Is it possible to run PPML without constant?

[Saqib Khan]

I am estimating the gravity equation and measure the trade creation and diversion using the data of the three FTAs, i.e. NAFTA, U and ASEAN. I am estimating my model with Fixed effect PPML with a set of dummies. When I use the foolowing command,
PPML exports border language distance GDPi POPULATIONi GDPj Populationj EUm EUn NAFTAm Naftan ASEANm ASEANn Exporter(duumy) Importer(dummy), cluster (dist)
it drop one of the FTA dummy (EUm EUn NAFTAm Naftan ASEANm ASEANn ,) due to collinearity. This drop dummy result is important for me. I want to drop the constant instead of the dummy. So my question is that how can I direct stata to drop the constant instead of the dummy?
Thanks in advance for your consideration.

Regards
Saqib

[Joao Santos Silva]

Dear Saqib,

If you install the latest version of ppml from SSC, you will see that it has an option to estimate the model without constant. Note, however, that because you are including exporter and importer dummies, you will not be able to identify the coefficients on importer and exporter characteristics such as GDP and population.

Best wishes

Joao

[Saqib Khan]

Dear Professor Joao,
Thanks for your reply. Now I have two issues. (1) If I run with constant, it drops the dummy variable (used for trade creation and diversion) saying that the variable is dropped due to collinearity. The result of these dummies is important for me as I have to decide whether the FTAs are trade creating or diverting.
(2) As you mentioned, I will not be able to identify the coefficients of exporter and importers characteristics if I run the regression without constant. So, here I need your suggestion, i.e. what should I do here?. I have used the exporters and importers dummies to control for the MTR.
Regards
Saqib

[Joao Santos Silva]

Saqui,

If you include the MRT dummies, you cannot include things like importer GDP or a dummy if the importer (or exporter) is part of an FTA. The only things you can include are variables that are specific to the pair, for example distance or a dummy indicating that both countries are part of a FTA.

Best wishes,

Joao

[Saqib Khan]

Dear Professor Joao,
Thanks for your response and for the guidance. I am now only estimating the model with the variables that are specific to the pair. However, it drops one of the dummy variable when I run the command, showing that the variable is omitted due to collinearity. However, when I check the collinearity through correlation matrix, it does not indicate a one-one relationship. Now I run the following command
PPML exports border language colony commoncolony distance EUm EUn NAFTAm Naftan ASEANm ASEANn Exporter(dummy) Importer(dummy), cluster (dist).
When I run the above command, it drops the FTA dummy. I need the results of this dummy to decide whether the specific FTA is a trade creating or trade diverting. I will be highly obliged if you can guide me regarding this issue.

Regards
Saqib

[Joao Santos Silva]

What are variables like EUm EUn? If they are dummies indicating that the importer or the exporter are in a FTA, they drop when you include MRT. The only FTA dummies you can include are those indicating whether both partners are in the same FTA. In other words, I believe that with MRTs we cannot see if a particular FTA is trade creating or trade diverting.

Best wishes,

Joao

[Saqib Khan]

Dear Professor Joao,
The variables EUm is a dummy variable that shows the trade between the two EU members, whereas the EUn shows the trade between a member and non-member state. Furthermore, I need your guidance regarding the estimation of gravity equation measuring the trade creation and trade diversion. I mean what should be my estimation specification to capture the impact of FAT and to control for the MRT. I will be highly obliged if you could guide me in this regard.

Saqib

[Joao Santos Silva]

Dear Saqib,

As far as I understand, that is not possible, but I am not really an expert on that.

Best wishes,

Joao

[Saqib Khan]

Dear Professor Joao,
Thank you once again for your reply and for your precious time. I have decided to include MRT calculated through the method of Baier and Bergstrand instead of using country-pair fixed effect. However, I have some issues regarding the commands and I will be highly obliged if you could guide me. I found the following command for the first order Taylor series approximation of MRT in a guide "Estimating the gravity model". I want to clear my conception regarding these commands. In the following commands, we use by (exp sector), by (imp sector) and by (sector) respectively. Here as I have variables with a different name, I want to know that what we mean by (exp sector), (Imp sector) and by sector? As I don't have the variable sector in my data. Should I use exporters country for (exp sector) and importers countries for (Imp sector)?
So by (exp sector), it should mean the exporting countries? And what should I use for the sector in the 3rd command?


commands:
egen temp1 = mean(ln_distance), by (exp sector)
egen temp2 = mean(ln_distance), by (imp sector)
egen temp3 = sum(ln_distance), by (sector)
gen ln_distance_star = ln_distance - temp1 - temp2 + (1/(58*58))*temp3


Thanks in advance for your consideration.
Regards: Saqib

[Joao Santos Silva]

Sorry, Saqib Khan, I am not familiar with that.

Best wishes,

Joao