This material is a research tool provided on a confidential, access-controlled basis. It is
not investment advice, not an offer or solicitation, and carries no performance guarantee.
The index expresses a well-calibrated distribution of outcomes, not a point forecast.
Start here — how to read the dashboard
Four ideas unlock everything else on the page.
GMQ-FCI
The Global Macro Quant Financial Conditions Index — our proprietary read on how
loose or tight financial conditions are, and what they imply for the range of
future US growth. Unlike most indices, it outputs a full distribution of outcomes, not a
single number.
Every series on the dashboard is normalized so a positive (red) reading means
conditions are tighter than normal (a headwind), and a negative (green) reading
means looser than normal (a tailwind). This is applied consistently — public indices
disagree on sign, so we standardize it. A value of +1 means roughly one
standard deviation tighter than its own history.
Standardized / z-score
Each input is expressed relative to its own history (how many standard deviations from
average), so different series — interest rates, spreads, the dollar — can be compared on
one scale. 0 = normal; ±2 = unusually loose/tight.
Horizons: 1Q · 1Y · ~2.5Y
How far ahead we look: one quarter, one year, and the medium-term financial cycle
(~2–3 years). The same model is rolled forward to each horizon.
The Growth-at-Risk framework
Why we publish a distribution of growth, not a single forecast.
Conditional growth distribution
Given today's financial conditions, the range of plausible future GDP growth outcomes
and how likely each is. The point is the shape — especially the downside —
because that is what matters for risk and positioning.
Growth-at-Risk (GaR)
The 5th-percentile of that distribution: a bad-but-plausible growth outcome —
only a 1-in-20 chance growth comes in worse. It is the headline downside-risk number.
A 1Q GaR of −3% means: if the next quarter goes badly (5th percentile),
annualized growth would be about −3%.
Median (50th percentile)
The central outcome — growth is equally likely to land above or below it. The gap
between the median and the GaR measures how skewed the downside is right now.
Quantile / percentile
A point in the distribution. The 5th percentile is exceeded (to the downside) 5% of the
time; the 95th, only 5% to the upside.
Fan (the bar under each GaR card)
A compact picture of the distribution: the line spans the 5th–95th percentile range, the
shaded box is the 25th–75th (the likely zone), and the gold tick is the median. A long
left tail = elevated downside risk.
Regimes
The economy behaves differently in calm vs. crisis — the model adapts.
Regime (calm · tightening · stress)
A market environment the model infers from the data. In calm, conditions are
benign; in tightening, they are deteriorating but orderly; in stress, the
system is under strain and the credit/leverage channels dominate the growth downside.
The relationship between conditions and growth is allowed to differ by regime — the key
way this index improves on fixed-weight predecessors.
Regime probabilities
The model's confidence that we are in each regime (they sum to 100%). The stacked chart
shows how this has evolved — note the red stress spikes in 2008–09 and 2020.
Inputs & channels
The seven families of data the index reads — including channels most indices ignore.
Rates
Policy rate, Treasury yields, the yield-curve slope, real (inflation-adjusted) yields.
Credit — price
The cost of corporate borrowing: investment-grade and high-yield spreads, and the Excess Bond Premium.
Excess Bond Premium (EBP)
The part of credit spreads not explained by default risk — i.e. investors'
risk appetite. A clean, long-history gauge of credit-market sentiment; one of the most
informative inputs.
Credit — quantity often omitted elsewhere
How much credit is actually flowing — bank lending growth and loan-officer survey
tightening. Captures the credit channel that price-only indices miss.
Leverage often omitted elsewhere
Borrowing and balance-sheet stretch in the non-bank/"shadow" system — an early-warning
dimension for financial stress.
Equity
Broad equity returns and equity volatility (VIX).
FX / funding
The trade-weighted dollar and emerging-market spreads — the global/USD-funding channel.
Stress
The OFR Financial Stress Index and its components, used as features and as a public comparison.
How we validate it — and what we do not claim
Honesty about what the index can and cannot do.
Climatology (the benchmark)
The naive forecast that "the future looks like the historical average." It is
deliberately hard to beat. We hold our model to this bar and report the result honestly,
rather than flattering the index.
Calibration vs. forecast accuracy
We do not claim to predict the exact number. We claim the distribution
is well-calibrated: when it says "5% chance of a bad outcome," that happens about 5% of
the time. A reliable sense of the odds is the deliverable — not a crystal ball.
CRPS
A score for how good the whole predicted distribution was against what actually
happened (lower = better). Our primary accuracy measure.
PIT calibration
A check that the probabilities are honest — that outcomes fall across the predicted
range as often as claimed. A well-calibrated model passes this test.
Point-in-time (vintage) testing
We test using only the data that was actually available at each past date — no
hindsight from later data revisions. This is stricter and more honest than most published
backtests. e.g. 2008Q4 growth was known as −3.9% at the time vs −8.9%
after revisions — we use the −3.9%.
Methodology, briefly
For the technically inclined.
World model
Rather than a fixed formula, the engine learns the dynamics of the
financial-macro system and simulates it forward to produce the growth distribution. This
also enables scenario questions ("what if credit spreads doubled?"). The current
dashboard shows a transparent baseline; the world-model core is the next build phase.
Recursive self-improvement
The system scores its own accuracy over time and proposes refinements — but every change
is human-reviewed before it affects the published index. It does not silently rewrite
itself.