What the number is
A single 0–1 score summarising market conditions associated
with credit-cycle stress — an experimental index, not a measurement of every private
balance-sheet exposure. It combines four families of public market driver:
the term structure of interest rates, credit stress, market volatility, and an equity valuation level. Higher means a higher combined
market-conditions score under this method — not a higher probability of any
event, and not a reading of the debt-service, refinancing, contagion, external or
fiscal buffers dimensions named above, which this index does not feed and do not feed
it. It does not directly observe household debt service, corporate refinancing needs,
cash buffers or bank positions; those belong to separate, separately named models.
The model recomputes the score daily and publishes it to cloud
storage; this page reads that published copy (with a baked fallback that is labelled
when stale). The inputs have different cadences: the market legs publish daily, the stress index about two business days after the date it describes, while the valuation leg is a quarterly value held until the next one, so a daily publication is not four daily observations.
Daily movement enters only through quarter-to-date driver means — a single day cannot
swing the reading on its own, which is why the line is smooth and why a still-open
quarter is marked provisional.
| Input | Plain explanation | Limitation to retain |
| Treasury yield curve |
The difference between long- and shorter-term Treasury
yields; recent flat/inverted conditions raise this component. From Treasury's daily par yield curve. |
Retained history can keep it elevated after the current
curve changes. |
| Credit stress |
The credit component of the Office of Financial Research's Financial Stress Index, built from corporate bond spreads in the United States, the euro area and Japan. |
A market financing indicator, not the amount of debt or a measure of total balance-sheet losses; it is global, not US-only. |
| Market volatility |
The volatility component of the same index, built from expected volatility in equity, interest-rate, currency and oil markets. |
Often responds during a sell-off; it does not necessarily
lead the shock. |
| Equity valuation |
The market value of US nonfinancial companies' equity, including foreign direct investment equity, as a percentage of their net worth, from the Federal Reserve's Z.1 accounts. |
Quarterly, used two quarters after its date because it is published about ten weeks after each quarter ends; high valuation alone does not identify when a decline will occur. |
The yield-curve input is the 10-year Treasury yield minus the
2-year yield. See the rates, their spread and the market index together in our
yield-curve guide.
Sources. The index is computed from public-domain data: the U.S. Department of the Treasury's daily par yield curve, the Office of Financial Research's Financial Stress Index and the Federal Reserve Board's Z.1 Financial Accounts. Its earlier edition used FRED series, including Moody's and Cboe data, and Shiller's CAPE; forecasts sealed on that edition are graded on it. The history shown is this edition's replay from 2000.
What the levels mean
The index has four levels on its 0 to 1 scale: benign below 0.25, watch from 0.25, elevated from 0.40, fragile from 0.55.
Each edge is the lower bound of its level, so a reading of exactly 0.40 is elevated.
The levels are bands of the combined score, not a count of stretched legs: a
weighted composite does not require a fixed number of the four drivers to be stretched
at any level. Elevated starts at 0.40, the boundary the forecasts sealed on the
earlier edition use; the edges at 0.25 and 0.55 are authored: they were chosen by
this model's authors. Crossing an edge is a statement about the combined index, not a
calibrated probability of any event.
Each level's record since 2000: the quarters at that level and the dates of each spell, counted from the index's quarterly history (108 quarters, 2000Q1 to 2026Q4; the last is provisional). The page recounts it from the latest published history when that loads.
- Benign, below 0.25: 40 quarters
- 2003Q2–2007Q2; 2010Q1–2011Q2; 2012Q3–2015Q2; 2017Q2–2018Q2
- Watch, 0.25 to under 0.40: 30 quarters
- 2002Q4–2003Q1; 2007Q3–2007Q4; 2009Q3–2009Q4; 2011Q3–2012Q2; 2015Q3–2017Q1; 2018Q3–2019Q4; 2021Q1–2022Q1; 2026Q3–2026Q4
- Elevated, 0.40 to under 0.55: 17 quarters
- 2000Q1; 2002Q2–2002Q3; 2008Q1–2009Q2; 2020Q1–2020Q4; 2022Q2; 2025Q4–2026Q2
- Fragile, 0.55 and above: 21 quarters
- 2000Q2–2002Q1; 2022Q3–2025Q3
A level describes
this index, not the economy. The index stayed below fragile through the 2008–2009
financial crisis and through 2020, so read fragile as a high reading of four market
signals, not as a sign that a crisis is under way.
What it is not
It is not a market call, not a recession probability, not a
timing signal, and not a demonstrated forecasting record. It is model-conditional:
it says what this model reads from these drivers. Its retrospective record is short and mixed (stated above), and it says nothing about exogenous
shocks — a war, a pandemic, a policy accident — that arrive from outside its
market drivers.
Why the US line is different
The US reading uses four public-domain series: the Treasury yield curve and the stress index's credit and volatility components, each daily, and a quarterly valuation series from the Federal Reserve. Its credit and volatility legs are global measures; its yield curve and valuation are the United States' own. Forecasts sealed on the earlier edition of this index stand in the ledger and are graded on that edition; this edition has no sealed forecast yet. Sealing makes a record checkable; it does not, by itself, make the model right.
Why the other lines are approximate
No other economy has daily public drivers of its own on this route. The comparison lines use the OECD's monthly long and short interest rates for the yield curve, the same global credit and volatility components as the US line, and market capitalisation over GDP as the valuation leg, normalised against that country's own history. Each substitution is
listed on the country card on the home page.
That makes each of them useful for watching its own economy's
direction of travel. It does not make them the same instrument as the US index,
and it does not support ranking economies against each other — the same numeric value
need not represent the same exposure in two countries. They are
approximate: not sealed, not scored, and never
published as predictions.
Where we publish nothing. A country reading is not
the US instrument pointed at other data: with legs missing, and monthly, lagged inputs,
it would be a different model with a different calibration — and it would need its own
sealed lanes before it deserved to be called a prediction. Showing one under the same
name is exactly the failure this ledger exists to prevent.