Supply-Chain / Network ABM Model
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?
Understand the mathematics
Follow the same equations and assumptions one section at a time or in a continuous reading.
Hover or focus a symbol defined in this model to find its exact matches. Select it to keep the highlight; Escape clears the selection.
Odd+D Protocol: Model Specification
The ODD+D (Overview, Design concepts, Details + Decision-making) protocol is the standard documentation format for agent-based models, introduced by Grimm et al. (2006, 2010) and extended by Muller et al. (2013) for decision-making. The protocol is applied here to the supply-chain / production-network ABM family.
PURPOSE: The model explores how a localized supply disruption propagates through a production network with Leontief-style complementarities, heterogeneous inventory buffers, and adaptive firm behavior. The target outputs are the supply-chain multiplier (ratio of aggregate output loss to direct damage), the propagation speed (how fast the cascade reaches a given fraction of firms), and the price pass-through dynamics. The target audience is disaster-risk agencies, central bank supply-side analysts, and policymakers evaluating supply-chain resilience measures (inventory mandates, dual-sourcing subsidies, strategic reserves).
ENTITIES, STATE VARIABLES, AND SCALES: The model contains three entity types plus a network structure. Firms ( to depending on resolution) carry state variables: production capacity (continuous, positive), output level (continuous, non-negative), inventory vector (one entry per input, continuous, non-negative), price (continuous, positive), supplier list (edges in the network), customer list (edges), and labor allocation (continuous). Final-demand agents ( aggregate block, or to household agents) carry: income, consumption demand, and savings. The policy block carries: procurement rules, tariff rates, and subsidy parameters. The network structure is a directed acyclic graph (or near-DAG with limited cycles) where each edge carries an input-output coefficient and a delivery lag. Temporal scale is weekly periods over 26 to 104 weeks; spatial scale ranges from a single-country IO table to a multi-country global trade network.
PROCESS OVERVIEW AND SCHEDULING: Each period proceeds in fixed order: (1) exogenous shocks arrive (capacity disruptions, demand shocks, labor constraints), (2) firms forecast demand and compute desired orders from inventory levels, (3) orders are transmitted upstream through the network, (4) the firms are then walked in topological order, upstream first, and at each node the firm produces from the deliveries it has already received THIS period and immediately allocates its output to its customers using the rationing rule, (5) inventories are updated (inflows minus usage), (6) prices adjust to unit cost and to the supply-demand gap at each node, (7) final demand is realized and aggregate statistics are recorded. The ordering is economically motivated on both counts. Orders must precede production because lead times mean firms order against anticipated, not realized, needs. And produce-then-ship must be interleaved node by node rather than run as two separate sweeps: a sweep that produces every firm first and allocates afterwards forces every firm to build with last period's deliveries, which inserts a one-period lag into every single supplier relationship whether or not the technology has one.
Firm 's output is the minimum of capacity, the binding input constraint (Leontief technology), and orders received - all scaled by a multiplicative labor factor, which is where labor enters throughout this file (it is a utilization factor on feasible output, not another argument inside the min). The min over inputs captures complementarity: excess of one input cannot substitute for shortage of another. The demand argument is what gives the model backward propagation: without it, a firm whose customers have all shut down would keep producing at capacity.
Inventory of input at firm evolves as last period's stock plus deliveries received minus inputs consumed in production, floored at zero. When deliveries fall short of consumption, inventory depletes - the buffer is being drawn down. The floor is not cosmetic: the production rule already caps usage at available stock, so a negative inventory would be a bookkeeping error that silently lends the firm inputs it never received.
Proportional rationing: when firm 's output is less than total orders, each customer receives a pro-rata share. This is the simplest rationing rule; priority-based and price-based variants redistribute the same total output differently.
Leontief vs. CES: a parameter, not a philosophy
The choice between Leontief () and CES () technology is not a binary modeling decision - it is a calibration question. Empirical estimates of short-run substitution elasticities between intermediate inputs cluster between 0.05 and 0.5 for manufacturing (Atalay 2017, Boehm et al. 2019). The Leontief limit is a reasonable approximation for weeks-to-months disruptions in manufacturing but overstates rigidity in services where inputs are more fungible.
Why topological order matters
In a directed acyclic production network, walking the firms in topological order (upstream before downstream) lets each firm produce from the deliveries it actually received this period rather than from stale values. But the ordering only buys you that if allocation is interleaved with production node by node: each firm ships to its customers as soon as it has produced, before those customers are processed. Two separate sweeps - produce everyone, then allocate everyone - reintroduce exactly the artifact the ordering was meant to remove, because every firm is then building from last period's deliveries and a phantom one-period lag is hardwired into every supplier relationship. The pseudocode on the Graph tab is written the interleaved way for this reason. Cycles in the network (mutual supply relationships) break the topological order and force a genuine lag on the back edges; report how many such edges the network has.
Rationing rules redistribute cascade impact
Under proportional rationing, a 50% output shortfall at a supplier means every customer gets half its order. Under priority rationing (e.g., largest customer first), small customers get zero while the largest customer is fully served. The aggregate output loss is the same, but the distribution of the cascade across downstream firms is completely different. Empirical evidence from the 2011 Thai flood (Haraguchi and Lall 2015) suggests that large customers receive priority treatment, concentrating disruption on smaller firms.
Transition checks
Agent update
The aggregate is computed after individual states are updated. If the rule is nonlinear or stochastic, the aggregate can change abruptly even when the plotted line connects adjacent periods.
Each point is an aggregate after one model update. The connecting line helps the eye, but the law is discrete.
Do not read the line as a continuous differential equation unless one is stated.
Cutoff and contagion
The indicator changes value discontinuously at . Aggregates inherit that discontinuity when enough agents sit near the cutoff.
A cutoff creates a kink or jump because the update rule changes at the threshold.
Do not force a smooth explanation onto a chart section driven by threshold crossing.
Variable timing
Agent state
Agent states update at dates. Abrupt aggregate movement can come from many small state changes applied at once.
Aggregate readout
The aggregate can jump when the distribution changes shape or many agents cross a boundary.
Threshold indicator
The indicator turns on at the cutoff. That creates a discontinuity in the micro rule and possibly the aggregate.