📈 economics

Efficient liability assignment under shock propagation

This paper characterizes a family of liability rules that achieve efficient path selection in shock propagation networks by assigning agent payments proportional to total losses using weights derived from the Shapley value of a path-counting cooperative game, which can be computed in polynomial time.

Original authors: Jens Gudmundsson, Jens Leth Hougaard, Kohmei Makihara, Alexandros Rigos

Published 2026-05-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jens Gudmundsson, Jens Leth Hougaard, Kohmei Makihara, Alexandros Rigos

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a supply chain as a giant, interconnected web of friends passing a heavy backpack down a line. Each person in the line represents a company or a worker, and the backpack represents a "shock" or a problem (like a broken machine or a missing shipment).

When the person at the very start (the "Source") gets hit by a shock, they have to drop the backpack. But here's the catch: they can't just drop it on the ground. They have to hand it to one of their neighbors. That neighbor then has to drop it and hand it to their neighbor, and so on, until the backpack finally reaches the end of the line (the "Sink").

Every time the backpack is passed along a specific route, it causes damage. Maybe the first hand-off breaks a vase, the second cracks a window, and the third spills coffee. The total damage is the sum of all these broken things.

The Problem: Who Pays?
In the real world, when things go wrong, people argue about who should pay.

The "Local" Way (The Status Quo): Usually, the person who dropped the backpack pays for the damage they caused. If I drop it and break a vase, I pay for the vase. If my neighbor drops it next and breaks a window, they pay for the window.
The Paper's Insight: The authors say this "Local" way is actually dangerous. It creates a trap. Imagine a path where I drop the backpack, causing a tiny scratch (low cost), but my neighbor then has to drop it down a path that shatters a million-dollar statue (huge cost). If I only pay for my scratch, I might choose that path because it's "cheap" for me, even though it destroys the company's total wealth. I don't care about the million-dollar statue because it's not my bill.

The Solution: The "Fixed-Weight" Rule
The paper proposes a new way to split the bill, called a Fixed-Weight Rule.

Think of it like a group insurance policy that everyone signed before the accident happened.

The Total Bill: First, you add up all the damage caused by the entire chain of events, from start to finish.
The Split: Instead of paying for what you personally broke, everyone pays a tiny, pre-determined slice of the total bill.
The "Weight": Who pays what slice? It depends on how important you are to the network, not on what you did in that specific moment.
- If you are a "hub" (a person who connects to many others), you pay a slightly larger slice.
- If you are on the edge, you pay a smaller slice.
- Crucially, everyone pays a little bit, even if they weren't directly involved in the specific path the backpack took.

Why is this better?
The authors prove mathematically that this method forces everyone to make the "smart" choice.

Because I know I will pay a small percentage of the total damage, I will naturally try to choose the path that causes the least total damage. I won't pick the path with the million-dollar statue just to save myself a few dollars, because I'll still have to pay a share of that million dollars.
It stops people from "colluding" (cheating together). If two friends try to game the system to lower their specific bills, the math shows they can't. The only way to lower the total bill for everyone is to pick the most efficient path.

How do we decide the "Slices"?
The paper suggests a clever way to calculate these slices using a concept from game theory called the Shapley Value.

Imagine counting every possible route the backpack could have taken.
If a person appears on many of those possible routes, they get a higher "weight" (a bigger slice of the bill).
If they are rarely part of a route, their slice is tiny.
The cool part is that the authors found a fast, computer-friendly way to calculate these slices, even for huge networks with thousands of people.

The Simulation
The authors ran a computer simulation of a supply chain with 90 people.

Under the old "Local" rule: People paid wildly different amounts. Some paid nothing, others paid huge sums. The total damage to the system was high because people kept picking the "cheap for me, expensive for everyone" paths.
Under the new "Fixed-Weight" rule: Everyone paid a small, predictable amount. The total damage to the system dropped significantly (by about 50% in their simulation). The risk was spread out like a safety net, so no single person got crushed by a massive bill.

In a Nutshell
The paper argues that when a shock ripples through a network, we shouldn't just blame the person who dropped the ball. Instead, we should treat the total damage as a shared responsibility. By assigning everyone a small, fixed share of the total cost based on their role in the network, we align everyone's incentives to stop the damage before it gets out of control. It turns a game of "who can dodge the bill" into a game of "how can we all minimize the total mess."

Technical Summary: Efficient Liability Assignment under Shock Propagation

1. Problem Statement

The paper addresses the challenge of aligning individual incentives with system-wide efficiency in networked environments subject to shock propagation. Specifically, it models a scenario where an exogenous shock hits a source node in a directed acyclic network (e.g., a supply chain), forcing a cascade of cancellations. When an agent is affected by a cancellation, they must cancel one of their own outgoing edges. This process continues sequentially until a terminal sink node is reached, forming a specific path through the network.

Each edge in the network carries an associated economic loss. The total systemic cost is the sum of losses along the realized path. The core problem is to design a liability rule—a mechanism for assigning the total realized loss among the agents—that induces agents to select the efficient path (the path with the minimum total loss) in a decentralized, non-cooperative setting. The authors aim to avoid rules that rely on unverifiable counterfactuals or that lead to unbounded inefficiency, such as "local liability" rules where agents are only responsible for the direct loss they cause to their immediate successor.

2. Methodology

The authors employ an axiomatic approach combined with game-theoretic analysis.

Model Primitives

Network: A fixed, directed, acyclic graph $(N, E)$ with a unique source $s$ and at least one sink.
Shock and Cancellation: A shock at $s$ initiates a sequential game. At each step, the current agent chooses which outgoing edge to cancel, determining the next agent in the chain.
Loss Function: A function $\ell: E \to \mathbb{R}_{\geq 0}$ assigns a non-negative loss to each edge. The total loss of a path $P$ is $\ell(P) = \sum_{e \in P} \ell(e)$ .
Liability Rule: A function $\phi: \mathcal{P} \times \mathcal{L} \to \mathbb{R}^N_{\geq 0}$ assigns a liability $\phi_i(P, \ell)$ to each agent $i$ given the canceled path $P$ and loss function $\ell$ , such that $\sum_i \phi_i(P, \ell) = \ell(P)$ .
Game Formulation: The interaction is modeled as an extensive-form game of perfect information. Agents choose strategies to minimize their own liability. The solution concept is the subgame-perfect equilibrium (SPE).

Axiomatic Framework

The authors characterize liability rules based on four key axioms:

Efficient Implementation: The set of SPE outcomes must coincide exactly with the set of efficient (minimum loss) paths for any loss function.
Realized-Loss Dependence: Liabilities must depend only on the losses of edges actually on the canceled path. Counterfactual losses (off-path) cannot influence liability assignments, ensuring enforceability.
Pairwise Collusion-Proofness: No pair of agents should be able to jointly deviate from an equilibrium path to a new path that makes both strictly better off (preventing side payments that undermine efficiency).
Scale Invariance: Scaling all losses by a positive factor $\alpha$ scales all liabilities by the same factor.

3. Key Contributions and Results

Characterization of Fixed-Weight Rules

The primary theoretical result (Theorem 1) establishes that the four axioms jointly characterize a unique family of rules known as fixed-weight rules.

Definition: A rule $\phi_w$ assigns to each agent $i$ a liability equal to $w_i \cdot \ell(P)$ , where $w$ is a pre-determined weight vector ( $\sum w_i = 1, w_i \geq 0$ ) and $\ell(P)$ is the total loss of the canceled path.
Implication: Liabilities are independent of the specific distribution of losses along the path; they depend solely on the total path loss and the agent's fixed weight.
Necessity of Positive Weights: To ensure efficient implementation, any agent with multiple outgoing edges (who can make a strategic choice) must be assigned a strictly positive weight. If an agent has zero weight, they are indifferent to path choices, potentially allowing inefficient equilibria.

Interpretation of the Results

The paper argues that fixed-weight rules treat losses as a common, systemic responsibility. Unlike "local liability" rules (where an agent pays only for the direct damage to their successor), fixed-weight rules require agents to internalize the full downstream consequences of their decisions, even if they are not on the realized path. This "broad solidarity" extends liability to off-path agents, which is necessary to prevent unbounded inefficiency (as illustrated by the counterexample in Figure 1).

Selection of Weights via Cooperative Game Theory

To select a specific member of the fixed-weight family, the authors propose a weight vector $w^*$ derived from the network structure:

Path-Counting Procedure: Weights are assigned based on the frequency and importance of an agent's presence across all possible source-sink paths. Specifically, each path contributes $1/|P|$ to the total weight, which is then divided equally among the non-sink agents on that path.
Shapley Value Connection (Theorem 2): These weights $w^*$ coincide with the Shapley value of an associated cooperative "path-counting" game, where the worth of a coalition is the fraction of paths entirely contained within that coalition.
Computational Tractability: While calculating Shapley values is generally #P-complete, the authors demonstrate that for this specific network structure, the weights can be computed in polynomial time using a dynamic programming approach (forward and backward passes to count subpaths of specific lengths).

Simulation Study

A simulation on a tiered supply-chain network (with an "hourglass" structure) compares the proposed fixed-weight rule ( $\phi^*$ ) against a "local-liability" benchmark ( $\hat{\phi}$ ).

Efficiency: The fixed-weight rule induces significantly lower total losses (approx. 50% reduction in the simulation) by encouraging shorter paths.
Risk Distribution: The fixed-weight rule results in lower variance in individual liabilities (lower squared liabilities) and a more equal distribution (lower Gini coefficient). It acts as an insurance mechanism where many agents bear small shares rather than a few bearing the full burden.
Agent Welfare: In the simulation, 74 out of 90 non-sink agents are better off under the fixed-weight rule compared to the local-liability benchmark.

4. Significance and Claims

The paper claims that its approach offers a tractable and transparent method for allocating responsibility in cascading network failures.

Robustness: The fixed-weight rules are robust to various generalizations, including incomplete information (where agents only know their own edge losses) and generalized cancellation games (where agents can cancel multiple edges or partial capacities). In these extended settings, the rules continue to ensure efficient implementation.
Practicality: The reliance on realized losses and the polynomial-time computability of the weights make the rule suitable for real-world implementation, potentially via smart contracts.
Theoretical Insight: The work highlights that efficient implementation in decentralized networks often requires a form of "global solidarity" where agents bear liability for losses they did not directly cause, challenging the intuition that liability should be strictly local to the decision-maker's immediate impact.

The authors do not claim to solve all network disruption problems but rather provide a specific axiomatic solution for the problem of shock propagation along paths in directed acyclic networks, emphasizing the necessity of systemic liability assignment over local attribution.