# The Compound and Friends: Semis Fell 30% That Week, and Six Sectors Sat Within 3% of All-Time Highs Source: Realpha Blog (blog.getrealpha.com) Original article and charts: https://blog.getrealpha.com/en/blog/compound-2026-11-10-can-the-stock-market-survive-a-tech-meltdown/ > Notes on The Compound and Friends recorded on 10 November 2026, where Michael and chart analyst Chart Kid Matt take last week's selloff apart chart by chart: 57 S&P 500 stocks fell more than 10% in two days, 32 of them technology, and semis — the index's largest industry group — drew down about 30%, while over the same stretch insurance rose 13%, staples 15% and commercial and professional services 18%, with six sectors within 3% of record highs. Matt also re-ran 2000: tech plus communication services fell 70% from the March 2000 peak to the October 2002 trough while the rest of the market rose 3% — a comparison that holds only without a deep economic recession. Every number is as stated at the time of recording. Educational listening notes only — no investment advice, no price targets, and the judgements here may be wrong. Published: 2026-11-12 Locale: en Tags: sector rotation, podcast-notes, technology stocks, value stocks, historical analogues ![A riverbank the morning after a storm, a great oak lying uprooted across the mud in the foreground with its root ball still clutching soil, behind it an unbroken stand of reeds still standing and all bent the same way, the bank receding into the distance under a pale grey sky](/covers/compound-2026-11-10-can-the-stock-market-survive-a-tech-meltdown-cover.png) > I bend, and I do not break. > > —— La Fontaine, Fables, "The Oak and the Reed" (1668, my own translation) In this episode of The Compound and Friends, recorded around midday on 10 November 2026, Michael and the chart analyst known as Chart Kid Matt take last week's selloff apart chart by chart (Josh wasn't there). Here is what they counted: 57 S&P 500 stocks fell more than 10% in two days, 32 of them technology and 12 industrials, and semiconductors — the index's largest industry group — drew down about 30%. Over the same stretch, insurance stocks rose 13%, staples 15% and commercial and professional services 18%, and six sectors sat within 3% of record highs. Matt then re-ran 2000: technology plus communication services fell 70% from the March 2000 peak to the October 2002 trough while the rest of the market rose 3%, and the episode is explicit about the condition attached to that comparison — the dot-com bust came without a deep economic recession; it was a stock market recession. ## He went to bed red and woke up green The night the election result became clear, futures fell straight away, with Nasdaq 100 futures down about 3% at the lows. The story the market believed was simple: this result would slow the AI buildout, the money for data centres would be blocked, so sell first. Michael went to bed. He woke up to futures a bright green. He says he smiled at that, because the flipped version also made sense: everyone had been complaining about the AI bill anyway — free cash flow eaten up, share issuance, circular financing, no room left in the debt market, the whole thing moving too fast. If demand isn't slowing, then easing off the gas at 150 miles an hour just stretches the same road out over a longer run, and the stretched version is the one that keeps the cash flow intact. Michael's usual position is that the headlines explaining daily moves are nonsense: one version in the morning, a different one eight hours later, nothing actually changed in between. But that Tuesday-into-Wednesday round-trip from red to green had a narrative that lined up, and he says those weeks are rare. ## Thursday looked like somebody liquidating all day Thursday was a different animal. Matt says that around 10:45 that morning he and a colleague were watching the intraday tape in the office, and it looked like someone was being liquidated — and kept being liquidated all day, tick after tick lower, closing near the lows. So they pulled up a chart and counted: 57 S&P 500 names down more than 10% over two days, 32 in technology and 12 in industrials. Micron fell 23%. NVIDIA fell 12% across the two days (9%, then 3%). Michael says he had expected the name at the centre of the storm to be destroyed, and instead NVIDIA was a relative winner among the semis — though a 12% two-day move in the largest stock in the world is already a big step. Semiconductors are the biggest industry group in the index and drew down about 30% in this episode, while the S&P 500 itself had a maximum drawdown of roughly 6% last week. ![Four downward bars show drawdowns of increasing depth, from the S&P 500's 6% at the shallow end to 30% for the semiconductor group at the deep end.](/figures/compound-1110-depth-differs-fivefold-en.svg) ## Counting it showed the damage sitting in one neighbourhood Matt cited a table from Rob Anderson at Ned Davis Research: this is the sixth Nasdaq selloff of more than 9% since the 2020 low, and leaving aside the tariff episode they call Liberation Day, the deepest of the others ran about 13%. What those earlier drawdowns had in common was nowhere to hide — the tariff one especially, where the bottom of the table is a sheet of red. What's different this time is the rotation. Over the same stretch, insurance rose 13%, pharmaceuticals 10%, staples 15%, and commercial and professional services 18%. Matt then pulled the S&P 500 sectors sitting within 3% of all-time highs: there are six. Look at the two-year path of healthcare and staples, he says, and you'd have no idea anything happened last week. ![On the left, four upward bars mark the sectors that rose over the same stretch; on the right, six dots sit just under a line marking all-time highs.](/figures/compound-1110-what-was-rising-en.svg) Michael's conclusion: a portfolio spread across sectors hurts this week without taking a knife to the heart, while anyone holding something far away from the cap-weighted index — which is a lot of people — is feeling something else entirely. He asked in passing whether even Google had fallen 30%, and Matt's answer was that it got hit too. There is an earnings leg under this rotation too. Matt's figure: the 493 names left after the seven big tech stocks (the Mag 7) are taken out are expected to grow earnings 32.6% year on year this quarter, against 22.6% for the Mag 7 — the first time in four quarters the 493 are expected to win. Michael didn't buy it on the spot: analysts aren't stupid and usually get close, he said, but 32% is an absurd number, so where is it coming from? Matt's answer was honest — he took the headline figures off Bloomberg and hasn't broken them down. ## He pulled up a stretch from twenty-six years ago and re-ran it Michael's assignment for Matt was this: if tech has peaked, what does that mean for tech itself, and what does it mean for the other 493? Matt went back and did 2000. First chart: 24 months on from the Nasdaq peak, 129 S&P 500 stocks were up more than 50% — more names up 50% than down 50% at that point. Michael interrupts here with what I think is the most human stretch of the episode. The dot-com bust traumatised a whole generation of investors, which is understandable; if his own stocks had fallen 90% he'd be singing a different tune too. But the impression that every stock got killed doesn't match what happened. Anything that didn't set its weights by market cap — equal weight, value, that family of approaches — did spectacularly well after the bubble burst, which is why for more than a decade afterwards so many investors were waiting for that scene to come round again — back then, as long as you avoided those five stocks, you were a genius. The second chart is harsher. Technology plus communication services fell 70% from the March 2000 peak to the October 2002 trough, while the rest of the market rose 3% — and before the final capitulation leg, the rest of the market was up almost 40%. Matt's line: people called it a tech bubble, and that was exactly the right name. ![Two lines split from the same peak: tech plus communication services falls all the way to minus 70%, while the rest of the market first gains nearly 40% before settling back at plus 3%.](/figures/compound-1110-two-thousand-split-en.svg) The third chart is the distribution of damage: 99% of technology stocks entered a bear market during that stretch, while staples, financials and utilities each had only about a third of their names fall 20%, and materials 45%. The fourth: at the market's low, about half the names in financials and in consumer discretionary were trading higher than at the bubble peak. ![The top four bars show the share of members down more than 20%, with tech nearly filling the panel; the bottom two bars show the share still above their bubble peak on the trough day, each about half.](/figures/compound-1110-breadth-not-magnitude-en.svg) The episode put the limits on it before anyone asked. All of this is clear only with hindsight — stop the chart at month 16, show it to someone living through it, and they'd say that looks like capitulation too. And the whole comparison rests on not walking into a large-scale global recession. In 2000 there wasn't one; that was a stock market recession, and the economy itself held up. ## They put a date in the calendar Last Wednesday threw up a combination: the equal-weight S&P 500 — the version that flattens all 500 weights instead of sizing them by market cap — hit an all-time high while the Nasdaq 100 sat 10% below its own. Matt went looking for the same setup in history and measured how long the Nasdaq took to catch up and make a new high. Excluding the tech bubble — which, he says, distorts the average past usefulness — the average is 113 trading days, with the Nasdaq up an average of 8.1% three months later and 21.6% twelve months later. Historically, he says, the order of events is the Nasdaq catching up to equal weight, and then a new high. ![A line at the top marks all-time highs, with the equal-weight S&P sitting on it and the Nasdaq 10% below, and an arrow spanning 113 trading days rising back to that line.](/figures/compound-1110-113-trading-days-en.svg) Michael's handling of this is what I liked: put the date 113 trading days out in the calendar, roughly five months, come back in the spring, update that bar on the chart and see what actually happened over those 113 days. That day also brought a hot CPI print. One detail worth keeping: there was no figure for last October because of the government shutdown, so there is no annual rate to compute, and Bloomberg's workaround was to average September and November into an estimate, leaving everyone with the monthly numbers. After the release, Fed funds futures moved the implied rate for early 2028 from 4.37% to 4.54%. Matt mentioned something he learned from Tom Lee at Fundstrat: with a lot of economic releases the real signal is in the market's reaction, with the number itself secondary, and he built a matrix on that line to measure three-month forward returns for each combination. That day never reached the cell he wanted to look at, though: the tape was still falling as they recorded, NVIDIA was down 6%, and Michael said the index move had nothing to do with CPI. ## "After a fall like this I want to hide in value" — you may still be buying tech This is the stretch I most wanted to forward to a friend. The episode lines up two ETFs that both carry the word value and compares their year-to-date returns: VLUE up 37%, IVE up 14%, a gap of more than twenty percentage points. ![Two bars both carry the name value, one reaching 37% and the other only 14%, with the gap of more than twenty percentage points marked between them.](/figures/compound-1110-same-name-different-thing-en.svg) The difference is in the ingredients. Matt turned the S&P 500 Value index's sector weights into a pie chart, and the biggest slice is technology — which startled him too; financials 15%, healthcare 13%, industrials 10.8%, consumer discretionary 8.5%, a mix that looks uncomfortably like the parent index. Stranger still is the construction: a single stock's weight can be split, part of it counted into the growth index and part into value, which is how names like Tesla and Apple turn up in value's top ten. Michael's reaction to that table was "in what universe". Matt's explanation is that the starting universe is too small — these indices mostly pick from the 500 names in the S&P, and there aren't enough of them to assemble a basket that is both cheap and diversified. ![A row of blocks of differing widths stands for the sector weights of the value index, the widest being tech, whose figure the show never gave, with the remaining blocks narrowing in turn.](/figures/compound-1110-inside-the-value-index-en.svg) The way I use this: before buying any ETF named after a style, open its top ten holdings and sector weights and read them; if the top ten overlaps heavily with what you already own, it is diversifying you less than you think. That takes five minutes, and this year it decided more than twenty percentage points. There's a related test in the episode, from Michael: if the AI trade has peaked, then value's relative strength keeps running — which cuts the other way too, because the AI trade continuing while value keeps beating growth is a pair that is hard to have both of at once. The value-to-growth ratio has gone roughly nowhere from mid-2020 to now, about six years, while last week's few days were value's largest winning margin over growth since 2000. Whether the AI trade has peaked, he says, he doesn't know, and neither do I. ## "Should I get out now" — write down the number that would change your mind One set of numbers in the episode stayed with me. Matt says analysts haven't cut earnings estimates so far, and NVIDIA's forward twelve-month earnings estimate hit an all-time high on Friday, while the stock is down close to 30% from a record high set about a month and 25 trading sessions ago. He put two lines side by side going back to the launch of ChatGPT: forward earnings estimates up 3246%, about 30x, and the share price up 944%, about 9x — estimates compounding three times faster than the price. ![On the left, two bars show forward earnings estimates standing more than three times as tall as price; on the right, two arrows point opposite ways, estimates rising to a record while price falls about 30%.](/figures/compound-1110-estimates-outran-price-en.svg) Michael's reply is the heart of it: estimates could fall 6% and the stock could fall 40%, because that's how sensitive the price is to earnings growth. It runs the other way too — estimates get revised down, the multiple goes from 10 to 16, and the stock still goes up. So "estimates are still making new highs" on its own can't carry an entry or an exit. ![In the top row a very short block leads through an arrow to a much longer one, labelled estimates down 6% and price down 40%; the bottom row runs the other way, with estimates cut while price still rises.](/figures/compound-1110-small-input-big-output-en.svg) What does carry one? Matt gave an honest version. Every time he posts NVIDIA's forward earnings publicly, someone tells him the estimates are inflated; maybe they're right, he doesn't know. What he can do is line up, for every quarter since 2021, the estimate made a year earlier against the twelve months of earnings that actually arrived — and since 2024 the realised number has come in above the old estimate in most quarters. That record gets tested again by the real figure every quarter. Two points pulling the other way stayed with me as well. Of the Mag 7 names that had reported this quarter, four of six beat — and those four fell an average of 3% the next day, on a chart that has been mostly positive bars since the first quarter of 2018. Michael tried to find a flattering reading for that (a wall of worry, something like it), admitted halfway through that he was grasping at straws, and scored the point for the bears. The other point was about himself: he'd publicly called NVIDIA the fattest pitch in the market the previous Tuesday. Then he said something I liked — the stock market is the ultimate equaliser; ask Druckenmiller where it goes next week and then ask his own wife, and he'd respectfully defer to the former, but nobody knows. So my own practice now, before asking whether to get out, is to write down a sentence that data can contradict: "If X happens, I was wrong." X might be two consecutive quarters of downward revisions, or a major customer's capex guidance turning. If I can't write X, the thing I'm deciding with is just how I feel that day. ## Further reading - The Compound and Friends, recorded 10 November 2026 (the chart special with Michael and Chart Kid Matt) — the source for this piece - Matt's charts: chartkidmatt.com - Ned Davis Research's comparison of the six Nasdaq drawdowns of 9% or more since the 2020 low - The published constituents and weights of the S&P 500 Value index, and the holdings of the various value ETFs - The hot CPI print discussed in the episode, and the Fed funds futures path implied the same day ## The one thing worth taking with you **To judge a selloff, count its breadth.** The whole episode does one thing: it replaces "the market is collapsing" with things you can count, and once table after table has been counted, "everything is falling" isn't left standing. Depth is what frightens you; breadth is what tells you whether this is one neighbourhood or the whole city. Here's something I've tried that works outside investing too. Take a sentence you heard recently and believed at the time, one that starts with "everyone" — everyone at work wants to leave, everyone my age has bought a flat by now, the whole industry is laying people off, all my classmates are doing better than me. On a sheet of paper write down ten people you can name, and mark each one yes, no, or don't know. Then look at the three columns. I did this once, and the longest column was don't know. ![Three stacked columns of small cells, with yes and no holding only two or three cells each and the taller don't-know column holding five, ten in all.](/figures/compound-1110-count-the-three-columns-en.svg)