Market Breadth Data******************************

Saturday, November 21, 2026

The AI Trade Broke. The Fed Didn't Blink.

 

Regime verdict: risk-off, de-grossing phase. The market's engine — the AI capex complex — failed its two biggest tests of the year in the same week (a clean Nvidia earnings beat and the largest IPO in history), and the Federal Reserve is still hiking into the selloff. That combination — a broken leadership story plus tightening policy into a correction — argues for defense, cash, and staged re-entry, not for catching the first bounce. The base case is a grinding correction into the December 8–9 FOMC, not a V-bottom.

1. REGIME & MACRO SYNTHESIS

Growth: not the problem — yet. Q3 GDP printed +3.4% (Oct 29). October payrolls +165K vs +65K expected, with +84K of upward revisions. Retail sales control group +0.6% vs +0.4% expected. Consumption is resilient. But the forward-looking surveys have turned: Empire State 7.8 (from 25.5), Philly Fed 6.3 (from 20.9, six-month low), and the Conference Board Leading Economic Index −0.3%. Growth is decelerating from strong, not collapsing — the slowdown story is a 2027 question, not a today question.

Inflation: sticky at the level that matters. October CPI ran hot (headline +0.4% m/m, core +0.3% for a third straight month). September PCE was 3.6% y/y, core 3.1%. But the expectations channel just moved the other way: Michigan's final 1-year inflation expectations plunged to 3.7% from 4.7% in October (4.3% prelim), the lowest since February; 5-year expectations eased to 3.3% from 3.5%. The market ignored it entirely on Friday — equities fell into the close anyway. When a dovish expectations print can't lift stocks, the tape is trading a different risk.

Policy: the Fed is hiking into a correction, and the debate has moved past December. A December hike is ~94% priced for 25bp (~6% for 50bp, per Barron's/CME FedWatch) — essentially unchanged through the rout. Governor Waller has put 50bp on the table ("you have to go 50 in December... to get it done"), and Governor Cook, speaking Friday, gave her first public account of her October dissent: the September 25bp hike "was not enough," rates are "not high enough" to dent the AI investment surge, which she names as a significant driver of inflation this year, and "additional and larger rate increases may be required later." She did not directly address December, per the WSJ — so Waller remains isolated on 50bp. Treasury Secretary Bessent pushed back on CNBC Friday, saying he would encourage the Fed not to embrace a half-point move. The live question is no longer December's 25bp; it is the path beyond — JPMorgan and Goldman now expect January and March hikes as well, with Goldman's first cut pushed to December 2027. Wednesday's October PCE is the next arbiter.

Liquidity & credit: the bond market is buying safety, not pricing a Fed pivot. The 10-year fell 14bp Friday to 4.91% (−24bp on the week), the 2-year 4.61% (−26bp on the week, its largest one-week drop since September 2024), the 30-year 5.30%. Yields fell on a flight to safety while the December hike stayed fully priced — that divergence is the tell. The dollar held near 19-month highs (DXY 103.84), with the safe-haven bid rotating partly into the yen. Credit is the dog that hasn't barked yet: no verified spread blowout accompanied the equity move. If high-yield spreads widen materially from here, the bear case accelerates; until then, this is an equity-leadership unwind, not a credit event. Watch it as the tripwire.

Market internals: liquidation, not rotation. Friday: S&P 500 6,910.39, −3.70% — seventh straight down day (longest since February 2020), −8.6% on the week (worst since April 2025), −11.6% from the October 6 record close of 7,819.04, barely positive for 2026. Nasdaq 22,617.73, −5.28% — second-largest one-day points drop in its history, −10% on the week (worst since March 2020), −18.0% from its October 6 record, with the Nasdaq-100 briefly in bear-market territory intraday; 2026 gains erased (−2.69% for the year). Dow 48,147.33, −1.93%, −8% on the week (worst since March 2020), in correction. Russell 2000 2,581.79, −2.76%. VIX 35.89, +26.2% on the day, spiking to 40 intraday, +86% on the week. Ten of eleven sectors fell; only Consumer Staples (+0.67%) gained. Technology −7.27% was worst. Gold fell with everything — spot $3,548.53, −2.0% Friday, a fresh 14-month low — and Bitcoin broke below $60,000 ($58,952 Friday; ~$57,938 Saturday). When gold, crypto, and equities fall together, portfolios are de-grossing. That is liquidation, not rotation.

The AI break — the regime's center of gravity. Nvidia reported a clean beat Tuesday (revenue $115.8bn vs ~$109.1bn consensus, Q4 guide $130bn vs ~$125bn, FY2028 growth raised to ~75%) and the stock still fell. Anthropic priced above range at $83, opened at $111.50, and closed its second day at $88.91 — its intraday low was $83.05, five cents above the IPO price; the broken-IPO branch did not fire, but the $100 hold test failed decisively. (Correction to Friday's close briefing, which carried one live-blog feed's $88.30: the official close is $88.91, −8.5%.) Then Reuters reported Friday that SEC enforcement staff are probing Nvidia's accounting, alongside a delayed 10-Q filing and a California attorney general subpoena. The probe's scope is unknown — Reuters could not establish when it began or what it covers, the SEC declined to comment, Nvidia did not immediately respond — and an investigation is not evidence of wrongdoing. But reported new detail frames the overhang: ~$104bn of 2025 revenue booked through Nvidia's Singapore unit, a change in geographic revenue reporting, receivables at 64 days of sales. The House also released BIS data showing 879,000 H200/MI325X export licenses approved for China (reported). Huang is set for closed-door House interviews December 1, and has agreed to testify at a public Senate Banking hearing; the board has ruled out reinstating him and retained Heidrick & Struggles for the CEO search. Nvidia closed $125.53, −14.01% Friday, −22.9% on the week (worst since October 2008), −46% from its October 6 top, roughly $2.7tn of market value gone. The SOX semiconductor index fell 11% Friday; AMD −13%+, Broadcom ~−12%, Micron −13.9% (−24% on the week), and memory got its own fork — TrendForce turned bearish on NAND for 2027. The results were strong; this selloff is about trust and price, not a business slowing down. That distinction matters for the recovery path: trust rebuilds slower than multiples.

2. ALLOCATION — WHAT THE REGIME PAYS YOU TO OWN, AND TO AVOID

For a balanced allocation. These are positioning judgments for the regime described above, not personal advice. Conviction: high / medium / low.

Equities — UNDERWEIGHT overall (medium conviction). The correction is 11.6% deep on the S&P and 18% on the Nasdaq, but no framework yet confirms a bottom (see Section 3). Within equities:

·        OVERWEIGHT defensives (high): Consumer Staples (Friday's only gainer), Utilities, Health Care. They held up in the worst tape of the year; they are where the relative-strength money is going.

·        UNDERWEIGHT growth / momentum / AI hardware (high): the crowded trade is unwinding and the catalyst calendar (SEC scope, 10-Q, CEO search, capex decisions) runs against it into December. Do not short into a VIX of 36 — the trade is crowded on both sides now.

·        OVERWEIGHT value over growth (medium): the Dow fell 1.93% Friday against the Nasdaq's 5.28%. Value's relative resilience is the market telling you where the air is coming out.

·        NEUTRAL small caps: Russell −2.76% Friday; no edge either way until credit confirms or denies the stress.

Duration — NEUTRAL to modest OVERWEIGHT Treasuries (medium). The 10-year at 4.91% after a 24bp weekly rally is a hedge, not a rates call: if the bear case fires, duration pays; if the Fed stays hawkish, the 2-year at 4.61% still prices the hikes. Own duration as portfolio insurance, sized as such.

Cash — OVERWEIGHT (high). T-bill yields still north of 4%, and cash is the only asset that gains optionality from volatility. In a de-grossing regime, dry powder is a position.

Credit — NEUTRAL investment grade, UNDERWEIGHT high yield (low conviction — data-limited). No verified spread blowout yet; the underweight is precautionary, not confirmed. If spreads widen, this becomes the highest-conviction underweight in the book.

Commodities — NEUTRAL gold (medium), UNDERWEIGHT oil (medium). Gold fell in a liquidation week, which argues against chasing it as a haven here; it remains a hold, not a buy, at $3,548. Oil: WTI $83.98, −5.2% on the week and −19.2% in November; Brent $92.57, a third straight weekly decline. The IEA supply-surplus theme and recovering Hormuz flows press crude; the old $111–$115 guard levels are stale — re-anchor to the current tape.

Crypto — UNDERWEIGHT (medium). Bitcoin broke $60,000 for the first time since July ($58,952; ~$57,938 Saturday) and ETH fell 7.8% Friday to $1,618. Risk appetite's canary is not singing.

USD — NEUTRAL to modest LONG (low). DXY 103.84 near 19-month highs; the dollar is doing its job as the funding currency in a risk-off move, but the yen is sharing the safe-haven bid.

3. TIMING, INFLECTION & POSITIONING

 The turn did not come — and that is the signal. Friday completed the 45-day Gann window from the October 6 peak on monthly options expiry, and the tape accelerated down through it instead of turning. A time window that produces acceleration rather than a low is a bearish resolution. The next window, November 20–24, now converges with the Hurst ~40-day trough window. A low forming there is the first place a turn could credibly appear; nothing yet says it has.

·        Gann: S&P closed below the round 7,000 level and below the 7,100–7,200 shelf that was support on Thursday. Support now 6,800–6,900, then 6,500. Resistance: 7,000, then 7,175.83 (Thursday's close), 7,300, then the 7,600 gate — now ~690 points overhead. Nasdaq sits below 23,000; support 22,000, then 21,000.

·        Elliott: the decline from the October 6 high reads as an unfolding corrective leg — most plausibly a wave 4 of the larger advance, or the A-B-C of a deeper correction. Two consecutive failed upside catalysts (Nvidia earnings, the Anthropic debut) fit an unfolding C-wave leg, not a completed correction. A daily close back above 7,600 would argue the correction is over; a close below 6,800 opens the path to 6,500. Price is well below its 200-day moving average, which must be reclaimed before any recovery is taken seriously.

·        Hurst: down across the board. The ~20-day cycle is in its down phase with no confirmed trough; the ~40-day trough window is due late November; the ~80-day cycle peaked with the market top in early October and its trough is not due until early 2027; the ~20-week cycle is declining. No cycle is yet pointing up.

·        Sentiment & flows: VIX 35.89 with a 40 intraday spike has the shape of a capitulation attempt, not yet a confirmed trough. Friday's −3.70% S&P move was bigger than on 100% of trading days in the past year. Small counter-evidence: ETFs tracking the Dow, S&P and Nasdaq-100 climbed 0.9%, 1.0% and 1.5% respectively on Friday night, and Nvidia retraced ~2–3% after hours (~$128–$129.65) — thin, unconfirmed, and not a reversal. Goldman Sachs, in Bloomberg's Friday evening edition, said the selloff has further to run. Thanksgiving week is historically one of the strongest of the year, and the tape is down 8.6% into it — the seasonal tailwind has not held in November. Futures reopen Sunday at 3:00 PM PT.

·        Synthesis: all three frameworks agree this is a corrective phase, not a confirmed bottom. The Nov 20–24 Gann/Hurst convergence is the test; Elliott still needs 7,600 to call it done.

4. SCENARIOS — BASE, BULL, BEAR

Probabilities are judgment, not forecasts. Each case names its catalyst, its invalidation, and what to do.

BASE — Grinding correction into the FOMC (50%)

The S&P trades 6,500–7,200 into December 8–9. The December hike lands at 25bp; the AI capex story stays intact but multiples compress; no single headline clears the trust overhang, and no single print breaks the hawkish path. A low forms in the Nov 20–24 window or into early December, and the recovery waits on FOMC clarity.

·        Catalysts: Wednesday's PCE comes in near consensus; Williams stays neutral Monday; the SEC probe stays scope-unknown; no hyperscaler cuts capex.

·        Invalidation: a daily close above 7,300 shifts this toward the bull case; a close below 6,800 shifts it toward the bear case.

·        Positioning: hold the defensive overweight and cash; add equity exposure only in thirds, on closes that hold above 7,000, not on intraday bounces. Hedge with put spreads or collars rather than outright puts — at VIX 36, outright vol is expensive.

BULL — Washout low already in (20%)

Friday was the capitulation: seven straight down days, VIX 40 intraday, the year's biggest one-day move, and the Friday-night ETF bounce holds into Monday. Wednesday's PCE comes in soft, Williams leans dovish, the SEC probe fizzles or narrows, and Nvidia names a permanent CEO and files the 10-Q.

·        Catalysts: PCE below consensus; a dovish Williams; SEC scope clarifies as narrow; 10-Q filed; a hyperscaler reaffirms capex.

·        Invalidation: a close back above 7,300, then 7,600, confirms the correction is over. Below 7,000 on a closing basis kills it.

·        Positioning: add the second third on a close above 7,175.83 (Thursday's close); rotate back toward quality growth only after 7,300 reclaims. Do not chase Monday's open.

BEAR — The trust shock becomes a funding shock (30%)

The SEC probe widens or the 10-Q delay extends; a hyperscaler cuts AI capex; credit spreads blow out; or the Fed surprises hawkish (a 50bp December move, or a hold framed as "more to come"). S&P toward 6,500, Nasdaq toward 21,000, and the correction becomes a bear-market conversation.

·        Catalysts: SEC scope widens to revenue recognition; 10-Q delay extends past early December; a major AI customer cuts spending; HY spreads widen materially; December delivers 50bp.

·        Invalidation: a close above 7,600 ends the bear case outright; a close above 7,300 weakens it substantially.

·        Positioning: raise cash toward the top of the range; keep hedges on through the FOMC; do not bottom-fish Nvidia or the SOX until a CEO is named and the 10-Q is filed — the stock is −46% from its top and still has no permanent chief executive.

What would change my mind fastest: a reclaim of 7,600 (Elliott's all-clear), a soft Wednesday PCE, the SEC probe closing without action, or Nvidia filing the 10-Q and naming a CEO. Any two of those together would move the base case toward the bull case.

5. CATALYST MAP — THANKSGIVING WEEK

·        Monday Nov 23: S&P Global flash PMIs (6:45 AM PT; manufacturing prev. 54.9, services 56.1) — first hard read on November activity after a week of soft regional surveys; a services print below 55 feeds the growth-worry side, a hot print hardens the hawkish case into Wednesday's PCE. Chicago Fed National Activity Index (5:30 AM PT, forecast ~0.4 vs 0.07 prior). Fed's Williams speaks (12:00 PM PT) — first Fed voice after Waller's 50bp push and Bessent's pushback. 2-year note auction (10:00 AM PT). Earnings: HPE, Agilent, Zoom.

·        Tuesday Nov 24: CB consumer confidence, Case-Shiller home prices. Earnings: Dell, HP, Analog Devices, Best Buy, Dollar Tree, Abercrombie & Fitch — the AI-hardware and consumer tests continue.

·        Wednesday Nov 25: the big day — October PCE (the Fed's preferred inflation gauge), Q3 GDP second estimate, durable goods, jobless claims, personal income/spending, Beige Book.

·        Thursday Nov 26: Thanksgiving — markets closed.

·        Friday Nov 27: early close; Black Friday spending reads.

·        Beyond: December 1 — Huang's closed-door House interviews. December 8–9 — FOMC decision (the odds watch ends there). December 10 — November CPI.

 

Bottom line: the regime changed this week — from a valuation wobble in a bull market to a trust shock in the market's leadership stock, with the Fed still hiking. Defense and cash are paid to wait; the Nov 20–24 window and Wednesday's PCE are the next two tests; 7,600 is the level that would say the correction is over, and 6,800 is the level that would say it is deepening. Until one of them breaks, trade the range and the catalysts, not the hope.

 

Friday, November 20, 2026

Heavy Selling Is Clustering: What To Expect Next

Two different volume readings flashed the same message this week: the selling is coming in clusters, not single days. One measures the whole NYSE, the other SPY’s own trading. Below is what each one shows and what SPY did after similar stretches before.

1. NYSE down volume: four heavy days in six sessions

On Thursday (Nov 19), 86% of NYSE volume went into falling stocks and 82% of issues declined. On its own that is a heavy day, not a rare one. What stands out is the clustering: it was the fourth day of 78% or more NYSE down volume in the last six sessions (Nov 12: 80%, Nov 16: 78%, Nov 17: 79%, Nov 19: 86%).

What SPY did after heavy NYSE selling, 1998 to 2026

Signal (NYSE down volume)

Cases

After 5 days

After 20 days

After 60 days

All days

7,010

+0.2% (58% up)

+0.8% (64%)

+2.3% (70%)

85% or more

189

+0.4% (60% up)

+1.3% (66%)

+3.4% (70%)

90% or more

107

+0.5% (55% up)

+1.2% (64%)

+4.4% (72%)

3rd day of 78%+ within 6 sessions

71

+1.4% (70% up)

+1.9% (67%)

+5.2% (74%)







2. SPY’s own volume: three heavy down days in four sessions

SPY fell 2.0% on Thursday to $715.87 on 76 million shares, 1.65 times its 50-day average. It was the third down day in four sessions with volume at 1.5 times the average or more (Nov 16: 1.54x, Nov 17: 1.57x, Nov 19: 1.65x).

SPY’s volume shows three such clusters since June. In June, SPY made its low (about 725) inside the cluster and rallied to 778 by early August. The Oct 30 to Nov 5 cluster included the biggest volume day of the period (126 million shares, 3 times the average) and was followed by a bounce to 760. This one is different: all three heavy days are down days, and SPY is at a new low for the period with no bounce yet.

What SPY did after its own heavy-volume selling, 1993 to 2026

Signal (SPY volume)

Cases

After 5 days

After 20 days

After 60 days

All days

8,400+

+0.2% (58% up)

+0.8% (64%)

+2.3% (70%)

3rd heavy-volume down day within 6 sessions

63

+0.3% (65% up)

+1.5% (68%)

+3.9% (73%)

What stands out

Both say the same thing

Whether you count heavy selling across the NYSE or in SPY itself, a cluster of heavy days was followed by better-than-normal returns: about twice the normal 20-day gain and better odds. Repeated selling tends to use up the sellers.

More selling usually comes first

After an 85%+ NYSE down-volume day, another one followed within 10 sessions more than half the time (106 of 189), and the median low over the next 20 days was 1.9% below that day’s close. In SPY’s own clusters the median case also saw a close 2.4% lower within 20 days. Feb 25, 2020 was one of those cases, and it came before the worst of that crash.

Not a guaranteed low

The edge shows up over one to three months, not in a straight line. The fall of 2023 had a similar stretch: after the third heavy NYSE day (Oct 3), more heavy days followed and SPY fell another 2.6% over three weeks before the low on Oct 27.

What to watch

An up day on heavy volume

In SPY, a heavy-volume up day would be the first sign the selling is being absorbed, as on Jun 11 and Nov 4. More heavy down days at lower closes would mean the cluster is still building.

The oversold gauge turning up inside the cluster

The oversold gauge we track is deep in its low zone, in the lowest 1.3% of all readings since 2004. Its strongest buy signal in back-tests is the day its slower reading turns back up from oversold, and that has not happened yet.








When the turn came during a NYSE selling cluster (three or more days of 78%+ NYSE down volume within six sessions), SPY was higher 20 days later in 22 of 25 cases since 2007, by about 5% on average (3.7% without the 2009 and 2020 rebounds), against about 3% for the turn on its own. The misses were Oct 2008 (-8.6%), Jun 2010 (-2.0%) and Oct 2023 (-0.5%). 25 cases is a small sample, so I treat it as a confirmation, not a system.

Timing matters

The six-session window moves forward each day and older heavy days drop out of it. With Nov 16, 17 and 19 in the window, a turn on Friday (Nov 20) or Monday (Nov 23) would still count. From Tuesday on it needs fresh 78%+ days to keep the cluster alive.

This is a record of what happened, not a forecast. For information only. Not investment advice.

 

Terms of Use

All rights reserved by the author. The material contained herein is original content and is the sole property of the author. Any commercial use or reproduction - either in part or whole - is strictly forbidden without the author's prior consent.

Disclaimer: The information provided here is for educational purposes only and does not constitute trading advice nor an invitation to buy or sell securities. The views are the personal views of the author. Before acting on any of the ideas expressed, the reader should seek professional advice to determine the suitability in view of his or her personal circumstances.