Closing Briefing — Thursday, August 20, 2026

1. The buyback trade round-tripped in one session, and the shape of the round-trip is the information. Wednesday's rally was 5, 6, 11 and 9 bp at the 7-, 10-, 20- and 30-year. Thursday gave back 5, 4, 3 and 4. The 7-year fully reversed, the 10-year gave back two-thirds, the 30-year gave back 44% and the 20-year gave back only 27%, keeping 8 of its 11 bp. The 5-year, which had rallied 2 bp Wednesday, sold off 4 and is now 2 bp cheaper than before the announcement. That is not a market rejecting the policy. It is a market pricing the policy exactly where the policy said it would operate — the 10-to-20 and 20-to-30-year buckets — and taking the premium back out of everything else. Bessent escalated on CNBC ("we're going to increase the size of the buyback… it could be more than the 4 billion per issue") and the belly still cheapened. Forward catalyst: the operation schedule, and the first execution on 9 September.

2. Walmart broke the consumer trade, and the sector table shows how completely. The stock fell 9.15% to $103.84 on 82.75m shares, its worst session since 17 May 2022, after U.S. comparable sales grew 2.6% against a 3.5% FactSet consensus and third-quarter adjusted EPS guidance came in at $0.62–$0.64 against $0.68 expected. Full-year guidance was raised to $2.80–$2.87 and still sits below the $2.90 consensus. CFO John David Rainey's line — the model "is only getting stronger and more durable" — did not survive contact with the comp. Consumer defensive fell 2.16%, the worst of eleven Finviz groups, and consumer cyclical fell 1.40%. WSJ tied the two together: mixed retail results "offered more gloom on the health of the American consumer." Forward catalyst: Dollar General, Dollar Tree and Best Buy on 27 August (§5).

3. Semiconductors were the only thing that worked, four sessions after being the only thing that didn't. SOX +0.53% against an S&P down 0.87% is a 1.40-point spread, and the composition is the memory complex: Micron +3.97% to $974.29, SanDisk +2.02% to $1,600.62, SK Hynix's U.S. line +4.43%, Marvell +5.82% to $251.09, AMD +0.67%, Broadcom +0.49%. The proximate cause is Korean: SK Hynix rose about 13% in Seoul on the acceleration of its ₩40tn ($28.7bn) buyback-and-cancellation programme and a shareholder-return commitment above 50% of 2025–27 cumulative free cash flow, and Samsung Electronics rose 9.49%, taking the Kospi up 5.89%. Note the non-participant: Nvidia fell 0.33% to $216.85 on 86.7m shares, underperforming its own index by 86 bp six sessions before it reports. Forward catalyst: Nvidia, 26 August after the close.

4. Moderna's give-back is the cleanest measure of how thin this tape is. The stock closed −23.55% at $133.32 on 96.67m shares, having ranged $128.61–$155.00, one session after closing +176.97% at $174.38 on 182.8m shares. Nothing new was published about INTerpath-001. What changed is that TD Cowen kept a hold rating on Merck with a $137 price target — below Wednesday's $152.20 close — writing that "success is not unduly surprising given strong Phase II data" and that the "breadth of potential [is] still unknown." Merck fell 1.98% to $149.18 and the healthcare group fell 1.80%, giving back 53% of Wednesday's 3.37% gain.

5. Bitcoin hit a two-and-a-half-month high on a day equities fell 0.9% — and Bloomberg says that is neither a risk signal nor a liquidity signal. The coin cleared $71,000, its best since 1 June, on a two-day gain of nearly 11%, after the White House push for the Clarity Act. Coinbase closed +7.65% at $172.45, Strategy +7.81% at $112.39; ether reached its best level since 12 May and solana since 16 May. Bloomberg's markets desk published the counter-read the same evening: "Bitcoin's Short Squeeze Leaves Rally Hunting for Real Buyers." A squeeze that runs 11% in two sessions while the Russell falls 1.27% and VIX rises 7.5% is a positioning unwind in one asset, not a broad liquidity bid.

6. Oil is a geopolitical instrument again — and the crack spread says the market does not believe the supply story. WTI settled +2.14% at $86.20 and Brent +2.36% at $93.78 after President Trump vowed "economic warfare" on Iran and the United Arab Emirates suspended trade with Tehran. Bloomberg: "Oil Set for Weekly Surge as US Seeks to Throttle Iran's Economy." Against a 2.14% crude move, RBOB gasoline rose only 0.29% and heating oil 0.59% — both cracks compressed for a second consecutive session. A genuine physical-supply scare widens product cracks because refiners cannot replace the barrel. This one narrowed them, which says the bid is in the paper barrel.

7. Deere delivered the best industrial print of the month into the worst industrial tape of the month. Adjusted EPS $5.10 against $4.70 expected; net sales and revenues +5% to $12.608bn; net income $1.38bn against $1.29bn a year earlier; construction and forestry +18% to $3.6bn; small agriculture and turf +12% to $3.4bn; and production and precision agriculture, the largest division, −6% to just under $4bn. Full-year net income guided to $4.75bn–$5.0bn. CEO John C. May: 2026 "will mark the bottom of the current ag equipment cycle." The stock closed +6.94% at $620.94 — and the industrials group still fell 1.73%, its second-worst of the eleven, and is −4.02% on the week, the worst weekly group on the board.

8. The Philadelphia Fed printed the best manufacturing number in five years and the market ignored it, twice over. The headline index jumped 6 points to 47.4 against a 25 consensus, the highest since April 2021; the employment sub-index surged 18 points to 27.9, the best since April 2022; and prices paid and prices received both fell to their lowest since February, before the U.S.–Israel strikes on Iran. That is a growth-positive, inflation-negative surprise, and the September hike probability moved against it, rising to 36.2% from 33.1% on CME's own board. When a disinflationary beat raises hike odds, the market is trading the supply of duration, not the data.

9. Broadcom is trying to raise more than $60bn of debt for AI compute and its stock went up 0.49%. Bloomberg reported talks with lenders for more than $60bn of senior debt, potentially paired with roughly $30bn of junior financing through a special-purpose vehicle, taking the package toward $100bn, with Blackstone and Apollo in discussions, following the three-way partnership struck in June. Proceeds would support Anthropic and other large-scale compute build-outs. Set that against BofA's recent flag of $370bn of AI-related debt across the complex. The equity's indifference is the tell: the market has stopped pricing AI financing as dilution and started pricing it as revenue visibility. That is a credit question before it is an equity one (§9).

10. Europe again refused to follow, and this time the direction was the other way. On Bloomberg's board the U.S. 10-year rose 5 bp to 4.70% while Germany was unchanged at 3.26%, France +1 bp to 4.12%, Italy +1 bp to 4.07%, Spain +1 bp to 3.70%, the U.K. +2 bp to 5.07%. European cash closes are stamped 11:59 a.m. ET, so Bessent's morning escalation was inside their session. Two consecutive days of a 5–6 bp American move against a flat bund is the market pricing an American duration problem in both directions. Equity Europe finished lower — DAX −0.42% at 25,983.04, CAC 40 −0.57% at 8,453.09, Euro Stoxx 50 −0.37% at 6,420.75, with the FTSE 100 +0.04% at 10,748.16.

Source: Finviz group screener, Performance table view, read after the close. YTD reconciliation: compounding each group's 19 August YTD by Thursday's one-day move reproduces the published YTD to within 0.15 percentage points across all eleven groups. Worked examples: energy 1.3798 × 1.0068 = 1.3892 → +38.92%, published +38.92%, deviation zero; financials 1.0717 × 0.9916 = 1.0627 → +6.27%, deviation zero; healthcare 1.1301 × 0.9820 = 1.1098 → +10.98%, deviation zero. The two drifters are real estate (computed +11.19% against a published +11.04%, 0.15 pp) and consumer cyclical (computed −3.00% against −2.91%, 0.09 pp).

The rotation inverted Wednesday's rotation, group for group, and that is the finding. Wednesday's four best were basic materials, healthcare, consumer cyclical and real estate; Thursday's three worst were consumer defensive, healthcare and industrials. Only basic materials (+0.75%) and energy (+0.68%) rose, and both are commodity-price pass-throughs rather than demand signals. The genuinely interesting cell is technology at −0.15% on a day the S&P fell 0.87%, because the same Finviz bucket fell 0.69% on Wednesday when the index rose — two sessions, opposite index directions, technology outperformed both times. Keep the Finviz-versus-GICS caveat in view: Walmart sits in consumer defensive, so a −2.16% group print is substantially one stock, and Amazon (−2.13%) sits in consumer cyclical, which is why that group fell 1.40% on a day the retail news was about supermarkets.

Higher

• Deere (DE) +6.94% to $620.94 — adjusted EPS $5.10 vs $4.70 expected; net sales and revenues +5% to $12.608bn; construction and forestry +18%, small ag and turf +12%, production and precision ag −6%; full-year net income guided $4.75bn–$5.0bn.

• Marvell (MRVL) +5.82% to $251.09 on 25.32m shares — a second consecutive strong session after Bloomberg reported the Google equity right; the stock is now +16.2% in two days.

• SK Hynix's U.S. line +4.43% to $163.08 — the Seoul listing rose about 13% on the accelerated ₩40tn ($28.7bn) buyback and cancellation programme.

• Micron (MU) +3.97% to $974.29 on 23.41m shares — the most actively traded name on Investing.com's board and the largest single contributor to a green SOX.

• Strategy (MSTR) +7.81% to $112.39 and Coinbase (COIN) +7.65% to $172.45 — bitcoin above $71,000, its best since 1 June.

• SanDisk (SNDK) +2.02% to $1,600.62; Broadcom (AVGO) +0.49% to $364.25; AMD +0.67% to $469.54.

• Alibaba ADR (BABA) +1.26% to $130.53 — closed higher after falling nearly 3% pre-market on a 75% drop in June-quarter profit driven by AI spending. Not an S&P 500 member. Range $121.88–$130.62, a 7.1% intraday recovery off the low.

• Ross Stores (ROST) −2.43% to $228.99 in the session, then +8.74% to $249.00 after hours — second-quarter results and a raised annual profit outlook. The Street was modelling $1.82 on $5.98bn against $1.56 on $5.53bn a year earlier.

• Nordson (NDSN) — reported after Wednesday's close and traded up 8% in extended hours on full-year adjusted guidance of $11.80–$12.00 against an $11.60 FactSet consensus.

Lower

• Walmart (WMT) −9.15% to $103.84 on 82.75m shares — worst day since 17 May 2022. U.S. comps +2.6% vs 3.5% (FactSet); Q3 adjusted EPS guidance $0.62–$0.64 vs $0.68; FY raised to $2.80–$2.87, still under the $2.90 consensus. Revenue beat.

• Moderna (MRNA) −23.55% to $133.32 on 96.67m shares — no new disclosure; a positioning unwind after Wednesday's 176.97%.

• CrowdStrike (CRWD) −5.60% to $190.34 — Axios reported CTO Elia Zaitsev is leaving to start an AI-focused cyber venture fund called Cognition. Range $189.93–$200.74: the stock closed 5.2% below its intraday high and at the low.

• SpaceX (SPCX) −4.10% to $133.93 on 113.8m shares, the heaviest volume on the board — Bloomberg reported Anthropic expects to match or top SpaceX's record IPO size, and added Citigroup to its top IPO banks. Rocket Lab (RKLB) −3.82%; neither is an S&P 500 member.

• Eli Lilly −2.71% to $1,245.69, Amazon −2.13% to $260.19, Merck −1.98% to $149.18, Apple −1.75% to $311.30, Tesla −1.66% to $345.28, Pfizer −1.51%, Oracle −1.21% to $142.07, Alphabet A −1.19% to $340.62, Microsoft −0.46% to $481.20.

• Nvidia (NVDA) −0.33% to $216.85 — the notable non-participant in a semiconductor index that rose 0.53%.

• Robinhood (HOOD) −0.69% to $95.11 — the fade of the session: an intraday high of $101.61 and a close 6.4% beneath it, on a day crypto equities rose 7–8%. Meta −0.03% to $545.89 was flat, and the only megacap that was.

Analyst actions

• JPMorgan, Overweight on Broadcom. Harlan Sur: the market "continues to underestimate Broadcom's significant dominance/lead (18 months+), chip/package design leadership, aggressive cadence of new designs, IP portfolio, and track record of execution," noting Broadcom has helped Google bring 14 advanced chip designs to market over twelve years, and modelling a 180% annual jump in AI sales.

• JPMorgan reiterated Overweight on Analog Devices. Sur: "Broad-based strength drives another beat-and-raise; data center growth likely to outstrip current expectations." Fiscal-Q3 EPS $3.45 ex-items on $4.02bn against FactSet's $3.34 on $3.92bn. The stock rose about 1% Thursday and is +39% year to date.

• TD Cowen, Hold on Merck, price target $137 — roughly 10% below Wednesday's $152.20 close. Steve Scala: "Success is not unduly surprising given strong Phase II data … but breadth of potential still unknown."

• Truist, Hold on Target. Scot Ciccarelli: the changes of the past six to nine months are "resonating with their customers, as comps and transactions have improved," but at 17–18× the stock trades at a 15–20% discount to the market against a three-year average discount of about 33%. Separately, Citi turned negative on the dollar after warning on U.S. buyback risk.

Mon 8/17 — completed. No S&P 500 reporter on either bucket.

Tue 8/18 — completed. BMO: Home Depot (HD) [6:00] — closed −0.12% on the print. AMC: Keysight Technologies (KEYS) [4:05] — closed −6.29% at $319.55 the following session. Jack Henry & Associates (JKHY) [4:15].

Wed 8/19 — completed. BMO: Lowe's (LOW) [6:00] — closed +2.02% at $220.00. Estée Lauder (EL) [6:00] — closed +16.30% at $98.01. Target (TGT) [6:30] — closed +4.28% at $159.00; Truist maintained a hold on Thursday citing valuation. Analog Devices (ADI) [7:00] — closed −0.89% at $373.26, then rose about 1% Thursday. TJX Companies (TJX) [7:30] — closed −4.21% at $144.50. AMC: Nordson (NDSN) [4:30] — raised full-year adjusted guidance to $11.80–$12.00 and traded up 8% after hours.

Thu 8/20 — completed. BMO: Deere & Company (DE) [6:20] — beat and raised the low end of the full-year range; closed +6.94% at $620.94. Walmart (WMT) [7:00] — comps and guidance short; closed −9.15% at $103.84, its worst session since May 2022. AMC: Ross Stores (ROST) [4:00] — raised the annual profit outlook; closed −2.43% at $228.99 and traded +8.74% to $249.00 after hours.

Fri 8/21. No S&P 500 reporter on either bucket.

Mon 8/24. No S&P 500 reporter on either bucket.

Tue 8/25. AMC: Intuit (INTU).

Wed 8/26. BMO: J.M. Smucker (SJM), Williams-Sonoma (WSM). AMC: Nvidia (NVDA), Salesforce (CRM), CrowdStrike (CRWD), Synopsys (SNPS), Agilent Technologies (A), Veeva Systems (VEEV), HP Inc. (HPQ).

Thu 8/27. BMO: Dollar General (DG), Dollar Tree (DLTR), Best Buy (BBY), Hormel Foods (HRL). AMC: Marvell Technology (MRVL), Autodesk (ADSK), Workday (WDAY), Ulta Beauty (ULTA).

Fri 8/28. No S&P 500 reporter on either bucket.

Changes vs. the prior calendar (8/19 report). One change, and it is a timing bucket, not a roster addition. Williams-Sonoma (WSM) now publishes as before-open on 26 August, where the prior edition carried it after-close from the 14 August Earnings Whispers pull and flagged that Nasdaq had returned "not-supplied." The name and the date are unchanged; confirm with company IR before positioning around the print. No additions and no removals anywhere else in the ten-day window.

Non-members on the same dates, listed so nobody mistakes their absence for an omission: PDD, XPeng on 8/24; Bank of Montreal, Bank of Nova Scotia, Semtech, Box, nCino on 8/25; Trip.com, Li Auto, Dycom, Donaldson on 8/26; Royal Bank of Canada, Toronto-Dominion, CIBC, Rubrik, IREN, Harmony Gold, Bilibili on 8/27; Ubiquiti, Miniso on 8/28.

What the forward calendar hands the desk. The current week is finished and the reaction function it produced is eight prints deep and violently dispersed: Home Depot beat and closed flat; Target beat and raised and closed +4.28%; Lowe's guided down and closed +2.02%; Estée Lauder closed +16.30%; TJX closed −4.21% on no headline; Deere beat and closed +6.94%; Walmart missed on comps and closed −9.15%; Ross Stores closed −2.43% and then traded +8.74% after hours. That is a 25.5-point range between the best and worst single-session outcomes inside one week, and in two cases the after-hours mark contradicted the cash close. Next week concentrates the risk again: seven S&P 500 names report on Wednesday 26 August, all but two after the close, and Nvidia is one of them.

Name the shape: a belly-led bear steepener with the front end pinned for a third day. The belly is the diagnostic. The 5- and 7-year rose 4 and 5 bp, the largest moves on the curve; the 10-year rose 4; the 20- and 30-year rose 3 and 4; and the 2-year was exactly unchanged for a third consecutive session at 4.19%. Read that against Wednesday's mirror. Wednesday the 20-year fell 11 bp and the 5-year fell 2. Thursday the 20-year rose 3 and the 5-year rose 4. The 5-year is now 2 bp cheaper than it was before the buyback was announced; the 20-year is still 8 bp richer. A programme that names the 10-to-20 and 20-to-30-year buckets should, if it is believed at all, hold its premium precisely there and nowhere else — and that is what happened. The 20s30s spread widened another basis point to +3, extending Wednesday's flip from flat, so the 20-year has outperformed the 30-year on two consecutive sessions in opposite directions. That is a durable relative-value signature, not a one-day artefact.

The cross-market check again isolates the move to America — this time on the way up. On Bloomberg's evening board the U.S. 10-year rose 5 bp to 4.70% while Germany was unchanged at 3.26%, France, Italy, Spain, Portugal and Greece each rose 1 bp, the Netherlands was unchanged, and the U.K. rose 2 bp to 5.07%. Canada rose 6 bp, Brazil 6 and Mexico 6 — the Americas moved together and Europe did not. Two sessions ago the same test isolated a 6 bp American rally; today it isolates a 5 bp American selloff. The American term premium is now trading as an idiosyncratic asset, which is the condition a buyback programme is designed to address and the condition it has not yet changed.

Read the bill curve separately, because it moved and the coupon curve's story does not explain it. The 1-month bill rose 3 bp to 3.80% while the 2-month fell 2 bp to 3.79% — the front of the bill curve is inverted by a basis point, with 1-month above both 1.5-month (3.77%) and 2-month. The 3-month rose 1 bp to 3.87% and the 6-month did not move. A 3 bp jolt at the very front, with the 2-year pinned, is not a policy signal; it is a bill-supply and month-end financing signal, and it belongs with the funding data in §9, where SOFR has just printed 3 bp through the administered rate for the first time this month.

Provenance of every column, stated. The NOW column is a live read of the CME FedWatch card, which printed "Data as of 20 Aug 2026 06:55:05 CT" — an evening stamp, taken while the page rendered at approximately 7:55 p.m. ET. A live CME read after 5:00 p.m. ET is indicative, not a settlement snapshot. For a second consecutive session CME published a full numeric four-column table rather than only the comparison chart, so 1 DAY, 1 WEEK and 1 MONTH are CME's own published figures with CME's own reference dates — not chart-read, not carried from a prior edition and not reconstructed.

Correction to the prior edition, as the method requires. Wednesday's report published a live CME NOW of 67.3% hold / 32.7% hike for September, disclosed at the time as indicative. CME's own 1 DAY column now carries 66.9% / 33.1% for 19 August. The live read was 0.4 percentage points too dovish and is corrected here; no conclusion in that edition changes sign.

CME versus Investing.com, reconciled and quantified. Investing.com's Fed Rate Monitor, timestamped 20 Aug 2026 07:45 p.m. EDT, puts September at 64.7% hold / 35.3% hike off a quoted ZQU6 price of 96.325. CME puts it at 63.8% / 36.2%. The gap is 0.9 percentage points and it is a price gap, not a methodology gap. The September meeting falls on the 16th, so under CME's day-weighting only 14 of the month's 30 days carry the post-decision rate; sensitivity is therefore 0.25 × (14/30) = 11.67 bp of contract yield per unit of probability, making each 0.001 of ZQU6 price worth about 0.86 percentage points of hike probability. A 0.9-point gap is about 0.00105 of contract price — roughly a tenth of a basis point, or the bid-offer.

Momentum, and the week-over-week arc — the direction changed. The September hike probability has gone 55.0% (20 July) → 33.9% (13 August) → 33.1% (19 August) → 36.2% (now). That is +3.1 points on the day and +2.3 points on the week — the first increase in either window in this reporting run, after a month in which the series fell more than twenty points. The 50 bp bucket remains 0.0% and has been since at least 13 August, having been 9.2% a month ago. The composition of Thursday's move is what makes it interesting: it happened on a session where claims beat, the Philadelphia Fed's price sub-indices fell to a six-month low, and the belly cheapened 4–5 bp. A disinflationary data day that raises hike odds is not a data day. It is the front end being dragged by the same term-premium repricing that moved the 5- and 7-year, and it is the second consecutive session in which the fed funds strip and the mandate have pointed in opposite directions.

Source: Investing.com Fed Rate Monitor, all cards timestamped 20 Aug 2026 07:45 p.m. EDT. Cut probability at every 2026 meeting: 0.0% — stated explicitly, as it has been every edition.

The multi-day momentum read. All three 2026 meetings repriced toward tightening on the day — cumulative hike +5.0 points at September, +3.9 at October, +2.5 at December — and all three are only marginally changed on the week (+1.0, +0.8, +0.4). Read the day and the week together and the arc is a round-trip, not a trend reversal: the strip de-priced hike risk into and through Wednesday's buyback rally and re-priced it back on Thursday's selloff, ending the week within a point of where it started at every meeting. The one asymmetry worth holding is at the back of the year — December's cumulative hike at 66.7% against September's 35.3% means the market prices roughly a two-in-three chance that one hike is delivered by year-end while giving the September meeting itself only a one-in-three chance. Note the provenance conflict, which recurs. Investing's "previous day" column for September reads 69.7% hold, against the 67.7% that same vendor's current column printed on Wednesday evening and against CME's 66.9%. Investing's previous-day snapshot is a fixed daily stamp taken at a different hour; the CME 1 DAY column is the one to use for day-on-day change, and the Investing prev-day cells are shown for completeness rather than relied on.

Mark the prior edition's call honestly first. Wednesday's report set the test in writing: "a 20 August IG print at 84 bp or wider would clear Tuesday's 83 bp threshold with room and confirm five sessions of drift as a trend." The print — the 19 August observation published Thursday — came in at 81 bp, one basis point tighter. The five-session widening streak of 78 → 79 → 80 → 81 → 82 ended at the sixth attempt. The trend is broken, not confirmed, and the report should say so plainly having spent three editions building the case for it. HY tightened alongside, −2 bp to 273. What did not reverse is the tail: CCC widened another 3 bp to 1,030 bp, its widest of the month, taking the CCC-minus-HY differential to 757 bp, +153 bp on the year while HY itself remains 8 bp tighter than it started 2026. So the correct reading of Wednesday's session in credit is: the buyback rally tightened the core and did nothing for the tail.

CDX retrieval note — the six-step ladder, worked and named. (1) Bloomberg in Chrome: /markets and /markets/rates-bonds both rendered; the Fixed Income Indices panel publishes Global Aggregate (501.69), U.S. Aggregate (2,344.80), Asian-Pacific (193.81), Pan-Euro (226.24) and EM USD Aggregate (1,405.16), and a programmatic search returned no CDX string anywhere. (2) WSJ Market Data: the front page and /market-data/bonds both rendered; no instrument table carrying a CDX level appeared. (3) Cbonds / ICE / S&P Global: the Cbonds CDX.NA.IG 5Y page returned a Cloudflare bot-verification interstitial, which this session did not attempt to bypass; ICE and S&P Dow Jones publish methodology and index-news documents, not levels. (4) FT Markets Data and Reuters credit wraps: no instrument page and no wire wrap quoting a level in the reviewed material. (5) TradingView / Barchart / CME CDS index products: documentation only. A general web search surfaced an undated third-party digest asserting "81 basis points" for IG CDX; under this report's standing rule an undated, unattributed digest number is not a CDX level and it is withheld. (6) Cash-market proxy: LQD −0.48% and HYG −0.19% against a 4 bp rise in the 10-year, published as a proxy for the direction of cash credit and not as a CDX level. Conventions restated: CDX IG 5y is quoted in basis points of spread; CDX HY 5y in price points, where a rising price means tightening credit spreads.

The MOVE series updated same-day for the first time in this reporting window. It also fills in Wednesday's gap. MOVE printed 73.18 on 20 August, +2.69%, after 71.26 on 19 August, −4.96% — a level the prior edition could not publish because the series had not updated. The two-day picture is now complete: rates volatility collapsed 5% on the buyback announcement and recovered 2.7% when the buyback failed to hold, ending 2.4% below where it began the pair of sessions. Against a VIX of 16.01 on the same date, MOVE/VIX is 4.57× — a clean matched-date ratio requiring no vintage caveat, and down from 4.79× on a matched 19 August basis.

That compression is the wrong kind. It happened because equity volatility rose 7.52%, not because rates volatility fell. VIX has gone 14.89 → 16.01 while MOVE has gone 74.98 → 71.26 → 73.18; the gap narrowed because the calmer market got nervous, not because the nervous one calmed down. A ratio near 4.6× against a five-year average closer to 4× still says the rates market carries the larger risk premium, and Thursday is the reason it should: an official intervention in the long end was announced, escalated on live television, and reversed by the market inside twenty-four hours. No verified 2y, 10y or 30y swap-spread level was obtainable this session — the vendors publishing them sit behind entitlement walls and no wire quoted a level in the reviewed material. That is a gap and is recorded as one for a second consecutive edition.

The take — the havens would not bid, and that is the whole session in one line. The S&P fell 0.87%, the Dow lost 703 points, VIX rose 7.52%, and the two currencies that are supposed to be bought when that happens both weakened: USD/JPY rose 0.59% through 159 and USD/CHF rose 0.39% back above 0.80. Gold, meanwhile, went up 0.82%. That combination — equities down, volatility up, yen and franc down, gold up, dollar flat — does not describe risk aversion. It describes a real-yield and term-premium event: American long rates rose 4–5 bp with no European follow-through (§6), which mechanically supports the dollar against the funding currencies and simultaneously supports the debasement asset against the currency complex as a whole. A haven that will not bid on a 700-point Dow day is telling you what the market thinks the risk actually is — and it is not equity risk.

The Korean tell held out of sample, in the opposite direction. The prior edition observed the won strengthening 1.82% as the Kospi fell 5.80% and proposed a mechanism: Korean equity flows are domestically levered, so a drawdown forces deleveraging that repatriates capital and lifts the won. The falsifiable prediction is that a Kospi rally should weaken the won. Thursday: Kospi +5.89%, won −0.21%. The magnitudes are asymmetric — a 5.8% decline moved the currency 1.82%, a 5.9% rally moved it 0.21%, roughly a ninth as much — which is what you expect if deleveraging is forced and re-leveraging is voluntary. The signal is confirmed and the asymmetry is now the tradeable part: buy the won into Korean equity drawdowns, do not sell it into Korean equity rallies.

And read what did not move at all. USD/TWD moved 0.03% and USD/MXN 0.01% on a session with a 700-point Dow decline, a 2.1% crude rally and a 13% move in the largest memory stock in Asia. The Australian dollar fell 0.04% while silver rose 3.73% — for the second consecutive session the metals bid was monetary rather than industrial, and copper's −0.09% confirms it. If you want the precious complex expressed in FX this week, neither the Australian dollar nor the Canadian dollar has been the vehicle; the dollar itself, doing nothing while gold rises, is the expression.

Basis caveats, stated before the analysis. Settles are Investing.com per-contract historical boards for the front contract; weekly, monthly and YTD columns are TradingEconomics on its Aug/21 dated rows, header order verified as Price | Chg | %Chg | Weekly | Monthly | YTD | YoY | Date before any YTD was quoted. Contract month: Investing.com's front CL board is the October contract at $86.20 while CNBC quoted the September contract at $87.56 — a $1.36 calendar spread, unchanged in size from the prior edition; TradingEconomics' crude level of $86.258 sits 0.07% from the October settle and is within tolerance and not withheld this session. Gold: TradingEconomics prints spot at $4,520.38, 1.28% below Bloomberg's 4,578.90, exceeding this report's 1% tolerance, so the TradingEconomics gold level is withheld and only its ratio columns are used; the level quoted is the Comex settle, corroborated by Bloomberg to within 0.08%. Volume stamps: the 20 August rows for gold (5.46K), silver (1.14K), copper (0.30K), platinum (0.48K), WTI (0.44K), natural gas (0.28K) and heating oil (0.02K) are thin electronic-session rows, so those settles are treated as indicative closes; only Brent, at 298,170 lots, carries a full pit-session stamp. Investing.com's metals boards have also re-based since the prior edition, so every day-on-day change above is computed within this session's board.

The take — the geopolitical bid is in the paper barrel, and the crack spread is the polygraph. President Trump vowed "economic warfare" on Iran, said no negotiations were planned, and the United Arab Emirates suspended trade with Iran after reporting two missiles fired in its direction. Brent added 2.36%. Now apply the test. A genuine physical-supply threat widens product cracks, because a refiner facing a scarcer barrel cannot pass the scarcity through instantly. Thursday did the opposite for a second consecutive session: crude +2.14%, gasoline +0.29%, heating oil +0.59%. Both cracks compressed while the barrel rallied on a war headline. The market is buying crude as a macro and sanctions-risk asset and is not buying the physical scarcity a real Hormuz interruption would create. Positioning supports that: crude is +50.22% year to date but −0.66% on the month — a six-session bounce inside a flat month, not a trend extension.

The precious complex kept going and the industrial one still refuses. Gold added 0.82% and silver 3.73% on a day the dollar was flat, the 10-year cheapened 4 bp and equities fell — a combination that is textbook bearish for a zero-coupon asset. It rose anyway, for the second session in three. Copper fell 0.09%. Over the two sessions since the buyback announcement, silver has gained 8.7% and copper has lost 0.15%; gold's year-to-date has gone from +0.49% to +4.69% and silver's from −11.04% to −4.35%, recovering roughly two-thirds of 2026's loss in three sessions. This is a monetary bid with a squeeze inside it, and the tell is that it no longer requires the dollar to fall. Wednesday gold rose on a 0.89% dollar decline; Thursday it rose on a flat dollar and higher real yields. That is a stronger signal than Wednesday's, not a weaker one.

Desk-style ideas for institutional investors. Each carries an explicit catalyst and an explicit invalidation. These are not personalized investment advice; verify independently and size to your own mandate before acting.

1. The rates trade — mark the old one, take the rest of the profit, and re-strike along the calendar

Mark on the prior edition's rates idea. The book is long ZQZ6 against short ZQZ7, DV01-matched one-for-one, entered at 96.165 / 96.060 for a spread of 10.5 bp, with half the original position remaining after two quarter-reductions. Thursday's mark: ZQZ6 96.160, ZQZ7 96.000, a spread of 16.0 bp — exactly unchanged on the day and still +5.5 bp from entry, worth $229.19 per contract pair at $41.67 of DV01 per basis point. No invalidation triggered. But the trade has gone two sessions without adding a basis point while the strip round-tripped underneath it, which is what a fully valued spread looks like. Action: close the remaining half at 16.0 bp and book the 5.5 bp.

The modal path, the base case and the tails. Modal path: hold on 16 September (CME 63.8%, Investing 64.7%, ease 0.0%); hold on 28 October at 51.3% modal against a cumulative hike of 48.7% — effectively a coin flip; one 25 bp hike delivered by 9 December, 3.75–4.00% modal at 44.9% with cumulative hike 66.7%; terminal 4.020% drawn by the September and October 2027 contracts at 95.980; year-end 2027 modal 3.75–4.00% at 34.3% with cumulative easing of just 1.7%. Base case: the hike keeps sliding right while the terminal stays pinned — the trough moved from July/September 2027 to September/October 2027 in one session without the rate changing — because the front end is driven by duration supply rather than by the mandate, and because a 206,000 claims print removes the labour-market route to easing. Tail one, dovish: a core PCE deflator at or below +0.2% m/m on 26 August, landing on top of a Philadelphia Fed price series already at a six-month low, would put a 2027 cut back into a strip that has just taken one out. Tail two, hawkish: a Jackson Hole framing that treats 3.50–3.75% as a floor rather than a destination puts September through 50% for the first time since July. Practical implication: the December-2026-versus-December-2027 spread has priced most of the "pushed right, not cancelled" story; the meeting-to-meeting calendar inside 2026 has not.

The new expression. Expression: long ZQU6 (September 2026) against short ZQZ6 (December 2026), DV01-matched one-for-one at $41.67 per basis point per contract, entered at 96.325 / 96.160, a spread of 16.5 bp. Thesis: the strip prices 35.3% cumulative hike at September and 66.7% at December — a 31.4-point gap across two meetings — and the calendar between now and 16 September contains exactly two events that can move it. Every session in which the hike slides right lifts the front contract against the back. Catalyst: PCE 8/26 at 8:30; Jackson Hole 27–29 August; claims 8/27; the first buyback operation on 9 September; the 16 September FOMC. Invalidation: the spread back through 13.0 bp; or the September cumulative hike printing above 50% on either vendor; or any 2026 meeting showing a non-zero cut probability, which would compress the spread from the dovish side. Sizing: one-for-one DV01, half the notional of the position just closed. The plausible range is roughly 13–22 bp, so this risks 3.5 bp to make 5.5.

2. Long the 20-year against the 30-year, on the buyback's own map — hold, and it is working exactly as specified

Expression: long the 20-year Treasury point against short the 30-year, DV01-matched, small. Mark: entered Wednesday with 20s30s at +2 bp; Thursday it widened to +3 bp, for +1 bp on the position. On a session when the whole curve cheapened, the 20-year rose only 3 bp against the 30-year's 4 bp and the 7-year's 5 bp, and over the two sessions since the announcement the 20-year has kept 8 of its 11 bp while the 7-year has kept none. Catalyst: the operation schedule and the first execution on 9 September; the refunding statement that sizes the buckets; Bessent's escalation that operations "could be more than the 4 billion per issue." Invalidation, unchanged: 20s30s back through −2 bp; or Treasury spreading operations evenly across the 10–30y complex; or a 30-year auction tailing more than 2 bp. Sizing: small, unchanged.

3. Long healthcare against the semiconductor complex — closed at the stated invalidation, on the stated test

Mark: healthcare −1.80%, SOX +0.53% — the pair lost 2.33 points in a session after gaining 5.5 the day before. The invalidation was written as "healthcare giving back more than half of Wednesday's gain." Wednesday's gain was 3.37%; Thursday gave back 1.80%, or 53.4%. The test triggered on the first session it could and the position is closed. What the trade taught: the long leg was never a sector view, it was one Phase 3 readout in one indication, and a single-trial gain reverses on nothing more than a TD Cowen hold with a $137 target on Merck. The short leg inverted the moment the funding-structure story turned positive — SK Hynix's ₩40tn buyback did more for the semiconductor index in one night than four sessions of AI-financing anxiety had done against it. A pair built on one catalyst per leg is two event trades wearing a factor costume.

4. New — long the memory and storage complex against short megacap platform technology

Expression: long an equal-weighted memory and storage basket (Micron, SanDisk, and the Korean memory complex where mandate allows) against an equal-weighted megacap platform basket (Apple, Amazon, Alphabet, Microsoft), dollar-neutral, small. Thesis: Thursday split the technology complex cleanly along the balance sheet. Memory rose — MU +3.97%, SNDK +2.02%, SK Hynix's U.S. line +4.43% — because its capital cycle is now returning cash: SK Hynix is accelerating a ₩40tn ($28.7bn) buyback-and-cancellation and committing more than half of 2025–27 free cash flow to shareholders. The platforms fell — AAPL −1.75%, AMZN −2.13%, GOOGL −1.19%, MSFT −0.46% — because they fund the build-out, and the market has just been shown what that costs in the Broadcom $60–100bn structure (§9). SOX outperformed NDX by 125 bp on the day. Catalyst: Nvidia, 26 August after the close; Marvell, 27 August; the pricing terms of the Broadcom package; Korean buyback execution at third-quarter earnings. Invalidation: the megacap basket outperforming the memory basket by more than 4% over five sessions; or a Nvidia print that re-rates the platforms on demand rather than capex; or any memory name cutting or deferring its buyback. Sizing: small — the long leg has run 9–13% in Seoul in a single night and is not a fresh entry point.

5. Protection on the CCC cohort funded in IG — hold at a half; both legs worked for the first time

Mark: the protection leg widened again — CCC +3 bp to 1,030 bp on 19 August, the widest of the month — and, for the first edition in four, the funding leg helped rather than hurt: IG tightened 1 bp to 81 bp, breaking a five-session widening run. The CCC-minus-HY differential is 757 bp, +5 bp on the day and +153 bp on the year. Catalyst: the 21 August FRED update; the record $145.2bn August IG calendar clearing into a cheaper curve; the NY Fed Corporate Bond Market Distress Index on 26 August. Invalidation, unchanged: the differential back through 735 bp, or IG widening beyond 85 bp. Sizing: a half, unchanged.

6. On-balance-sheet AI funding against off-balance-sheet AI funding — hold; the long leg is carrying it

Mark: the long leg is +16.2% in two sessions — Marvell +9.85% Wednesday and +5.82% Thursday to $251.09 — on the Google $12.2bn equity right. The short leg's test case arrived Thursday: Broadcom is seeking more than $60bn of senior debt plus roughly $30bn of junior financing through an SPV, with Blackstone and Apollo in talks, and the stock rose 0.49%. So the short leg is flat and the pair is up on the long leg alone — an honest description rather than a vindication. What decides the trade: if a $60–100bn self-funded structure can be announced and the equity does not fall, the market has stopped punishing self-funding and the short leg has no edge. Catalyst: Nvidia 26 August, Marvell 27 August, the pricing terms of the Broadcom package and of the Nebius convertible. Invalidation: a customer-funded structure trading below its announcement price within ten sessions; or a self-funded AI financing above $50bn pricing inside indicative terms with the equity up. Sizing: small, unchanged.

7. Short the bank complex against the S&P, into the flattener — cut to a token; the driver reversed

Mark: the thesis was that a Treasury buyback programme is structurally a flattening instrument and every basis point of flattening is a debit to net interest margin. The curve steepened instead: 2s10s +4 bp to 50, 2s30s +4 bp to 104, the 2-year still pinned at 4.19%. The pair lost, though barely — financials −0.84% against an S&P −0.87%, so the short leg cost 3 bp of relative performance. The written invalidation, "2s10s back through 55 bp," did not trigger at 50 bp. The honest reading: the invalidation is intact and the mechanism is not. One session of buyback enthusiasm produced the flattener; one session of scepticism reversed it; the programme has not started and will not until 9 September. Action: cut to a token, keep the 55 bp stop, and re-size only if the curve flattens through 46 bp on an actual operation rather than an announcement.

Two prior closes, marked forward. The consumer-discretionary put spread was closed Wednesday at its stated invalidation, roughly flat — and Thursday would have paid, with consumer defensive −2.16% and consumer cyclical −1.40% on Walmart's 9.15% decline. Closing at a written stop one session before the thesis worked is the cost of having a written stop, and it is cheaper than not having one. The distillate crack against the barrel was closed Wednesday at a loss, and Thursday vindicated the exit: crude rose 2.14% while gasoline rose 0.29% and heating oil 0.59%, compressing both cracks for a second consecutive session. Neither position is being re-entered.

The vol note. VIX closed at 16.01, +7.52%, and MOVE at 73.18, +2.69%, for a matched-date MOVE/VIX of 4.57×. Equity volatility finally moved — but look at what it took: a 700-point Dow decline, the largest single-name retail drawdown since 2022, a failed sovereign intervention and a 2.25% distance from the record high, all to get VIX one point above 15. The eight-day binary calendar the prior edition flagged is now five days long and has lost its two easiest legs: claims came in at 206,000 and Walmart is out of the way. What remains is concentrated and simultaneous — PCE and Nvidia both on Wednesday 26 August, Jackson Hole 27–29 August — with no "Very high" macro item at all before Monday. That argues for owning gamma dated into 26–29 August rather than direction, financed out of the quiet Friday and Monday rather than out of the wings.