Eight-lens monitor of the credit cycle plus a single composite index — the Credit-Cycle Melt-Up Index (CCMI). Each lens fires when historical melt-up criteria are met. The composite synthesizes 5 risk-on lenses into one continuous z-score, calibrated against forward 12-month equity returns on a train-only OLS. The monitor's job is not to call tops — it's to disclose the regime shape so the reader can see when public credit is substituting for private, when equity is racing credit, or when spread compression is selective. Updated .
ICE BAML option-adjusted spreads, last 5 years on log axis. HY at bps (th pct since 2000); CCC at bps (th pct); IG at bps (th pct). The HY-IG differential of bps is the cleanest "credit risk-on" reading: when IG and HY converge, the market is dismissing default risk; when CCC blows out separately, that's *selective* credit melt-up — the quality tier compresses while the worst-quality tier prices stress. Source: .
NVDA at $ (); HY OAS at bps (). When mega-cap equity leadership and credit risk-on travel together, the regime is melt-up. The April 2025 wobble (HY peak on tariff stress) tested the linkage; both series resumed their rise once the policy backdrop stabilized. Sources: .
US private non-financial credit-to-GDP gap (Borio-Lowe canonical metric, BIS). Currently
pp as of
. Above +10 pp historically precedes financial stress events; below −10 pp is deep deleveraging. The US has been below trend since the post-GFC repair — this cycle's "melt-up" is happening *despite* private credit deleveraging, not because of it. Globally
only of countries are above trend (Saudi, Japan, Argentina, Israel) — see
Multi-Polar World for the cross-country dispersion. Source:
.
US Federal debt held by the public, % of GDP. Currently
%. The 100% threshold (dashed) is where rating agencies and bond markets historically begin pricing fiscal-dominance risk. Public credit expansion is the inverse of private deleveraging: when households and firms repair balance sheets, governments substitute via deficit spending. The combination of high public debt + low private credit gap + strong asset prices is the classic late-stage fiscal-dominance regime — captured in detail by the FDI sub-index on the
Dollar System page. Source:
.
S&P 500 YoY return minus nonfinancial business credit YoY growth, monthly. Currently pp — equity outpacing credit by pp. SPX up % YoY; business credit growing only % YoY. Above +10 pp (dashed) is melt-up territory historically — equity multiple expansion outpacing the underlying credit aggregate that should support it. The divergence resolves either by credit catching up (the bull-extension scenario) or by equity correcting (the cycle-end scenario). Source: .
Z.1 nonfinancial corporate equities market value / nominal GDP. Currently % as of — th percentile since 1990. Historic peaks for context: dotcom 2000 ~210%, 2007 ~140%, 2021 ~280%. Above the 75th historical percentile fires the elevated-multiple lens. Buffett's "best single measure" of equity market valuation (Fortune, 2001). The current reading places the equity multiple in textbook melt-up territory — even allowing for buyback-driven mechanical compression of the denominator over time. Source: .
Broker-dealer margin loans (FRED Z.1) / nominal GDP. Currently % as of — th percentile since 1990. Late-cycle peaks: 2000, 2007, 2021. Investor leverage rising into a melt-up is a textbook reflexive feedback loop — when leverage is elevated, drawdowns force liquidations that amplify the move down. Above the 75th historical percentile fires the elevated-leverage lens. Source: .
10-year minus 2-year Treasury yield, last 5 years. Currently pp (). 1Q change: pp. Days since last inversion: . Note: yield-curve dynamics are tracked here as context, not as a melt-up flag. Curve steepening from prolonged inversion is a textbook late-cycle inflection — historically signaling either (a) the Fed easing into recession, or (b) the long end pricing fiscal-dominance term premium. Either interpretation supports the asymmetric / fiscal-dominance reading already firing across other lenses. Source: .
Three cross-asset risk dials. VIX . NFCI measures financial conditions vs. average — , which is itself a credit-cycle signal: when NFCI is deeply negative AND public debt is high AND equity is racing credit, that's the late-cycle "everything bid + cheap funding" regime. 10Y real % sets the discount rate against which the entire risk-on regime is priced. Source: .
The Credit-Cycle Melt-Up Index (CCMI) is the equal-weighted z-mean of 5 risk-on lenses validated by Phase 0 signal probe:
negated HY OAS, negated CCC OAS, negated VIX, Buffett ratio, margin debt / GDP. Each lens is z-scored on an expanding window (no full-sample peek). Latest z =
(
,
th percentile since 2000). Historic peaks for context: dotcom 2000, pre-GFC 2007, post-COVID 2021.
v2 calibration (rolling-origin 60-mo OLS). The current calibration applies the same rolling-origin pattern proven on the inflation nowcast — at each historical month, the model is fit on the prior 60 months of (z, forward_12m_return) pairs and used to predict 12m forward return. The "current" nowcast uses the same protocol on the most recent 60 months: yt+12m = × zt, train window → . With current z = σ, the nowcast = %.
Rolling-origin OOS metrics (n=, calibration started after 60-mo composite-history burn-in): corr , MAE pp, bias pp, R² .
v1 → v2 transition. The earlier static train-only OLS (calibrated on data ≤ 2022-12-31) had MAE ~10.8 pp and a systematic bias of −10.8 pp on the post-2022 OOS sample — the model trained pre-disinflation didn't generalize to the AI-rally / disinflation regime that followed. v2 with rolling refit reduces bias to near zero (− pp) and MAE to pp. The v1 metrics live on the Calibration Ledger as the documented track record of the prior model; v2 predictions are scored from this point forward under target ccmi_12m_return_v2.
Read this carefully: the nowcast is a conditional expected return given the composite at z = σ — not a deterministic forecast. R² ≈ % means the composite explains roughly that fraction of forward-return variance; the rest is noise, regime change, and unmodeled factors. MAE = pp is the typical absolute error to expect on any single forecast.
CCMI tells you where credit conditions sit on the melt-up scale. The Cycle Clock tells you which phase the broader business cycle is in. The two together carve up forward 12-month equity returns into very different regimes. The matrix below: each cell is the mean forward 12m S&P 500 return when both CCMI and Cycle Position landed in that combination historically.
Black border = current setup ( CCMI × ). Red border = danger zone (HIGH CCMI × MID CYCLE) — historically the lowest forward-return regime. Cells show mean 12m forward S&P 500 return for that combination, with sample size. Empty cells (n=0) had no historical observations.
The "danger zone" is the cell where CCMI is in melt-up territory (z > +0.5σ) and the cycle is already in MID CYCLE — credit conditions are most extended at the same time the cycle is most exposed. Historically this combination produced the lowest forward returns. The asymmetry: when you're not in the danger zone, returns are reasonable; when you are, the next 12 months disappoint. The danger zone fires when CCMI crosses +0.5σ while the Cycle Clock holds in MID CYCLE. Worth setting an alert for.
| Signal |
Pearson ρ |
Read |
| Cycle level (Trend + Stress) | | strongest single signal — when broad health is high, forward returns are lower |
| Direction (CP momentum) | | peak-momentum-on-the-way-down predicts peak-of-cycle |
| Cycle score (0-100 percentile) | | regime-stable summary of cycle level |
| CCMI z (credit melt-up) | | credit cycle peak nowcast |
All four signals are negatively correlated with forward 12m equity returns over the 2008-2024 sample. They're partially correlated with each other (cycle level, score, and direction all derive from the same indicator panel) so a linear combination doesn't materially improve on the strongest single signal. Where the combination does add value is at the extremes — bottom decile of (−CCMI + 0.5×level + 0.5×direction) returned +27.8% on average, top decile returned +4.5%.
| Lens |
Threshold |
Source |
| HY OAS |
≤ 25th historical percentile = compressed |
FRED BAMLH0A0HYM2 |
| CCC OAS |
≤ 25th historical percentile = compressed |
FRED BAMLH0A3HYC |
| US private credit gap |
> 0 pp = above trend (melt-up) |
BIS WS_CREDIT_GAP |
| Public debt / GDP |
> 100% = high (fiscal substitution) |
FRED GFDEGDQ188S |
| Asset / credit divergence |
> +10 pp (SPX YoY − biz credit YoY) |
FRED SP500 + BCNSDODNS |
| VIX percentile |
≤ 25th historical percentile = low-vol regime |
FRED VIXCLS |
| Equity multiple (Buffett) |
≥ 75th historical percentile = elevated multiple |
FRED NCBEILQ027S + GDP |
| Margin debt / GDP |
≥ 75th historical percentile = elevated leverage |
FRED BOGZ1FL663067003Q |
| Yield curve 2s10s |
(context only — no flag fired) |
FRED T10Y2Y |
Why six lenses, not one: the credit-cycle melt-up isn't a single indicator. Different cycles have different shapes — sometimes spread-led (1996-2000, 2014-2015), sometimes equity-led (1999, 2021), sometimes public-debt-led (2010-2015 EM crisis). The monitor's job is to disclose the shape of the current regime so the reader can see which lenses are firing and which are not, rather than collapsing the assessment into one number.
A regime where 5+ lenses fire simultaneously is historically a setup for a stress event 12-24 months out. A regime where 3-4 fire is "elevated" — neither a clean melt-up nor a tightening — and the dispersion across lenses is itself the signal: in the current reading, the equity-credit divergence and public-credit substitution lenses are firing, but private credit and CCC spreads are not — that's the asymmetric / fiscal-dominance shape, not the uniform-melt-up shape.
Cross-references:
CCMI's 12-month return nowcast was calibrated by static train-only OLS on data ≤ 2022-12-31. The post-2022 OOS portion (currently 28 realized predictions) shows MAE of ~10.8 pp and a systematic bias of ~−10.8 pp — the model consistently under-predicts forward returns. The reason is identical to the inflation nowcast's earlier sign-flip: a static OLS trained on one regime can't extrapolate to a regime it didn't see (the 2023-2024 disinflation + AI-driven equity-multiple expansion). The fix that worked for inflation — 60-month rolling-origin OLS that refits monthly — is the v2 path for CCMI as well. Until rebuilt, treat the directional signal as informative but the magnitude as biased.