Information Constraints, Benchmark Dispersion, and Border Misreporting: Evidence from Japanese Customs Data

Border taxes are effective only insofar as customs can verify declarations at the border. We study whether higher border tax rates induce importers to understate the declared tax base, and whether such responses are concentrated where customs has less information from past transactions. Using confidential Japanese customs microdata linked to UN Comtrade, we construct HS6 product–partner–year cells for 2014–2021 and estimate how tax rates affect the trade-cost-adjusted log gap between partnerreported exports and Japan-reported imports, a reduced-form proxy for trade misreporting. The average semi-elasticity of this gap with respect to the tax rate is positive but statistically indistinguishable from zero. Yet the average masks sharp heterogeneity: responses are close to zero in information-rich lanes but economically meaningful in thin-information lanes, and somewhat larger where implied-unit-value dispersion is greater. A decomposition of mirror-data outcomes shows that the tax-responsive component of these discrepancies appears mainly in quantities rather than in implied unit values. The results imply that, even in a high-capacity setting, the effective incidence of border taxes depends on the lane-level information available for enforcement.