Global food trade network grows more connected yet still hinges on fragile hubs

The world’s food trade network has become steadily more connected and more evenly balanced over the past two decades, but its vulnerability still hinges on a small number of core countries, according to a new study published in Agricultural and Food Economics. Fan Zhang, Zhuang Chen, Shengen He and Di Cheng analyzed the global food trade network for soybean, rice, wheat and maize from 2000 to 2023, and found that overall network vulnerability declined as connectivity rose, even though the concentration of exports among a few dominant suppliers has not fundamentally changed.

The study, published on 10 October 2026, introduces a vulnerability assessment framework that combines static structural analysis with dynamic shock simulation in a single unified design. The authors argue that most previous research has treated these two approaches separately: some studies describe the topology of trade networks using indicators such as density and centrality, while others simulate how networks degrade under disruptions such as trade restrictions, production shocks or node failures. By integrating both, the new framework aims not only to measure how a network performs when it is attacked, but also to identify the structural roots of that performance.

At the heart of the static assessment is the idea that connectivity level and connectivity distribution are the fundamental determinants of network vulnerability. Connectivity level is evaluated at three scales—overall, cluster and individual—through three dimensions the authors call relevance, agglomeration and accessibility. Relevance captures the breadth of connections across the whole network using average degree and network density; agglomeration measures how densely nodes cluster into groups using the average clustering coefficient and the relative size of the largest connected subgraph; and accessibility reflects the efficiency of trade pathways between individual countries, using network diameter and average path length. The three dimensions are combined multiplicatively, a choice the authors say makes the composite indicator sensitive to weak links, so that a pronounced shortcoming in one dimension cannot be masked by high values in the others.

The data come from the United Nations Comtrade Database, with bilateral trade observations for 2000, 2005, 2010, 2015, 2020 and 2023. Because importers levy tariffs and their records are generally more complete, the study uses data reported by importing countries, and it assesses vulnerability from the perspective of importing countries, since supply disruptions are generally regarded as riskier for importers than for exporters.

The structural evolution of the four commodity networks between 2000 and 2023 was striking. The number of nodes changed only modestly—the rice and soybean networks grew slightly, from 194 and 155 countries to 196 and 169, while the wheat and maize networks declined marginally—but the number of edges expanded dramatically. The rice network, the densest of the four, grew from 2,077 edges in 2000 to 3,087 in 2023, while the maize, wheat and soybean networks reached 2,091, 1,437 and 1,277 edges respectively by 2023. Average degree rose in every network: in the rice network it climbed from 21.4 to 31.5, meaning the average country gained roughly ten additional trade partners, and network density rose from 0.111 to 0.162. The maize and wheat networks increased from average degrees of 13.1 and 11.3 to 23.2 and 17.1, and the soybean network from 9.9 to 15.1, although soybean remained the sparsest of the four.

The networks also showed small-world characteristics. Average path length stayed broadly within the range of 2 to 3 across all four commodities, meaning that on average a food commodity needs only about three steps to move from one country to another. The average clustering coefficient of the rice network rose from 0.54 in 2000 to 0.599 in 2023, indicating that roughly 59.9 percent of neighboring nodes were mutually connected, and the other three networks showed similar upward trends in local cohesion.

Across the whole study period, the connectivity levels of the four networks consistently ranked, from highest to lowest, rice, maize, wheat and soybean, and all four showed an upward trend. The authors attribute the ranking primarily to differences in relevance—the breadth of overall connections—while agglomeration and accessibility played secondary roles. The rice network benefits from a supply and demand structure with stable demand, numerous participants and frequent interactions: rice is the staple food for more than half of the world’s population, and East and Southeast Asia have long served as core regions of global rice trade, with India, Thailand, Vietnam and Pakistan acting as key supply centers. Many Asian countries also maintain policies such as stock regulation and import quotas that reinforce the continuity of rice trade relationships. The soybean network, by contrast, is the most vulnerable because its trade is highly concentrated: Brazil, the United States and Paraguay together account for 91.3 percent of the global export market, while demand is concentrated in crushing and feed use, with China as the most important import market. Once a core exporting country is disrupted, importing countries may be unable to find alternatives. The wheat and maize networks occupy intermediate positions, with maize consistently performing better overall than wheat; the wheat network’s relatively low connectivity stems not from poor accessibility but from the limited breadth of its global trade expansion.

Connectivity distribution—the question of whether trade linkages and influence are spread evenly across countries—also moved in a more favorable direction. Using double-logarithmic rank-size curves of degree and eigenvector centrality, the authors found that the absolute values of the distribution slopes declined continuously from 2000 to 2023 in all four networks, indicating weaker hierarchy and a narrowing gap between core and peripheral countries. The soybean network remained the steepest and most hierarchical throughout, with its degree-distribution slope falling only from 1.18 to 1.06, while the rice network showed the flattest and most balanced structure, its eigenvector centrality slope dropping from 0.86 to 0.53. A robustness check using Gini coefficients confirmed these patterns, ranking the networks from most to least hierarchical as soybean, wheat, maize and rice, with all four showing declining inequality over time.

The dynamic half of the framework simulated random and intentional attacks on the 2023 networks, measuring how network efficiency changed as nodes were progressively removed. Under random attack, the decline was gradual: removing 20 percent of nodes reduced network efficiency by only 6 percent, and removing 50 percent reduced it by only 10 percent, with a pronounced decline appearing only after more than half the nodes were gone. The authors interpret this as evidence that the network contains multiple alternative paths, giving it a certain capacity to withstand random risks such as climate-related disturbances. Intentional attacks targeting core countries—removing nodes in descending order of degree centrality, betweenness centrality, closeness centrality and PageRank—produced far sharper declines. In the soybean network, once roughly 25 percent of nodes were disrupted, network efficiency fell below 0.1, and when 40 percent were attacked it approached zero. The wheat and maize networks showed stronger resistance but still tended toward failure when 30 to 40 percent of nodes were attacked, while the rice network declined most slowly, keeping its efficiency above 0.1 even when fewer than 50 percent of nodes were attacked. These dynamic results were broadly consistent with the static assessment, cross-validating the framework.

The authors caution that their analysis has limitations. The network is built from bilateral trade relationships between countries and does not capture finer-grained supply chain linkages such as production, processing and transportation, and the analysis considers only the decline in performance under shocks, not the recovery phase afterward. They also note that despite the overall improvement, the high concentration of exports and dependence on core countries and key routes have not fundamentally changed: Russia remains a major wheat exporter, the Black Sea region occupies a prominent position as a supply hub, and disruptions such as the Russia–Ukraine conflict, the Red Sea crisis and disturbances to the Suez Canal have raised the cost of global food transportation. Countries in the Global South that depend heavily on food imports while lacking sufficient capacity to withstand risk remain particularly exposed. Based on their findings, the authors suggest promoting diversified import sources, strengthening regional cooperation through trade agreements and reserve systems, and maintaining strategic redundancy in alternative trade partners and transport routes to reduce the network’s vulnerability to sudden shocks.

Subject of Research: Vulnerability of the global food trade network for soybean, rice, wheat and maize, assessed through combined static and dynamic network analysis

Article Title: Multi-scale connectivity and distribution: vulnerability in global food trade networks via static-dynamic integration

Article References: Zhang, F., Chen, Z., He, S., & Cheng, D. (2026). Multi-scale connectivity and distribution: vulnerability in global food trade networks via static-dynamic integration. Agricultural and Food Economics, 14(1), Article 81. https://doi.org/10.1186/s40100-026-00523-5

Image Credits: AI Generated

DOI: 10.1186/s40100-026-00523-5

Keywords: food trade network, network vulnerability, connectivity, soybean, rice, wheat, maize, risk shock simulation, global food security, trade hubs, complex networks, UN Comtrade