Diversification of global food trade partners increased inequalities in the exposure to shock risks

Abstract

Recent global food trade disruptions have evidenced how local shocks can cascade into global security threats. While the capacity of food systems to absorb spillovers depends heavily on its underlying trade networks, few studies quantify how their temporal evolution reshapes systemic vulnerability over time. Here, we evaluate how changes in global connectivity from 1986 to 2022 reshaped responses to production shocks. Using FAO data, we built yearly multiplex representations of the food trade system and quantified robustness through a stochastic shock-propagation model with dynamic export bans. We find that while increasing globalization intensified inter-dependencies and amplified cascades, robustness trends remain heterogeneous. Grain trade has become more decentralized and resilient to targeted shocks; conversely, Animal and Vegetable Fats exhibit growing centralization and fragility around key exporters like Indonesia and Malaysia. These structural transformations caused diverging shifts in systemic vulnerability, disproportionately threatening already vulnerable regions such as Africa and Southern Asia.

Figures (4)

• Figure 1: Schematic illustration of the food trade multiplex and the stochastic shock–propagation model. For each year, the food trade system is represented as a multiplex network with 12 layers, each corresponding to a food category, where nodes are countries and directed links are trade flows. Shock propagation is simulated independently within each layer. A production shortfall is introduced in one country (e.g., the Netherlands, NDL), which aims to compensate for it by reducing exports to a subset of its partners ($N_i$) selected through a stochastic mechanism weighted by partner GDP. In the example, reducing exports to Indonesia (IDN), Spain (ESP), and Italy (ITA) is sufficient for NDL to fully compensate its deficit at $t=1$, allowing the United States of America (USA) to maintain its regular supplies. At $t=2$, the countries affected by export reductions experience supply shortages, and thus, they will apply their own compensation mechanisms. Even after reducing all exports, both Italy and Indonesia will fail to cover their demand. Thus, we consider them to be "Disrupted" by the shock. Contrarily, applying its own export reductions allows Spain to cover most of the missing imports, bringing the shortage into a "tolerable" range and preventing the disruption of the country. Even if Spain is not considered disrupted by the shock, it had to implement export reductions, thus spillover effects will still propagate to Canada (CAN) at $t=3$. The shock propagates through the network until no more countries are disrupted by it.
• Figure 2: Temporal evolution of the food trade multiplex between 1986 and 2022. The left panels show the position of countries in the Multiplex participation coefficient ($P_i$) and weighted overlapping degree ($Z(o_i)$) plane, both for exports (upper) and imports (lower). These two panels show the position of all countries for the first and last year studied, 1986 and 2022 (pink and blue, respectively), and the complete developmental trajectories for two example countries, Brazil (BRA) and the United States of America (USA). The $Z(o_i)-P_i$ plane is divided into six sectors, labeled following the categories defined in Sec. \ref{['sec:plane']}: A1 "Localized nodes", A2 "Mixed nodes", A3 "Multi-product nodes", A4 nodes are considered "Localized hubs", A5 "Mixed hubs", A6 "Multi-product hubs", NO "In-existing countries in that year" (i.e., countries not yet been created /countries that disappeared). Meanwhile, the right panels present the flow of countries across the 6 sectors of the $Z(o_i)-P_i$ plane at four different years between 1986 and 2022. Note that this analysis has been performed using a multiplex where edges are weighted by the monetary value of the transaction (in 1000 USD).
• Figure 3: Temporal evolution of the largest cascades (top 5%) of each year between 1986 and 2022 for the Grain products and Animal and Vegetable Fats layers (upper and lower panels, respectively). Results are based on 20 stochastic simulations of single-country shocks initiated in each producer country for each year and food category. The left panels show the temporal evolution of the cascade size. Color indicates the probability of having a cascade with a specific size. Meanwhile, the right panels show which countries are triggering the top 5% cascades in each year. In this case, color indicates the average cascade size of the top cascades initiated by that country. For visualization purposes, the right panels have been limited to the top 19 countries (complete panels are available in Supplementary Fig. S5).
• Figure 4: Analysis of the temporal evolution of vulnerability ($\beta_{20}$) for all UN geographical sub-regions. Upper panels show the change in average $\beta_{20}$ in each UN subregion between the first (white) and last (colored) year of data available. For most cases, this corresponds to 1986 and 2022. When trade data was not available, we considered the first year with data: Central Asia (1992), Micronesia (1996), and Polynesia (1989) for Grains, Central Asia (1993) and Micronesia (1989) for Fats; and Central Asia (1992) for all food categories. Arrows are colored based on the direction of the vulnerability change, blue for an increase and pink for a decrease. The lower panels show the temporal evolution of the vulnerability ranking between 1986 and 2022. Gray lines reflect the individual trends of each country, while we have colored the average response in some sub-regions of interest. The darker gray area reflects the countries that are not disrupted by any spillover cascade, thus have $\beta_{20}=0$. In both analyses, geographical sub-regions follow the definitions from the United Nations (UN), see Supplementary Sec. 3.2 for more details.