How far upstream does each economy sit in world production?

A tonne of iron ore passes through smelting, casting, stamping, assembly and retail before anyone consumes it; a haircut is consumed the moment it is produced. Upstreamness makes that difference a number: the expected count of production stages between an industry's output and final demand. This page computes it from the full inter-industry structure of the ADB Multi-Regional Input-Output tables, 62 economies plus a Rest of World block, 35 sectors, 19 years, and asks which industries and which economies sell farthest from the final consumer.

What upstreamness measures

Antras, Chor, Fally and Hillberry (2012, American Economic Review Papers and Proceedings102(3): 412-416) define upstreamness U as the expected number of production stages between an industry's output and final demand: U = 1 for an industry whose entire output goes straight to final consumers, 2 for one whose output enters exactly one further stage, and so on, with the full distribution of paths through the input-output matrix resolved by a Leontief-type inversion. Here U is computed from the Asian Development Bank's Multi-Regional Input-Output tables (ADB Key Indicators Database, kidb.adb.org), a closed world system of 62 economies plus a Rest of World block, each split into 35 sectors, so no open-economy trade correction is needed. Industries with gross output at or below 15 million USD are dropped from the reported panel (their table entries are balancing allocations, not measured flows, and produce phantom outliers); above that floor, U in this panel runs from 1.00 to 10.58 (the maximum is LVA's mining and quarrying sector in 2000).

The world's most and least upstream industries

The 2024 ordering reads like a walk down a supply chain. At the top sit the industries whose output almost never meets a consumer directly: mining and quarrying (U = 4.02), electricity, gas and water (3.49), and basic and fabricated metals (3.34), each sold overwhelmingly to other producers who process it further. At the bottom sit the services delivered face to face: health and social work (U = 1.09, almost pure final demand), education (1.21) and public administration (1.24). Construction (1.52) is the interesting low case: it devours steel, cement and machinery, but its own output, a finished building, is final demand, so it anchors long chains without being upstream itself. The measure is about where output goes, not how much processing came before.

Has world production gotten more roundabout?

Yes, and the timing tracks the fragmentation of manufacturing. The cross-sector world mean rose from 2.13 in 2000 to 2.24 by 2007, climbed unevenly through the 2010s, and peaked at 2.46 in 2022, the year supply chains were stretched by the post-pandemic restocking surge and the commodity-price spike that inflated the upstream tiers. By 2024 it had eased back to 2.37, a step down from the peak that returns the series to its 2013-2021 range, not to pre-crisis levels. Production today passes through 0.23 more stages than in 2000 on this measure: modest per sector, but applied to the whole world economy it is an enormous volume of intermediate shipments that did not exist a generation ago.

Where economies sit: structure, not income

Average each economy's sector-level U (simple mean across its reported sectors, since the panel carries no output weights at this level) and the 2024 ranking is led by MNG (2.86), KOR (2.79) and CHN (2.75): a mining economy, a heavy-industry exporter and the world's largest producer of intermediate goods. The bottom holds BTN (1.69), PAK (1.71) and BGD (1.71), economies whose garments, crops and services sell close to final consumers. Income explains little: the United States, at $85K per head, ranks 54 of 62 with a mean U of 1.90, because a service economy consumes most of its own output one step from production. The honest reading of the scatter is that industrial structure, not development level, sets chain position. One caveat the simple mean imposes: it weights all sectors equally, so a hydrocarbon exporter like Brunei averages just 1.73 even though its mining sector alone scores 3.41; the mean describes the breadth of an economy's industrial structure, not the composition of its output.

Movers, 2007-2024

Position changes slowly, but it changes. Comparing each economy's mean U in 2007 and 2024 on the set of sectors observed in both years, the largest upstream moves belong to small hub economies and mining exporters: MLT (+0.52), LAO (+0.52), CYP (+0.48) and MNG (+0.41), with NLD (+0.39) and KOR (+0.30) showing that large industrial economies drifted the same way as their intermediate and re-export business deepened. The largest downstream moves are BRN (-0.47), KAZ (-0.38), MYS (-0.29) and THA (-0.28): hydrocarbon and commodity economies alongside Southeast Asian manufacturers, all with sector-level chain positions that shifted toward final demand over the period.

Upstreamness is not the mirror of downstreamness

The same tables yield a second measure: downstreamness D, the expected number of stages embodied in an industry's production on the input side (D = 1 for an industry using primary factors only). The two are far from mirror images. Mining sells into long chains (U = 4.02) but buys short ones (D = 2.19); transport equipment is the reverse, close to final demand on the sales side (U = 2.03) yet embodying more prior stages than almost any other sector (D = 3.04), the signature of complex final assembly. Electrical and optical equipment is long on both dimensions (U = 2.66, D = 3.03), a mid-chain sector that both absorbs and feeds elaborate networks, while real estate is short on both (U = 1.56, D = 1.51). Construction pairs one of the lowest U values (1.52) with a high D (2.69): the end of many chains, the start of none.

Where to go next

The GVC pages carry a position measure for each economy, but it is a different object: the per-country profiles on /gvc infer position from trade flows (the balance of foreign value added in exports against domestic value added re-exported by partners), a proxy built from what crosses borders. The measure on this page is computed from the full inter-industry structure, every domestic and cross-border transaction in the MRIO system, so it counts the stages themselves rather than inferring them. The two agree on the broad map and disagree instructively where domestic processing is long but trade is thin. For the rest of the stack, the supply-chain hub collects concentration, chokepoints and vulnerability, and critical inputs asks which upstream products each economy cannot export without.