I am currently trying to replicate Section 4.1. of a paper by "On the Heterogeneous Effects of Sanctions on Trade and Welfare: Evidence from the Sanctions on Iran and a New Database" by Felbermayr et al. (2020). This paper tries to see the effect of various sanctions on trade flows, using the Global Sanctions Database (a big database of trade sanctions at different time periods).

Following the authors of the paper, I conducted a regression where I want to see the effect of sanctions against Iran on the trade flows of the sanctionning country.

Thus, I created a big number of dummy variables that work this way :

COUNTRY_IRN => Takes the value 1 if the sanction in question is a sanction of "Country" (replace with any country) on Iranian imports.

COUNTRY_IRN => Takes the value 1 if the sanction in question is a sanction of "Country" (replace with any country) on Iranian exports.

I have such dummies for 13 different countries, making it 26 dummies in total. When I run my regression (my other independant variables are dummies that identify complete/partial trade sancitons, as well as for exportation and exportation sanctions towards Iran. Also, a bunch of time invariant controls for country-pairs), I get these results :


Array




I used a Poisson Pseudo Maximum Likelihood regression model with country-pair fixed effect, time-varying exporting-country dummies and time-varying destination-country dummy variables, The command I used to get these results is :

Code:
ppmlhdfe tradeflow_comtrade_o TRADE_SANCT_COMPL TRADE_SANCT_PARTL OTHER_SANCT SANCT_IRAN_EXP SANCT_IRAN_IMP IRN_USA USA_IRN IRN_CAN CAN_IRN IRN_AUS AUS_IRN IRN_CHE CHE_IRN IRN_CHN CHN_IRN IRN_TUR TUR_IRN IRN_BRA BRA_IRN IRN_ARE ARE_IRN IRN_RUS RUS_IRN IRN_IND IND_IRN IRN_ZAF ZAF_IRN IRN_JPN JPN_IRN IRN_SGP SGP_IRN log_dist comlang_off sibling_ever contig RTA, a(CountryPairs Sanctioning_time_fixed Sanctioned_time_fixed)

As you can see, all country-specific dummies, save for the USA, have been omitted from the regression because of collinearity (as stated by Stata), and I cannot find out why. Is it linked to the fixed effects I used ? Is it because all of these dummies are mutually exclusive ? What can I do to get coefficients for all these variables, without Stata omitting them ?

I hope I have been clear enough, thank you in advance for your answers.