U.S. Pre-Market Brief | 2026-08-20 (ET) Thursday

Coverage window: 2026-08-19 16:00 ET regular-session close → 2026-08-20 08:28 ET (includes the 8/19 after-hours session, the overnight Asia/Europe sessions, and this morning's pre-market). About 60 minutes to the open (09:30 ET).

Quote conventions:

• For single stocks and ETFs, the prior close (= 8/19 close), same-day percentage change, pre-market price, and pre-market volume are taken from the stockanalysis.com quote API, read at 08:12–08:28 ET, timestamped item by item in the text. Note: the update timestamps differ by ticker, and pre-market volume is a cumulative figure — so wherever this brief compares activity across tickers, it uses a single re-pull taken at 08:27–08:28 ET and converts to "percent of average daily volume (%ADV)" and notional turnover, never raw share counts.
• Index closes, futures, Treasury yields, VIX, and crude are taken from the CNBC quote API, read at 08:08–08:19 ET, timestamped item by item.
• All earnings figures come from company press releases / SEC 8-K originals, with links in the text. Every figure this brief derives itself is labeled "estimated" and is never mixed in with company disclosure.

---

⚠️ Read This First: The Most Important Fact This Morning Is a Timeliness Correction

Most financial media this morning (including Yahoo Finance's live blog) are still writing "Treasury yields are falling: 10Y down 5bp to 4.65%, 30Y down 9bp to 5.19%." Those are the 8/19 closing numbers, not this morning's.

This brief took two readings 11 minutes apart, and the levels agree:

	Pre-announcement (8/18 close, back-solved from the 8/19 close and that day's change)	8/19 close (day of the support operation)	08:08:22 ET	08:19:39 ET	Retracement
10-year	4.704%	4.647% (−5.7bp)	4.704%	4.706%	~100% (a full round trip)
30-year	5.286%	5.196% (−9.0bp)	5.253%	5.258%	~69%
2-year	—	4.179%	4.190%	4.190% (+1.1bp)	—

Table scrolls horizontally

The only accurate statement is this one: the 10-year has made a full round trip back to its pre-announcement level (4.706% vs 4.704% pre-announcement, 0.2bp higher — effectively flat); the 30-year has retraced about 69% and is still 2.8bp below its pre-announcement level.

⚠️ Three things this brief explicitly does not say (all three are easy to write off the cuff and none of them holds up):

• Not "yields hit a new high": the 10Y touched roughly 4.74% intraday on 8/19, and today's 4.706% is still below that peak.
• Not "the entire 30-year decline has been erased": it has retraced only about 69%, which is not the same thing as the 10-year.
• Not "TLT provides cross-verification": TLT is −1.06% pre-market (08:19 ET, 954,000 shares), consistent in magnitude with the 30Y move (duration ~17), but what it tracks is long-end Treasuries themselves — it and the yield reading are not two independent pieces of evidence. It can only show the quotes are fresh; it cannot add weight to the "rates are rising" call.

⚠️ Also not saying "the rescue failed": ① the tool itself is a bond-market liquidity management tool, and a per-operation cap of $2 billion → $4 billion is tiny relative to the stock of Treasuries outstanding (which just crossed $40 trillion), so it should never have been expected to have lasting overnight effect; ② this morning's bear steepening (2Y +1.1bp, 30Y +6.2bp) can equally be explained by long-duration supply and pre-data risk aversion, and this brief has not ruled that out. "The single-day effect of the buyback did not persist" is a fact; "the buyback tool failed" is an interpretation, not a conclusion.

Why this still leads the brief: 8/19's sector action — homebuilders (DHI +4.24%, TOL +4.00%, PHM +3.67%) and precious metals (GLD +3.84%, SLV +4.47%) — was widely attributed to "falling yields," but this brief has not verified that correlation, so the attribution is itself just a narrative. What is certain: if you open the session assuming "yields are still falling," that assumption no longer holds this morning for the 10-year.

---

0. One-Sentence Summary

Today is a split tape: "the crypto short squeeze is the only direction with genuine volume behind it, but not a single name in it has improving fundamentals; and the two retail/farm-machinery 'beats' were both built on the same tariff refund."

① The strongest theme is digital assets. Bitcoin pushed above $70,000 overnight (this brief measured CoinGecko at $71,909, 24h +11.6% at 08:06 ET, the highest since 6/2), with roughly $2.7 billion of short liquidations in 24 hours — the largest forced-liquidation wave on record since 2021, with over 160,000 traders liquidated; about $1 billion of that was bitcoin shorts wiped out within roughly an hour. Over the same period, Trump met with executives from Coinbase, Kraken, and Blockchain.com at the White House.

On "volume," the method first: cross-ticker pre-market activity can only be compared using "pre-market volume as a percent of each name's own average daily volume (ADV)" or notional turnover — never share counts — AZO trades at roughly $2,960, so 1,000 shares equals 17,500 shares of COIN. The table below is a single re-pull taken at 08:27–08:28 ET (pre-market volume is cumulative, so snapshots taken at different moments are not comparable, hence the uniform re-pull):

⚠️ This table overturns one of this brief's own first-draft claims. The first draft said "crypto is the only direction in the whole market with genuine volume" — on a %ADV basis that statement is false: the two most active names are actually FUTU (38.96%) and WMT (19.82%), both trading on their own earnings; the largest notional is MRNA ($1.034 billion).

After the unit correction, the only formulation that still holds is this one: crypto is the only sector today where multiple tickers simultaneously show 4%–10% %ADV, move in the same direction, and none of them has a company-specific catalyst of its own. The high activity in FUTU, WMT, and MRNA is single-stock event driven and does not constitute a sector. That distinction is the part that carries information.

② But here is this brief's most important judgment: not one name in this theme has improving fundamentals — and the one that is improving happens to have the smallest crypto exposure.

Conclusions after sub-agents checked the SEC originals one by one (details in §5.2):

• COIN: Q2'26 net revenue $1.154 billion, −17.3% YoY; operating loss −$160.7 million; three consecutive loss-making quarters, TTM net loss $987.8 million (no P/E). More telling — stablecoin revenue of $292.2 million, which should be counter-cyclical, is also −5.4% YoY, while operating expenses rose +16% against the trend.
• MSTR: holds 840,447 BTC at an average cost of $75,385. At BTC $71,909 it is still down about 4.6% on those holdings — nowhere near back to cost. The 8-K also discloses that on 7/1–7/5 it sold 2,225 BTC at $60,773 (below cost) to fund preferred dividends — preferred dividends run about $1.68 billion a year, while the software business generates only about $330 million of annualized gross profit. (Note: that sale was six weeks ago and is not a catalyst for today; and the holdings data is as of 8/16, so a newer weekly disclosure may already exist.)
• MARA: Q2 revenue −27%, net loss −$611.3 million, BTC holdings −29%, plus a post-quarter $600 million bitcoin-collateralized loan (weighted cost 7.56%).
• HOOD: Q2 revenue $1.308 billion (+32.3%), net income $561 million, operating margin 43.9% — the only one with both revenue and profit growing. And its crypto trading revenue is just 7.6% of the total (16.2% a year ago).

So the correct reading is: this is a short squeeze beta ignited by crowded positioning, not a fundamental turn. It is tradeable, but anyone treating it as an allocation case is buying a sector whose most recently disclosed quarter was all losses or revenue declines. And every name is still 36%–69% below its 52-week high — this is a bounce inside a deep bear, not a new-high structure.

③ The two "beats" from Walmart and Deere were built on the same tariff refund. The origin is the refund wave triggered by the 2026-02-20 Supreme Court ruling striking down IEEPA tariffs, which has already contaminated headline numbers this week at HD (8/18), TGT/LOW/TJX (8/19), and WMT and DE (this morning):

The difference between the two is honesty, not degree of contamination: Deere put the amount in its press release, Walmart did not. Reading only the 8-K understates the contamination; reading only the headline turns a one-off into an operating trend.

WMT is −6.00% pre-market ($114.30 → about $107.44, pre-market volume 6.80 million shares = 19.82% of average daily volume, notional $730 million — the second most active name this morning, so the gap is credibly backed). The real damage is not in comps, it is in the second-half guidance: Q3 adjusted EPS guidance of $0.62–0.64 against $0.62 in the year-ago quarter is 0% to +3.2%; back-solving from the company's own base, Q4 is implied at $0.70–0.79 against $0.73 last Q4 — the full-year raise comes entirely from a first half that has already happened, and the second half has been guided to zero growth (estimated; method in §5.1).

④ The extreme divergence between semis/AI hardware and software is still running this morning. On 8/19: the Philadelphia Semiconductor Index (SOX) fell −2.12% while the S&P rose +0.21% on the same day. AI hardware was hit across the board — STX −7.87%, WDC −6.87%, DELL −6.64%, LRCX −6.33%, COHR −6.19%, FN −5.81%, LITE −5.23%; yet software surged that same day — NOW +6.45%, CRM +5.07%, WDAY +4.08%, ADBE +3.55%. This is capital rotating from "training-cluster hardware" toward "inference efficiency / AI agent monetization." Note: yields fell that same day — if this were "duration de-leveraging," software could not have rallied hard in the opposite direction, so a rates-based attribution does not hold here.

⑤ Geopolitics is lifting crude, and energy names began following this morning. On 8/19 Trump announced an "ECONOMIC D-DAY" against Iran, threatening sanctions on any country providing Iran financial, airport, or corporate access; the day before, the UAE had already suspended all trade with Iran after being hit by two Iranian ballistic missiles. WTI (September) $88.67, +3.31%; Brent (October) $94.36, +2.99%. Worth noting: XOM's pre-market volume expanded from 56,000 shares to 480,000 shares in 40 minutes (+1.66%), with XLE +1.20% — one of the few directions this morning where volume is accumulating.

Driver types: ① liquidity/policy (crypto squeeze + White House meeting, S) ② earnings and guidance (retail + farm machinery, S) ③ macro (long-end yield rebound + 08:30 ET jobless claims, A+) ④ geopolitics (economic war on Iran, A) ⑤ rotation (hardware → software, A).

Pre-market state (futures 08:09 ET / everything else 08:18–08:19 ET): S&P 500 futures (September) 7,688.50 (−0.52%), Nasdaq 100 futures 29,295.00 (−0.74%), Dow futures 53,134 (−0.74%); on the ETF side: SPY −0.56%, QQQ −0.84%, DIA −0.74%, IWM −0.80%. Futures have weakened steadily over the past 20 minutes (ES from −0.33% to −0.52%). VIX 15.99, +7.39% — a modest index decline alongside a nearly 7.4% jump in volatility is the divergence most worth watching this morning. Gold via GLD is −1.05% pre-market.

---

1. News Overview

Three Known Gaps in This Table (please note)

① This brief contains no analyst rating changes dated today (8/20). That is a real coverage gap, not a claim that "there were no rating changes today." The "Wall Street's top analyst calls" style articles returned by search were all discarded after checking their publication dates one by one — the widely cited piece covering "Oppenheimer downgrades Walmart, Deutsche Bank upgrades Palantir to a $200 target, JPMorgan downgrades Nike" was actually published 2026-08-04, 16 days ago. Recurring columns like these have near-identical headlines and URLs across different dates and are extremely easy to mistake for same-day catalysts. (Item 14, Marvell/Google, was cross-checked against the SEC filing and confirmed as 8/19, so it is usable.)

② This brief contains no intraday data for today: the 08:30 ET jobless claims, plus the Philadelphia Fed and existing-home sales prints possibly scheduled for today, are all released after this brief is finalized.

③ All consensus figures come from third-party sell-side aggregators; this brief did not obtain primary consensus detail, and different sources contradict each other — for WMT's FY27 EPS consensus, one source gives $2.90 and another $2.97; for Deere's equipment revenue expectation, one gives $10.73 billion and another $10.916 billion. Any comparison involving consensus in this brief should be treated as an order-of-magnitude reference, not a precise conclusion; company disclosures and guidance, by contrast, all come from SEC originals and can be relied on.

---

2. Strongest Themes, Descending

---

3. Master Stock Strength Ranking

The scoring model is in §4. The "pre-market" column format is: percentage change / pre-market volume / read time (ET).

3A. Positive Side

3B. Negative Side

---

4. Stock Scoring Model (100 pts total)

Shown component by component using this brief's number-one name, HOOD (74 pts), against the downgraded COIN (62 pts) — the two are in the same theme, and the entire gap comes from fundamentals and the expectation gap:

Why WMT is not on the positive list: source authority 15/15 and catalyst directness 20/20 (both from its own 8-K), but earnings elasticity is near zero because the second half is guided to zero growth, the expectation gap is negative, and the risk deduction is maxed out (−15) — the total lands in negative territory.

---

5. Top Stocks — Detailed Analysis

5.1 WMT — the "full-year guidance raise" is real, but everything raised is a first half that already happened

• Related news: 08-20 07:00 ET, SEC 8-K EX-99.1 (original⁴).
• Primary figures (all from the 8-K):
  • Total revenue $187.937 billion (+5.9%, +5.1% at constant currency); net sales $186.100 billion.
  • GAAP EPS $0.80 (vs $0.88 a year ago, −9.1% YoY); adjusted EPS $0.81 (vs $0.68, +19.1%). Only two adjusting items: excluding $0.12 of net losses on equity and other investments, and excluding $0.11 of net benefit from a tax matter.
  • ⚠️ The tariff refund was not excluded from adjusted EPS. The company's own wording: "gross margin improved 96bp… primarily driven by tariff refunds"; "operating income +28.8%, adjusted constant-currency +17.4%, which includes the impact of tariff refunds received… setting that net impact aside, underlying operating income growth sits at the upper end of our guidance range." The press release does not disclose the refund amount; CFO Rainey told CNBC the company can collect roughly $2.9 billion (CNN¹⁶) — note that this is a collectible/cash figure, not the amount recognized in the Q2 P&L.
  • U.S. comps +2.6% (vs +4.6% a year ago): traffic +1.5% (exactly flat versus +1.5% a year ago), ticket +1.1% (vs +3.1% a year ago); e-commerce contributed roughly 510bp to comps (about 420bp a year ago).
  • Two different drag measures, presented here as-is without reconciling them ourselves: the summary page says "including an 80bp drag from health and wellness"; the U.S. segment page says "125bp of drag from prescription drug deflation (the maximum fair price rules effective January 1)."
  • U.S. segment operating income $8.120 billion (+20.6%), gross margin 29.4% vs 27.9% (+158bp, including the refund), with the expense ratio deteriorating 72bp.
  • Sam's Club comps (ex fuel) +4.4%: traffic +7.0%, but ticket −2.5% — that is a trade-down signal.
  • Global e-commerce +23%, global advertising +38% (Walmart Connect ex VIZIO +43%), global membership income +17%.
  • H1 operating cash flow $19.710 billion (up $1.4 billion), capex $14.181 billion (+24%), free cash flow $5.529 billion, down $1.4 billion YoY. Cash $11.5 billion, total debt $57.2 billion.
• Catalyst logic (the decisive layer):
  • Q3 guidance: net sales +3.0% to +3.75% (cc), adjusted operating income +2.0% to +4.0%, adjusted EPS $0.62–0.64. The $0.63 midpoint against $0.62 in the year-ago quarter is 0% to +3.2%.
  • All three FY27 guidance items were raised: net sales +3.5–4.5% → +4.0–5.0%; adjusted operating income +6.0–8.0% → +7.0–8.5%; adjusted EPS $2.75–2.85 → $2.80–2.87. But capex was simultaneously raised from about 3.5% of net sales to about 4.0% (roughly $3.5 billion more), which is a net negative for free cash flow.
  • ⚠️ The "raise" is relative to the company's own May guidance, not to sell-side consensus. Third-party aggregated FY27 consensus is $2.90 from one source and $2.97 from another — on either one, the $2.87 top of guidance sits below it.
  • [Estimated] The second half is guided to zero growth: back-solving from the company's own base, the first half delivered $1.46; adding Q3 guidance of $0.62–0.64 against the full year of $2.80–2.87 → Q4 is implied at $0.70–0.79, versus $0.73 last Q4. In other words, the full-year raise comes entirely from a first half that already happened.
  • The CFO's own words: "Our operating income outlook reflects reinvesting the tariff refunds received in Q2 into customer experience and price investment in the second half. For that reason I'd encourage you to look at Q2 and Q3 together to assess the underlying growth of the business." — the company itself concedes that neither Q2's +17.4% nor Q3's +2–4% is a true growth rate.
  • There is also a technical factor: the Q3 sales guidance includes "a shift in the timing of the Flipkart big-billion event between Q3 and Q4, creating more than 100bp of growth headwind," so the underlying figure behind the 3.0–3.75% is roughly 4.0–4.75%.
• ⚠️ On "was the comp miss caused by regulation" — this brief has to be precise, and there are three layers here:
  • ① What is certain (all from the 8-K original): traffic +1.5%, exactly flat versus a year ago; the entire deceleration came from ticket (+1.1% vs +3.1% a year ago). The company explicitly attributes 125bp of comp drag to the maximum fair price rules for prescription drugs effective January 1. "Traffic did not break" is the hardest fact of the quarter.
  • ② What this brief does not do, and asks clients not to do either: add the 125bp back to get 3.85% and declare "it was actually a beat." There are two independent errors here: first, the rules have been in effect for more than two quarters since January 1, so sell-side models most likely already embed them and adding back double-counts; second and more fundamentally — the 3.7% consensus is built on a standard basis that does not exclude the item, so comparing a custom basis against a standard one is invalid on its face, regardless of whether the rules were already modeled.
  • ③ And the bridge itself does not balance: comps decelerated from 4.6% to 2.6%, a 200bp deceleration; the 125bp explains only about 60% of it, leaving roughly 75bp unexplained. At the same time the 8-K body says "125bp of prescription drug deflation" while the lead paragraph says "80bp from health and wellness," and the company itself does not reconcile the two measures, so this brief will not reconcile them either. So even "it was mostly caused by regulation" overstates it; the accurate phrasing is "more than half."
  • There is only one correct conclusion: the character of this deceleration leans toward "price/mix" rather than "traffic loss," but it neither overturns the fact that comps missed nor supports the claim that it was actually a beat.
• Valuation (read at 08:19 ET): prior close $114.30, pre-market $107.60. On TTM GAAP EPS of roughly $2.77, the P/E is about 38.8× (at the pre-market price); on the FY27 guidance midpoint of $2.835, the forward P/E is about 37.9×; P/S about 1.16×. And the EPS growth implied by guidance is only about +7.4% — an asset with 6% revenue growth and 7% earnings growth trading near 38× has almost no room for guidance error. That is the complete explanation for the −6%.
• Pre-market and technicals: −6.00%, pre-market volume 6.80 million shares (08:28 ET), %ADV 19.82%, notional $730 million — the second most active name this morning after FUTU (38.96%). The gap has genuine absorption behind it; it is not a low-activity head fake. The 52-week range is $95.42–$135.155.
• Final call: watch only, and do not short. The negative is hard (zero-growth second-half guidance plus a near-38× multiple); but −6% has already priced a fair amount, and the layer of "traffic did not break + advertising +38% / membership income +17% / e-commerce +23%, three high-margin businesses completely unscathed" clearly has not been read by the market this morning, so the risk of an intraday retracement is not low.

5.2 Digital assets — the strongest theme, but please separate "trading" from "allocation"

• Related news: ① BTC pushed above $70,000 overnight; this brief measured CoinGecko at $71,909 (24h +11.6%) at 08:06 ET, the highest since 6/2; ② roughly $2.7 billion of short liquidations in 24 hours, the largest on record since 2021, with more than 160,000 traders liquidated, including about $1 billion of BTC shorts closed out within roughly an hour; ③ Ethereum's largest single-day gain was about 12%; ④ on 8/19 Trump met Coinbase, Kraken, and Blockchain.com executives at the White House.
• Catalyst logic: this is a short squeeze ignited jointly by crowded shorts, improving liquidity expectations, and a friendly policy signal — not trend buying from new money. The Treasury expanding long-bond buybacks was read by the market as a liquidity injection, and that was the macro trigger — but that trigger already stopped working this morning (see the correction box at the top), while the crypto rally continues, which means the squeeze has entered a self-reinforcing phase and has decoupled from its original macro rationale.
• ⚠️ Fundamental verification (sub-agents checked the SEC originals one by one — this is the core of this section):

• ⚠️ One premise to correct regarding CRCL: many commentaries tie CRCL to "the 10Y yield," which is wrong. Its reserves are short-duration Treasuries/repo, anchored to fed funds/SOFR, not the 10Y. The company itself quantifies the sensitivity in its 10-Q: Q2 reserve revenue rose $33.5 million YoY, of which +$147.4 million came from +25.2% circulation growth and −$113.9 million from a 66bp decline in the average yield → every 1bp ≈ $1.7 million per quarter, roughly $6.9 million per year. And the 3-month bill now sits at 3.797%, so rate cuts are happening. That means this morning's long-end rebound does nothing for CRCL, while the front-end decline is a genuine negative — it is the name in the group whose 8/19 gain has the weakest attribution.
• Theme stage: mid-fermentation, leaning crowded. The criterion is "yesterday it rose on volume, and this morning it is still rising on volume"; but the whole group is still 36%–69% below its 52-week high (MSTR −69%, BMNR −67%, COIN −58%, MARA −56.5%, CRCL −49%, IBIT −43.5%, HOOD −36%) — a deep-bear bounce, not a new-high structure. That determines that positions here are trades, not allocations.
• Final calls:
  • HOOD, priority deep-dive — the only one in the group with improving fundamentals, and it rose the least precisely because the market is not pricing it as a crypto stock.
  • IBIT, watch closely — the cleanest instrument for expressing this theme, with no single-company operating risk.
  • COIN, watch closely (downgraded from "priority deep-dive") — the highest catalyst directness, but three straight loss quarters plus shrinking stablecoin revenue plus rising expenses; a BTC rally can repair volume, it cannot repair the cost structure.
  • MSTR / CRCL / GLXY watch only; MARA / BMNR / SBET avoid.

5.3 DE — the "big beat" is distorted in two places, and this brief downgrades it to watch only

• Related news: 08-20 pre-market, SEC 8-K (EX-99.1 press release⁵, EX-99.2 call materials⁶).

• ⚠️ Distortion one: mismatched revenue definitions (verified and confirmed). Several aggregator sites are circulating "revenue $12.608 billion vs $10.916 billion expected" and calling it a "15% beat." The press release is explicit: "worldwide net sales and revenue increased 5% to $12.608 billion… equipment operations net sales for the quarter were $10.999 billion." The three segments sum ($3.998 + $3.383 + $3.618 billion) to $10.999 billion, matching the equipment definition exactly — the sell-side consensus corresponds to equipment operations net sales. The actual $10.999 billion against expectations (of $10.73 billion or $10.916 billion, depending on the source) is a beat of only 0.8%–2.5%, not 15%. Comparing total revenue against an equipment-basis expectation conjures roughly $1.7 billion of beat out of thin air.

• ⚠️ Distortion two: almost all of the YoY growth comes from the tariff refund. The press release states it plainly: "the company recognized tariff recoveries of $110 million and $382 million for the third quarter and first nine months, respectively" (stemming from the 2026-02-20 Supreme Court ruling striking down IEEPA tariffs; $272 million was booked in Q2, and 272 + 110 = 382 ties out).

  • [Estimated, not company disclosure] $110 million pre-tax, at a 25% tax rate and 270.4 million diluted shares ≈ $0.30/share; ex that, FQ3 EPS ≈ $4.79, only +0.9% against $4.75 a year ago (versus a headline +7.4%). On the same basis, net income is up only about +0.6% YoY — lower than EPS's +0.9%, with the difference coming from buyback-driven share count reduction; the two measures should be read together.
  • ⚠️ This estimate rests on three unverified assumptions, so please do not cite it as fact: ① 25% is an assumed tax rate, and Deere's actual effective rate is usually lower (the company's full-year tax rate guidance is 24–26%); ② only the tariff "recoveries" were removed, not the tariff "costs" still being incurred in the period — that is a one-sided treatment of a one-off item, and the net impact is unknown; ③ it was not verified whether the year-ago quarter contained similar one-off items, and if the prior-year base is also unclean, the YoY comparison is equally distorted. The accurate net one-off impact has to wait for the call and the 10-Q.

• Confirmable primary figures: FQ3 net income $1.379 billion / EPS $5.10 (vs $1.289 billion / $4.75 a year ago); full-year net income guidance narrowed and raised from $4.5–5.0 billion to $4.75–5.00 billion — only the floor was lifted, the ceiling was untouched. [Estimated] Back-solving from the $3.808 billion already delivered over 9M, Q4 is implied at $940 million–$1.19 billion of net income, versus $1.068 billion last Q4 — essentially flat at the midpoint.

• Segment breakdown — all the growth is outside large agriculture:

The core large-ag business is still contracting; C&F operating profit nearly doubled, which the company attributes "primarily to favorable price realization" — that is price increases, not volume. And within the full-year segment outlook, the only one cut was P&PA: from "down 5–10%" to "down about 10%"; South American ag went from "about −15%" to "−15 to −20%," Europe from "flat to +5%" to "flat," and global forestry from "about −5%" to "about −10%." The agriculture-side outlook was cut across the board.

• On "2026 is the bottom of the farm machinery cycle" — this brief's previous version got this wrong, and it is corrected here: this is the company's own statement, not a sell-side inference. The third bullet of the press release reads "order book trends confirm 2026 as the bottom of this agricultural machinery cycle," and CEO John C. May explicitly says "we continue to believe 2026 will mark the bottom of this agricultural cycle," citing early order program trends, improving used equipment inventory, and rising precision-ag adoption.
  • But there is counter-evidence in the same materials: as of July 2026, Deere's North American dealer new-machine inventory as a percent of trailing-twelve-month retail sales was 33% for 100+ hp tractors, above 31% a year ago. The company is talking about "used" inventory improving, while the "new" channel inventory it discloses is actually higher — the two do not contradict each other, but they are extremely easy to conflate. The cycle-bottom judgment rests on an order book whose detail is not public and cannot be verified externally; it is a company opinion.
• Cash flow quality is on the weak side: 9M consolidated operating cash flow of $3.250 billion versus $3.464 billion a year ago (−6%), against 9M net income of $3.802 billion → OCF/net income of just 0.85×. The main drag is "receivables related to sales" at −$1.252 billion (versus −$494 million a year ago). A cycle-bottom year should normally release working capital; that is not happening here.
• Valuation (read at 08:15 ET): prior close $580.63 (−1.37%), pre-market +2.15% (08:27 ET) on 30,100 shares = 1.89% of average daily volume — very low activity, so this gain is far less credible than WMT's (%ADV 19.82%). TTM P/E about 33×; on FY26 guidance ($17.57–18.49), about 32–34×. Note: the "forward P/E of about 28×" often quoted externally corresponds to FY2027 (ending October 2027), a recovery-year estimate, not FY26 — the multiple the market is paying rests entirely on "2026 is the bottom, 2027 recovers." The 52-week range is $433–$674.19.
• Final call: watch only (downgraded from "watch closely"). Three reasons: ① after correcting for the definition mismatch, the beat is only 0.8–2.5%; ② ex the tariff recovery, EPS is up only about +0.9% YoY (estimated); ③ core large-ag revenue is −6% and operating profit −9%, and it is the only segment whose full-year outlook was cut. The results themselves are not bad, but the phrase "big beat" does not hold up. Verification point: management's specific language on the order book at the 10:00 ET call — that is the only route by which "cycle bottom" moves from company opinion to verifiable fact.

5.4 Long-end rates — the item most easily distorted by the media today

• Chain of facts: after long-end yields hit their highest levels since 2007 on 8/19, the Treasury announced it would "at least double" the size of its 10–30 year liquidity support buyback operations (per-operation cap $2 billion → at least $4 billion); the 10Y closed that day at 4.647% (−5.7bp) and the 30Y at 5.196% (−9bp). Since the outbreak of the Iran conflict the 10Y has risen nearly 70bp in total, touching roughly 4.74% intraday, and has pushed the 30-year mortgage rate to about 6.75%.
• Measured this morning (CNBC quote API, two readings): 08:08:22 ET → 10Y 4.704%, 30Y 5.253%; 08:19:39 ET → 10Y 4.706%, 30Y 5.258%, 2Y 4.190%. The two readings 11 minutes apart agree on level, which rules out a single-pull data failure.
• ⚠️ One data flaw that must be disclosed: CNBC's change field does not reconcile with "level minus the 8/19 close as reported by the media," differing by about 0.6bp (the API gives 10Y +0.053, while 4.706 − 4.647 = +5.9bp) — indicating a small difference between CNBC's internal prior close and the media convention. This brief uses levels exclusively and back-solves changes itself; it does not cite the change field.
• Retracement (back-solved by this brief): working back from the 8/19 close and that day's change, the pre-announcement (8/18) closes were 10Y 4.704% and 30Y 5.286%. Therefore:
  • 10-year: now 4.706%, a completed round trip (~100% retraced), 0.2bp above pre-announcement — effectively flat.
  • 30-year: now 5.258%, about 69% retraced, still 2.8bp below pre-announcement.
  • These two are not the same thing; please do not merge them into "the decline has been entirely erased."
• On TLT: −1.06% pre-market (08:19 ET, 954,000 shares), consistent in magnitude with the 30Y move (duration ~17). But what TLT tracks is long-end Treasuries themselves — it can show the quotes are fresh, but it is not an independent second source of evidence. This brief does not describe it as "cross-verification."
• Implications (and the competing explanations this brief has not ruled out): the move is concentrated at the long end (30Y > 10Y > 2Y) and the curve keeps steepening. That looks more like term premium than rate-hike expectations — a single $4 billion buyback against a stock of Treasuries that just crossed $40 trillion is more symbolic than substantive. But note: this morning's bear steepening can equally be explained by long-duration supply and by position adjustment ahead of the 08:30 data, and this brief has not ruled those out. "The single-day effect of the buyback did not persist" is an observable fact; "the buyback tool failed" is only one interpretation. (Note: Jackson Hole is on 8/27–29, not this week, so it is not an explanatory factor for today.)
• Sector implications: the groups that rose on 8/19 and were widely attributed to "falling yields" — homebuilders (DHI +4.24%, TOL +4.00%, PHM +3.67%, LEN +2.78%) and precious metals (GLD +3.84%, SLV +4.47%) — carry an attribution this brief has not verified for correlation, and precious metals in particular are driven more by real rates and the dollar. What can be observed: this group is indeed giving it back pre-market this morning (GLD −1.05%, SLV −1.07%, LEN −0.87%, TOL −1.06%). Conversely, the banks that fell on 8/19 (JPM −1.65%, GS −1.81%, KRE −2.41%) still show no reaction pre-market this morning (JPM −0.25%, KRE −0.12%, both with %ADV under 0.3%, so the evidence is very weak).
• ⚠️ The main falsification point for this item: 08:30 ET initial jobless claims. The prior print was 209,000 (above the 202,000 expected). If this print rises materially again, the long end could turn lower on the spot and this call would be overturned — go with the actual data.

5.5 MRNA — do not describe it with a single-point quote

• Related news: there is no new negative. On 8/19 Merck/Moderna announced that the Phase III INTerpath-001 trial of the personalized mRNA cancer vaccine intismeran autogene in combination with Keytruda met its primary endpoint, and MRNA closed +176.97% at $174.38, the largest single-day gain in company history (199 million shares traded); MRK closed +12.6%, BNTX +21.96%, XBI +5.90%, XLV +3.51%. This morning is pure profit taking.
• ⚠️ The pre-market quote must be given as a range: MRNA fell as low as $151.78 (about −13.0%) in this morning's pre-market, and this brief measured $156.00 (−10.54%, pre-market volume 6.27 million shares) at 08:19 ET. Any single-point citation will mislead; where the open lands within that range cannot be predicted.
• Fundamental reminder: the company posted a net loss of $782 million in Q2 2026, with persistently negative operating cash flow. The Phase III trial has so far published no quantitative efficacy data (no HR, no absolute benefit) — the market is pricing Phase III off the HR=0.51 from the Phase II KEYNOTE-942, while the Phase III has n=1,137 and includes lower-risk IIB/IIC patients. That evidence gap existed yesterday and still exists today.
• Final call: avoid. Not because the logic broke, but because on a name that ran +177% and then gave back 13% within 24 hours, pre-market pricing carries no information at all.

---

6. Negative / Avoid List

---

7. Within-Theme Rankings

Theme 1: Digital asset short squeeze

Theme 4: AI hardware → software rotation

Theme 5: Economic war on Iran lifts crude

---

8. Open-Verification Signals

Pre-market (before 09:30)

1. Whether the crypto group's volume holds: IBIT's pre-market volume rose from 11.21 million shares to 12.59 million shares over the 21 minutes from 08:07 to 08:28 (ADV 141 million shares, %ADV 8.94%), still accumulating but with a flattening slope. If it is pushed above 15 million shares (%ADV about 10.6%) before the open, the theme holds; if it stalls near 13 million shares, the bid dried up after 08:00.
2. Whether WMT is gap-and-go or gap-fill: 6.80 million pre-market shares at %ADV 19.82% is hard volume. Key observation: whether it can reclaim $110 (half the gap) in the first 30 minutes. If it can, the market understood "traffic did not break"; if it cannot, the zero-growth second-half guidance has the upper hand.
3. Whether the gap between energy stocks and crude closes: crude +3.31% versus XOM +1.66% / XLE +1.20%, a gap of roughly 1.6–2.1 percentage points. XOM's pre-market volume has already expanded from 56,000 to 480,000 shares — the single most trackable volume change this morning.

Intraday (09:30–10:00)

4. The 08:30 ET initial jobless claims print is today's first hard data and this brief's main falsification point. The prior print was 209,000 (above the 202,000 expected). If this print rises again, the long end could turn lower and the opening section's "the rescue-driven decline has been erased" call would need on-the-spot revision.
5. Sector ETF confirmation: if the retail selloff is real, you should see XLP / XLY fall on volume (pre-market XLP −1.17% / 16,900 shares, XLY −0.67% / 367 shares — far too thin to constitute evidence at this point).
6. Whether banks catch up: if the long end keeps rising while KRE / JPM still do not, the market is worried about credit rather than margins, which is the more negative read.
7. DE's 10:00 ET earnings call: management's specific language on the order book determines whether "2026 is the cycle bottom" can move from company opinion to verifiable fact.

Options Sentiment

8. The divergence between VIX at 15.99 (+7.39%) and futures down only 0.52–0.74% is the item most worth watching this morning. A small index decline alongside a volatility spike usually means money is buying protection for a known event — today that is the 08:30 claims print, and next week it is NVDA (8/26) plus Jackson Hole (8/27–29).
9. Crypto names have extremely high IV after the squeeze, so chasing calls offers poor value; MRNA's IV is at extreme levels after +177%, and IV crush risk applies to options in either direction.

Risks

10. Reversal after a gap: MRNA (+177% yesterday), MSTR (+12.68% yesterday and another +8.75% this morning), and BULL (up more than 20% over two days) are classic candidates for gap fills.
11. A lone riser with no sector follow-through: NDSN (+4.28% / 9,966 shares) and DE (+2.47% / 30,100 shares) are far too thin with no peer follow-through — do not judge the sector from them.
12. Futures are deteriorating: ES went from −0.33% at 08:00 to −0.52% at 08:09, and YM from −0.46% to −0.74%. If that direction persists into the open, a lower open is the likely outcome today.
13. This brief's own biggest risk: all yield and quote readings were taken at 08:08–08:19 ET, more than 70 minutes before the open. Pre-market pricing accumulates continuously — go with your own real-time readings just before the open.

---

9. Final Conclusions

① The 5 Stocks Most Worth Watching Today

② The 3 Strongest Themes Today

③ Directions to Avoid Today + Why

1. AI hardware and storage (STX / WDC / LRCX / DELL / COHR / FN / LITE / ALAB / CRDO / VRT): down −4.2% to −7.9% across the board on 8/19, SOX −2.12%; this morning's bounce mostly shows %ADV of 0.3%–1.2% with almost no absorption; only COHR (2.07%) and LITE (1.86%) are exceptions.
2. Cybersecurity (CRWD / PANW / FTNT / S): CRWD's forward P/E >166x plus more than $200 million of executive selling plus 8/26 earnings — this is not the moment to build a position.
3. Everything that rose yesterday on "falling yields" (homebuilders DHI/LEN/TOL/PHM, precious metals GLD/SLV): the premise no longer holds this morning, and most of them have not fallen back yet.
4. The thin-volume broad decline in retail peers (TGT/COST/HD/LOW/DG/KR/ORLY/AZO): do not treat a −2% built on 10,000–20,000 shares as sector-wide pricing — this pattern is the easiest to falsify after the open.
5. The high-leverage end of the crypto chain (MARA / BMNR / SBET / CLSK): their most recent quarters were all revenue declines or large losses, and MARA's new borrowing is collateralized by BTC — that is positive feedback on the way down, not a cushion.
6. MRNA: +177% → −13% within 24 hours; pre-market pricing carries no information and both directions are dangerous.
7. IREN: the revenue mix has already shifted from mining to AI cloud, but the latest financials stop at 3/31 and are 5 months stale, and debt quadrupled over three quarters — stop buying it as a beneficiary of a BTC rally.

④ Final One-Sentence Call

There are really only three things to do today. First, do not take at face value the "falling yields" the media are still quoting this morning — this brief measured twice, at 08:08 and 08:19, and the 10-year has made a full round trip back to its pre-announcement 4.704% (now 4.706%), while the 30-year has retraced about 69% and is still 2.8bp below pre-announcement; both remain below the 8/19 intraday highs, so this is not a "new high," nor is it appropriate to flatly declare "the buyback tool failed" (this brief could not rule out the competing explanation of long-duration supply) — but the premise that "yesterday's rally was driven by falling yields" no longer holds this morning, at least for the 10-year. Second, crypto is the only sector today where multiple tickers simultaneously post 4%–10% %ADV, move in the same direction, and none has a company-specific catalyst, and it is worth trading — but be clear-eyed: not one name in this sector has improving fundamentals in its most recent quarter, COIN has three straight loss quarters with stablecoin revenue shrinking too, MSTR is still 4.6% underwater on its coins at BTC $71,909 (average cost $75,385), and the one name with both revenue and profit growing, HOOD, happens to be the one whose crypto exposure is down to 7.6%; this is squeeze beta, not a fundamental turn. Third, the two "beats" from Walmart and Deere were built on the same IEEPA tariff refund, with Deere spelling out $110 million and Walmart not saying a word — on this brief's estimates, ex-refund the former's EPS is up only about 0.9% and net income only about 0.6% YoY, while the latter's adjusted EPS is roughly $0.75–0.76 rather than $0.81; the two moved in opposite directions this morning, but what they genuinely share is that neither company's headline number this quarter can be used as-is.

---

Data-Sourcing and Execution Log (internal)

Data-sourcing failures

• yfinance returned YFRateLimitError across the board on this machine, and a sub-agent confirmed an IP-level ban (not a leftover process). All quotes in this brief were rerouted to stockanalysis.com (/api/quotes/s/<sym>) and the CNBC quote API, and financial data to SEC EDGAR originals and XBRL; not a single number here comes from yfinance.
• WebFetch returned 403 for cnbc.com, benzinga.com, and sec.gov alike. SEC originals were downloaded instead via curl with a User-Agent (work/wmt_q2fy27.htm), which worked; every key Walmart figure comes from that 8-K original.
• The official stockanalysis /api/screener route kept returning 404, so the pre-market gainers/losers list was rebuilt by reverse-engineering the SvelteKit flat array in the page's __data.json (work/sa_sk.py). That entry point's market-cap filter parameter does not work — adding ?f=marketCap-over-2000000000 still returns microcaps (SGLY +119%, RITR +78%). So this brief's large-cap coverage came from hand-building a theme-based candidate pool (about 170 names) and pulling quotes one by one, not from a whole-market ranking. That is a genuine difference in coverage method and may miss large-cap movers outside the theme pool.
• The CNBC futures symbols @ES.1 / @NQ.1 / @YM.1 / @RTY.1 returned empty records, so @SP.1 / @ND.1 / @DJ.1 were used instead and worked. Index closes used .SPX / .IXIC / .DJI / .RUT / .SOX, all carrying 8/19 timestamps, verified as not stale.
• stooq futures (es.f and friends) all returned 404; the stockanalysis index endpoint (/api/quotes/i/spx) returned 400. Both backup channels were unavailable, so indices and futures rely on CNBC alone.
• Ticker-level failures: BITF and SQ returned 404 (ticker changed/delisted, not an API failure).

Sub-agents

• fundamentals-analyst ×2 (crypto group / WMT+DE group), risk-auditor ×1. All three agents overturned substantive conclusions in this brief's first draft, confirming once again that the sub-agents are worth waiting for.

The two methodological errors caught by risk-auditor (the most valuable QA pass in this brief)

• ① The yield arithmetic contradicted itself. The first draft wrote "10Y +5.1bp, 30Y +5.9bp, the entire rescue-driven decline has been erased." The audit pointed out: the 10Y closed at 4.647% on 8/19 after a −5.7bp day ⇒ the pre-announcement (8/18) close was 4.704%, and the current reading is also 4.704–4.706%, so the 10Y is a "full round trip / flat," not "above the pre-rescue level"; while the 30Y has retraced only 69% (5.258% vs 5.286% pre-announcement) and is still 2.8bp below it — merging the two maturities into a single "entirely erased" is wrong. The audit also found that CNBC's change field differs by about 0.6bp from "level − media-convention prior close," so the brief was changed to use levels exclusively and back-solve changes itself. The claim that "TLT constitutes cross-verification" was also deleted (TLT tracks long-end Treasuries themselves and is not an independent source), "the rescue failed" was downgraded to "the single-day effect did not persist," and the unruled-out competing explanations were added.
• ② Cross-ticker volume comparison used incomparable units. The first draft computed "pre-market volume / average daily volume" for the crypto group while reporting only absolute share counts for the retail group, and concluded from that "crypto is the only direction in the market with real volume." The audit called this a double standard plus a dimensional error (AZO trades at roughly $2,960 and its share count is not comparable to COIN's), and noted that "thin volume is not evidence" was applied only to unfavorable samples. Everything was recomputed on %ADV plus notional (a single re-pull at 08:27–08:28), and the result directly overturned the original conclusion: the most active names are FUTU at 38.96% and WMT at 19.82%, both earnings stocks. The surviving formulation was changed to "the only sector where multiple tickers simultaneously post 4%–10% %ADV, move in the same direction, and none has a catalyst of its own." Along the way it emerged that retail peers cannot be treated as one bucket: COST 3.67%, KR 2.59%, and TJX 2.45% are an order of magnitude away from TGT 0.21% and LOW 0.13%, so they were split into two groups.
• Other audit items adopted: for WMT, that the 125bp "explains only about 60% of the 200bp deceleration" and that "comparing a custom basis against a standard basis is invalid on its face"; for Deere, three qualifiers added to the estimate — "net income +0.6% YoY," "removing the recoveries one-sidedly without removing the costs," and "the tax rate is an assumption"; for MSTR, adding "the data is as of 8/16 and may be stale" and "the July coin sale is six-week-old news."

The four claims this brief's own checks overturned (important)

1. The first draft ranked COIN first and labeled it "priority deep-dive," on the grounds that "subscription revenue provides a buffer." After a sub-agent read the 10-Q, that was disproven: stablecoin revenue is also −5.4%, operating expenses are +16% against the trend, and there have been three straight loss quarters. COIN was downgraded to sixth and "watch closely," and HOOD was promoted to first.
2. The first draft said Deere's "EPS +7.4% YoY, delivering growth in an industry guided to −15 to −20%, is high quality." The 8-K original discloses $110 million of Q3 tariff recoveries, and ex those it is only +0.9% (estimated); moreover the core P&PA segment is the only one whose full-year outlook was cut. DE was downgraded from "watch closely" to "watch only."
3. The first draft said "'2026 is the farm machinery cycle bottom' looks more like a sell-side inference than a company statement" — that was wrong. Both the third bullet of the press release headline points and the CEO's own words say exactly that. It has been corrected, and the counter-evidence from the company's own materials ("new-machine channel inventory is actually higher") has been added.
4. The first draft added WMT's 125bp back to get 3.85% and called it "actually a beat." The sub-agent pointed out that the rules took effect January 1 and sell-side models very likely already include them, so adding back double-counts. It was changed to state only the hard fact — "traffic +1.5%, flat versus a year ago, with the entire deceleration in ticket" — with no add-back.

Other

• Yahoo Finance's live blog this morning quoting "10Y 4.65% (−5bp) / 30Y 5.19% (−9bp)" was identified as a restatement of the 8/19 closing numbers, disproven by two timestamped CNBC readings plus TLT pre-market, and elevated into the correction box at the top of the brief. And per risk-auditor's direction, a self-imposed limit was adopted: do not write "the 10Y is already above the pre-rescue level," because it reached 4.74% intraday on 8/19.
• The "Walmart downgraded by Oppenheimer / Palantir upgraded by Deutsche Bank" article was checked via WebFetch and found to have a publication date of 2026-08-04, and was removed wholesale. This is the second hit on the recurring-column-headlines-hide-date class of trap; recommend hard-coding "analyst-rating articles must have their publication date WebFetch-verified first" into the pre-market workflow.
• The mutually contradictory consensus sources are disclosed in the body: WMT FY27 EPS has both a $2.90 and a $2.97 version; DE equipment revenue expectations have both a $10.73 billion and a $10.916 billion version. No primary consensus detail was obtained for either.

Data not obtained (declared honestly in the body) ① Today's (8/20) analyst rating changes; ② today's schedule-confirmed macro release list and consensus figures (other than jobless claims); ③ WMT's Q2 P&L recognition amount for the tariff refund (not disclosed in the press release; requires the call or the 10-Q); ④ primary consensus detail for WMT / DE; ⑤ a standalone balance sheet and operating cash flow for Deere's equipment operations (the consolidated figures are distorted by financial services; requires the 10-Q).

---

⚠️ Risk disclosure: this list is pre-market information review and observation only and does not constitute investment advice. U.S. equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals are common. Automatically generated content may contain timeliness gaps or factual errors — rely on company disclosures / SEC filings, and do not use this directly as a basis for trading.