From Local Shocks to Global Cascades: An End-to-End Framework for Simulating Complex Contagion in Financial Networks
This study proposes a hybrid static-dynamic framework that integrates community detection with a non-linear, multi-path risk coupling model to simulate and quantify the propagation of complex contagion, including AI-associated risks, from local shocks to global cascades within financial networks.
Original paper licensed under CC BY 4.0 (https://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
In the modern financial world, companies do not operate in isolation. They are woven together by a dense web of relationships: a parent company owns a subsidiary, a supplier relies on a manufacturer, and a bank lends to a corporation. These connections allow capital to flow efficiently, but they also create pathways for trouble to spread. When one part of this system stumbles, the shock can travel along these links, potentially toppling neighbors and triggering a chain reaction that threatens the entire market. This phenomenon is known as contagion. For decades, experts have tried to map these networks to predict where a crisis might start and how far it could travel. However, traditional maps are often static snapshots; they show the connections as they exist at a single moment but fail to capture how a shock actually moves, accelerates, or fades over time. Furthermore, the rise of artificial intelligence has introduced new, unpredictable risks. Algorithms that trade or manage assets can sometimes make confident but factually incorrect decisions, creating sudden, localized glitches that could ripple outward in ways old models cannot see.
A team of researchers from Nanjing Audit University and the Jinling Institute of Technology has developed a new way to study these risks, moving beyond static maps to simulate the actual journey of a financial shock. Their work focuses on a network of 2,052 listed companies in China, treating each firm as a node in a vast, interconnected system. The researchers first built a detailed model of how these companies relate to one another, accounting for ten different types of ties, from ownership and control to supply chain dependencies. To understand the strength of each connection, they used a large language model to assign a score to every relationship, determining how likely a problem at one company would impact another. For instance, a parent company's failure is far more likely to drag down its subsidiary than a supplier's failure would impact a distant client. This created a weighted map where every link had a specific intensity, reflecting the real-world logic of risk transmission.
The study then introduced a dynamic simulation to watch how a risk event would unfold. Unlike older methods that assume a shock spreads in a simple, straight line, this new framework accounts for "complex contagion." This concept recognizes that a company is more likely to fail if it is hit by trouble from multiple directions at once. If a firm receives bad news from two different suppliers simultaneously, the combined pressure is far greater than the sum of the two individual shocks. The researchers designed an algorithm that tracks this multi-path influence, calculating how risk accumulates as it travels through the network. They also built in a mechanism for recovery, acknowledging that companies can sometimes repair themselves or be shielded by regulators, effectively becoming immune to further infection after a crisis passes.
To test their framework, the researchers simulated a shock starting at a single company and watched how it spread through the network over several stages. They found that the behavior of the system was far more complex than simple maps suggested. Some companies that appeared to be the most dangerous based on their number of connections turned out to be limited in how far they could spread trouble. For example, one major company, identified as V1145, had hundreds of direct connections and infected hundreds of others immediately. However, because those newly infected companies had few connections of their own, the shock died out quickly, contained within a small local group. Conversely, another company, V53, did not appear to be a major hub in the static data. Yet, when the simulation ran, this seemingly ordinary node triggered a massive cascade. Its initial shock rippled through the network, infecting a chain of other firms that then went on to infect even more, eventually affecting over 300 companies and causing 138 to fail.
The research highlights a critical blind spot in how financial regulators currently view risk. Relying solely on static data, which counts connections and measures influence at a single point in time, can lead to a false sense of security. It might flag the wrong companies as the biggest threats while missing the hidden "super-spreaders" that only reveal their danger when the system is in motion. The study also identified specific clusters of companies that act as isolated islands or high-risk bridges. By using a community detection method, the researchers were able to group the network into 13 distinct communities. They found that while some communities were tightly knit and isolated, others acted as dangerous hubs where risk could easily spill over into the rest of the market.
Ultimately, this work provides a new toolkit for understanding the hidden dynamics of financial stability. It demonstrates that the path a risk takes is just as important as the starting point. A shock that begins small can grow into a systemic crisis if it finds the right sequence of connections, while a massive shock might fizzle out if it hits a dead end. By simulating these journeys, regulators can move beyond guessing and start targeting their monitoring efforts more precisely. They can identify which specific links to watch, which communities to isolate, and which seemingly minor nodes might actually hold the key to preventing a global cascade. In an era where artificial intelligence adds new layers of unpredictability to the market, this ability to trace the full life cycle of a risk event—from a local glitch to a global crisis—offers a vital step toward keeping the financial system stable.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.