I'm trying to construct a gravity model for EU trade flows using a panel dataset, but am suffering from what appears to be a persistent omitted variable problem. My residuals display a queer log-like, almost polynomial type pattern when plotted against the log of trade volumes.
I have searched the literature and included every variable in the model that has appeared in previous studies, including the logs of; - gdp per capita - real and nominal gdp - distance - multiple different measures. - Real and nominal exchange rates. - Price levels - various measures. - Proxies for product differentiation. - Exporter / importer fixed effects. - Dummies for things like Schengen / Euro membership, home trade, etc. Still, the curvilinear / log pattern remains.
Can anyone conjecture as to what the omitted variable may be? I have exhausted my knowledge of the theory.
Any help would be very gratefully appreciated.