💰 quantitative finance

An Operator-Based Visual Analytics Pipeline for Synthetic Systemic Risk Dynamics

This paper presents a modular, operator-based visual analytics pipeline that transforms synthetic financial data into dynamic visualizations of systemic risk dynamics, treating visualization as an integral analytical stage rather than a post-processing step to enhance the exploration and communication of complex risk processes.

Original authors: Ana Isabel Castillo Pereda

Published 2026-09-22✓ Author reviewed ⓘ

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Ana Isabel Castillo Pereda

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 by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Modern financial systems are vast, intricate webs where banks, investors, and markets are constantly connected. When one part of this web stumbles, the shock can ripple outward, potentially causing a chain reaction that threatens the entire economy. This phenomenon, known as systemic risk, is difficult to study because it involves countless moving parts that change over time. To understand how a crisis might spread, researchers often build computer models that simulate these connections. However, a common problem with these models is that the math happens in one place, the network analysis in another, and the final picture is often just a static snapshot drawn at the very end. This separation can make it hard to see how a small change in risk might evolve into a large-scale collapse as time passes.

A new approach developed by researchers at the University of São Paulo seeks to fix this disconnect by treating the entire process as a single, continuous flow. Instead of building a model and then trying to visualize the results later, the team created a system where the math, the network structure, and the visual representation are all linked together in a seamless chain. Imagine a factory assembly line where raw materials enter one end and a finished product comes out the other; in this case, the "raw materials" are simple computer-generated numbers, and the "finished product" is a moving, dynamic picture of how financial distress spreads. The researchers did not invent a new way to predict crashes or a new theory of how banks fail. Instead, they built a transparent, modular framework that lets scientists watch the entire process unfold in real time, from the initial spark of risk to the final visual display.

The experiment began by feeding the computer a large set of synthetic data—520 made-up observations that represented different financial conditions. These inputs were not random noise; they were transformed through a series of steps designed to mimic how real-world risk might behave. First, the computer converted these raw numbers into a hidden, or "latent," risk signal. Think of this as taking a complex weather pattern and distilling it into a single number that represents the likelihood of a storm. Next, the latent signal was mapped into a continuous synthetic probabilistic risk score between zero and one. These scores are components of the controlled synthetic experiment and should not be interpreted as calibrated probabilities of real-world financial distress.

Once the risk scores were established, the researchers built a digital financial network. This network consisted of 24 nodes, representing different financial institutions, connected by 31 weighted links that showed how they were related. Each node was given a specific vulnerability score, representing how easily it could be knocked over, while the connections between them determined how easily trouble could travel from one to another. The researchers then introduced a controlled shock at node 5, which serves as the prescribed initial shock source in the reference experiment. This shock did not jump randomly; it spread outward along the shortest paths through the network, much like a ripple moving across a pond, but with a key difference: the intensity of the shock faded the further it traveled from the source.

The most significant innovation in this work is how the researchers handled the visualization. In traditional studies, the visual part is often an afterthought, added only after the math is done. Here, the visual representation is built into the pipeline itself. As the shock moved through the network, the computer simultaneously updated a dynamic display. Nodes changed color to show their rising distress, grew larger to indicate their importance, and shifted position to reflect their changing state. The result was a living dashboard that showed four different views at once: the underlying risk landscape, the changing probability of a crisis, the structure of the network, and the spread of the contagion. This allowed the researchers to see how the network's structure influenced the speed and reach of the shock, independent of the initial risk scores.

The simulations revealed a clear distinction between two types of risk. The overall probability of a crisis, calculated from the initial data, remained relatively stable over time. However, once the shock hit the network, the network-mediated contagion responded strongly to the shock and expanded progressively through the network. The number of institutions affected increased in distinct steps as the shock reached new layers of the network, showing that the structure of the connections mattered just as much as the initial risk levels. The experiment also illustrates that assigned vulnerability, network connectivity, and contagion intensity are distinct quantities and need not rank nodes in the same way.

This work serves as a proof of concept, demonstrating that complex financial dynamics can be studied more effectively when the mathematical modeling and the visual representation are treated as parts of the same process. The researchers used a specific set of parameters for their simulation, including a decay factor that controlled how quickly the shock faded with distance, but the value of the work lies in the architecture itself. Because the system is built from interchangeable parts, other researchers can swap out the risk models, the network structures, or the contagion rules without having to rebuild the entire system. This modularity means that the framework can be adapted to use real-world data in the future, or to test more sophisticated theories of how financial crises spread.

The study does not claim to have solved the problem of predicting financial crises, nor does it offer a new method for forecasting market behavior. Its contribution is structural: it provides a clear, reproducible way to connect the dots between abstract risk models and the visual tools used to understand them. By making the flow of information explicit, the researchers have created a tool that allows scientists to explore how different components of a financial system interact. The ability to watch a shock propagate through a network in a coordinated visual display offers a new way to understand the hidden mechanics of systemic risk, turning a static calculation into a dynamic story that can be seen and understood all at once.

Technical Summary: An Operator-Based Visual Analytics Pipeline for Synthetic Systemic Risk Dynamics

Problem Statement

Modern financial systems involve complex, high-dimensional interactions among institutions and assets. While quantitative models for systemic risk and network contagion (e.g., Eisenberg–Noe, Gai–Kapadia) are well-established, the exploration and communication of their results often rely on static plots and isolated diagrams. These representations frequently treat risk modeling, network analysis, contagion simulation, and visualization as separate, fragmented stages. Consequently, the temporal and interdependent nature of processes like evolving risk scores and shock propagation is often obscured, and visualization is typically relegated to a post-processing step rather than being integrated into the analytical workflow.

Methodology

The paper proposes a unified operator-based visual analytics pipeline that treats the transformation of synthetic financial data into dynamic scientific visualizations as a sequential composition of mathematical operators. The framework is defined by the function:

$F = \Pi \circ \Psi \circ C \circ G \circ P \circ D$

This pipeline consists of six distinct operators:

Latent Risk Mapping ( $D$ ): Constructs a nonlinear latent risk representation from synthetic multidimensional observations ( $X_i$ ). A specific nonlinear surface is defined using trigonometric and polynomial interactions of input features to generate heterogeneous risk regions.
Probabilistic Score Generation ( $P$ ): Transforms the latent representation into probabilistic risk scores ( $p_i$ ) using a logistic (sigmoid) mapping. Time-dependent perturbations are added to simulate evolving risk states, generating dynamic scores $p_i(t)$ .
Financial Network Construction ( $G$ ): Embeds the probabilistic scores into a weighted financial network graph $G=(V, E, W)$ . Nodes represent financial entities with assigned sector labels and vulnerability scores. Edges are generated probabilistically, favoring connections within the same sector.
Distance-Based Contagion Dynamics ( $C$ ): Models the propagation of distress from an initial shock source ( $v_0$ ) using shortest-path distances ( $d_i$ ) on the graph. The contagion intensity $C_i(t)$ decays exponentially with distance, modulated by a time-dependent shock intensity function $S(t)$ .
Visual Encoding ( $\Psi$ ): Explicitly formalizes visualization as a mapping operator that transforms quantitative systemic states (topology, vulnerabilities, scores, contagion intensities) into visual attributes: node positions ( $r$ ), sizes ( $s$ ), opacity ( $\alpha$ ), and color ( $\chi$ ).
Perspective Projection ( $\Pi$ ): Maps the three-dimensional visual coordinates generated by $\Psi$ onto a two-dimensional visualization domain using a depth-dependent scaling factor to create a perspective effect.

The implementation relies on controlled synthetic experiments. It does not utilize real-world financial data or trained machine learning models. Instead, it generates synthetic observations and networks to demonstrate the architectural consistency and information flow of the pipeline.

Key Contributions

The paper identifies four primary contributions:

Modular Architecture: The formulation of a pipeline that connects latent risk mapping, probabilistic scoring, network construction, contagion dynamics, and scientific visualization through a unified operator composition.
Explicit Mathematical Representation: A formal definition of the entire process as a sequence of operators transforming synthetic information into dynamic visual representations, making information flow explicit.
Reproducible Implementation: A computational realization based on synthetic experiments that combines nonlinear risk surfaces, shortest-path contagion, and time-dependent visual encoding.
Integration of Visualization: The treatment of scientific visualization as an integral, explicit analytical stage within the pipeline, rather than a purely post-processing step.

Results

The experimental configuration utilized $N=520$ synthetic observations and a network of $M=24$ nodes with $E=31$ edges. Key findings from the synthetic experiments include:

Separation of Dynamics: The temporal analysis demonstrated a clear distinction between the evolution of the underlying probabilistic risk field (which remained relatively stable in the mean) and the network-mediated contagion (which exhibited discrete, step-wise increases in cascade size as the shock propagated).
Distinct Quantities: The framework successfully maintained the separation between assigned node vulnerability, topological connectivity (degree), and dynamic contagion intensity. High vulnerability did not necessarily correlate with high connectivity or immediate contagion, highlighting the independent roles of these variables.
Visual Coherence: The resulting dashboard provided coordinated views of latent risk, probabilistic distributions, network topology, and temporal indicators, illustrating how the operator pipeline preserves relationships between quantitative states generated at different stages.

Significance and Claims

The paper explicitly positions its contribution as architectural rather than predictive. The author states that the work does not:

Introduce new predictive models or machine learning algorithms.
Propose new financial contagion mechanisms (the distance-attenuation rule is a controlled synthetic mechanism, not a replacement for established models like clearing or threshold-based contagion).
Claim empirical forecasting performance or validation on real-world data.

Instead, the significance lies in providing a transparent and modular computational layer. The framework allows generative risk models, network structures, and contagion mechanisms to be composed, examined, and communicated within a coherent visual analytics structure. By defining operators with compatible input and output spaces, the architecture facilitates the replacement of individual components (e.g., swapping the synthetic scoring for an empirical model) without redesigning the entire pipeline. The work serves as a proof-of-concept demonstrating how heterogeneous analytical components can be integrated while preserving their distinct mathematical roles, thereby supporting reproducibility and the exploration of complex systemic-risk processes in synthetic settings.