Live monitor · connecting…
Will the system collapse?
Right now
CHECKING…
Reading 130 public data feeds from central banks, the US Treasury and markets.
System stress score
--/100
Will it? What followed months like this one
Every month since 1973 that scored within 3 points of today, and what happened next, against the average month.
The four pillars
What the score is made of
Each pillar is the average percentile of its inputs against their own history: 90 means worse than 90% of all months on record.
Right now
Red flags, and what's still holding
Every input at the extremes of its own history. A red flag is a reading worse than at least 85% of all months on record.
Red flags
Reassuring
Reading the gauge
When should you worry?
Each band comes from the backtest: how often, since 1973, a reading that high came with a crisis already under way or one that began within the following 12 months.
Backtest · 1973 to today
Fifty years of the gauge
What the score would have said at the time, month by month. Each point uses only data available then: percentiles against the history so far, and a recession model refitted every January on outcomes already known.
crisis episodes US recessions (NBER) start of a 20% stock-market fall
Concrete examples
What it said before, during and after
Ten moments from the last fifty years, scored with only the data that existed at the time.
The early-warning model
Recession probability, next 12 months: --
The early-warning pillar is a forecasting model built on the two signals with the best record for calling US recessions a year ahead: the yield curve and financial conditions, plus the credit market's risk appetite.
- Yield curve. When 3-month Treasury bills pay more than 10-year bonds, the market expects the Fed to cut because growth is about to weaken. Every US recession since 1969 was preceded by an inversion, usually 6–24 months earlier and often after the curve had started to un-invert, so the model also looks at the deepest inversion of the last 18 months.
- Financial conditions. The Chicago Fed's NFCI combines 105 measures of money-market, debt and equity conditions and of bank lending. Tight and tightening conditions starve the economy of credit.
- Excess bond premium. The part of corporate bond spreads not explained by default risk: when it rises, lenders are losing their nerve.
Logistic regression with a ridge penalty, trained on monthly data since 1971 to answer "does a recession start within 12 months?". Refitted each January using only outcomes already known; backtested out of sample from 1979.
Today's inputs
Out-of-sample track record
Learning from past crises
What moves before a crisis
Every crisis and 20% stock-market fall since 1973, lined up on the month it started. Market gauges stay ordinary until the month trouble hits; the signals that move earlier are in the economy and the credit cycle. None of them can see an overnight crash coming.
Average percentile around the start of episodes
Pre-crisis checklist
AUC: how well a high reading flagged an event starting within 12 months (0.5 = coin flip). We tested these and more inside the score itself: none improved it, because the four pillars already carry their information, so they're shown here as context. The recession probability didn't move either: adding claims, permits, lending standards, unemployment or inflation to its model left today's reading between 7% and 9% and none improved its out-of-sample record.
Inflation, the Fed and the worst cases
Stagflation, or worse?
Stagflation is inflation that stays high while the economy weakens: the 1970s, 1980–82, briefly 1990 and 2008. We define it as inflation of 5% or more while unemployment has risen at least half a point from its 12-month low.
Stagflation index
--/100
Inflation: official and alternative
What about "real" inflation? The best-known alternative, ShadowStats, is a paid service that doesn't recompute the data: its creator has said it adds a constant to the official figures (Econbrowser, 2008), and BLS economists have rebutted its claims (Monthly Labor Review, 2008). Instead we track the measures Federal Reserve banks build to see through noise. Producer prices and oil show the pressure still in the pipeline.
The Fed's stance
Hyperinflation checklist
Conditions seen before hyperinflations elsewhere (Germany 1923, Hungary 1946, Zimbabwe 2008, Venezuela 2018). The US has never had one, so this can't be backtested here.
Same metal, many markets
Gold, priced everywhere at once
No government sets the price of gold. It trades around the clock in London, New York, Shanghai, Zurich, Mumbai and on-chain, and arbitrage keeps every venue within a few dollars of the others. When a venue drifts, metal flows to it; a dislocation is an early sign that a market is breaking.
Live venues
Tokenised gold (PAXG, XAUT) is backed by allocated London bars and trades 24/7; central-bank references are once-a-day fixes.
Gold in ten currencies
The fiscal fuse
Debt keeps compounding
US federal debt · live estimate
$--
US net foreign debt, % of GDP
Gross external debt, % of GDP
World Bank / IMF Quarterly External Debt Statistics (CC BY 4.0). Debt owed abroad by governments, banks and companies. High numbers are normal for financial centres like London and Paris; what matters is the trend.
Government debt, % of GDP
Central-bank gold
IMF International Liquidity data, monthly, about two months behind.
How it works
The method, in plain English
Collect
Every 15 minutes a script on our server pulls 130 public series: Federal Reserve and FRED data, the US Treasury, the Office of Financial Research, the ECB, the IMF, the World Bank, the central banks of Argentina and Poland, and public exchange tickers. Each is stored with its full history, in some cases back to 1919.
Compare with history
A VIX of 30 or a debt ratio of 120% means little on its own. Each measure is turned into a percentile: the share of all past months in which it was lower. 90 means "higher than 90% of history".
Four questions
Stress now: is a crisis happening? Early warning: is a recession coming? Inflation & money: is money losing value? Debt & fragility: how much could break? Inside each pillar, inputs measuring the same thing are averaged first, so no theme is counted twice.
One score, two readings
Score = (2 × stress + warning + money + fiscal) ÷ 5. Stress counts double because it is the only pillar that says a crisis is actually happening. "Crisis now" = (2 × stress + warning) ÷ 3; "structural risk" = the average of the two slow pillars.
What it is good at
What it can't do
It can't see shocks that come from outside finance, like a pandemic, a war or a sudden crash with no build-up (Black Monday in 1987 barely registered beforehand). The yield curve can cry wolf: its deepest inversion since 1981, in 2022–24, has not been followed by a recession. Debt doesn't time crises at all; it decides how much damage one can do. And a backtest with a dozen crises is a small sample: treat the score as a thermometer with a weather forecast attached, not a prophecy.
How the formula evolved
- v1 · Market stress. Stress indices, credit spreads, VIX, plus a blend of the yield curve, debt and gold. Good at saying a crisis is happening; weak at saying one is coming.
- v2 · The recession model. The yield curve and financial conditions became a proper forecasting model, refitted every year on outcomes already known. Recession warning a year ahead improved to an AUC of about 0.9.
- v3 · Inflation, the Fed and debt. Added official and alternative inflation, inflation expectations, money growth, the dollar, the deficit, r − g, foreign debt, G7 debt and bond sell-offs, plus unemployment. Tested and rejected for the forecast: the Fed funds rate, its pace of hikes, the real policy rate and the 10-year minus 2-year curve, which added nothing the 10-year minus 3-month curve didn't already know (it is, in effect, the market's read of Fed policy). Debt measures don't time crises at all, so they shape structural risk rather than the warning.
- v4 · Valuations, leverage and a consistency fix. Added stock-market value against GDP (a record 255%) and corporate debt against GDP, one of the better predictors of 20% stock falls (AUC 0.77 a year ahead). Switched r − g to the rate new borrowing costs, and debt growth to after-inflation terms, so the high-inflation 1970s no longer flatter today. Fixed the history to use the same latest-known inputs as the live reading (it had been dropping slow quarterly series in recent months) and a gap left by the October 2025 shutdown, when no CPI or unemployment data was published. Every target backtests as well or better.
- v5 · Fairer averaging and calibrated labels. Inputs that measure the same thing are now averaged together first (eight consumer-inflation measures had been outvoting producer prices, expectations and the dollar; four stress indices outvoted the rest), which improved crisis warning a year ahead from 0.73 to 0.75. Bands are now set where the historical odds actually change, and the page shows those odds: the old "calm" label covered months whose odds were average. Tested and rejected: weights fitted by a model (worse out of sample: a dozen crises are too few to fit weights on) and an oil-shock signal (no added power; oil today is just below its three-year high).
- Research after v5 · signals before crises. An event study of 20 crisis and crash starts showed market gauges (stress indices, VIX, credit spreads, stock trends) don't move until the month trouble hits; if anything, volatility is unusually low three months before. Building permits, bank lending standards and jobless claims did move earlier and are shown in the pre-crisis checklist, but adding them to the score or the recession model didn't improve either out of sample, so the formula is unchanged.
Inspired by the long-gone didthesystemcollapse.com, which compared London and Shanghai gold prices; its "NOT YET" was a switch flipped by hand. This one is computed. Nothing here is financial advice.
Data feeds
Sources, live status
Everything on this page comes from one public JSON file, rebuilt every 15 minutes:/live/collapse.json. Each series stays under its publisher's terms.
$ connecting to willthesystemcollapse.com/live/collapse.json …