Problem with the multicollinearity in trade gravity model estimated by ppml

[Thuy Dang]

Hi everyone,

I'm using ppmlhdfe in STATA to estimate my trade gravity model.

This is my code in STATA:
ppmlhdfe trade fta log_tariff log_ntmA log_ntmD log_ntmE i.groupID i.imp_p_HS6, cluster(iso3num_d)

in which,
- non-tariff measures are disaggregated by type (A,B,E,D) and are count variables.
- fta is a binary variable
- i.groupID: importer-year fixed effects
- i.imp_p_HS6: importer-product fixed effects
- robust SE is clustered by importer

I do not include any fixed effects for exporter, because my data only has one exporter.

If I exclude log_tariff variable, the coefficient of log_ntmA becomes significant (p<0.01). However, whenever I include log_tariff, the coefficient of log_ntmA becomes insignificant, though it does not flip sign.

Can I exclude my log_tariff variable from the regression? Or what could I do to fix it?

Please kindly help! Thank you very much!

P/s: Joao Santos Silva I follow your answer on different threads on this gravity model topic, if you have some times, please kindly help me. Thank you!

[Joao Santos Silva]

Dear Thuy Dang,

I am not sure if I understand what the problem is. It is natural that when you drop a variable the results for the others change, so there is nothing special about what you find. As for deciding whether to keep the log_tariff variable, that is up to you and depends on what your objective is. As an aside, I suggest you absorb the fixed effects; check the help file for examples.

Best wishes,

Joao

[Thuy Dang]

Dear Joao Santos Silva ,

Thank you for your reply. I figured it out, but encountered another problem.
I want to estimate the effects of non-tariff measures on the coffee exports from Vietnam to EU. I only have one exporter, and 28 EU importers, and 5 product categories but in the same HSCode 4-digit 0901. Since the EU non-tariff measures applied by EU are for all coffee exporters to EU and my product categories are just the sub-cat HS 6-digit of HS0901, the non-tariff measures variables in my data set only vary by year.

I have gone through literature about PPML with fixed effects, but I wonder whether these high-dimensional fixed effects can be applied. By now, to account for the multilateral resistances, I try to use importer-year fixed effects and product fixed effects. The result turns out that the coefficients of log_ntm can still be estimated, but insignificant. I am afraid that the importer-year fixed effects will absorb some of the variation in my ntm data.

I look forward to your comments.

Thank you for your time.

Best regards,
Thuy.

[Joao Santos Silva]

Dear Thuy Dang,

Indeed, it may be that your data does not have enough variation to precisely identify the coefficient of interest. Can you some how expand the dataset to alleviate the problem?

Best wishes,

Joao

[Thuy Dang]

Dear Joao Santos Silva,

Could you give me some suggestions on how to expand my dataset? Since the EU non-tariff measures are applied to all exporters and importers homogeneously, it may not work if I expand my data to other countries who export coffee to EU. What I am thinking is to expand to include the data of other countries that import coffee from Vietnam. However, from this option, how could I estimate the impacts of non-tariff measures in each importing market?
And I'm not sure whether I need to add more than one exporter?

[Joao Santos Silva]

Dear Thuy Dang,

How to expand the data will depend on what you want to do, but at least it would be good to have data for more years, including the period before the NT measures were adopted.

Best wishes.

Joao

[Thuy Dang]

Dear Joao Santos Silva ,

Thank you for your advice. I appreciate your help with this matter.

Thanks again,
Thuy.