A 12-phase empirical study built a way to mathematically separate two things that financial media constantly confuses — using nothing but free, public data.
When Treasury prices swing violently, most observers reach for the same explanation: fear, panic, dysfunction. This research shows that's often the wrong diagnosis — because the market has two separate "problems" that can exist independently of each other.
How violently Treasury prices and interest rates are moving. This is built from real, measured volatility of key Treasury yields — the 2-year, 5-year, and 10-year — plus the volatility of the yield curve's shape. Think of this as "how choppy the ocean is" on the surface.
How strained the financial plumbing is — the overnight financing markets, settlement systems, and balance-sheet capacity that dealers use to function. This is built from repo rate spreads published by the Federal Reserve Bank of New York. Think of this as "how well the engine room is running" beneath the surface.
These aren't aspirational goals. They are direct outcomes of what the 12-phase study built and tested.
Financial media lumps all market turbulence together. Big price swings get labeled "panic," "fear," or "broken market" — regardless of whether the underlying financial plumbing is actually under stress. This framework provides a measured, reproducible basis to separate "prices are choppy, but the financial machinery is fine" from "prices are choppy and the machinery is actually breaking down." That distinction matters enormously for policy, for investors, and for public understanding of what's actually happening during market events.
Historically, measuring this kind of market stress required proprietary tools like the ICE BofA MOVE Index — instruments that cost money and aren't fully transparent. This project deliberately replaced those dependencies with three open, reproducible measures built from free public data: realized Treasury yield volatility, yield-curve slope instability, and a return-based instability measure. The entire methodology is documented, auditable, and can be rebuilt from scratch by anyone. That lowers the barrier for independent researchers and policymakers to monitor Treasury market health without institutional gatekeeping.
A common misread: when dealers are borrowing heavily in overnight repo markets, observers often interpret that as a distress signal. The research found evidence that this is frequently wrong. High repo borrowing by primary dealers can simply mean the market is active — dealers are doing their job, financing positions and providing liquidity. It is not, by itself, evidence that the system is under strain. Separating that "active market" signal from genuine funding stress is one of the framework's concrete contributions.
The framework wasn't just theorized — it was checked against three historical events where we already know what actually happened, to see if the two-axis model tells the right story.
Overnight repo rates suddenly spiked to nearly 10%, catching markets off guard. This was a genuine funding-plumbing event: the constraint axis should have been elevated. The instability axis, by contrast, reflects how much Treasury prices themselves were moving — a separate question. The framework's two measures behaved differently, consistent with a stress event in the funding system rather than pure price panic.
March 2020 saw the closest the Treasury market came to a full breakdown in modern memory. Investors were selling everything — even the safest assets — to raise cash, and dealers couldn't absorb the flow. Both axes were expected to be simultaneously elevated, representing a worst-case "both surface and engine" event. This is the episode the framework is most designed to identify correctly.
The collapse of Silicon Valley Bank and Signature Bank created significant Treasury price volatility — rates moved sharply as investors rushed into government bonds as a "safe haven." But was this a Treasury market plumbing failure, or just aggressive repricing? The framework allows that question to be asked precisely, separating the price turbulence from any underlying funding-channel strain.
"The project constructs a weekly dataset and then creates two composite indices... capturing 'how choppy the rate market is' and 'how constrained the financing and intermediation channel looks.' The operating model then asks a narrow question: if we already know the market is volatile, does the constraint axis still explain anything additional?"
The project went through 12 documented phases over roughly a year, each with locked specifications, transformation logs, and quality audits. Governance locks were applied at key stages to prevent the kind of "result shopping" that plagues informal research — where findings get quietly adjusted until they look good.
Data sources are entirely public: Federal Reserve Bank of New York reference rates (SOFR, TGCR, BGCR), Federal Reserve primary dealer statistics, and Treasury yield data. The processed datasets include a daily panel of over 8,500 rows and a Friday-aligned weekly panel of over 1,700 rows.
The statistical tests were pre-registered — meaning the hypotheses and methods were written down before the analysis ran, making it impossible to change the question after seeing the answer. A Hidden Markov Model was used to identify probable market regimes, treated as a secondary interpretive tool rather than a definitive classifier.
An independent "red team" adjudication phase (Phase 7) was included specifically to challenge and attempt to break the model's assumptions before any final conclusions were drawn.
One of the most rigorous aspects of this project is its explicit documentation of what it does and doesn't prove. Those boundaries are worth being clear about.
Every data source used in this research is freely available. Every transformation is logged. Every design decision is documented in a dedicated memo. The project was built to be audited, challenged, and rebuilt by anyone. That's not a feature — it's the point. The goal was to demonstrate that institutional-grade Treasury market monitoring doesn't require institutional resources or proprietary data subscriptions.