World trade grew more connected—but its feedback loops weakened

Global trade moved more goods across borders between 1995 and 2023, but the world economy did not become more tightly bound by circular chains of influence. In the measure introduced in this paper, the diffusion determinant rose from 0.12 in 1995 to 0.19 in 2023. That rise signals weaker circular interdependence: influence was less likely to travel through a sequence of trade relationships and return to its starting point.

It is a counterintuitive result because trade networks are often described through growth, connectivity and the importance of hubs. But a network can become larger or more connected without becoming more mutually dependent. Didier Lebert’s paper argues that the difference matters. Its framework, Economic Dominance Theory (EDT), asks not simply which countries trade, or how central each country is, but how much trade influence circulates back through the network.

The findings offer a different lens on globalization. Short feedback loops—especially direct two-country exchanges—remain geographically concentrated. Longer loops, by contrast, cross regions overwhelmingly. And most of the feedback beyond direct reciprocal trade runs across product categories, not within a single product network. Those patterns suggest that the structure of dependence is not captured by a single map of countries or a single score of connectivity.

Trade feedback weakened overall, with reversals along the way

Label	Value
1995	0.12
2006	0.21
2013	0.16
2023	0.19

Reported values only; higher diffusion determinant indicates weaker feedback.

The Science

The World Trade Web is the network formed by countries’ exports and imports. A great deal of network research describes it using familiar measures: how many trading partners each country has, how much trade passes through it, whether countries cluster into communities, or how similar trade patterns are across products and years. These measures answer useful questions about connections and position. Lebert’s central claim is that they do not directly measure whether influence loops back through the system.

Economic Dominance Theory, first introduced by Lantner in 1974, treats trade as directed flows. An export from country A to country B is not interchangeable with an export from B to A. The approach normalizes bilateral flows into a Ghosh matrix, a representation of how activity in one part of an economy propagates toward other parts. The paper then uses the determinant of that matrix to summarize circular interdependence: the extent to which a chain of directed influences can eventually return to its origin.

A simple way to picture the distinction is to imagine a chain of trade links. A country sells to a second country, which sells to a third. That is a path through the network. If the third country’s trade, in turn, helps sustain a flow that returns to the first, the path closes into a feedback circuit. EDT emphasizes the closed circuit, not just the countries’ positions along the path.

The mathematical workhorse is a log-determinant and its trace expansion. In plain language, this lets one overall quantity be separated into contributions from feedback circuits of different lengths. A short loop might involve reciprocal trade between two countries; longer loops pass through more nodes before returning. The same accounting can be decomposed by removing an arc or a node, separating product layers, comparing regions, or tracking change over time. The paper’s methodological promise is exact decomposition: instead of inventing a distance between two complex graphs, it can identify which circuits or layers account for a change in the measure.

The application uses CEPII BACI international trade data from 1995 to 2023, arranged into 96 product categories at the Harmonized System’s two-digit level (HS2). The supplied paper summary reports the headline outcomes, but not the underlying tables or figure files. So the charts here reproduce only the reported values and rounded shares; they do not infer missing years or country-level estimates.

What They Found

The diffusion determinant—the paper’s indicator that increases as feedback weakens—was 0.12 in 1995, 0.21 in 2006, 0.16 in 2013 and 0.19 in 2023. The path was not a steady climb. It rose through 2006, fell by 2013, and then rose again. The paper’s overall interpretation is that circular interdependence did not deepen alongside the expansion of trade: by 2023 the indicator remained above its 1995 value.

That qualification matters. The measure is not a direct tally of trade volume or a verdict that the world became “less global.” It concerns a particular structural feature: how much directed trade influence closes into feedback loops. A higher diffusion determinant, in the paper’s interpretation, means weaker circular interdependence. The result therefore complicates any story in which more international exchange automatically means more mutual dependence.

Cross-product feedback dominates short loops

Label	Value
Order two	90 percent (more than)
Order three	99 percent (more than)

Order two is reported as more than 90%; order three as about 99%.

The strongest feedback does not sit neatly inside individual product categories. Cross-product feedback accounts for more than 90% of order-two feedback and about 99% of order-three feedback, according to the paper. Order two and order three refer to loops of different lengths in the trace expansion: the key point is that feedback involving multiple products dominates even among relatively short circuits. Trade in one category can be connected to the return path through another.

This is a useful corrective to treating product networks as separate worlds. A country’s role in food trade, machinery trade or chemicals trade may be described in distinct layers, but the feedback sustaining those flows can cross the boundaries between them. The result is not simply that products are “connected.” EDT offers an accounting framework for how much of the measured circularity belongs to within-layer versus cross-layer circuits.

Geography also changes with the length of the loop. Short reciprocal loops remain regional, while long feedback is overwhelmingly transregional. A direct, two-country exchange is more likely to be embedded in a regional relationship; once influence circulates through longer chains, those chains tend to span regions. The world trade network therefore has no single geography. The answer depends on whether the question concerns immediate reciprocity or the longer routes by which influence returns.

Together, these results replace a flat image of global connectedness with a layered one. Trade feedback has a time profile, a product structure and a geography that varies by loop length. The paper’s central methodological claim is that these are not separate snapshots stitched together after the fact: they can be expressed as components of one determinant-based object.

Why This Changes Things

The appeal of EDT is not that it makes other network measures obsolete. It is that it targets a different question. Centrality tells us which nodes occupy important positions according to a chosen definition. Community detection identifies groups with dense connections. Economic complexity focuses on productive capabilities. Input-output and value-chain analysis often trace positions along production chains. EDT shares some mathematical foundations with input-output analysis, Lebert notes, but shifts attention from where activity sits along a chain to whether flows close into circuits.

That distinction can affect how we interpret resilience and exposure. A country may have many trading partners, or occupy a prominent position in a network, without being strongly embedded in feedback loops. Conversely, a small number of carefully structured relationships could create circular dependencies that a partner count would miss. The study does not establish that one structure is safer or more dangerous than another. It gives researchers a way to ask the question more precisely: which flows return influence to their source, and how does that closure change?

The product result also has potential relevance for debates about supply-chain diversification. A policy that shifts sourcing within one sector might leave cross-sector feedback largely intact; another intervention could alter a longer circuit that spans products and regions. This paper does not evaluate policy interventions or identify specific vulnerable goods. Its contribution is a measurement framework that could help later work distinguish those cases rather than treating trade links as an undifferentiated web.

The geographic finding carries a similar implication. Regional trade arrangements may matter especially for short reciprocal relationships, while longer circuits connect more distant parts of the world. That gives analysts a reason to study regional integration and global interdependence together, rather than assuming that one simply replaces the other. A region can be tightly connected at short range while remaining part of a broader transregional feedback system.

Most strikingly, the headline time series cautions against equating trade growth with deeper interdependence. The diffusion determinant does not show a smooth, one-way historical trend; it rises, falls and rises. This is a reminder that a global system can expand in scale while changing in the structure of its mutual dependencies. Trade volume alone cannot tell us whether the world’s circuits of influence are becoming more closed.

What’s Next

The findings are a first application of the framework to a long, product-layered trade record, not a final account of what caused the observed shifts. The reported values identify when the diffusion determinant changed, but the supplied summary does not attribute those movements to particular countries, products, crises or policy decisions. Explaining the rises and falls would require tracing the decompositions in the full analysis and connecting them to historical events without confusing correlation with cause.

There are other questions to test. How sensitive are the results to how trade flows are normalized, how products are grouped, or which countries are included? Do the same patterns appear in services, investment or input-output data? Do particular kinds of trade shock weaken or strengthen feedback closure? And can the exact decomposition help identify which arcs or product layers account for a shift, rather than merely describing the overall direction?

Those questions matter because “dependence” is not automatically a bad thing. Feedback can mean reciprocal support, shared exposure, or both. A closed circuit might help transmit benefits through a system, but it might also allow disruption to travel back to its source. The determinant is a structural measure, not a measure of welfare, fairness or resilience on its own. Interpreting it requires combining network structure with evidence about what flows through the network and what happens when they are interrupted.

Still, the paper opens a valuable route beyond the usual choice between maps of connections and lists of influential countries. By measuring how trade influence returns through a network—and by separating that feedback by length, product, region and time—EDT makes a hidden structural property measurable. The next step is to use that accounting to explain which changes matter for real economies, and when a more connected world is also a more mutually dependent one.