From Shipments to Supply Chains: Mining Input-Output Links from Firm-level Trade Flows

We develop a method to recover a granular, product-level input-output structure from firm-level customs transactions. Building on an association-rule mining algorithm, we exploit systematic co-occurrences between what firms import and what they export to infer input use in production. Applying the approach to data from several countries, we show that the inferred mappings closely resemble conventional input-output relationships and perform well in external validation exercises. We illustrate the value of these linkages in two applications. First, following China's WTO entry, export growth is accompanied by a pronounced rise in imports of the inputs our mapping associates with those exports. Second, in the 2018-2019 U.S.-China trade war, tariffed Chinese exports to the United States contract, and the shock propagates upstream: China's imports of exposed inputs fall, and third-country suppliers that specialize in those inputs experience the declines in their exports to China, especially if China is a key destination market. Overall, simple pattern recognition tools and micro-level trade data can deliver high-resolution input–output linkages that improve exposure measures and empirical assessments of supply-chain propagation.