LoanFlow Map

Inspiration

Modern loan markets are highly dynamic. Loans are originated, sold, pooled, and transferred across institutions at scale. While this improves liquidity, it also makes risk harder to see.

What inspired this project was a simple question that doesn’t have a simple answer today:

Once a loan is sold, where does the risk actually go?

Regulators and risk managers often rely on static reports or institution-level disclosures, which can miss how exposure builds up across the system over time. We wanted to explore whether this movement of loans and the resulting concentration of risk could be made visible and intuitive, rather than buried in spreadsheets

What it does

LoanFlow Map is an interactive visualization that shows how loans move through the financial system and how risk accumulates as a result.

• Institutions are represented as nodes.

• Loans flow between institutions over time.

• Node size reflects total exposure.

• Node color reflects risk-weighted exposure.

• The timeline allows users to replay how risk evolves.

• Risk hotspots highlight dangerous concentration.

• Stress scenarios simulate partial loan defaults.

• Anomaly indicators flag unusual or abrupt shifts in exposure.

Rather than focusing only on defaults, the system highlights systemic fragility that emerges purely from loan redistribution.

How we built it

The project was built as a web-based interactive prototype.

• Frontend: React + SVG for custom graph rendering and animations.

• Data layer: Synthetic loan, institution, and transfer datasets designed to reflect realistic market behavior.

• Graph logic: Ownership is recomputed at each timestep based on transfer history.

• Risk aggregation: Exposure and risk-weighted exposure are recalculated dynamically.

• Stress engine: Applies configurable default scenarios to loans.

• Cascade logic: Propagates stress effects through connected institutions.

• Anomaly detection: Flags sudden concentration or unusual transfer patterns.

All logic runs locally in the browser, making the system easy to demo and explore without external dependencies.

Challenges we ran into

One of the biggest challenges was designing data that actually revealed meaningful behavior. Early versions of our synthetic dataset looked realistic on paper but failed to surface risk patterns visually, which forced us to rethink how transfers, timing, and concentration were modeled. Small inconsistencies—such as multiple transfers at the same timestep or overly uniform risk scores—made the system appear stable even when it shouldn’t have been.

Another challenge was balancing accuracy with clarity. Financial networks can become visually overwhelming very quickly, so we had to carefully simplify without misrepresenting the underlying logic. Deciding what not to show was often harder than deciding what to include.

Finally, translating complex risk concepts into an intuitive visual language took several iterations. Features like risk hotspots, cascading impact, and anomaly indicators had to be visually distinct yet cohesive, ensuring that users could understand what was happening without needing technical explanations.

These challenges ultimately shaped the project into something clearer, more focused, and more usable.

Accomplishments that we're proud of

• Building a clear visual language for a complex financial concept.

• Making systemic risk visible before any defaults occur.

• Designing interactions that non-technical users can understand quickly.

• Successfully integrating stress testing, cascading impact, and anomaly detection into a single coherent flow.

• Keeping the interface calm and readable despite dense information.

What we learned

• Risk is often a network problem, not an institution-level one.

• Visual pacing and storytelling matter as much as technical correctness.

• Small design decisions (color, animation timing, pauses) dramatically affect comprehension.

• Synthetic data must be carefully designed, bad data can hide the very patterns you want to show.

• Explaining complex systems clearly is harder than building them.

What's next for LoanFlow Map

With more time and real-world integration, LoanFlow Map could be extended to:

• Ingest anonymized real loan transaction data.

• Support scenario libraries aligned with regulatory stress tests.

• Add explainability for why anomalies are flagged.

• Provide comparative views across markets or time periods.

• Integrate directly into supervisory workflows.

The long-term vision is a tool that helps regulators and market participants see risk forming early, not just respond after it materializes.

Built With

• html/css
• javascript
• react
• svg
• vite