WILL MARKET CRASHCrash Risk Monitor

Outlook · October 2026

What the numbers cannot see. Tariffs, wars, the Fed, scored by judgment and kept on record.

The crash score on the front page reads only measured series, so a tariff, a war or a Fed surprise reaches it only after it has moved a price. This page scores those things directly: eight named factors, each rated 0 to 4 under a written rubric fixed before the first reading, with the reasons spelled out. It has no history and no calibration. Every reading is logged before its outcome is known, and the record below is the only test it gets.

Written 17 Oct 2026 by Claude Fable 5.1Data score as of 16 Oct 2026Rubric v1.0.0
Outlook scorejudgment · 0–100
0/100
Elevated

Eight factors, weighted and scored by hand under a fixed rubric. Same axis as the data score, no observed frequencies behind the band yet.

Chance of a 20% fall in the S&P 500 within six months, as judged with this reading: 25%. A normal month is 13%; months scoring like the data score today were followed by one 6% of the time.

Data score, S&P 500measured · 0–100 · as of 16 Oct 2026
0/100
Guarded

The front page's score that day: seven measured warning signs. Months like it were followed by a 20% fall 6% of the time. Nasdaq: 50 (Elevated).

Where they differ

What the data cannot see, and how much it matters

The data score says 47 for the S&P 500, Guarded, up four and a half points in a fortnight, and 50 for the Nasdaq, its first Elevated reading since June; most of the move is Friday's fear gauge, added by hand because the public series runs a day behind, and the rest is the Chicago Fed's weekly gauges tightening four weeks running. This page says 59, Elevated, and for the first time the two are converging rather than arguing. Where they still differ is what the data score cannot see at all: the price of CoreWeave's bonds, the redemption queue at a software lender, and a frontier lab's valuation being cut a quarter in a week. Its credit block reads a monthly bond-yield spread that has not moved, so a 55-basis-point widening in junk since September is worth two points to it. Its rates block went up this fortnight because long rates fell, by the same slope rule that sent it down when they rose. What has improved is genuine and the page says so: the plumbing passed its two tests, the Fed is priced to hold, and the strikes are paused. That is why the probability moved three points and not six. Where the two agree is on direction: both rose, and both now say the risk is above where it was a month ago. If the fear gauge holds above 20 and the hyperscaler prints disappoint in the last week of October, the data score will follow within a month and be right for the right reasons, and this page will only have been early.
The factors

Eight things the numbers are late to, scored 0 to 4

Ordered by weight. Each score is one of five fixed readings: 0 quiet; 1 background; 2 active; 3 acute; 4 shock under way. The lines under each say what would move it, so the next reading can be checked against this one.

AI spending and earningsweight 15%3acute

Raised from 2 to 3. For the first time the AI trade's debt repriced, not only its equity. CoreWeave, which has borrowed about $50 billion to build data centres for Microsoft and OpenAI, saw its bonds fall below 80 cents on the dollar with yields of 15-16% and credit-default swaps near 1,200 basis points; its shares and Nebius's lost between a sixth and a quarter on the week, and Gartner put out an estimate that 80% of such 'neoclouds' will not survive independently to 2030 (Reuters via Investing.com, Fortune, Investing.com, 16 October). The customers' numbers moved the other way from the valuations: OpenAI told investors its annualised revenue was about $50 billion at the end of September, not the $68 billion reported a fortnight earlier, because it books only its net share of partner sales, and Nvidia, Oracle and Broadcom fell 3-5% on the day (CNBC, Admiral Markets, 8 October); Anthropic reported third-quarter revenue of $16.8 billion, up 45% on the quarter after more than doubling the quarter before, and an adjusted profit, and some of its investors now want the November listing priced at $1.5 trillion rather than $2 trillion (Yahoo Finance, 15 and 16 October). The Nasdaq lost 4.5% on the week, its worst since June, the Philadelphia semiconductor index 11.4%, its worst since April 2025, with memory stocks down 15-24% on reports of Chinese progress. Two of the four raising triggers are close to met: an AI listing being repriced a quarter below its target, and a data-centre borrower's debt trading as distressed. Scored 3, acute, because there is now a plausible path to forced selling inside six months through the financing of the buildout; not 4 because no one has cut capex, the listing has not been pulled, and Nvidia has not guided down. The third-quarter prints from the hyperscalers in the last week of October are the test.

Up if: CoreWeave or another AI borrower misses a payment, restructures or fails to place debt; a hyperscaler cuts capex or reports cloud growth well below guidance; Nvidia guides down; the Anthropic listing is pulled or priced under $1.5 trillion.Down if: Third-quarter prints show AI revenue growing faster than AI capex at two or more of the big four; CoreWeave's bonds back above 90; the Anthropic listing prices at or above $1.5 trillion and trades up.
The Fed's pathweight 15%3acute

Unchanged at 3, after a fortnight in which the October decision was priced three different ways. The September minutes (7 October) had most participants expecting another rise 'by year end'. September payrolls of 29,000 against 90,000 expected, with unemployment at 4.2% (2 October), took the October odds to about 24%; the CPI of 14 October, 3.7% headline and a hot 0.3% core, took them to 51.5% in a day; then Warsh, at the IMF meetings on 16 October, refused to prejudge October or December, said of guidance 'we're not in that business anymore', praised recent tightening abroad, and the odds fell to 15-17% (InvestmentNews, 14 October; CNBC, Axios, Investing.com, 16 October). None of the three raising triggers fired: the 10-year sold at 5.30% on 7 October, its highest since 2000, but with 2.77 times cover and 80% to indirects, and the 30-year's 5.618% on 8 October was as well bid; neither CPI nor payrolls left October above 60%. The lowering trigger of a 10-year under 5% came within a tenth: 5.08% on 16 October from 5.31% on 5 October, as money left tech for bonds. Held at 3 because the committee's own stated path is another hike and the chair would not rule it out six days before an election, with headline inflation at a five-month high. It goes to 2 if October passes with a hold and the 10-year holds under 5%.

Up if: A hike on 28 October against the pricing; the 30-year through 5.75% or the 10-year through 5.5%; a tail of more than 3 basis points at the November refunding; September PCE on 30 October above 3.5% core.Down if: October passes with a hold and no guidance for December; the 10-year under 5% for two weeks; a one-and-done signal from Warsh.
War and geopoliticsweight 15%3acute

Unchanged at 3, with the raising trigger from the last reading fired and a reason not to raise. The attacks on shipping came back: ten vessels were struck in the week to 14 October, the Singapore-flagged container ship GFS Pride sank on 13 October and the Cape Dao, hit on 23 September, sank the next day; the Houthis put a missile into the top floors of Aramco's headquarters in Dhahran on 14 October and were intercepted over Riyadh (gCaptain, 14 October; Anchorage Daily News, 14 October; Fox News, 14 and 15 October). Brent is about $106 and WTI about $98, up 3% and 9% from the last reading and still below the 28 September spike (Trading Economics, 16 October). The reason the score is not 4: on 8 October Trump wrote that the United States 'will not be attacking Iran at any time prior to the Midterm Elections', said talks were 'productive', and Tehran has a U.S. counter-proposal it says it will answer within days, while a third carrier arrives in the region by the end of October (WLOX/AP, 8 October). Prices are moving, but on shipping losses rather than on a new front, and the one dated commitment on the table is a pause. Held at 3, direction up: the war is being fought on tankers now.

Up if: Brent through $110 or WTI through $105; the pause is broken by either side before 3 November; Saudi or Emirati export infrastructure is hit, not just offices; Hormuz transits fall below 50 a day.Down if: Iran accepts the counter-proposal with a verified reopening of Hormuz; Brent under $90 for a month; the pause holds through the election and becomes a ceasefire.
Trade and tariffsweight 15%2active

Unchanged at 2. The Busan truce runs to 10 January 2027, and the fortnight since the last reading produced complaints rather than actions: Bessent has said China is about $17 billion short on its farm-purchase commitment and that rare-earth shipments have not returned to agreed levels, which is why the extension was two months rather than the six Washington asked for; Beijing said on 15 October that rare earths will keep flowing to Europe whatever happens to the U.S. deal (CNBC, 15 October; Startup Fortune, October). The presidents are reported to meet again in Shenzhen in November and Miami in December, so neither side wants the truce to lapse before then (Manila Times, 12 October). The pharmaceutical tariff is in force and has not moved prices. Scored 2 because every deadline is scheduled and priced; 3 needs a new control or a lapse.

Up if: The 10 January expiry passes without a deal or a further extension; new chip or rare-earth controls; a fresh sectoral tariff on semiconductors or autos; the November meeting is cancelled.Down if: The product lists take effect and a longer extension is written before January; the pharmaceutical tariff is narrowed or its exemptions widened; a Canada deal rolls back the 50% metals rates.
Credit and private lendingweight 10%2active

Unchanged at 2, with the raising trigger touched by a repeat rather than a new name. High-yield spreads are 323 basis points, from 308 at the last reading and 268 on 22 September, with 325 on 14 October the widest since April; investment grade is flat at 84 (FRED). That is 27 short of the 350 line and nowhere near 100 in a month. Blue Owl again capped withdrawals at 5% from its two retail credit funds after investors asked for 39% of the technology fund and 17% of the income fund in the third quarter, and Blackstone held its 5% cap with about 10% requested (Bloomberg, Benzinga, 2 October): the same funds that gated in the second quarter, which the last reading already counted, with the requests still rising at the one most exposed to software. Bank results were mixed in the right places: JPMorgan missed for the first time in 17 quarters and added $539 million to reserves, more against corporate loans and less against cards, where charge-offs fell; Wells Fargo's provision rose to $1.03 billion; Goldman's trading made the quarter (JPMorgan, 13 October; Investing.com, 14 October). Fitch's private-credit default rate stands at a record 6.1%. Scored 2 because no lender has taken a mark and no new fund has gated; 3 if spreads pass 350, a bank writes down private-credit exposure, or CoreWeave becomes a default rather than a price.

Up if: High-yield spreads through 350 basis points or 100 wider in a month; a fund that has not gated before gates, or a BDC cuts its dividend on marks; a bank writes down private-credit or data-centre exposure; a large high-yield deal is pulled.Down if: Spreads back under 290 for a month; redemption requests at the gated funds fall for two quarters; the hiking cycle stops at one.
Fiscal and tax policyweight 10%2active

Unchanged at 2, and the factor that improved most without a change in score. The continuing resolution runs to 11 December, so there is no shutdown inside the quarter, and the auctions the raising trigger watches were the strongest of the year: the 10-year at 5.300% drew 2.77 times cover with 80.3% to indirect bidders, the 30-year at 5.618% drew 2.54 against a 2.41 average with dealers left holding 6.8% against 10.3% (The Vault Report, Vantage Markets, 9 October). Buyers wanted the highest yields since 2000. Since then the 30-year has come back to 5.43%. Government debt is still named in bond coverage alongside the data-centre borrowing that competes for the same money, and tariff receipts still run well above last year. Scored 2 because the deficit is being financed in an orderly way at a price; 3 needs a failed auction or a post-election fight over the 11 December bill, which a divided result on 3 November would make likelier.

Up if: A tail of more than 3 basis points at a 10- or 30-year auction; a post-election Congress that will not pass a clean extension by 11 December; a fresh unfunded tax cut; the 30-year back through 5.75%.Down if: Full-year appropriations pass in November; the November refunding clears like October's; a credible deficit-reduction package.
Politics and institutionsweight 10%2active

Unchanged at 2, with words but no act. Trump told Time on 1 October that Warsh 'should have voted against' the September hike and 'I probably would have voted against the board if I were him', a step up from backing him personally while calling the committee political, and the committee's next decision is on 28 October, six days before the midterms (Bloomberg, Quartz, 1 October). Warsh's answer on 16 October was to refuse guidance in either direction. No move against a governor, no legislation, no legal action. The president's pause on Iran strikes until after the election is a reminder that the calendar is now organising policy. Scored 2 because the conflict is open and the pre-election decision is the obvious flashpoint; 3 needs an act, not a post. It goes to 1 if October passes without a White House campaign against the decision and the election produces a clear result.

Up if: An attempt to remove or override a Fed governor or to change the committee's structure; a Treasury statement rejecting Fed policy in policy terms; a disputed midterm result; a rule-of-law shock that moves the dollar.Down if: The October decision passes without a White House campaign against it; the midterms produce a clear result; the White House and the Fed settle into public silence.
Market plumbing and positioningweight 10%1background

Lowered from 2 to 1, as the last reading said it would be if the quarter-end and the October auctions both passed quietly. They did. The 30 September settlement of $202 billion went through with SOFR at 3.88%, under the rate the Fed pays on reserves, and it has sat there since; the standing repo facility was not needed (SOFR Rate, 14 October). The 10-year and 30-year auctions were the best bid of the year with dealers taking the smallest shares. The basis trade has kept shrinking, to about $900 billion from $1.26 trillion at the start of the year, concentrated in the two- and five-year, an unwind in the orderly direction (Hedgeweek, October). The one stress test this factor watches that did happen, a 20-basis-point rally in the 10-year on a tech sell-off, passed without a reported margin call or a swap-spread break. Scored 1, background: the mechanism is present and discussed, but nothing is scheduled inside six months that would test it, and the positions that would be tested are smaller than they were. It goes back to 2 if the November refunding leans on dealers or the VIX holds above 20 into the Fed meeting.

Up if: SOFR or repo rates spike relative to the funds rate; CME raises Treasury futures margins; swap spreads or the basis move sharply in a day; the November refunding tails; the VIX above 25 forces a volatility-targeted unwind.Down if: The November refunding clears like October's; the basis trade keeps shrinking; the VIX back under 16.
Method

How this score is made

A rubric written before the first reading. Eight factors with fixed weights, registered 16 Sep 2026 as version 1.0.0, and one five-point scale. The composite is the weighted mean of the eight scores times 25, so it runs 0 to 100 on the same axis and the same band thresholds (35, 50, 65, 80) as the data score. Unlike the data score, nothing behind these bands has been measured against a century of history: the bands are a reading aid, not odds.

0quietnothing live; the factor is not in the news for a reason that matters to prices
1backgroundpresent and discussed, no concrete action or deadline inside six months
2activeconcrete action taken or scheduled, markets have started to price it
3acuteescalating, with a plausible path to forced selling inside six months
4shock under waythe event is happening and prices are already moving on it

Weights. AI spending and earnings 15%; The Fed's path 15%; War and geopolitics 15%; Trade and tariffs 15%; Credit and private lending 10%; Fiscal and tax policy 10%; Politics and institutions 10%; Market plumbing and positioning 10%.

The probability line is a separate judgment. It starts from the base rate and the data band's observed rate, and the reasoning for where it lands is written with each reading: Start from 13%, the share of all months since 1928 followed by a 20% fall within six months. The data score today is 47 for the S&P 500, Guarded, and months like it were followed by a fall about 6% of the time; the Nasdaq's score crossed into Elevated at 50 on Friday's volatility, and Elevated months were followed by a fall 14% of the time, about the base rate. I raise the chance from 22% to 25%. One factor went up and one went down, so the composite is 58.8 from 57.5; the three points come from what changed in kind. Two weeks ago the things the eight factors do not score were breadth, spreads, the long bond and confidence. Breadth recovered to the line, the long bond rallied 20 basis points and spreads widened another 15, so that list is about even. What is new is that the AI buildout's debt was repriced as distressed for the first time, the Nasdaq had its worst week since June with the fear gauge at 20 and semiconductors down 11%, and the labour market stalled with 29,000 jobs and 4.2% unemployment while inflation went back to 3.7%: the mix in which a central bank cannot cut into a sell-off. Pulling the other way, and why this is 25 and not 30: the quarter-end and the auctions were clean, the Fed is now priced to hold, the 10-year is nearly back under 5%, the president has taken military escalation off the table until 3 November, and Anthropic reported an adjusted profit. Most six-month windows that look like this still do not end in a 20% fall, and the window for grading this reading closes on 17 April 2027.

What this score never does. It never feeds the data score, and the data score never reads it. The front page's method, spec and record are untouched. A reading is never edited after publication; a change of view is a new reading with a new date, and both stay on the record.

Who writes it. The author named on each reading, from the sources listed below, on the day of a data refresh or on the day of a large event. A change of author is visible in the record.

The record

Every outlook reading, and what happened next

The record so far
ReadingOutlookP(20% fall)Data score that dayWindow closesWhat happened
16 Sep 202658Elevated20%39Guarded · 6%16 Mar 2027open
23 Sep 202658Elevated20%41Guarded · 6%23 Mar 2027open
30 Sep 202658Elevated22%43Guarded · 6%30 Mar 2027open
17 Oct 202659Elevated25%48Guarded · 6%17 Apr 2027open

4 readings published since 16 Sep 2026, none with a closed window yet. The first closes on 16 Mar 2027. Each reading was published before its outcome was known, a window closes six months after its reading, and a 20% fall means the same thing it means for the data score. The Brier score is the average squared gap between a forecast probability and what happened, so a forecast that said 10% before a fall pays more than one that said 40%.

Sources

What this reading was written from