Gravity model for euro dummy

[Angela Cojoc]

Hey everyone,
I have been trying to analyse trade effects of using the euro for my bachelor thesis but I am having trouble getting some good results. I am doing a Pooled OLS, Fixed-effects for importer-exporter-time, and a ppml, yet my results change all the time and I don't know what is wrong with my data.
Most of my coefficients for the euro_both dummy, which is equal to 1 if both countries are using the euro at the same time, are even very big, or even negative. I have attached the screenshots of the results from the OLS, Fixed effects and ppml. If you have any ideas of what I could change or check in my data to get more normal results, I would appreciate it.
These are the codes i used for the:
- ppml: . ppmlhdfe tradeflow_baci euro_both ln_dist ln_gdp_o ln_gdp_d euro_onesided, absorb(iso3num_o is
> o3num_d year) cluster(dist)
-fixed-effects: reghdfe ln_tradeflow_baci euro_onesided euro_both ln_gdp_o ln_gdp_d ln_dist, absorb(iso3num_o iso3num_d year) vce(cluster iso3num_
> o#iso3num_d)
-OLS: reg ln_tradeflow_baci euro_both euro_onesided eu_o eu_d ln_dist ln_gdp_o ln_gdp_d contig

[Joao Santos Silva]

Dear Angela Cojoc,

There are several issues here. First, just use PPML because the other methods are unreliable in this context. Second, you need to include pair-fixed effects to take into account that countries self-select into the euro. Finally, your importer and exporter fixed effects need to be interacted with year (and do not include year fixed effects as these would be redundant).

Best wishes and let us know how you get on,

Joao

[Angela Cojoc]

Thank you Joao Santos Silva
I wanted to do different methods to compare them in my thesis and to show why PPML is the most reliable one. I redid it with your recommendations but my z value is lower than 1.96 for all my variables. And the pair_id absorbed FE are all redundant. Do you have any more ideas to what I could change in order to get some significant results?

[Joao Santos Silva]

Dear Angela Cojoc,

I do not think you are creating the pair_id variable correctly. Maybe the examples here will help? Also, please note that finding significant explanatory variables is generally not a legitimate research goal.

Best wishes,

Joao

[Angela Cojoc]

Thank you for your suggestions Joao Santos Silva
The material you linked was quite helpful so I used that database instead and I added the year the country joined the eurozone. My results show a decrease in trade if both countries are using the euro which does not really allign with past research. Should i add other variables or change anything about the model?

[Joao Santos Silva]

You need to cluster your standard errors; you will probably find that the variable is not significant.

[Angela Cojoc]

This is what I get when I cluster(pair_id) Joao Santos Silva

[Joao Santos Silva]

You removed RTA; I would include it, and possibly a separate dummy for EU.

[Angela Cojoc]

it does not change much

[Joao Santos Silva]

That is probably as far as you want to go for an undergrad dissertation. One problem you have with that database is that it does not include intra-national trade, and that may explain the results you get.