Trading with the Machines: Partner-Country AI Innovation and African Export Performance

Trading with the Machines: Partner-Country AI Innovation and African Export Performance." The paper estimates whether growth in AI patenting among destination economies raised African merchandise exports between 2016 and 2023, using a shift-share instrument built from destinations' baseline composition of AI patenting across technology fields. UN Comtrade Database, UN Statistics Division, extracted 10 August 2026. Eight annual files, 2016–2023, split by year because a single query is capped at 100,000 records. Parameters: goods, annual, HS as reported, AG2 commodity codes, all 54 African economies as reporters, all partners, second partner World, exports only, total modes and customs codes, Plus breakdown. Appended: 629,649 records, 49 reporters, 238 partners, no duplicates. The WORLD_TOTALS file is a separate extract with partner set to World; it supplies export-share denominators and does not enter the estimation sample. Encoding is latin-1. Rows carry a trailing comma, creating one empty variable on import. Import all columns as strings so chapter codes 01–09 keep leading zeros. Country AI Activity Metrics, Emerging Technology Observatory, CSET, Georgetown University, extracted March 2026. ai_patents_subfield.csv reports AI patent applications by country, field and year and supplies the treatment; ai_patents_granted_subfield.csv reports granted patents and is retained but unused. Attribution follows the priority jurisdiction, the first office at which protection was sought, so the data describe where patents are filed rather than where inventors reside. CSET flags counts incomplete from 2022, reflecting the filing-to-publication lag; baseline shares are fixed at 2016–2018. Nine of eleven fields map onto merchandise trade. cset_patents_iso3.csv adds ISO3 codes and removes aggregates and patent offices, leaving 69 countries; the estimation code reads this file. cset_pubs_iso3.csv is the publications equivalent, unused, as its field taxonomy has no sectoral interpretation. afcfta_ratification.csv gives deposit dates for 49 AfCFTA signatories from the tralac Ratification Barometer (June 2026); unused, because by 2023 the never-treated pool was too thin to identify anything. code/ — Stata do-files for construction, estimation and figures. Run 00_master_v2.do. Bilateral geography variables from CEPII, not redistributed under CEPII's terms. Download from cepii.fr; the merge is documented in 02_merge_gravity_v2.do. These are absorbed by exporter-destination fixed effects in the headline specification.