Data structure for ppml

[iris lee]

Dear all,

I'm not familiar with the topic about trade and I will appreciate a lot if anyone could give any suggestions.

I have a city-pair trade flow data and another city-level variable data and I want to see that if this city-level variable has an influence on the trade among cities.

Actually, I'd like to know that
(1) whether the ppml estimator has any requirement on the structure of data. For example, there are 3 cities and therefore 6 city pairs. Some of city pairs have zero trade so the observations can be 4. Should I keep the zero-trade city-pair and make up the zero value or just drop them? Does this make any differences on the result theoretically?
(2) what kind of fixed effect should be control? The city-pair fixed effect, the start-city fixed effect or the end-city fixed effect.

I’m not sure if I explain the questions clearly. Looking forward to the response.

Regards,
Iris

[iris lee]

hello all,

I have tried both data structure and there is no difference. I also find that if I join both start-city fixed effect and end-city fixed effect, the coefficient will be omitted. But I'm not sure if city-pair fixed effect is necessary.

Looking forward to the response.

Thanks a lot.

[Joao Santos Silva]

Dear iris lee

1) you should keep the zeros.
2) only you can know what fixed effects are needed, if any.

Best wishes,

Joao

[iris lee]

Dear iris lee

1) you should keep the zeros.
2) only you can know what fixed effects are needed, if any.

Best wishes,

Joao

Dear @Joao,

Thank you so much! This helps a lot.

Best regards,
Iris