Two networks, same countries

The question

Ask which economies are central to world trade and the usual answer counts gross exports: the full value of everything that crosses a border. That number has a known problem. An input crossing three borders before final assembly is counted three times, so gross trade measures shipments, not production. A country that imports parts, adds a little, and re-exports looks larger than it is.

The value-added view asks a different question: of what a country ships, how much did it actually make? Both are legitimate. They are simply not the same question, and confusing them is the single most common mistake in trade commentary.

What the data says

Both networks below come from the same Eurostat FIGARO accounts, for the same countries and the same years, so nothing here turns on mixing sources. Nodes are countries, edges are bilateral flows.

The top of the ranking barely moves, which is itself worth noticing: for the largest economies the two measures broadly agree. The interesting part is the ratio between them.

Slovakia at 40 per cent is not a poor performer. It is an assembly economy: it imports parts and exports cars, and the gross figure credits it with value created in Germany and Czechia. Luxembourg at 33 per cent is a different story again, mostly financial and merchanting flows. The measure separates two things gross trade blends together.

One mistake this exercise makes easy to see

The first version of these numbers was wrong, and instructively so. FIGARO reports TOTAL and NEU27_2020 as partners beside individual countries. Left in the edge list, every country's largest trading partner came out as "TOTAL" at forty per cent, and outward strength roughly doubled, because the sum of the flows was being counted alongside the flows. Aggregates in a partner dimension are not a tidying problem; they silently produce a plausible, wrong answer. The published method says which codes are excluded and why.

The limits — read these before citing anything above

• This is not the world. FIGARO covers 49 economies: the EU and its main partners. China, the United States and Japan appear as partners but the network is built from an EU vantage point. Statements about global trade centrality cannot be made from it.
• One industry: all of them. Everything here is the TOTAL industry aggregate. Value chains are industry phenomena — the automotive network and the pharmaceutical network look nothing alike — and averaging them hides exactly what a value-chain analysis is for.
• Domestic value added is not a full decomposition. The literature after Koopman, Wang and Wei splits gross exports into many terms: value added absorbed abroad, value added that returns home, and pure double counting. This page uses one of those terms. It is the right first cut and it is not the accounting identity.
• The network is nearly complete. Almost every country trades with almost every other, so the topology carries little information and the weights carry nearly all of it. That is why the measures here are weight-based, and why betweenness and eigenvector centrality are absent: on a dense weighted graph they depend more on the threshold you chose than on the data.
• Two units, one comparison. The gross table is published in millions of euro and the value-added table in thousands. Comparing them without converting is wrong by a factor of a thousand — a mistake that produces numbers which look plausible.
• It ends in 2022. Inter-country input-output accounts take years to assemble. Nothing here says anything about last quarter.

Why a more comprehensive approach is needed

Each limit above points at the same missing thing. A country-to-country table is a projection of something larger, and the questions people actually ask live in the larger object.

"Which industry depends on which industry" needs an industry-by-industry matrix, not a country-by-country one. "Which product, from where, to where" needs product-level bilateral data. "How does a shock propagate" needs the full inter-country input-output structure, because propagation runs along industry links, not country links. All three are the same missing dimension: this database can hold indicator by area by partner by period, and a matrix needs an industry pair as well.

That is a deliberate gap rather than an oversight. It is recorded in the source register beside the sources that would fill it — OECD ICIO, EXIOBASE, WIOD — and the decision to add it should be made once and serve trade, value added, finance and product data together, rather than three times in three shapes.

Until then, the honest description of this page is: a first cut, on a partial network, of one term of a larger decomposition, for one industry aggregate. It is enough to show that the two ways of counting disagree, and to show where. It is not enough to rank the world.

Answer quality: Indicative — Suggestive rather than conclusive. Read the limits before citing.

Coverage

49 economies from an EU vantage point, not the world. One industry aggregate, not the industry detail value chains actually live in.

Directness

One term of a larger decomposition. Domestic value added is the right first cut and is not the full accounting identity.

Currency

Ends in 2022. Inter-country input-output accounts take years to assemble.

Provenance

Eurostat FIGARO throughout, so both networks come from one set of accounts; the derived measures carry a written method.

Enough to show that the two ways of counting disagree, and where. Not enough to rank the world.

This assessment is written by the author of the page, not computed. A score derived from a formula would look objective and would not be.