ETF market structure & leverage froth Forced deleveraging set a near-term bottom; leveraged retail products are the fault line

Sohn (Jun 29, Strategas ETF strategist): 'ETFs are the market' — ~30% of US exchange volume (40-45% in stress), and mutual funds have been in cumulative outflow since 1984 (−$5.1T from active equity funds), the money flooding into ETFs of every kind (core, factor, active, levered). The froth signal: leveraged + levered-single-stock ETFs are a ~$200B record-usage pile whose end-of-day rebalancing adds market volatility (they buy more into up days); levered single-stock products (2x Nvidia, even 2x SpaceX) didn't exist before 2022 and now number in the hundreds on ever-smaller names ('like a lotto ticket'), bought commission-free on phones; thematic ETFs average a ~32% three-year drawdown with 70% carrying a Sharpe below 1. 'Chasing single themes can be detrimental… things are getting frothy.' Paulo Macro (Jul 8): retail has shifted out of cash equities into 0DTE options + leveraged ETFs concentrated in semis/tech — products whose rebalancing creates enormous flow and whose vol drag means they "eventually return to their natural state, i.e. zero" — while margin debt has peaked and started rolling over, "a signal that precedes all major bear markets this century" (Simon White); the leverage demand was also what tightened equity funding from late May. Halftime committee (Jul 9): Brown sizes the same froth from the desk — ~$200B sits in leveraged ETFs, much of it 2× single-stock vehicles, equivalent to ~$500B of notional exposure, so derivative players with "no meaningful point of view" (playing down-7%-yesterday-maybe-up-7%-today) exacerbate any natural selling: "this has changed market structure. We all have to get accustomed to it." The flip side is a clean momentum reset, not a top — zero 52-week highs in SMH but its RSI cooled to a healthy 49 and the average semi's 10-day rate-of-change bottomed at −15%, handing sidelined buyers an entry. Santoli's caution: don't declare the unwind finished just because earnings held up — the pullback came off "one of the most aggressive, crowded crescendos of buying in history" (semis doubled in under 3 months off the Mar-30 low, then gave back 18%) — though the benign tape (volatility internal, index-level vol suppressed, trend holding) keeps tactical longs viable. Eisman (Jul 10): the diversification illusion quantified — with 50%+ of the S&P tech/tech-related, owning the index "does not provide diversification"; nor does 60/40, since 15% of all existing and ~50% of newly-originated 2026 corporate debt is AI-related — the stock and bond markets are making the same single AI trade. Polomny (Jul 11): "a bubble probably bigger than the internet bubble," sustained by passive flows + money printing — he entertains but rejects the "higher plateau for stocks" rationalization ("all bubbles eventually pop") and won't chase; his discipline is waiting for quality to be repriced by an exogenous drop (the CBOE/CME/ICE exchange washout his live example). Halftime committee (Jul 10): the CapEx fear has capitulated into an asset — the day after Micron's blowout the AI spenders were sold while memory ran, which Baruch reads as capitulation on the CapEx fear (base case: Mag-7 outperforms in 2H), and Zuckerberg's excess-compute-as-resellable-cloud reframe ("sell it long term like AWS or Azure") flipped the two-year Meta-monetization bear case — Simpson bought new, Weiss built his biggest-ever position, Harrington keeps it as "the only Mag-7 we still invest in" (20×, $200B social-media revenue growing 28%). Goldman Sachs (Asia research, Jul 10, the 48-page primer): institutional confirmation of the Chinese-open-source erosion Singh/Eisman flagged — Chinese open-weight models have hit a "critical point of intelligence," good enough for agentic/coding work at 10–25% of US SOTA pricing (~$1 vs $4–8 per 1M tokens), with the US hyperscalers themselves hosting them (AWS Bedrock, Google Model Garden, Copilot considering DeepSeek); enterprises pivoting from "token-maxxing" to ROI-first structurally routes typical workloads to the cheap models, eroding frontier-lab pricing power. The sovereignty kicker: Meituan's LongCat 2.0 (a 1.6T-parameter MoE) trained END-TO-END on a domestic 50k-card Huawei cluster — Chinese silicon now handles frontier-class training, blunting the export-control chokehold; GS models China AI-model revenue Rmb35bn→Rmb879bn by 2030 (~25× token growth). Zhipu initiated Neutral at $110bn; MiniMax the rated Buy. Barron's Roundtable (Jul 10): Giroux's three mega-themes — AI capex growth slows, the mix shifts short-cycle, and "diversity of compute" ends Nvidia's monopoly (Anthropic's best LLM already runs on Google/Amazon ASICs; inference favors ASICs/CPUs) — he's negative the long-cycle AI derivatives (CAT at 38× ≈ ~230× its actual data-center power business; GEV 50× "isn't sustainable" per Jain too; ETN; Dell/SMCI/HPE hollowed by hyperscaler in-sourcing) and owns the compute-diversity winners (AMD/AVGO/TSM) plus hyperscalers heading for a 2030–32 FCF "shock" upward. Jain's accounting kicker: AI capex books INSTANTLY as the recipients' revenue but depreciates over 5–7 years for the spenders — earnings optically inflated on both legs (OpenAI losing ~$39B on $13B revenue; open-source now ~2/3 of tokens). Ellenbogen: Anthropic's revenue went ~$9B→$50B+ in six months and its confidential IPO filing will force the market to re-sort AI ownership by cash-flow duration. 2026-AUG-02 (Jay Singh SSR call): Leopold Aschenbrenner's Situational Awareness — launched 2024, $45B AUM, eight people, +439% net through June, ~4x levered long AI infrastructure / short enterprise software — fell 67% in July when both legs went against it in three weeks, and sold its entire public book to Citadel on July 30 at a discount (retaining a $10B private book incl. Anthropic; still +80% YTD). The unwind triggered a +10.7% one-day move in the Morgan Stanley momentum index and +19.1% in its TMT momentum index, the largest on record ex-dot-com. Dan Niles: "forced liquidations due to excess leverage typically mark near-term historical bottoms" (SOX −28.6%, MS TMT momentum −53.5% from the Jun 22 peak; no excess memory inventory unlike 1995). Precedents: Sowood (2007), Amaranth (2006), LTCM (1998), Archegos (2021) — all absorbed at a discount. In Korea, single-stock 3x leveraged ETPs drove the Kospi −44%, 320,000+ accounts liquidated in one day, ₩2.15T (~$14.5B) of retail leveraged-ETF losses with 62% in the 20-30 age bracket, and an Irish 3x SK Hynix ETP −96% in a month; the Finance Minister publicly apologised for approving them. "Leveraged ETFs are a trap, especially the single-stock ones." 2026-JUL-31 (Horizon Kinetics Q2 2026 commentary): not hostile to indexation as Bogle practiced it, but once indexed dollars exceeded actively managed dollars each new ETF dollar had to buy more of the very same shares, so ETFs "crossed over from passively participating in price discovery, to having a hand in making the prices" — the benchmark is now set by the funds tracking it. Inclusion rules built for trillions created an artificial bifurcation: companies too small or too illiquid get less flow, higher cost of capital and less research coverage, while the largest constituents take a disproportionate share of inflows. HK frames the resulting discounts as its hunting ground — the ETF Divide (discounts imposed for non-economic reasons) and the Equity Yield Curve (value realization too distant for a relative-return industry) — citing AutoNation (13.7%/yr per-share earnings growth since 2000 at 11x, 58% of shares retired in five years) and Penske Automotive (14% 15-yr EPS growth at 11x on a ~$3B float), both smaller than the two smallest S&P 500 members, which carry 0.1% and 0.0% weights. 2026-AUG-02 — Mohnish Pabrai (New Money): asked whether passive flows have broken price discovery, he declines the question on principle — it goes in the too-hard pile: "why ask the address of a home when you're never going to visit?" His guess is "we're very far away from that point," but he won't hold a view on a market he doesn't bet in. 2026-AUG-04 — Rick Rule (Stansberry Investor Hour): passive mechanics as an actionable trade — index construction rewards "trading liquidity, simple market cap and heft," so a small/large valuation gap resolves only two ways (arbed away, or the big buys the small): "front running the ETFs, if you give yourself a two-year or three-year lag time, is really truly simple." Two M&A types: strategic (Agnico buying a deposit within trucking distance of its mills) and tactical — manufacturing index members (Equinox/Orla: no operating synergy, but a million oz/yr wins inclusion "and passive buying every two weeks from workers who don't even know they own gold stocks"). Corollary: own the manager, not the product (Sprott — "would you rather pay or be paid?"). 2026-AUG-10 — Le Shrub (Paulo Macro, Fly on the Wall): the mega-IPO addendum to passive ("pacification") — issuers are "effectively gaming the system": mark the company up privately "with a view that you're going to put it in the index at a stupid valuation and the passive flows will maintain that stupid valuation." "Now we have like peak passive where we're getting trillion dollar companies added to the index on IPOs — like SpaceX." 2026-JUN-16 (McDonald, historical): Passive at 60–65% of the market "becomes evil because it gets too big and too easy to game": insiders front-run index inclusion and dump on the add — CoreWeave added to the Nasdaq while "financially a mess," Lululemon added at all-time highs and now −60%, SpaceX fed into S&P/Nasdaq/Russell on a ~4% float with an aggressive unlock — "the little guy gets holding the bag." 2026-AUG-17 — Jay Singh (David Lin Report): the July drawdown was Korean leveraged-ETF speculation meeting a government clampdown — a 45% drawdown in those names, a 50% drawdown in momentum stocks (biggest in several years), 1.2M accounts margin-called. It left memory names at four times earnings, which he bought "just for a trade." His own rule (Klarman): never margin, ~20% cash, amplify only via call spreads — the 4×-levered Situational Awareness blow-up as the cautionary tale. 2026-AUG-19 (Ted Oakley, Oxbow Advisors, The Real Story): Mike Green's number, cited approvingly — passive ownership is within 5–7% of the level where the flow reverses, "if everybody owns the same thing and then you turn the other way," with boomer demographics as the mechanism. 2026-AUG-23 — Jay Singh (SSR call, 2026-AUG-23): US ETF inflows are on track to top $2 trillion in 2026, a 40% jump over 2025, with more than $1T in the first half alone — "with retail euphoria driving much of this." Over 1,100 new ETFs launched year to date, taking the total past 6,000: "there are going to be more ETFs than stocks in the Russell indices, which is unbelievable," driven by levered funds, tax-efficient wrappers and structured derivatives. Slegers 2026-JUL-21 (guest: TJ Terwilliger, quoting Terry Smith’s shareholder letter): the price-setting asymmetry, quantified. Index funds now “manage more than 60% of all assets under management”, but per Cboe Global Markets active managers’ share of trades has fallen from 80% in the 1990s to just 10% — so the marginal price is set overwhelmingly by money that “doesn’t care about price… buy or sell based on money flowing in or out.” The consequence is two self-reinforcing loops (inflow → buying → higher prices → better measured performance → more inflow, and its mirror for active funds), which makes part of the passive/active performance gap mechanical: “Vanguard’s UK All Share tracker has made 66% over five years, trouncing the average UK equity fund’s return of just 32%.” Bogle’s founding assumption — that index funds stay small and free-ride on active price discovery — is named as broken. Slegers 2026-MAY-10: notes that the three largest ETFs in the world all track the S&P 500, together holding more than $2trn, and argues the passive default is now a concentrated single-driver bet rather than diversification. Proposes Berkshire Hathaway as the substitute, claimed on the index's own four merits: 26 listed holdings plus 60+ wholly-owned businesses; "there are no management fees to own Berkshire Hathaway stock" against any expense ratio; winners left to run (Coca-Cola since 1988, See's since 1972); and a better long-run record — plus $300bn+ of cash and a buyback started by Greg Abel. Christopher Bloomstran's published estimate: B shares worth $560-580 against a $475 price. Pieter Slegers (Compounding Quality), 2026-MAR-31: a flow observation on the timing of the switch into index products — "Every time stocks fall for a little while, investors give up on them. They switch to investing in ETFs. This always happens when individual stocks are trading at very attractive valuation levels. Then when everything goes back up, investors switch their focus again to individual stocks." Offered as a contrarian marker on single-stock valuations rather than as a criticism of indexing, and worth reading against the conflict: the same publication runs a parallel ETF portfolio, so it is describing a rotation between two of its own products. 2026-SEP-02 (Hay/Haymaker): the forced-seller mechanism named precisely. A flow reversal "has the very real potential to place significant downward pressure on the fully invested U.S. index funds which are also highly concentrated in AI-related shares" — both halves load-bearing: fully invested means no cash buffer to meet redemptions, and cap-weighted concentration means forced sales land hardest on the most-crowded megacap AI names. The vehicle that absorbed the record inflow is the same one with no discretion when it reverses. (2026-SEP-03, Dan Niles, Excess Returns) The mechanics of the July low: Situational Awareness "wasn't necessarily the positions they had. The problem was they were running it at 4x leverage… if you have a 25% drawdown, but you have 4x leverage on it, then you're down 100%" — squeezed on both sides, software shorts rallying while semiconductor longs fell, the same shape as Long-Term Capital Management. Because the fundamentals held, the forced sale to Citadel was risk-reducing: "the market obviously ripped the next day." The retail analogue is Korean accounts that "not only go on margin but then invest in levered ETFs on margin," which were "absolutely destroyed" when SK Hynix and Samsung imploded. Dillian (2026-SEP-03): agrees deleveraging marks a low but adds the sequencing caveat — “anytime you have a leverage player that goes t.u., that usually marks a bottom,” and Citadel “got the cleanup print” on the Situational Awareness liquidation (block trades announced 27–28 July, the bulk resold by late August). The caveat: “when Bear Stearns blew up, it was March 17th of 2008 … the S&P rallied 17% over the next three months” before Lehman. “The most leverage player gets taken out first, but there’s still so much leverage in the system … my guess is there’s another Situational Awareness coming in the months down the line.” He also reassigns the crowding after Farley’s Vanda data: retail semis positioning is at a two-year low; the length sits with institutions and the multi-strategy pods. 2026-SEP-07 (Excess Returns — Dave Nadig clip, + host Matt Zeigler): the retail-froth channel is now sports betting, and the danger is its shape. Nadig: "it is rarely the big loss that's the problem… they're betting on a hundred baseball games a summer and slowly losing 10% a week, then a 20% win and they feel like heroes and the cycle repeats" — "grinding despair… much harder to recover from than the one big mistake," and "a grinding tax on math literacy." His structural worry, seconded by Zeigler: the products "live right next to financial products literally in the same Robinhood app, or pretty soon on the same exchange at the CME… if there's these qualities of gambling risk that are different than investment risk, polluting those against each other just seems like the opportunity for problems." Forehand generalises the shape to portfolios: "any strategy where the losses are very small and they accumulate over time… are the most deadly, because you don't see the moment where you feel the loss" — overtrading being the same pattern in a brokerage account. 2026-SEP-06 (Jared Dillian, ex-Lehman ETF trader, on Talking Billions): indexing went from 2% of AUM in the late 1990s to ~56–60% today, and the top seven names are ~35% of the index — so index buying is mechanically top-seven buying, and the only large-cap managers beating the benchmark are those overweighting the same seven. He names the dynamic reflexive and credits Michael Green with the deeper work. The tail he fears is liquidity, not valuation: the pandemic's 35%-in-a-month fall was "a mass liquidation of index funds… there's 150 million people who are doing the same thing." A related structural point — open-end funds' once-a-day NAV was a behavioural feature that ETFs removed. Wigglesworth (The Meb Faber Show, 2026-SEP-11) — bond ETFs passed the liquidity test; "income" labels didn't. Against Carl Icahn's warning to Larry Fink that a hyperliquid wrapper on illiquid credit would be lethal, fixed-income ETFs are "if not solving that at least ameliorating" liquidity — a flywheel for electronic, portfolio and systematic bond trading pushing credit toward equity-like behaviour ("fascinating and a little bit scary"). The fault lines he does see: products sold as income that is really return of capital (covered-call funds, Saylor's "bitcoin yield") — "dressing up a fox to look like a goose" — and same-day liquidity itself, "a massive systemic danger." Paulo Macro (Sep-12): ETF share-count vs short-interest as a positioning tell — price up + shares down + short interest up (BNO) means longs redeeming while shorts borrow from a shrinking float, a sharper squeeze setup than the creation-fed shorting of ARKK in 2022. 2026-SEP-14 (CNBC ETF Edge, Justin Schack/Rosenblatt): 8 of August's 10 most active ETPs were leveraged/inverse, the other two bitcoin; retail reaching for high-risk products because "maybe I can't afford that first house"; hourly-reset levered ETFs are in the pipeline. Peterffy, 2026-SEP-16: system leverage is the risk he watches: margin loans are "very large" and bank OTC books are an unmapped "spaghetti" of counterparty exposure with no central view. But "I'm not really worried about it at this time." Michael Green (Thiel Macro, How I Invest, 2026-SEP-11): leveraged ETFs create their own buying and selling — $100 in a 3x semis ETF (SOXL) must buy another $60 after a 10% rise. Retail has used 3x products as conviction holds since February; volatility drag puts break-even for someone buying in steadily at about 150% a year. Levered SpaceX ETFs had to buy twice their exposure at launch, adding to the listing's squeeze. Professionals have historically shorted both sides of levered ETFs to harvest the drag, which damps market volatility. Jenny Harrington (Gilman Hill, Dividend Stockpile 2026-SEP-21): calls option-income 'Boomer Candy' ETFs high-fee structured products that cap both downside and upside; the trap is return of principal ($100k at 10% a year can leave you with zero), so income never grows and market value erodes. Ask an LLM about fees, leverage/derivatives, how the income is generated, and how the fund fared in the pandemic, the GFC and rising rates; many dividend ETFs 'look a lot better than they really are.' Harley Bassman on MacroVoices #550, 2026-SEP-17: there are now more ETFs than stocks. Linear (futures) leverage is fine to hold; daily-reset leverage is a day-to-week tool, and volatility drag 'will eventually burn that ETF to zero'. 10%+ yields when junk pays 6-8% usually mean return of capital. Derivative and illiquid-underlying ETFs carry a NAV liquidity mismatch that bites in a vol spike (VIX ~17, realized 10-11; MOVE realized ~68). See his commentary 'Looking Under the Hood of ETFs'.