Diffusion CapitalDiffusion Capital
Back to insights

Research · October 9, 2026

What will move markets in Q4 2026

Earnings can carry the S&P 500 through Q4 if AI spending keeps turning into profits, but the index rests on a few AI companies and has little cushion against a ten-year Treasury above 5%. The main risks are Fed tightening in response to supply-driven inflation, the AI labs' revenue and France.

Maximilian Ruess

Research

What will move markets in Q4 2026

A4 PDF · 6 pages · 657 KB · 9 October 2026

Download PDF

A record with a third of stocks in a bear market

The S&P 500 closed at a record on 6 October, but most of the stocks in it are not taking part in the rally. Only 30.1% trade above their 50-day average, a common measure of short-term trend, and 34.5% are in a bear market, at least 20% below their 52-week closing high. On the 352 record closes since 2016 that share averaged 14.9%, and only 8, all in August 2020, had more. In the third quarter the typical stock returned −4.0%. The gain came from a few large companies, almost all tied to the AI trade: Microsoft, Nvidia, Apple and Meta produced 192% of the Q3 gain in our index reconstruction, so without them it would have fallen. The ten largest companies now make up 40.4% of the index's market value, more than the 26.6% they held at the peak of the 2000 tech bubble, and Nvidia alone makes up 7.9%.

The index sets a record while most stocks lag

S&P 500 as % of its record close; % of members above their 50-day average

Over the past year the S&P 500 stayed within about ten per cent of its record and ended at a record on 6 October, while the share of members above their 50-day average fell from 69 per cent at the August record to 21 per cent on 30 September and stood at 30 per cent on 6 October.

S&P 500, % of record closeMembers above 50-day average, %
020406080100%Jan 26Apr 26Jul 26Oct 26

Members as of each date; a company with two share classes counts once. One year to 6 October 2026, when the index closed at a record (highest close since 1950).

Source: Financial Modeling Prep - S&P 500 member daily prices, profiles, statements and consensus, Financial Modeling Prep - daily index, ETF, commodity and FX closes and the US Treasury par curve for the Q4 2026 outlook · 2026-10-09

Four stocks delivered nearly twice the index's gain

Contribution to the S&P 500's Q3 2026 total return, percentage points

Microsoft added 1.5 points, Nvidia 1.0, Apple 0.9 and Meta 0.6, a combined 4.1 points or 192 per cent of the gain, while the other 496 members subtracted 2.0, leaving the index up 2.1 per cent in the third quarter.

0.01.02.03.04.05.0Percentage pointsMicrosoft+1.5Nvidia+1.0Apple+0.9Meta+0.6Other 496 members-2.0S&P 500, Q3 total+2.1

Members on 30 June weighted by full market value, dividends reinvested. EA and AvalonBay, acquired in August, are valued at their last close as members.

Source: Financial Modeling Prep - S&P 500 member daily prices, profiles, statements and consensus · 2026-10-09

We expect volatility to rise in Q4

Over three months the ten largest companies moved about 1.4% for every 1% move in the index, while the other 490 moved only about 0.7%, so the leaders now set the index's direction. Their correlation with the rest is 0.04, against a ten-year average of 0.26. Cboe's index of the correlation priced into options on the largest stocks stands at 11.5, in the lowest 2% of days since 2006. Since 2006 the index fell 10% or more in 13% of the three-month spells that began with implied correlation in its bottom historical tenth, against 21% at other times. The VIX rose five points or more within three months after 82% of them, against 64% after other spells with a VIX as low. All 3 completed episodes were followed within three months by a drawdown of 10% or more; their endpoints are known only in hindsight. This one, at 322 trading days, is the longest yet. We expect volatility to rise in Q4, but a bear market is not our base case. We would retain equity exposure and buy protection against a fall.

The ten largest earn close to their share of value

At the 2000 peak the ten largest companies made up 26.6% of the index's value but earned only about 15.9% of its profits. At the end of September they made up 40.2% of the value and earned 35.6% of the profits, a premium of only 4.6 percentage points, below its average since 2017. Their share of profits has risen from 21.4% at the end of 2017, almost as fast as their share of market value. Analysts expect them to earn 37.6% of next year's profits. The comparison suggests that today's concentration is better supported by earnings than it was in 2000.

AI spending still pays, but breadth stays narrow

Companies making up 46.5% of the index's value are expected to report in the week of 26 October. Their growth depends on continued demand for AI. Google and OpenAI disclosures through July imply annualised token-volume growth of 4.9x to 6.2x, though token counts also rise with more computation per task. Returns on new capital remain above the cost of capital, but have fallen. The five biggest spenders raised capital spending 81% in a year. Each extra dollar of capital earned 17%, down from 31% a year earlier, against our estimated cost of capital of about 9.7%. If the annual profit increase stays unchanged, their spending plans imply about 11% on new capital and insufficient cash to restore buybacks without borrowing. We do not expect the market to broaden this quarter or next. Since 2016 the equal-weighted index has beaten the cap-weighted one in only 28% of quarters, and in only 9% of those that began with the ten largest at a record weight. We expect broader market gains later, as companies using AI cut costs.

Valuations are high on past profits but close to average on forecasts

The index trades at 41.6 times its average inflation-adjusted earnings over the past ten years, a level exceeded only in 16 months of the dot-com boom. Its forward valuation is closer to its historical average, at 19.3 times expected earnings, approximately the 71st percentile of months since 1995. Since 1996 the forward P/E has explained almost none of the variation in three-month returns (R² 0.00), while changes in the multiple have accounted for 76% of quarterly swings, against 24% for earnings. This year earnings growth explains much of the index's rise. Earnings rose +25.5% while the multiple moved -9.0%. The ten largest companies trade at 26.8 times trailing earnings, below their decade average of 29.6 times, while the other 490 trade at 22.2 times, above their average of 19.3 times. Yet analysts expect the other 490's profits to grow +20.7% next year against +29.0% for the ten largest.

Ten largest below their average P/E, the rest above

Trailing P/E of the ten largest S&P 500 members and the other 490, month-end

Since 2016 the ten largest companies' trailing P/E has ranged widely and now stands at 26.8, below its 29.6 decade average, while the other 490 trade at 22.2, above their 19.3 average.

Ten largestTen largest avgOther 490Other 490 avg
10152025303540Times earnings201620182020202220242026

Point-in-time members, operating earnings over the last four quarters. Dashed: each group's average, October 2016 to September 2026. Last point 6 October 2026.

Source: Financial Modeling Prep - S&P 500 member daily prices, profiles, statements and consensus, Financial Modeling Prep - per-symbol earnings history and S&P 500 constituents for the Q4 2026 outlook · 2026-10-09

Q2 beats were broad, while 2027 forecasts rely on AI

Q2 earnings grew 33.9% on a year earlier, excluding large one-off gains and charges, against 22.8% expected before companies reported. 86% of companies beat the earnings forecast and 77% the revenue forecast. Outside the 29 AI companies earnings grew 24.2% against 12.9% expected. Analysts raised their Q3 estimates by 1.4% during the quarter, according to FactSet, against an average cut of 2.2% over the past five years. Energy and technology led the estimate increases, while eight of the 11 sectors had their estimates cut. We expect Q3 growth in the middle of the 32.7% to 37.5% range implied by the last eight quarters' earnings surprises, with larger surprises outside AI. Outside AI consensus growth is 17.5% in Q3, 18.7% in Q4 and 12.4% in 2027, so the AI group's share of the expected increase rises from 52% in Q3 to 64% in 2027. Since June, forecasts for 2027 have risen +4.8%, with Nvidia accounting for 47% of the increase, and only +1.5% outside AI and energy. The 2027 forecasts also assume the net margin widens from 15.0% to 17.0% while revenue outside AI grows 4.8%. On the results calls, we will look for capital spending plans for 2027 and evidence that AI is lowering costs at companies outside the AI trade. On the July and August calls 69% of them mentioned AI, but only 11 put a number on what it did for their costs or earnings.

Consensus has growth outside AI slowing in 2027

S&P 500 EPS growth on a year earlier, Q2 2026 reported and today's consensus

Q2 2026 earnings grew 62 per cent for the 29 AI companies, 24 per cent for the rest and 34 per cent for the index. Consensus has the AI group at 52, 49 and 49 per cent for Q3, Q4 and 2027, the rest at 18, 19 and 12 per cent, and the index at 27, 28 and 24 per cent.

AI complex (29)Outside AI (~470)S&P 500
0%10%20%30%40%50%60%70%Year-on-year growthQ2 26 reportedQ3 26Q4 262027

AI complex: 29 builders, suppliers and power firms. 6 Oct members and share counts, large one-offs replaced. 2027 is fiscal years. Q4 is partly derived from annual forecasts.

Source: Financial Modeling Prep - per-symbol earnings history and S&P 500 constituents for the Q4 2026 outlook, Financial Modeling Prep - S&P 500 member daily prices, profiles, statements and consensus · 2026-10-09

We expect the S&P 500 to rise about 4% in Q4

Our model puts the S&P 500 at +4.3% above its 6 October close at year-end, mainly from earnings. Forward earnings rose +21.1% over the past two quarters. After two-quarter gains above 8%, the next quarter's median gain was +3.7%, supporting an earnings contribution of +3.5%. From 5.27%, the ten-year yield has been about as likely to rise as fall. Multiple changes contribute little to average returns but 86% of simulated variance, with a 31% loss probability and a one-in-ten outcome of −7.2% or worse. Since 2022, a 25bp yield rise has been associated with a −2.3% multiple change. Our 4.75% yield scenario gives a +8.5% median return; 5.77% gives +0.3%. Across 83 historical forecasts, 83% of returns fell within the 10th–90th percentile range, but direction forecasts did no better than historical frequencies. In 14 periods since 2018 with positioning like today's—trend-followers long, few puts traded, low VIX and expensive single-stock options—the worst subsequent three-month close averaged −7.2%, against −4.6% generally. A 10% earnings shortfall and a multiple one point lower would imply -14.7% for the index.

Earnings add about 3.5% whichever way yields go

Q4 S&P 500 price return by ten-year path; average contribution of each part, %

If the ten-year falls to 4.75 per cent (34 per cent odds), earnings add 2.8 points and the multiple 5.1, for about 8 per cent. If it holds near 5.3 per cent (31 per cent), earnings add 3.5 and the multiple 0.3. If it rises to 5.8 per cent (35 per cent), earnings add 4.1 and the multiple takes away 4.3, leaving the index roughly flat.

EarningsMultiplePrice return
-6.0-4.0-2.00.02.04.06.08.0% contributionYields fall (34%)Yields hold (31%)Yields rise (35%)

Our simulation, 20,000 draws, 6 Oct to 31 Dec. Yield paths and odds from quarters since 1962 that began near 5.3%. Bars are averages; medians in the text run slightly higher.

Source: J.P. Morgan Asset Management - Guide to the Markets U.S. (4Q 2026), Financial Modeling Prep - daily index, ETF, commodity and FX closes and the US Treasury par curve for the Q4 2026 outlook, FRED series in the Q4 2026 outlook snapshot (Federal Reserve Board, BLS, BEA, EIA, ICE and OECD series) · 2026-10-09

Chip designers are priced for less growth than the rest of the AI supply chain

We value the S&P 500's AI suppliers and spenders at an 8.8% discount rate. At that rate, a company with flat profits would trade at about 11.3 times expected earnings. Nvidia, Broadcom, AMD and Marvell trade at 19.8 times, which implies 1.8 years of growth at the 39% a year analysts expect, although their 2028 forecasts have risen +24% since June. At that growth rate, the price implies profits reaching 1.84 times today's expected level, about the same as for the index (1.86 times). Other suppliers are priced for longer periods of growth: chip equipment for 8.3 years, taking earnings to 3.83 times their forward level, networking and optics for 6.2 years, and power and cooling equipment for 6.7 years. Memory and storage makers trade at 6.8 times, below the value of flat profits, so their price assumes earnings fall 40%, which is common near the peak of a memory cycle. With a higher discount rate for the more volatile chip shares, power and cooling equipment is still priced for 4.2 years more growth than the chip designers. We favour the chip designers into the late-October results if the largest AI spenders issue guidance above consensus. Our model links chip earnings forecasts to customer capital spending growth of about +38% in 2027, in line with the consensus of about 38% that Goldman Sachs cited in late September. That consensus was up from 22% in June, while Goldman itself expects 54%.

Implied years of growth along the AI supply chain

Years of growth at consensus pace that today's price implies, 8.8% rate

At an 8.8 per cent discount rate the chip designers' price implies 1.8 years of growth at 39 per cent a year and memory and storage none, against 8.3 years for chip equipment, 6.2 for networking and optics, 6.7 for power and cooling equipment, 3.3 for power producers, 2.6 for servers, 4.9 for the five big spenders, 5.7 for other chips and 5.0 for the other 490.

0.02.04.06.08.010.01.8 yrs at 39%none; prices a 40% fall8.3 yrs at 18%6.2 yrs at 23%2.6 yrs at 19%6.7 yrs at 21%3.3 yrs at 21%4.9 yrs at 18%5.7 yrs at 23%5.0 yrs at 12%Years of growthChip designersMemory and storageChip equipmentNetworking, opticsServersPower, cooling gearPower producersFive big spendersOther chipsOther 490

Consensus net income growing at each group's 2026-29 pace for N years, then flat. Memory trades below the value of flat profits. S&P 500 members; rate: 10y yield + 3.5%.

Source: Financial Modeling Prep - S&P 500 member daily prices, profiles, statements and consensus, Financial Modeling Prep - daily index, ETF, commodity and FX closes and the US Treasury par curve for the Q4 2026 outlook, FRED series in the Q4 2026 outlook snapshot (Federal Reserve Board, BLS, BEA, EIA, ICE and OECD series), Robert Shiller - Irrational Exuberance online data (ie_data.xls) · 2026-10-09

The Fed drives the ten-year; supply drives the thirty-year

The ten-year Treasury yield hit 5.31% on 5 October, its highest since 2002. A Fed staff model attributes +59 bp of this year's +113 bp yield rise to expectations of higher Fed rates. It attributes +51 bp to the term premium, the extra return investors require for holding long bonds.

In our model, no further Fed hike leaves the year-end ten-year yield near 5.01%, and if it hikes twice, near 5.51%, against 5.27% on 6 October. Yields also move about 36bp in a typical quarter on days without scheduled news, so the two outcomes overlap. The thirty-year reacts much less to the Fed and more to bond supply. It ends about +20bp above 6 October in the larger-auction scenario and +19bp higher in the Japanese bond sell-off scenario.

At the ten-year auction on 7 October, a day after the cut-off for our other figures, investors bid 2.77 times the $39bn on offer, the most since January 2016, and dealers had to take only 2.5%, the smallest share in 215 auctions since 2009. At about 5.3% the Treasury can sell its current auction sizes, although strong auctions have in the past lowered yields for only a day.

Treasuries now do less to protect a portfolio of shares. Most of our cases raise yields, and the main exception, a flight to safety if AI spending disappoints, has worked less often since 2022. When shares fell 5% or more, the ten-year yield fell in 72% of cases from 1990 to 2021 but in only 29% since 2022, when equity sell-offs more often coincided with rising yields.

Yields and the S&P 500 in seven Q4 scenarios

Median change from the 6 October close to 31 December

A table of seven year-end scenarios. With no hike (26% market odds) the ten-year falls a median 26bp (4.57-5.49% range), the thirty-year 17bp and the S&P 500 rises 6.2%. One hike with steady auctions (58%) lifts the ten-year 7bp and the thirty-year 3bp, with shares up 3.5%; two hikes (17%) add 24bp and 13bp, shares up 1.8%. The conditional cases: larger auctions +22bp and +20bp; a flight to safety after an AI-spending miss -17bp and -9bp with shares down 10.0%; a JGB sell-off after the 30 October BoJ meeting +18bp and +19bp; Brent at $130 with firm core CPI +25bp and +17bp.
Odds10Y chg30Y chgS&P 500
No hike26%−26bp−17bp+6.2%
One hike, steady auctions58%+7bp+3bp+3.5%
Two hikes17%+24bp+13bp+1.8%
One hike, larger auctions–+22bp+20bp+2.3%
Flight to safety (AI miss)–−17bp−9bp−10.0%
JGB sell-off (30 Oct BoJ)–+18bp+19bp+2.7%
Brent $130, firm CPI–+25bp+17bp+1.8%
Fed-row odds are Polymarket and Kalshi at 6 October; the other rows are conditional cases with no market price. Auction sizes are our assumption. The AI row includes its shock.

The calendar points to a December hike, not an October one

At the 6 October close, Polymarket and Kalshi put the chance of a hike on 28 October at 17% and 18%, respectively. For 9 December, their probabilities were 76% and 72%. September CPI on 14 October is the only major release before the October meeting. Unless it comes in well above forecast, we do not expect the Fed to hike six days before the 3 November midterms, when a hike would look political. By 9 December the Fed will have two more payroll reports and October's inflation figures. In our scenarios, a December hike leaves the ten-year yield +36bp and the thirty-year yield +23bp higher at year-end than a hold would. A 25bp rise in the Fed's median 2027 rate projection adds only +4bp. This year each 10bp rise in the ten-year yield has been associated with an S&P 500 move of about −0.8%. We do not model two term-premium risks: the 3 November midterms, which bear on 2027-28 deficits more than on this quarter's $628bn of borrowing, and changes to Fed Treasury holdings, $4.35tn in the March staff-paper baseline.

Markets moved the next Fed hike to December

Probability of a hike at each FOMC meeting, average of Polymarket and Kalshi, %

The probability of a hike at the 28 October meeting rose from about 25 per cent in August to 69 per cent on 28 September, then fell to 17 per cent by 6 October. Most of the fall came on 29 September, when New York Fed President John Williams said there was no need for urgency; soft PCE and payroll data followed. The probability of a hike on 9 December rose from about 30 per cent to 78 per cent and stood at 74 per cent on 6 October.

December meetingOctober meeting
020406080100%Warsh at Jackson HoleWilliams: no urgencyPayrolls +29kSep 26Oct 26

New York 16:00 closes, 3 August to 6 October 2026, outcome prices scaled to sum to one. Lines mark speeches by Warsh (Jackson Hole) and Williams, and September payrolls.

Source: Polymarket - Fed Decision in October, December and January (FOMC 28 Oct 2026, 9 Dec 2026, 27 Jan 2027), Kalshi - KXFEDDECISION-26OCT, KXFEDDECISION-26DEC, KXFEDDECISION-27JAN and KXFED-26DEC markets, Federal Reserve Bank of New York - John C. Williams, Unwavering Dedication (remarks at the University at Buffalo, 29 September 2026) · 2026-10-09

Crude is getting through, but diesel is short going into winter

Gulf crude exports have recovered more than shipments of diesel and other refined products. Kpler tracked 10.1 mb/d of crude through the Strait of Hormuz in the week to 6 October, 74% of its pre-war rate, while exports via the Red Sea and Gulf of Oman partly offset the shortfall. Saudi Arabia's East-West pipeline, the largest of them, was hit by drones in September and was back above 80% of capacity by 5 October. Refined products through the strait have fallen to 0.7 mb/d from 3.6 mb/d, because damaged Gulf refineries run well below capacity, and Kpler does not expect them to recover fully before Q2 2027. Refiners elsewhere are paying up for crude to make diesel, so Dated Brent, the price of crude loading within weeks, averaged $114 in September against $101 for futures on 6 October. The G7 agreed on 2 October to release up to 100 mb of emergency stocks over four months, diesel first, but the US Strategic Petroleum Reserve already holds the least crude since 1982.

President Trump has said strikes on Iran could resume after the 3 November midterms if there is no deal, and a third US carrier group is heading to the region. Polymarket gives a US-Iran agreement on the strait by the end of the year a 37% chance. We expect a stalemate through Q4, with Brent near $100, close to futures and below the EIA's $105, set before the G7 release. A deal that lifts the US blockade of Iran's ports would take Brent towards $80, although diesel would stay short until the Gulf refineries are repaired. Renewed strikes would take it towards $130, and Iran's fractured leadership leaves room for factions to attack ships during talks.

US diesel stocks have been below normal since May

US diesel and heating oil stocks, million barrels; shaded: 2021-2025 range

US stocks of diesel and heating oil were near their 2021-2025 average until March, fell below the lowest level of those years for the same week in May and stood at 105 million barrels on 2 October, 11 per cent below average.

2021-2025 average2026
100110120130140150160170Million barrelsWar beginsAprJulOct

EIA data to 2 October 2026. The dotted line is the 2021-2025 average for the same week; on 2 October stocks were 105 mb, 11% below it.

Source: US EIA - Weekly Petroleum Status Report history workbooks (SPR, commercial crude, Cushing, gasoline, distillate, jet fuel stocks; refinery runs, product supplied, crude and distillate trade) · 2026-10-09

Fuel costs have risen alongside freight prices and airfares

New York Harbor diesel doubled from $105 a barrel in February to $210 at the end of September, and US diesel stocks were -11% against their 2021-2025 average on 2 October. Prices have eased a little since the G7 release, and futures price a gradual fall through the winter. We expect the gap between diesel and crude to stay above $70, twice its February level, until Gulf refineries return in spring. Trucking prices rose 2.0% in August, and airfares were 23% higher than a year earlier, consistent with the pattern after past diesel price rises. The Cleveland Fed's nowcast of 0.2% a month for core CPI in September and October suggests limited broader inflation pressure so far. The Fed raised rates by 25bp in September, and the minutes warn that the longer energy prices stay high, the greater the risk that costs spread to other prices. Markets give a December hike about three chances in four, so the risk for shares is that persistent diesel inflation prompts further hikes in 2027. The September and October CPI reports, both out before the 9 December meeting, will show whether other prices are following.

Diesel has risen much further than crude this year

New York Harbor diesel (dark) and Dated Brent (red), US$ per barrel, since 2025

New York Harbor diesel and Dated Brent in US dollars per barrel from January 2025. Through 2025 diesel ran about 25 to 35 dollars above Brent. After the war began in late February 2026 both rose, Brent to 138 dollars in April and diesel to about 190. Brent fell back to the 70s and 80s in June and July while diesel stayed near 160 to 180, so the gap widened. In September Brent rose to 114 dollars and diesel to 210, a gap of 96 dollars on 29 September.

50100150200250US$ per barrelWar beginsAprJulOctJanAprJul

The shaded gap is what diesel costs above crude. It was $96 on 29 September, above 99% of days in 2016-2025, against $34 in February and $117 at the 2022 peak.

Source: FRED series in the Q4 2026 outlook diesel snapshot (EIA and BLS series), FRED series in the Q4 2026 outlook snapshot (Federal Reserve Board, BLS, BEA, EIA, ICE and OECD series) · 2026-10-09

France's debt path is the problem, and the 2027 election is the test

French bonds are falling because France keeps running large deficits, and pensions are part of the reason. French men leave work at 61.9 on average, against 64.7 across the OECD. The 2023 reform that was to raise the pension age to 64 has been paused. The government now expects this year's deficit to reach 5.4% of GDP, more than last year's 5.1% and the 4.6% it first planned. For 2027 it plans a deficit before interest of 2.1% of GDP, which needs €54bn of spending cuts and tax rises that it lacks the votes in parliament to pass. Even if the plan holds, we calculate that France would borrow 1.9 points of GDP more than it can without its debt growing faster than its economy. The calculation uses the low average interest rate on its existing bonds. As old bonds are replaced with new ones at today's ten-year yield, France would need a surplus before interest of 2.4% of GDP.

The ECB can intervene and buy a country's bonds to stop yields rising without good reason, through its Transmission Protection Instrument, but only if the country's debt is sustainable and it follows EU fiscal rules. Scope cut France's rating to A+ on 18 September. The extra yield France pays over Germany closed above 100bp on 24 September for the first time since 2012. Moody's reviews its rating on 23 Oct, and the National Assembly votes on the budget on 17 Nov 2026. Both front-runners in the April 2027 presidential election worry bond investors. Marine Le Pen leads the polls at 35% and wants to return the pension age to 62. Jean-Luc Mélenchon has proposed cancelling the 18% of French debt held by the Banque de France, which EU treaties forbid.

France pays the most over Germany since 2012

French minus German 10-year government yield, daily, basis points

The French spread over Germany rose from about 50 basis points in early 2024 to about 80 after the June 2024 snap election, then from 78 to 115 in September 2026 to 144 on 2 October, the widest since January 2012, before 123 at the 6 October fixing.

406080100120140160Basis pointsSnap electionApr 24Jul 24Oct 24Jan 25Apr 25Jul 25Oct 25Jan 26Apr 26Jul 26Oct 26

Banque de France TEC 10 (11:00 fixing) minus the Bundesbank 10-year. Up 45bp since 31 August after a 144bp peak on 2 October; the daily record is 214bp, November 2011.

Source: Banque de France Webstat - French government bond constant-maturity rate, 10 years (TEC 10), daily, Deutsche Bundesbank - yields derived from the term structure of Federal securities, 10-year residual maturity · 2026-10-09

Healey needs about £12bn to restore his buffer and is likely to find less

Chancellor John Healey has promised to meet the fiscal rules in his first Budget on 28 Oct 2026. The binding rule requires revenue to cover day-to-day spending in 2029-30, and the OBR's March forecast left £23.6bn to spare. Since then the ten-year gilt yield has risen 0.85 percentage points to 5.33%, which on the OBR's rule of thumb adds about £7bn to that year's debt interest. With weaker growth and higher inflation, EY now puts the margin at £11.3bn. Restoring March's buffer would take about £12bn of tax rises or spending cuts, and matching the £29bn average buffer of 2010-2024 about £18bn. That is less than the £26bn raised in November 2025, but income tax, VAT and employee National Insurance, about 70% of tax revenue, are ruled out. The Treasury has sounded out investors on a buffer of £15-20bn. We expect it to aim for the low end, which needs about £4bn. A thin buffer would leave gilts exposed to the next rise in yields: the thirty-year reached 6.04% on 7 October, the highest since 1998, and markets price a Bank of England hike on 5 Nov at about 80%.

Healey must find £4-18bn, depending on his buffer

Tax rises or spending cuts needed in 2029-30 for each buffer, £bn

Starting from EY's estimate of £11.3bn of headroom, a £15bn buffer would need £3.7bn of tax rises or cuts, a £20bn buffer £8.7bn, restoring March's £23.6bn £12.3bn, and matching the £29bn average of 2010-2024 £17.7bn.

05.0101520£3.7bn£8.7bn£12.3bn£17.7bn£bnTreasury low end (£15bn)Treasury high end (£20bn)Restore March (£23.6bn)2010-24 average (£29bn)

Need = buffer minus EY's £11.3bn headroom estimate (5 Oct); dark bar is our view. The OBR left £23.6bn in March; higher gilt yields have since cost about £7bn of it.

Source: EY - UK Pre-Budget Fiscal Outlook October 2026, Bloomberg via Yahoo Finance - Healey considers smaller Budget headroom, Office for Budget Responsibility - Economic and fiscal outlook March 2026, Office for Budget Responsibility - March 2026 detailed forecast tables: debt interest ready reckoner, Bank of England - ten-year nominal par yield on British government stock (IUDMNPY), Resolution Foundation - New Chancellor faces £10 billion headroom headache · 2026-10-09

The midterms matter more for 2027 than for Q4

Prediction markets favour Democrats winning both chambers of Congress on 3 November. We find no statistically clear relationship between changes in those odds and asset prices so far. Republicans hold 53 Senate seats, so Democrats need a net gain of four. Polymarket gives them a 65% chance of winning the Senate, up from 25% in August 2025, and 91% for the House. The cost of living dominates the campaign, and in a KFF poll 60% of voters said they worry about gasoline and transport costs, tied with health costs as their top concern. About seven Senate races are close enough to decide control, among them Ohio, Maine, Texas, Iowa and Alaska. Alaska and Maine count voters' second choices when no candidate wins a majority, which takes time, so control may not be known for a week or more after the vote.

In weekly data since August 2025, none of fifteen sector and style funds has a statistically significant relationship with the Senate odds. The strongest link has a t-statistic of 1.5, below the conventional threshold of about 2. The ten-year yield has moved +3bp for each 10-point rise in the odds, also not statistically distinguishable from zero. Election results have moved markets little in the past, because investors usually price the likely winner before the vote. On the day after a midterm the index has moved 1.03% on average, against 0.67% on an ordinary day. Strong fourth quarters in midterm years are a weak guide too. Since 1950 the fourth quarter of a midterm year has returned +6.6% against +3.4% in other years. Since 1999 that lead has reversed to -1.1 pp, and it usually followed a weak summer, while this year the index returned +17.2% from April to September.

A Democratic House would mean divided government, preventing either party from passing a tax and spending bill alone in 2027 and 2028. Raising the debt limit, reached again in 2027, would require agreement between the parties. Before the new Congress sits in January, the outgoing Republican majority is likely to try to pass a budget bill that needs only a simple majority in the Senate, with $350bn for defence, in the same weeks as the 11 December funding deadline. The election does not itself remove tariffs imposed under existing trade laws. For the AI trade, Morgan Stanley's policy team argued on 7 October that the vote matters through local resistance to new data centres more than through federal AI rules. It sees a Republican sweep as the best outcome for investor confidence in AI construction and a Democratic sweep as the worst. The decisions that actually delay projects, such as grid connections, power prices for large users, zoning and water permits, are taken by states, utility commissions and local governments.

Markets expect a Democratic House; the Senate is open

Polymarket odds of the four House and Senate outcomes, %, normalised to 100

Over the past year the odds of Democrats winning both chambers rose from about 20 per cent to 64 per cent, mostly at the expense of a Democratic House with a Republican Senate, now 26 per cent.

D House, D SenateD House, R SenateR House, R SenateR House, D Senate
020406080100%Oct 25Jan 26Apr 26Jul 26Oct 26

A sweep is priced at 64%, above the 59% implied if the two chambers were independent, so markets expect one national swing to decide both. Kalshi: House 91%, Senate 63% D.

Source: Polymarket - Balance of Power: 2026 Midterms, Kalshi - CONTROLH-2026 and CONTROLS-2026 markets · 2026-10-09

The midterm autumn lead rests on a few strong years

Average S&P 500 price return, %, by month and year of the presidential cycle

A grid of average monthly S&P 500 returns for the four years of the presidential cycle. Midterm-year October and November stand out at about three per cent each, after a weak midterm summer.

JanFebMarAprMayJunJulAugSepOctNovDecElection yearElection year, Jan: +0.2+0.2Election year, Feb: -0.1-0.1Election year, Mar: +0.6+0.6Election year, Apr: +1.0+1.0Election year, May: +0.4+0.4Election year, Jun: +1.4+1.4Election year, Jul: +0.7+0.7Election year, Aug: +1.3+1.3Election year, Sep: -0.3-0.3Election year, Oct: -0.8-0.8Election year, Nov: +2.2+2.2Election year, Dec: +1.1+1.1Post-electionPost-election, Jan: +0.8+0.8Post-election, Feb: -1.3-1.3Post-election, Mar: +0.4+0.4Post-election, Apr: +1.6+1.6Post-election, May: +1.9+1.9Post-election, Jun: -0.2-0.2Post-election, Jul: +2.2+2.2Post-election, Aug: -1.0-1.0Post-election, Sep: -0.5-0.5Post-election, Oct: +1.4+1.4Post-election, Nov: +1.6+1.6Post-election, Dec: +0.7+0.7Midterm yearMidterm year, Jan: -1.0-1.0Midterm year, Feb: +0.3+0.3Midterm year, Mar: +1.2+1.2Midterm year, Apr: -0.3-0.3Midterm year, May: -0.6-0.6Midterm year, Jun: -2.1-2.1Midterm year, Jul: +1.3+1.3Midterm year, Aug: -0.4-0.4Midterm year, Sep: -0.8-0.8Midterm year, Oct: +3.0+3.0Midterm year, Nov: +2.7+2.7Midterm year, Dec: +0.8+0.8Pre-electionPre-election, Jan: +4.2+4.2Pre-election, Feb: +1.0+1.0Pre-election, Mar: +1.9+1.9Pre-election, Apr: +3.4+3.4Pre-election, May: -0.1-0.1Pre-election, Jun: +1.7+1.7Pre-election, Jul: +1.1+1.1Pre-election, Aug: +0.3+0.3Pre-election, Sep: -1.0-1.0Pre-election, Oct: +0.1+0.1Pre-election, Nov: +0.9+0.9Pre-election, Dec: +3.0+3.0

1950-2025, 18 to 19 years per cell. Midterm Octobers average +3.0% against +0.9% in all years, but dropping 1998 alone cuts the midterm Q4 lead by a quarter.

Source: Financial Modeling Prep - daily index, ETF, commodity and FX closes and the US Treasury par curve for the Q4 2026 outlook · 2026-10-09

AI spending plans should rise again, and the AI labs' revenue is the main risk

Microsoft, Alphabet, Meta and Amazon are likely to raise their spending plans at results expected on 28-29 October, because demand for computing power still runs ahead of supply. One-year rental prices for Nvidia's H100 chips rose about 40% between October and March, and OpenAI's API traffic grew 2.5 times in the five months to April. Consensus cited by Goldman Sachs puts 2027 spending growth for those four companies plus Oracle at about 38%. SemiAnalysis expects OpenAI and Anthropic to use 40-50% of new computing power in 2027, making their revenue growth a risk to these plans. AI shares fell after the Financial Times reported OpenAI's September annualised revenue at about $50bn on 8 October; differing treatment of cloud-partner sales complicates comparisons. TSMC's results on 15 Oct and Anthropic's reported November listing, raising up to $100bn at a valuation of up to $2tn, are the other tests. Post-results index moves have averaged 0.95% since 2023. The VIX at 15.0 measures 30-day volatility; late-October protection needs to be assessed at the relevant option prices.

World AI token use rose about 25-fold in five quarters

Tokens processed worldwide, trillions a day, quarterly average

Exponential View's estimate of world AI token use rose from about 34 trillion tokens a day in the first quarter of 2025 to about 880 trillion in the second quarter of 2026, with API traffic about two fifths of the total.

APISubscription and internal
02004006008001.0kTrillion tokens a dayQ1 25Q2 25Q3 25Q4 25Q1 26Q2 26

Exponential View estimate for all major providers including China, read from its June 2026 chart; June 2026 was its estimate. Subscription and internal use include chat apps.

Source: Exponential View (Azhar et al.) - The State of the AI Economy, June 2026 · 2026-10-09

Q3 results are unlikely to show AI spreading beyond tech

Profits outside tech show little evidence of gains from AI yet. Most of the 26.8% consensus Q3 earnings growth comes from chips, oil and a few tech firms, while FactSet expects financials to grow 3.0%. The banks report on 13-14 Oct. Quantified savings remain sparse outside tech, but the transcript sample includes non-bank companies too. Until those savings show in costs, the index stays dependent on a few AI companies.

Higher rates are hurting housing and borrowers, not the AI build-out

The largest AI spenders have continued borrowing despite higher Treasury yields. The 30-year mortgage rate reached 7.28% on 1 October, from 6.15% at the end of 2025, adding about $300 a month on a $400,000 loan. S&P 500 homebuilders are -12.0% this year against +14.2% for the index. A high-yield spread above 4%, against 3.12% now, would suggest private-credit strains are spreading to public markets.

Inflation should stay above target, and the Fed is likely to hike in December anyway

Higher rates cannot repair the energy supply disruptions contributing to inflation. We nevertheless expect the Fed to consider a December hike to limit their effect on other prices. Consumer prices rose 3.4% in the year to August, driven by energy and the war, while core prices rose 2.4%. Goldman Sachs estimated in July that AI's contribution to core PCE inflation would rise from 0.2 to 0.5 points by year-end. Warsh said in September that the Fed cannot affect any individual price and can only stop rises from spreading, so its hikes work mainly by cooling demand. September CPI on 14 October and core PCE on 29 October are the dates to watch, and markets give a December hike 76%. Long yields are unlikely to fall while governments across rich countries borrow more. The US term premium has moved +51 bp this year. The thirty-year gilt yield reached 6.04% and Japan's thirty-year yield hit a record. Markets assign a about one in four chance to a Bank of Japan hike on 29-30 October, leaving scope for a surprise.

Headline inflation has pulled away from core

US consumer prices, % change on a year earlier, monthly

Headline CPI inflation fell from about 9 per cent in 2022 to near 3 per cent in 2024-25, then rose to 3.4 per cent in August 2026 as energy prices jumped, while core inflation eased to 2.4 per cent.

All itemsCore, excluding food and energy
0.02.04.06.08.010.0%Jan 22Jan 23Jan 24Jan 25Jan 26

BLS CPI via FRED, to August 2026. Core leaves out food and energy; the gap since March is mostly energy and the war. Dashed: 2% reference; the Fed targets PCE, not CPI.

Source: FRED series in the Q4 2026 outlook snapshot (Federal Reserve Board, BLS, BEA, EIA, ICE and OECD series), Bureau of Labor Statistics, Consumer Price Index for August 2026 · 2026-10-09

France faces a downgrade, the China truce should hold and the midterms will change little

France should get through the quarter with a downgrade but no crisis. We think a Moody's cut on 23 Oct is more likely than not, and a failed budget vote on 17 Nov 2026 would leave the option of a special funding law. The gap over German yields, 123bp now, should stay between about 120 and 150bp. A censure motion that brings down the government would push it above 150bp and the euro lower. The China truce looks set to be extended. Watch the tariff exclusions expiring on 9-10 November and the expected Trump-Xi meeting in Shenzhen on 18-19 Nov. High inflation gives the US an incentive to avoid a renewed trade war. Democrats are likely to win the House, priced at 91%, while the Senate is closer at 65%, but past midterms moved the index only 1.03% the next day. The bigger risk is a lame-duck budget bill with $350bn for defence before the 11 December funding deadline, which would push long yields up. A short funding extension, with no shutdown, is the more likely outcome.

Q4 2026: the risk bunches in the last week of October

Events and reported plans, 13 October 2026 to 10 January 2027

A timeline of 26 events and reported plans. Late October includes peak earnings, the Fed, UK Budget, GDP, core PCE, ECB and Bank of Japan. The Anthropic bar covers November before Thanksgiving, not a confirmed listing date.

EarningsUS dataPoliticsCentral banksFiscal and trade
19 Oct2 Nov16 Nov30 Nov14 Dec28 Dec11 JanBig bank resultsUS CPI (Sep)TSMC resultsMoody's review of FrancePeak earnings (expected)FOMC decisionUK Autumn BudgetUS Q3 GDP and core PCEECB and Bank of JapanAnthropic IPO (reported)US midtermsTreasury refundingBank of EnglandUS payrolls (Oct)China exclusions expireUS tariff pause expiresUS CPI (Oct)French budget voteAPEC / Trump-XiNvidia (date unconfirmed)US payrolls (Nov)FOMC with projectionsUS CPI (Nov)US funding deadlineECB / BoE / BoJTruce end (China date)

Dates: BLS, BEA, central banks, HMT, Treasury and USTR. Company results dates estimated except TSMC; IPO month reported, not confirmed. Tariff expiries are as written.

Source: Financial Modeling Prep - economic calendar 2015-2026, Financial Modeling Prep - per-symbol earnings history and S&P 500 constituents for the Q4 2026 outlook, HM Treasury - Chancellor letter to the Treasury Select Committee on the Budget 2026 date, USTR - Notice of product exclusion extensions (FR Doc 2025-21671), NACo - Congress passes short-term funding bill through Dec. 11, Bloomberg - US and China to extend trade truce to Jan. 10 Bessent says, TSMC - Quarterly results, Reuters via Yahoo Finance - Moody's sees no quick resolution of France's political difficulties, Bank of England - MPC dates for 2026, European Central Bank - Schedules for the meetings of the Governing Council, Financial News (Korea) - APEC Leaders' Meeting in Shenzhen, Yahoo Finance - Anthropic targets pre-Thanksgiving IPO at $2 trillion valuation · 2026-10-09

Sources