The Structure of Ecological Unequal Exchange

Who appropriates whose materials, energy, land and labour through global trade — and, the question this project adds, how that appropriation is organised as a network.

The question

Ecologically unequal exchange is now measurable at global scale. High-income economies appropriate raw materials, energy, land and labour from lower-income economies in physical volumes far exceeding what they send back, while capturing the value added. That result, in the form established by Dorninger et al. (2021), is fundamentally bilateral: a set of net flows between countries and country groups.

Bilateral flows tell you that country A appropriates from country B. They do not tell you how the appropriation is structured:

• Is it organised through a small set of broker sectors intermediating between peripheral extraction and core consumption?
• Does the world system decompose into regional blocs with distinct ecological metabolisms, or into a single core–periphery hierarchy? These are competing hypotheses, not complementary ones.
• Is a country's position a property of its size, or of its structural role?
• How deep are the supply chains through which appropriation actually travels? A first-tier flow and a fifth-tier embodied flow describe very different economies.

Why a network, and not a metaphor

There is a precise reason network structure is the right object here rather than a borrowed image. Writing A for the matrix of technical coefficients, the Leontief inverse is a geometric series:

Because A is column-substochastic, this is a random walk with leakage, and the inverse is its walk-generating function. Leontief multipliers therefore are a path-counting centrality, in the literal sense — not an analogy to one. Computing centrality on this network computes the same quantity input–output economics has always computed, in a language that lets structural questions be asked of it.

Data and method

The project is built on the GLORIA multi-regional supply-use tables (release 060), converted to a Leontief input–output system and solved as an environmentally extended MRIO. Embodied resource flows follow F = q̂ L ŷ; trade in value added follows B = p̂ L ŷ.

Parsing is done in Julia via Juliora, with large matrices held outside R and only aggregates or edge lists crossing the boundary. The Leontief inverse is never materialised — every question takes the form of a backsolve against a stored factorization, which is what makes the analysis tractable on ordinary hardware rather than requiring cloud compute.

Measures deployed

• Global structure — edge-weight and strength distributions; density measured on a disparity-filter backbone at a stated threshold, since MRIO graphs are near-complete and raw density is uninformative.
• Per-country tail parameters against the global benchmark, extending Acemoglu et al. (2012) from a single national input–output network to 164 economies, with log-normal tested against power law rather than assumed.
• Centrality — strength, PageRank and betweenness over country-sector nodes, with flow weights inverted to distances before any shortest-path computation.
• Community structure and hierarchy, using direction-aware methods, since the entire claim of unequal exchange is directional.
• Core–periphery detection, arbitrated against modularity by a degree-corrected stochastic block model rather than reported alongside it.

Status

The 2019 cross-section is the primary deliverable. Temporal extension follows once the single-year structure is characterised — first a spread of years chosen around the endpoints and the pandemic disruption, then the full series.