Dear all,
I am doing a study on how Brexit can impact the EU export by using the gravity model. My sample is a panel data, in which 28 EU countries exports to all countries (including EU countries) across 41 sectors (as defined by GTAP) from 1988 to 2018. My independent variables are traditional gravity variables such as contiguity, distance, GDP. My dummy variables/interest variables are "EEA" dummy accounting for if an importer is an EEA (European Economic Area) member; "auto" dummy = 1 if the trade in the automotive industry and the interaction term EEA_auto.
I use PPML estimator and exporter- and importer-time fixed effects for the estimation. My code is as follows:
egen exp_time = group(exp year)
quietly tabulate exp_time, generate(exp_time_fe)
egen imp_time = group(imp year)
quietly tabulate imp_time, generate(imp_time_fe)
*(1) Fixed effects including GDP variables (Just to compare)
ppml tradevalue gravity_variables_withGDP EEA auto EEA_auto exp_time_fe* imp_time_fe*, cluster (dist)
*(2) Fixed effects without GDP variables (The specification is used for interpretation)
ppml tradevalue gravity_variables_noGDP EEA auto EEA_auto exp_time_fe* imp_time_fe*, cluster (dist)
The coefficient of the EEA dummy changes significantly from (1) to (2), which makes the effects of EEA dummy is too large that I could not interpret. I acknowledge that is a problem probably due to multicollinearity and I should not interpret it as my explanatory variable, but it is my interest variable, I would like to keep it but I don't know exactly how to fix this issue.
Then I also try to include only exporter and importer and year fixed effects instead of exporter- and importer-time fixed effects. My code is:
egen exp_1 = group(exp)
tabulate exp_1, generate(exp_fe)
egen imp_1 = group(imp)
tabulate imp_1, generate(imp_fe)
egen year_1 = group(year)
tabulate year_1, generate(year_fe)
*(3)
ppml tradevalue gravity_variables_withGDP EEA auto EEA_auto ///
exp_fe* imp_fe* year_fe* ///
, cluster(dist)
*(4)
ppml tradevalue gravity_variable_noGDP EEA auto EEA_auto ///
exp_fe* imp_fe* year_fe*, cluster(dist)
All parameters look much more consistent between (3) and (4) and the effects of my interest variables are perfectly fine to interpret.
My question is: Is it appropriate that I control for exporter and importer and year fixed effects instead of exporter- and importer-time fixed effects and use that result to interpret? If not, could you please give me some advice on how to still keep exporter- and importer-time fixed effects and obtain a better result?
Note: I already use ppml_panel_sg but it did not work since it will absorb my interest variables. Therefore, I sticked to the normal ppml command.
Thank you very much in advance for your time and consideration!
I am doing a study on how Brexit can impact the EU export by using the gravity model. My sample is a panel data, in which 28 EU countries exports to all countries (including EU countries) across 41 sectors (as defined by GTAP) from 1988 to 2018. My independent variables are traditional gravity variables such as contiguity, distance, GDP. My dummy variables/interest variables are "EEA" dummy accounting for if an importer is an EEA (European Economic Area) member; "auto" dummy = 1 if the trade in the automotive industry and the interaction term EEA_auto.
I use PPML estimator and exporter- and importer-time fixed effects for the estimation. My code is as follows:
egen exp_time = group(exp year)
quietly tabulate exp_time, generate(exp_time_fe)
egen imp_time = group(imp year)
quietly tabulate imp_time, generate(imp_time_fe)
*(1) Fixed effects including GDP variables (Just to compare)
ppml tradevalue gravity_variables_withGDP EEA auto EEA_auto exp_time_fe* imp_time_fe*, cluster (dist)
*(2) Fixed effects without GDP variables (The specification is used for interpretation)
ppml tradevalue gravity_variables_noGDP EEA auto EEA_auto exp_time_fe* imp_time_fe*, cluster (dist)
The coefficient of the EEA dummy changes significantly from (1) to (2), which makes the effects of EEA dummy is too large that I could not interpret. I acknowledge that is a problem probably due to multicollinearity and I should not interpret it as my explanatory variable, but it is my interest variable, I would like to keep it but I don't know exactly how to fix this issue.
Then I also try to include only exporter and importer and year fixed effects instead of exporter- and importer-time fixed effects. My code is:
egen exp_1 = group(exp)
tabulate exp_1, generate(exp_fe)
egen imp_1 = group(imp)
tabulate imp_1, generate(imp_fe)
egen year_1 = group(year)
tabulate year_1, generate(year_fe)
*(3)
ppml tradevalue gravity_variables_withGDP EEA auto EEA_auto ///
exp_fe* imp_fe* year_fe* ///
, cluster(dist)
*(4)
ppml tradevalue gravity_variable_noGDP EEA auto EEA_auto ///
exp_fe* imp_fe* year_fe*, cluster(dist)
All parameters look much more consistent between (3) and (4) and the effects of my interest variables are perfectly fine to interpret.
My question is: Is it appropriate that I control for exporter and importer and year fixed effects instead of exporter- and importer-time fixed effects and use that result to interpret? If not, could you please give me some advice on how to still keep exporter- and importer-time fixed effects and obtain a better result?
Note: I already use ppml_panel_sg but it did not work since it will absorb my interest variables. Therefore, I sticked to the normal ppml command.
Thank you very much in advance for your time and consideration!
Comment