Trade Index Update, October 2026

GINC Trade at trade.ginc.org is now our primary source of global trade data insights. It covers who trades what with whom for 234 economies and 5,020 products, every year from 1995 to 2024. This week we also published the first version of the GINC Trade Dependency Graph. It is a map of which products are made from which, built by assessing every possible pairing of those 5,020 products.

Why we built it. Trade is one of the clearest signals of how nations depend on each other, and of how those ties shift over time. Our aim is to help us all build a better understanding of global trade relationships, how they change and where vulnerability and leverage sits. Each nation and product category has its own page, with its trade over time and its primary import and export partners. The explorer on the home page shows any year’s flows for all goods or for a single category, and any view can be linked to and shared.

Sources. The site is built on BACI, the international trade database from the French research centre CEPII. BACI reconciles what exporters and importers each report to the UN’s Comtrade database into one figure for every flow. Each figure is a single year, exporter, importer and product, at the six-digit level of the Harmonized System (HS), the international product classification used by customs authorities. We use the 1992 revision of the classification, the only one that covers every year from 1995. Product names also draw on the Harvard Growth Lab’s classifications.

Where this is heading. The raw flows are the foundation. Building on the work of Comtrade, CEPII, the Harvard Growth Lab and others, we intend to develop composite indices that show more clearly what is happening over time. The Dependency Graph is one of the first steps towards new insights which we now intend to validate against prior work in the field such as that of AIPNET, an academic production network published by Fetzer, Lambert, Feld and Garg (2024).

The Dependency Graph. Trade data shows what moves between countries but not how products relate to each other. The graph fills that gap. An edge from one product to another means the first is a direct input to making the second: iron ore to pig iron, cocoa beans to cocoa paste. Inputs of inputs are not counted directly, but they can be traced by following the graph. Version 0.01 holds 399,510 edges between the 5,020 products.

Every combination tested. A language model screened all 25.2 million ordered pairs of products. It asked the question both ways round: is product A an input to product B, and is B made from A? It returned a probability for each answer. Uncertain or disputed pairs, about 357,000 of them, got a second and closer look. A separate pass then set 979 types of machinery, tools and instruments against every product.
Input and output shares. Each edge carries two figures. The input share is how much of a product’s dependence on its inputs comes from that one input, and a product’s input shares add up to 100 per cent. The output share is how much of an input’s direct uses go to that product, and those add up to 100 per cent across its uses. Each share comes with a confidence, and each edge has a probability that it is a real direct input and a role: principal material, component, process material, energy, capital equipment or packaging.
Dependence, not cost. These are shares of dependence, not of cost. Photoresist is a tiny fraction of the cost of a semiconductor and still essential to making one.
First results. Excluding machinery, the products the most others depend on directly include electricity, salt, sulphuric acid, caustic soda, crude petroleum and primary iron. Where a nation dominates the trade in products like these, it holds leverage over everyone who makes things from them.
Where to find it. Every product now has its own page showing what it is made from and what it is used to make.

This is an early version. One model family made these judgements, and more than half of the edges have estimated rather than individually judged importance. Expect the graph and the indices built on it to change as we add further reviewers and take in feedback. If a link looks wrong, or you know a supply chain better than the graph does, please leave us feedback.