Tackling discrepancies in trade data: the Harvard Growth Lab international trade datasets

Abstract

Bilateral trade data informs foreign and domestic policy decisions, serves as a growth indicator, determines tariffs, and is the basis for financial and investment decisions for corporations. Accurate trade data translates into better decision-making. However, the raw bilateral trade data reported by UN Comtrade suffer from two structural problems: reporting differences between country partners and countries reporting in different product classification systems, which require product-level harmonization to compare data across countries. In this paper, we address these challenges by combining a mirroring technique and a data-driven concordance method. Mirroring reconciles importer and exporter differences by imputing country reliability scores and applying a weighted country-pair average to calculate the estimated trade value. We harmonize product classifications across vintages by calculating conversion weights that reflect a product's market share. The resulting publicly available datasets mitigate issues in raw trade statistics, reducing reporting inconsistencies while maintaining product-level granularity across six decades.