed effect coefficients. However, dropping the importers' GDP makes the distance variable insignificant for almost all the services sector. What are your thoughts on this?

[Bushra Faizi]

Sorry, I did not mention before, the objective of constraining the importers' GDP is to isolate the GDP importer effect so that the coefficient on the importer fixed effect contains information on protection only.

[Joao Santos Silva]

Dear Bushra Faizi,

I cannot comment on the significance of the parameters because I do not know enough about your problem. If you can include the log GDP with the restricted coefficient if you use the option offset in ppml to include 0.8*log(GDP).

Best wishes,

Joao

[Bushra Faizi]

Thank you very much for your valuable comments. I read a lot about using offset in command, but every time I try, it gives me error.
Can you please tell me how I use the offset command with my this command: I mean what to write exactly? I tried offset(0.8*lgdp) did not work

Command 1: ppml TRD comcol ldist COMLANG_OFF lgdp_imp CONTIG imp_fe_* exp_fe_1-exp_fe_`=`NoC'-1', iter(30) noconst

Thank you.

[Joao Santos Silva]

You need to create a variable that is 0.8 of lgdp and use it as the offset.

[Bushra Faizi]

Pardon, If I am understanding correctly, (creating a variable that is 0.8 of lgdp and use it as the offset.) that will constraint the value of lgdp, but I want to constraint the co-efficient of lgdp.
Are we talking about the same thing?

[Joao Santos Silva]

Yes, because the offset variable has a coefficient of 1,

[Bushra Faizi]

Thank you very much for the prompt response. I appreciate.