Risk through market history
The score each December, 1928 to 2026
Computed December readings for selected years, and October 2026 at 47. Higher scores mean higher estimated crash risk.
The score each December, 1928 to 2026
Computed December readings for selected years, and October 2026 at 47. Higher scores mean higher estimated crash risk.
Each sign scores 0 to 100, and higher is worse. They do not count equally. Lending stress counts most, at 35%, because it has shown up before more crashes than anything else. Share price counts least, at 5%, because on its own it is worse than a coin flip at calling a crash. Hover any card for more.
Stock prices are near the highest they have ever been compared with company earnings, close to the 1999 dot-com peak. The index set a record on 6 October and has given back 2% since, which barely dents that. The whole market is worth more than twice the economy behind it, and tech, 38% of the S&P 500 after a week in which chipmakers fell eight days running, is still the most stretched part of all.
Investors have borrowed $1.45 trillion to buy stocks, within 3% of June's record and 37% more than a year ago; September's figure lands any day now. The Chicago Fed's weekly read of leverage across the financial system has risen four weeks running, which is what lifted this score. The crowded bond trade that worried regulators has kept shrinking, to about $900 billion from $1.26 trillion at the start of the year. And a lot of lending now happens out of public view: this week the bonds of one AI data-centre company with about $50 billion of debt fell to distressed prices, and a private lending fund got requests to redeem 39% of its shares.
The fear gauge jumped from 16 to 20 on Friday, touching 22 during the day for the first time in four months, as the Nasdaq fell 2% and chip stocks closed their worst week since April 2025. That one day is most of what lifted this score, and it is in the score because we added Friday's close by hand: the public series this score reads runs a day behind. Daily moves over the past month are still about half a percent, normal, so options are now priced for a good deal more movement than the market has been delivering. Twenty is not panic; it is the market paying attention. Whether calm is reassuring or a warning is a real argument, and we tested it: scoring low volatility as dangerous made the model worse, so it counts calm as calm.
The jobs picture turned: 29,000 jobs added in September against 90,000 expected, the two months before revised down by 60,000, unemployment up to 4.2%, and pay growth the slowest since 2021. The recession alarm moved from below zero to zero; it sounds at 0.50. Factories are still growing, for a ninth month, but paying the highest prices in two years, and inflation came back to 3.7%. Consumer sentiment in the Michigan survey hit the lowest reading for current conditions in its history. Participation, which crossed the worry line a fortnight ago, came back to it: about 52% of S&P 500 stocks were above their 200-day average on Thursday, before Friday's fall. A few giant stocks are still carrying the whole market, and this score reads the index, not its members.
The Fed raised rates on 16 September for the first time since 2023, to 3.75–4%. A hot inflation reading on 14 October briefly made a second rise in October a coin flip; two days later the Fed chair declined to signal anything and the odds fell to about one in six. The long end reversed: the ten-year sold at 5.30% at auction on 7 October, its highest since 2000, then fell back to about 5.1% as oil steadied and money left tech for bonds, and the thirty-year came off its 24-year high. That flattened the curve a little, so the bond market's recession signal moved back toward a warning and this score rose, by the model's own rule. It reads the slope, not the level: last fortnight a sell-off read as safer here, this fortnight a rally reads as riskier.
Cash is moving smoothly between banks. The quarter-end passed without a squeeze, overnight lending rates stayed just under the Fed's own rate, and the Treasury's October auctions drew the strongest demand of the year. What moved this score is the Chicago Fed's broader weekly gauge of financial conditions, which has tightened four weeks running from a very loose starting point. There is still no sign of the kind of squeeze that usually shows up when a crisis is brewing.
The extra cost for risky companies to borrow is 3.2 percentage points above safe borrowers, from 3.1 a fortnight ago and 2.7 in mid-September, its widest since April; safe companies pay what they did. That is still low by history, and the monthly series this score reads has not registered it, so the score moved only on the Chicago Fed's weekly credit gauge, which has tightened four weeks in a row. The line to watch is 4.5. The stress is in the places this score cannot see: the bonds of CoreWeave, which has borrowed about $50 billion to build AI data centres, fell below 80 cents on the dollar this week, and Blue Owl capped withdrawals from its technology lending fund again after investors asked for 39% of their money back.
Stress in lending 35% · Fear & mood 25% · Interest rates 15% · Money flow 10% · How expensive stocks are 5% · Borrowed money 5% · Economy & market health 5%. The weights come from checking which signs actually showed up before the last 14 crashes. Lending stress warned most often, so it counts most. Share price warned least, so it counts least. The Nasdaq scores 2.75 points higher because it leans on a handful of tech giants.
Scale: under 35 low · 35–50 guarded · 50–65 elevated · 65–80 high alert · over 80 crisis
These weak spots build slowly, and any one of them could turn an ordinary scare into a full-blown crash. None of them tells you when. Each one just makes the fall bigger once something else starts it.
Tech now makes up 39% of the S&P 500, the most ever. If one of these giants disappoints and investors start doubting the huge sums being spent on AI, both indexes get dragged down through the same few names.
Hits hardest → NasdaqOnly about half of S&P 500 stocks are in an uptrend, down from 73% in the middle of August and 66% at the start of September, while the index sits within half a percent of its record. When a rally rests on so few names, it is much easier to knock over.
Hits → Both indexesThe amount investors have borrowed to buy stocks hit an all-time high of $1.50 trillion in June 2026, dipped in July, and climbed back to $1.45 trillion in August. Borrowed money does not pick the day trouble starts, but it guarantees that when stocks fall, people are forced to sell, pushing prices down even further.
Hits → Anyone who borrowedBorrowing is near record highs and concentrated in the biggest funds, including about $2.4 trillion tied up in U.S. government bonds. If those bets had to be unwound in a hurry, the stress would spread to every market at once.
Hits → The whole systemA huge $256.8 trillion of lending worldwide now happens outside regular banks, and much of it, especially private loans, is hard to see into. Trouble there could build for a long time before anyone notices.
Hits → Loan-heavy cornersNasdaq-100 stocks cost about 32 times their yearly profits. Because so much of their value rests on profits far in the future, they are the part of the market most sensitive to any surprise jump in interest rates.
Hits → NasdaqA fragile market still needs something to set it off. These are the things to watch, grouped by how soon they could bite. The 6-month list is the tripwire: if those start flashing, the overall risk score climbs fast.
The score cannot see tariffs, wars or the Fed until they move a price. The outlook page scores them by judgment, under a written rubric fixed in advance, and keeps every reading on record so it can be graded against this one. Outlook score, written 17 Oct 2026: 59 (Elevated), with a stated 25% chance of a 20% fall within six months.
Every month since 1928 gets a score. This is what happened next. In a normal month, the chance of a 20% fall within six months is 13%. A score is only worth reading if it beats that.
Today the score is 47. Months like this were followed by a 20% fall 6% of the time. A normal month is 13%.
Lending stress alone does the same work as all seven signs put together. If you only watch one thing, watch that.
| Signal | Today | Starts to worry at |
|---|---|---|
| Cost for risky companies to borrow | 3.2% above safe borrowers, from 2.7% in mid-September. Still calm by history. | 4.5% |
| Fear gauge (VIX) | 20, up from 16 on Friday. Actual swings are half that. | 25 |
| Share of stocks in an uptrend | 52% on Thursday. Back at the line. | below 50% |
| Interest rates | 3.75–4.00%, raised on 16 September. October odds spiked to a coin flip on hot inflation, then fell to about one in six when the Fed chair would not signal. | a surprise rise |
| The score itself | 47. Below average risk. | above 65 |
| Reading | S&P 500 | Nasdaq | Window closes | What happened |
|---|---|---|---|---|
| 16 Jul 2026 | 45Guarded | 48Guarded | 16 Jan 2027 | open |
| 17 Jul 2026 | 45Guarded | 48Guarded | 17 Jan 2027 | open |
| 25 Jul 2026 | 46Guarded | 49Guarded | 25 Jan 2027 | open |
| 2 Aug 2026 | 45Guarded | 48Guarded | 2 Feb 2027 | open |
| 4 Aug 2026 | 45Guarded | 48Guarded | 4 Feb 2027 | open |
| 17 Aug 2026 | 43Guarded | 46Guarded | 17 Feb 2027 | open |
| 3 Sep 2026 | 39Guarded | 42Guarded | 3 Mar 2027 | open |
| 22 Sep 2026 | 41Guarded | 44Guarded | 22 Mar 2027 | open |
| 29 Sep 2026 | 43Guarded | 46Guarded | 29 Mar 2027 | open |
| 16 Oct 2026 | 48Guarded | 50Elevated | 16 Apr 2027 | open |
10 readings published since 16 Jul 2026, none with a closed window yet. The first closes on 16 Jan 2027. Each reading was published before its outcome was known, and a window closes six months after its reading. This is the one test of the score that no backtest can take for it.
The home page shows one more number beside the score: the chance of a 20% fall within six months, combining the score with the outlook. It is kept apart from the score because the two are different kinds of evidence, and it is for the S&P 500 only, because that is the index the score has been tested on.
The rule was written down before the first number. Take what followed months that scored like this one (the odds table above) and the outlook's own stated chance, and average them, half and half. Today: (6% + 25%) / 2 = 16%. The weights are fixed in model/spec/combined.json and change only with a version number.
The score's number is a measured frequency: of every month since 1928 that scored in today's band, the share followed by a 20% fall. The outlook's number is one reading of the things the score cannot see, tariffs, war, the Fed, AI spending, credit, market plumbing and politics, written under a fixed rubric and kept on record. Neither is adjusted to fit the other, and the average has no track record of its own yet. It becomes gradable when the first outlook window closes in March 2027, from the two public logs.
2000 was about overpriced stocks. 2008 was about too much borrowing. 1987 and 2020 gave almost no warning. No single sign catches all of them. Anyone promising a reliable crash alarm is selling something.
A sudden one-day crash
Warned early: stocks were expensive, had shot up fast, rates were rising, and fewer stocks were holding up.
Missed it: the bond market, jobs, and lending all looked fine.
The dot-com bubble bursts
Warned early: record-high prices, sky-high tech valuations, a narrow rally, a flood of new listings, a bond-market warning.
Missed it: stress in lending gave no early warning.
The debt & banking crash
Warned early: a bond-market warning, huge borrowing in housing and among non-bank lenders, and rising stress in lending.
Missed it: stock prices alone looked reasonable.
The COVID shock
Warned early: almost nothing. An outside shock hit an expensive, crowded market.
Missed it: jobs and factory data gave no lead.
Short, plain answers to the questions people ask most about a market crash. The numbers reflect the October 2026 reading.
The score is 47 out of 100 for the S&P 500 and 50 for the Nasdaq. Both are guarded. Stocks are very expensive and borrowing is near its record, which sounds bad. But lending stress is low, and that is the sign that actually warns you before a crash. Today looks safer than a normal month. That can change quickly.
About 6%. Of the 326 months since 1928 that scored where we are now, 6.1% were followed by a 20% fall within six months. A normal month is 13%, so today sits below the usual risk. The margin of error runs from 2% to 12%, and six months is a short window in which a lot can change.
Here, a crash means the market falling 20% or more from its recent high within the time period shown.
The biggest ones today are record-high stock prices versus company earnings, a near-record amount of borrowed money, and a market propped up by a few giant AI stocks. Calmer signs include low fear, a low jobless rate, and no stress in lending.
The fastest triggers to watch are a jump in lending stress, a spike in the fear gauge (VIX) above 25, fewer stocks holding up the market, a surprise interest-rate hike, or a disappointing AI earnings report.
Seven groups of signals each score 0 to 100, where higher is worse. They are combined into one number, weighted by how often each one warned before the last 14 crashes. Lending stress carries the most weight. Share price carries the least. We then look up what happened after every month in the last 100 years that scored the same way. The result is a base rate, not a forecast.