Tomáš Havránek, Zuzana Iršová
National borders reduce trade, but most estimates of the border effect seem puzzlingly large. We show that major methodological innovations of the last decade combine to shrink the border effect to a one-third reduction in international trade flows worldwide. The border effect varies across regions: it is substantial in emerging countries, but relatively small in OECD countries. For the computation we collect 1,271 estimates of the border effect reported in 61 studies, codify 32 aspects of study design that may influence the estimates, and use Bayesian model averaging to take into account model uncertainty in meta-analysis. Our results suggest that methods systematically affect the estimated border effects. Especially important is the level of aggregation, measurement of internal and external distance, control for multilateral resistance, and treatment of zero trade flows. We find no evidence of publication bias.
JEL codes: F14, F15
Keywords: Bayesian model averaging, bilateral trade, borders, gravity, meta-analysis, publication selection bias
Issued: March 2015
Download: CNB WP No. 1/2015 (pdf, 825 kB)