Hello!
I am estimating a PPML gravity-inspired model where my dependent variable (a production-related variable) varies at the exporter-sector-year level only - it takes the same value for all importers within a given exporter-sector-year cell. My key independent variable (trade policy) varies at the importer-year level.
My estimation code:
ppmlhdfe production_variable importer_tradepolicy importer_GDP, absorb(Exporter_id#importer_id#sector_id Exporter_id#year sector_id#year) nolog
All coefficients on importer-side variables return as machine zeros (e.g., coefficients of 10⁻¹², z = 0.00, p = 1.000). I believe this occurs because Exporter_id#year FEs absorb all variation in the dependent variable, leaving no residual variation for importer-side variables to explain.
My sample has ~960,000 observations: ~185 exporters, ~30 importers, 26 sectors, 8 years.
Is this diagnosis correct? Is there a specification that allows estimating the effect of an importer-side variable on an exporter-determined dependent variable within a gravity FE structure, or is this fundamentally unidentifiable?
Thank you in advance!
I am estimating a PPML gravity-inspired model where my dependent variable (a production-related variable) varies at the exporter-sector-year level only - it takes the same value for all importers within a given exporter-sector-year cell. My key independent variable (trade policy) varies at the importer-year level.
My estimation code:
ppmlhdfe production_variable importer_tradepolicy importer_GDP, absorb(Exporter_id#importer_id#sector_id Exporter_id#year sector_id#year) nolog
All coefficients on importer-side variables return as machine zeros (e.g., coefficients of 10⁻¹², z = 0.00, p = 1.000). I believe this occurs because Exporter_id#year FEs absorb all variation in the dependent variable, leaving no residual variation for importer-side variables to explain.
My sample has ~960,000 observations: ~185 exporters, ~30 importers, 26 sectors, 8 years.
Is this diagnosis correct? Is there a specification that allows estimating the effect of an importer-side variable on an exporter-determined dependent variable within a gravity FE structure, or is this fundamentally unidentifiable?
Thank you in advance!
Comment