Replication Data for: The Customs-BoP Gap is Not in the Customs — A Mirror-Trade Verdict on China's Missing Surplus

Replication package for a follow-up probe to Peters (2026), 'Decomposing the 2020s Asian Surplus' (Paper 35), triggered by Brad Setser's April 30 2026 tweet thread on Ma & Wei (2026, NBER WP 35056). Tests whether mirror trade data — Chinese-reported exports vs partner-reported imports — discriminates between Wei's two proposed channels for the Chinese customs-BoP gap (VAT-refund fraud and contract manufacturing) and Setser's alternative thesis (SAFE smooths errors-and-omissions toward zero). Uses UN Comtrade public preview API for HS-2 chapters 84/85/87, 12 major partners, 2018-2024, with re-pulled C00/motCode=0/partner2Code=0 totals for partners returning multi-row queries. CIF→FOB adjustment of 4% applied (IMF standard). Key findings: (1) USA-CHN bilateral mirror gap sums to +1.3% over 2018-2024 — VAT-refund over-invoicing at the scale needed to drive a $184bn customs-BoP gap is ruled out. (2) Aggregate mirror gap closed 2018→2024 ($106bn → $10bn) while customs-BoP gap opened (-$10bn → +$184bn) over the same window — the two gaps moved in opposite directions, meaning whatever drives the SAFE adjustment is invisible to partner customs offices. Both Wei's contract-manufacturing channel and Setser's smoothing thesis remain consistent with the data and cannot be discriminated by mirror trade alone. (3) HS-87 vehicles show a persistent negative gap (-5% to -14%, $2-8bn/year) plausibly attributable to sanctions-driven undercounting in RUS/BLR/KGZ, not to VAT fraud.