ppmlhdfe

[Tomas Pientka]

Dear all,

I'm currently writing a bachelor's where I am trying to use a gravity model to measure the impact of Slovakia joining the Eurozone on its exports. As I am fairly new to gravity models I am slightly struggling to run a calculation which would follow the findings of the literature.

I am using panel data spanning from 1995 to 2021 and more specifically I am looking at the automobile exports (I am pasting a screenshot of the data).

* Example generated by -dataex-. For more info, type help dataex
clear
input int(t i j) long k float v str13 q str3(iso3_o iso3_d) int dist byte eu_d double gdp_d byte eur
2019 703 251 870322 1941857.8 "   133319.423" "SVK" "FRA" 1092 1 2728870246.706 1
2018 703 842 870323 2387087.3 "   126416.106" "SVK" "USA" 6865 0    20527156026 0
2018 703 276 870323   1961001 "   125160.755" "SVK" "DEU"  553 1 3977289455.388 1
2018 703 251 870322   1534928 "   109528.554" "SVK" "FRA" 1092 1 2790956878.747 1
2020 703 251 870322 1732781.8 "   109034.343" "SVK" "FRA" 1092 1 2630317731.455 1
2011 703 156 870323   1614660 "    91122.784" "SVK" "CHN" 8451 0 7321892159.488 0
end

I figured that the best way to measure this data is by using the ppmlhdfe command however it seems to be showing me rather peculiar results of which I am not too sure about. I tried different versions of the code and looked at numerous questions on this forums and I keep on getting weird results. Is the ppmlhdf the best command to use for this project? is there anyway I can fix these errors I=with omitted variables and maybebe set the model up in a way that will give me the results that go in line with the literature?

This code for example:

ppmlhdfe v ln_dist ln_gdp_o ln_gdp_d eur eu_d contig comlang_off, a(pair t) cluster(pair)

gives me the results attached below


Thank you all so much in advance!

[Joao Santos Silva]

Dear Tomas Pientka,

Indeed, ppmlhdfe is the best command for what you are doing and the results do not show any "errors"; what happens is that you are trying to include in the model variables that are perfectly collinear with the fixed effects and therefore are dropped. You only need to estimate the the coefficients on the eurozone dummy, so it is fine that the other variables are dropped. On a side note, I suggest you think carefully about the set of fixed effects to use (e.g., consider the so-called 3-way specification).

Best wishes,

Joao

[Tomas Pientka]

Dear Joao Santos Silva

I greatly appreciate your feedback, and have implemented the 3-way specification into my ppmlhdfe code. I also decided to use data including other countries than Slovakia, and rather create a euro_SVK dummy. This regression works well on all of the dummies except the euro dummies which are always insignificant. I am aware of the study by Larch et al. (2018) which uncover these results for the effect of the Euro however Gunnella et al. (2023) or Kopecky (2024) still found statistically significant results although very small to be fair. I wanted to kindly ask you whether you could advise me if I am missing something important or if these results are reasonable? I have been checking my data and all seems to be in order.

I am using this code:

ppmlhdfe export euro_SVK euro_no_SVK cu_no_emu eu fta_no_eu, absorb(i.exporter#i.year i.importer#i.year i.exporter#i.importer) cluster (pair_id)

And constantly I am getting these results:

I would greatly appreciate your valuable advice, and I thank you in advance for it.
Kind regards,
Tomas Pientka

[Joao Santos Silva]

Dear Tomas Pientka,

I do not think you should use the i. notation in the inside absorb(), but I am not sure it makes a difference. Anyway, I also have an old paper where we did not find a significant effect (you can see a summary here), and this recent paper suggests that the results are very sensitive to the data used.

Best wishes,

Joao