Inferring firm-level supply chain networks with realistic systemic risk from industry sector-level data

Production networks form the structural backbone of modern economies, yet recent crises have exposed their intrinsic vulnerability. Quantifying the system-wide impact of local disruptions–i.e., systemic risk–requires detailed knowledge of supply chain networks, since such risks depend sensitively on their underlying topological patterns. In practice, however, microscopic network data are rarely available, forcing systemic-risk estimates to rely on coarse sector-level information, such as input/output tables, and aggregate production data at the firm-level. Here we assess the ability of maximum-entropy approaches to reconstruct supply chain networks with realistic level of systemic risk when only sector- and firm-level information is available. Leveraging the unique availability of the complete production network of Ecuador, we benchmark four reconstruction models with different formulations and data requirements, by comparing the inferred structural and systemic-risk properties (particularly, the Economic Systemic Risk Index) with those of the real network. We find that the stripe-corrected gravity model, which incorporates firm-specific input requirements disaggregated by sector, most accurately reproduces the systemic-risk content of the empirical network, underscoring the importance of capturing heterogeneity in firms’ input profiles. Inferring the latter using sectorial input/output tables–through the input-output gravity model we introduce here–still yields satisfactory estimates while requiring substantially less information. Our results identify the minimal sector-level information needed to statistically generate synthetic production networks that faithfully encode firm-level systemic risk.