How does a localized supply disruption propagate through a production network to become an aggregate output loss, and what determines whether the shock is absorbed locally or amplified into a systemic bottleneck?
Background
Standard macroeconomic models treat aggregate supply as a single production function mapping capital and labor into output. That compression erases the network topology that determines how a disruption in one sector reaches other sectors. A semiconductor shortage does not reduce GDP uniformly -- it hits auto manufacturing first, then cascades to dealers, rental fleets, and insurance premiums along specific input-output linkages. Supply-chain ABMs exist because the propagation path matters as much as the initial shock. Battiston et al. (2007) formalized cascade dynamics on credit networks, establishing the analytical template for network-transmitted failures. Hallegatte (2008) built the first explicit supply-chain disruption ABM for natural disaster assessment, modeling firm-level production rationing and inventory drawdown in a Leontief-style network. Henriet et al. (2012) extended this to a full production-network ABM with heterogeneous firms and adaptive inventory rules. Inoue and Todo (2019) calibrated a supply-chain ABM to the entire Japanese inter-firm transaction network (roughly one million firms) and simulated the propagation of the 2011 Tohoku earthquake. Pichler et al. (2022) adapted the framework to COVID-era supply disruptions with sector-specific lockdown shocks and labor constraints.
The core mechanism is a cascade through complementarities. A firm that cannot obtain a critical intermediate input cannot produce, even if all other inputs are available. This Leontief-style complementarity means that disruptions propagate forward (downstream customers lose supply) and backward (upstream suppliers lose demand). The propagation speed and reach depend on three factors: network topology (how concentrated are supplier relationships), inventory buffers (how many periods of stock does a firm hold), and substitution elasticity (can a firm switch to an alternative supplier when the primary one fails). When inventories are thin and substitution is difficult, a single-node disruption can cascade through the network and produce an aggregate output loss far larger than the direct damage at the initial node. This amplification is the supply-chain multiplier, and it is invisible to any model that does not track the network.
Central banks, finance ministries, and disaster-risk agencies use supply-chain ABMs operationally. The Bank of Japan funded the Inoue-Todo model to stress-test the Japanese economy against earthquake, typhoon, and pandemic scenarios. The European Central Bank's supply-chain task force used Pichler et al. (2022) to estimate the output cost of COVID lockdowns by sector. The World Bank and UNDRR use Hallegatte-type models in disaster damage assessments for developing countries, where the indirect (network-propagated) losses often exceed the direct physical damage by a factor of 2 to 5. The OECD employed production-network analysis to assess the output effects of the 2021-2022 semiconductor shortage. Academic use spans from pure network-cascade theory (Acemoglu et al. 2012, Baqaee-Farhi 2019) to applied policy evaluation for supply-chain resilience.
The model family has evolved along several axes. Early versions used fixed Leontief technology with no substitution; newer variants allow CES production functions where firms substitute between inputs at different elasticities. Inventory management has moved from passive buffer stocks to active (S, s) policies with stochastic lead times. Network topology has shifted from stylized random graphs to empirically calibrated firm-level transaction networks from VAT records or supply-chain databases (e.g., FactSet Revere, S&P Capital IQ). The active frontier includes multi-country supply-chain ABMs for trade-disruption analysis, climate-adaptation overlays (flood and heat stress on logistics nodes), and dual-sourcing strategies as a resilience mechanism.
How the Parts Fit Together
The economy is assembled from three agent populations embedded in a directed production network. Firms are the primary agents: each occupies a node in the network and carries a production function, an inventory stock for each intermediate input, a customer list, a supplier list, and a pricing rule. The production function is typically Leontief in intermediates (fixed proportions) with an optional CES substitution layer for alternative suppliers of the same input. Each firm draws labor from a shared labor market and produces a single output that it ships to downstream customers according to a rationing rule when supply is scarce. The firm population ranges from 100 to 1,000 in stylized models, 10,000 to 50,000 in calibrated sector-level models, and up to 1,000,000 in full-resolution firm-level implementations (Inoue-Todo 2019). The second population is a set of final-demand agents (households or a consolidated demand block) that purchase finished goods and supply labor. The third agent is a policy block: government procurement, import/export rules, and optionally a central bank setting interest rates that feed into inventory financing costs.
Interaction happens through the directed supply network. Each edge in the network represents a supplier-customer relationship with a specified input share. When a firm produces, it draws intermediates from upstream suppliers and ships output to downstream customers. If a firm cannot produce (because it lacks a critical input or its labor is constrained), its output drops, inventories at downstream customers deplete, and the disruption propagates forward. Simultaneously, the firm reduces its orders to upstream suppliers, and the demand shortfall propagates backward. Prices adjust through a markup rule: firms facing excess demand raise prices, and the price increase passes through the network as a cost-push wave. The network topology -- degree distribution, clustering, centrality of key nodes -- determines whether disruptions are contained in a local cluster or spread systemically.
State variables update each period in a fixed sequence: (1) demand orders arrive from downstream customers, (2) each firm checks inventory, places orders to upstream suppliers, (3) suppliers allocate output using a rationing rule (proportional, priority-based, or price-based) when demand exceeds supply, (4) firms produce using available inputs and labor, (5) output is shipped and inventories are updated, (6) prices adjust based on the supply-demand gap at each node, (7) final demand agents consume and the aggregate output statistic is recorded. This sequential structure means the propagation from a disruption to its network-wide consequence is traceable edge by edge rather than collapsed into an aggregate production function.
Applications
The Bank of Japan commissioned the Inoue-Todo supply-chain ABM to quantify the propagation of the 2011 Tohoku earthquake through the Japanese inter-firm transaction network. Using roughly one million firm nodes from the Tokyo Shoko Research database, the model estimated that the direct damage (factory destruction in Tohoku) produced a GDP loss of approximately 0.4%, but the network-propagated indirect loss was 1.2% of GDP -- a supply-chain multiplier of roughly 3x. The model identified Toyota's tier-1 supplier Renesas Electronics as the single most critical bottleneck: its semiconductor plant closure cascaded to all major automakers within two weeks. This analysis directly informed the Japanese government's supply-chain resilience strategy, including subsidies for dual-sourcing and strategic inventory buffers in critical sectors.
Pichler, Pangallo, del Rio-Chanona, Lafond, and Farmer (2022) at the Oxford Institute for New Economic Thinking adapted the production-network ABM to COVID-era supply disruptions. The model took sector-specific lockdown shocks (fraction of labor force unable to work) as inputs and simulated the propagation through the UK and US input-output networks. Key findings: sectors with high upstream centrality (chemicals, basic metals) amplified lockdown shocks beyond their direct employment weight, and the aggregate output loss was 2-3x the employment-weighted direct effect due to complementarities. The ECB's supply-chain task force cited this work in its 2021 assessment of euro-area supply disruptions.
The model breaks down in three settings. First, when disruptions are slow-moving and firms have time to rewire their supplier networks (months, not weeks), the fixed-topology assumption overstates propagation. A tariff phased in over two years gives firms time to relocate sourcing; the ABM's disruption logic is designed for sudden shocks. Second, when the production technology has high substitution elasticity (services, digital goods where inputs are fungible), the Leontief complementarity assumption is too strong and the model overpredicts cascades. Third, when the dominant channel is financial rather than physical (a firm fails because of cash-flow insolvency, not input shortage), the baseline model without financial frictions misses the primary propagation mechanism. For financial contagion, a bank-firm credit-network ABM is more appropriate.
Literature and Extensions
Key Papers
- Battiston et al. (2007) -- credit-network cascade dynamics establishing the analytical template for network-transmitted failures
- Hallegatte (2008) -- first supply-chain disruption ABM for natural disaster assessment with firm-level rationing
- Henriet et al. (2012) -- production-network ABM with heterogeneous firms, inventory rules, and substitution
- Inoue and Todo (2019) -- million-firm Japanese supply-chain ABM calibrated to actual inter-firm transaction data
- Pichler et al. (2022) -- COVID supply disruption ABM with sector-specific lockdown shocks and labor constraints
- Acemoglu et al. (2012) -- theoretical foundation: network origins of aggregate fluctuations from idiosyncratic shocks
Named Variants
- Multi-country supply-chain ABM (trade disruption, tariff simulation)
- Financial-physical hybrid ABM (Battiston credit cascades + Hallegatte physical cascades)
- Climate-overlay supply-chain ABM (flood/heat stress on logistics nodes and transport links)
- Endogenous network rewiring (firms search for alternative suppliers post-disruption)
- Bullwhip-effect ABM (demand-signal amplification through inventory ordering)
Open Questions
- What is the empirically relevant substitution elasticity between suppliers in manufacturing supply chains, and how does it vary by sector?
- Can supply-chain ABMs be calibrated to real-time shipping and logistics data (AIS vessel tracking, freight indices) for near-term disruption forecasting?
- How should dual-sourcing subsidies be sized relative to the network topology to minimize the expected supply-chain multiplier?
Components
A production unit with state vector (output capacity, inventory vector, supplier list, customer list, production function, price, labor allocation). Decisions: produce, order inputs, ration output, adjust price.
The quantity of input from firm j required to produce one unit of output at firm i. Defines the directed edge weight in the production network.
Firm i's stock of input j at time t. The first line of defense against upstream disruptions. Depletion triggers order amplification (bullwhip effect) or production curtailment.
CES elasticity between alternative suppliers of the same input class. At sigma = 0, production is Leontief (no substitution); as sigma rises, firms can reroute sourcing when a supplier fails.
The allocation mechanism firm i uses when its output is insufficient to fill all customer orders. Common rules: proportional (pro-rata), priority-based (largest customer first), price-based (highest bidder first).
Price of firm i's output at time t. Adjusts upward under excess demand, propagating cost-push inflation through the network. The aggregate price level emerges from individual price adjustments.
Total final-goods production across all firms serving final demand. The key emergent variable -- its deviation from the no-shock baseline measures the supply-chain multiplier.
Firm i's position in the network topology, typically measured by Katz-Bonacich centrality or the Leontief inverse column sum. High-centrality firms are potential systemic bottlenecks.
Assumptions
Each firm's production function requires fixed proportions of intermediate inputs. Output is limited by the scarcest input: x_i = min_j(z_{ij} / A_{ij}), where z_{ij} is available input j and A_{ij} is the required coefficient.
If violated: With high substitution elasticity (CES with sigma >> 1), firms reroute around disrupted suppliers and the cascade weakens dramatically. The amplification result depends on complementarity being strong relative to the time horizon of the shock.
The supplier-customer network does not rewire during the simulation horizon. Firms cannot establish new supplier relationships within the disruption window.
If violated: If firms can instantly rewire to alternative suppliers, the network-propagation mechanism is short-circuited. Endogenous rewiring extensions exist (Inoue-Todo 2019) but operate on a slower timescale than the disruption itself.
Each firm maintains inventory according to a target-stock rule: order when stock drops below s, order up to S. Inventory policy parameters are calibrated from industry data, not optimized in real time.
If violated: Optimal real-time inventory management with full network information would dramatically reduce cascade propagation, but no firm has that information set in practice.
When a firm's output is insufficient to fill all orders, it rations output across customers using a pre-specified rule rather than clearing through a Walrasian price mechanism.
If violated: Walrasian price clearing would require instantaneous price adjustment to equate supply and demand at every node, which contradicts the observed stickiness of intermediate-goods prices and the prevalence of quantity rationing in supply shortages.
Each firm produces one output. Multi-product firms are represented as multiple single-product nodes in the network.
If violated: Multi-product firms can cross-subsidize and shift production mix in response to input shortages, a margin of adjustment that single-product nodes cannot replicate.
Firms do not face credit constraints or bankruptcy risk from disruption-induced revenue losses. The disruption propagates through the physical input-output channel, not through balance sheets.
If violated: Adding financial frictions (Battiston et al. 2007) creates a second propagation channel: a firm that loses revenue cannot pay suppliers, triggering a backward financial cascade on top of the forward physical cascade. The combined model is substantially more complex and harder to calibrate.
Unlock Full Macro Model Library with Starter.
This feature is exclusively available to Starter, Research, and Pro. Upgrade when you need this workflow, review pricing, or send a question before changing plans.
What you keep on Free
- Create and edit one custom board
- Use up to 3 widgets on each Free board
- Browse indicators and calendar