GLOBALCRASH · METHODOLOGY
How the number is made
The Global Crisis Index is a weekly 0–100 composite of systemic financial fragility, built from 36 indicators across seven pillars. Higher means more dry tinder: stretched valuations, tightening credit, correlated stress. It measures how fragile conditions are — not when anything breaks.
Pillars & weights
| Pillar | Weight | Anchored in |
|---|---|---|
| Credit & liquidity | 25% | Five equal sub-buckets: Baa credit spread · FCI composites (NFCI, OFR FSI + safe-assets, St. Louis Fed, US CISS) · nonfinancial leverage · liquidity (net Fed liquidity, funding, M2) · bank credit supply (SLOOS) |
| Macro cycle | 20% | Yield curve + re-steepening, NY Fed recession probability, Sahm rule, jobless claims, inflation de-anchoring |
| Valuation & froth | 15% | Shiller CAPE, equity risk premium |
| Housing | 10% | Case-Shiller momentum, mortgage spread, building permits |
| Crypto | 10% | BTC MVRV, stablecoin supply, DeFi TVL, perp funding |
| Sentiment | 10% | VIX + term structure, SKEW |
| Geopolitics | 10% | GPR (Caldara–Iacoviello), EPU, euro-area CISS, OFR EM/advanced stress, oil shocks |
The math
Percentile ranks, not z-scores. Every indicator is scored as its rolling 20-year percentile — "credit spreads at the 74th percentile of two decades" is the native language. Fat-tailed financial data breaks Gaussian assumptions; percentiles don't care.
Complacency is a signal. Indicators that go quiet before storms (VIX, credit spreads, leverage build-up) are scored two-sided: extreme calm reads as mild risk, never as safety. Suppressed volatility preceded both 1998 and 2007.
Weekly print, smoothed. Pillar scores are equal-weighted within each pillar, combined by the fixed weights above, and smoothed with a 4-week EMA. The official print lands Fridays 20:00 UTC. A separate 13-week velocity tracks the slope — in fast crises, the slope screams before the level does.
Systemic breadth. A correlation-weighting layer adapted from the ECB's CISS (Hollo, Kremer & Lo Duca, ECB WP 1426) measures whether stress is broad-based across pillars or isolated to one corner. Stress that moves together is systemic; stress in one corner usually isn't.
Slow data carries honestly. Every member carries its last value forward (up to one quarter; ~17 weeks for structurally quarterly series) and is labelled with its true as-of date in the "show the math" table — never silently interpolated. Each pillar shows how many of its members are live; a pillar is never allowed to be impersonated by its single latest-updating member.
Thin windows can't fake extremes. Any member with under ten years of effective history has its percentile shrunk toward 50 in proportion to its window and capped at 99 — a two-year-old series may never print the scariest number on the page. Correlated official composites (four Fed/ECB stress indexes) are bucketed as one vote, not four.
One caveat on the correlation regime. It is a coincident-to-lagging read: "Narrow" describes today's cross-pillar correlation, it is not reassurance about tomorrow. And a note on records: the fragility peak (74.0, Mar 2001) is a different object from the systemic stress peak, which came in October 2008 — valuation-heavy fragility and funding-market panic are measured on different clocks.
Known blind spots. Private credit, pension/LDI leverage, insurance balance sheets, and China property have no free, timely data feeds — episodes driven from those corners (the 2022 gilt/LDI spiral, Evergrande) reach this index only through their echoes in the markets it does see. Named here so nobody discovers it the hard way.
Roadmap (v2). A dedicated bank-and-funding pillar (SOFR spreads, deposit flows, discount-window borrowing — all free Fed data), a companion acute-stress meter alongside this fragility gauge, empirically calibrated band edges frozen per annual release, and a fixed-roster 1990 backfill for fair historical grading.
What it cannot do
It cannot see exogenous shocks: pandemics, wars starting, plumbing accidents. It read low in January 2020. Expect roughly one false alarm for every one-to-two real crises — stress episodes that resolve without recession will move it. That trade-off is deliberate: this is an awareness instrument, not a trading signal, and it is never investment advice.
Design principles
Every animation on the page traces to a real data event — no synthetic motion, no fear-timed alerts, no paid rank, no "crash incoming" language, no infinite scroll. The instrument grades itself in public: see the Scorecard.
Build status. The weekly index is computed from live data at every print. The between-print tick stream shown on the Pulse is currently a preview animation; live per-tick market feeds arrive with the hosted release.
Data sources
FRED (St. Louis Fed) · OFR Financial Stress Index · ECB Data Portal (CISS) · Chicago Fed NFCI · CBOE · NY Fed · NBER · Shiller/multpl · FINRA · DefiLlama · Coin Metrics · Caldara–Iacoviello GPR · policyuncertainty.com · Binance