The S&P 500 finished July down 0.2%. On that number alone, nothing happened. Underneath it, the most crowded trade on earth was liquidated at four times leverage, Seoul hit circuit breakers on consecutive days for the first time in its history, and the US Treasury bought yen for the first time since 1998.
What changed in July, and what it means for how we are positioned.
Most monthly reviews are an exercise in describing weather. This one is not. July 2026 was the month the market stopped treating artificial intelligence as a demand story and started treating it as a financing story — and the transition was violent enough to break one of the best-performing funds in the world, halt a national exchange twice, and force two governments into a currency operation they had not attempted together in twenty-eight years.
The headline return of −0.2% is real, and it is also close to meaningless. It is the arithmetic result of two enormous forces passing each other in opposite directions: a collapse in high-beta, AI-adjacent equities, offset almost exactly by a rally in everything that had been ignored for five years. Fifty-nine per cent of S&P 500 constituents rose in a month the index fell. The equal-weight version of the same index gained 1.0% and beat the Nasdaq-100 by 7.5 percentage points — the widest monthly margin since the equal-weight Nasdaq series was created in 2005.
That distinction matters for how you read your own book. If you held the index, you had a dull month. If you held the consensus, you had a very bad one. If you held what nobody wanted — energy, small caps, community banks, Indian IT — you had an excellent one. The dispersion, not the direction, was the event.
What follows separates the three things that actually happened, because they are routinely conflated: a positioning unwind, which is now complete; a valuation debate about AI infrastructure, which has barely started; and a credit migration, which is scarcely being discussed and which we think matters most over the next eighteen months.
A fifty-point spread between the best and worst major exposures, in thirty-one days.
A negative July for the S&P, ending eleven consecutive positive Julys.
Equal weight over the Nasdaq-100, the largest gap since the series began in 2005.
Share of S&P constituents that finished the month higher.
The best-performing major asset class of 2026 so far.
Two supporting details give this rotation more weight than a single month usually deserves. First, it is not new: US small caps gained more than 22% in the first half — their best opening six months since 1991 — after five consecutive years of trailing large caps, matching the record losing streak of 1994–98. Second, the leadership is economically coherent rather than merely defensive. The Nasdaq Community Bank Index returned 23.9% year to date through July, roughly double the S&P's 10.1%. Regional lenders do not outperform into a credit contraction.
The uncomfortable counterweight is concentration. Ten stocks accounted for 78% of the S&P 500's year-to-date return of 10.21%. The index trades near 22.1 times estimated 2026 earnings against a five-year average of 19.9 and a ten-year average of 19.0 — but on forecast index earnings growth of roughly 24%, a rate not seen outside a post-recession rebound. Goldman's framing is the sharpest we have read: the risk here is an earnings bubble, not a valuation bubble. If 24% is delivered, 22 times is defensible. If it is not, the multiple is the least of the problem.
How a fund up 439% became a forced seller of everything in six days.
On 24 July, the Financial Times reported that Situational Awareness LP — run by Leopold Aschenbrenner, a former OpenAI researcher then aged 24 — was up 439% net through 30 June. Its first 13F, covering end-2024, had disclosed just $254.8m across six positions: Marvell at roughly 34%, Vistra 23%, Vertiv 20%, Talen Energy 11%. Backers reportedly included Patrick and John Collison, Nat Friedman, Daniel Gross and, unusually for a directional equity fund, Jane Street.
Six days after that FT report it had no public equity exposure at all.
The Q1 2026 13F, submitted on 15 May and published on the 18th, disclosed $13.68bn of reportable positions — more than double the $5.52bn shown three months earlier. Composition mattered more than the total, and this is where most summaries go wrong.
Roughly $3.85bn was common stock, concentrated in second-derivative infrastructure rather than the mega-caps: Bloom Energy at 22.8% of the equity book, SanDisk 18.8%, CoreWeave 14.4%, IREN 10.4% and Core Scientific 10.1% — the top five alone exceeding 76% of it. The fund owned 8.2% of Core Scientific's shares outstanding and roughly 4–5% of several other names in the same theme. The remaining $8.46bn was notional put exposure against essentially the entire semiconductor complex: $2bn against the SMH ETF, $1.6bn against Nvidia, with further positions in Broadcom, Oracle, AMD, Micron, ASML, Intel, Corning and TSMC. Across the whole filing, the top ten positions were 72.66% of the book.
So this was never a long-only infrastructure fund. It was a barbell: long the physical bottlenecks of the AI build-out, short the chipmakers supplying them — a bet that the value would migrate from silicon to power, storage and compute capacity.
A 13F reports US-listed long equity and listed options only. It does not show short sales, bonds, cash, private holdings or swap exposure — and notional option values can make a book look far larger than the capital behind it. Quiver Quantitative estimates the firm's assets at roughly $9.3bn from a recent ADV filing. Press coverage around the unwind described a book of $20bn or more, and some outlets have used the $13.68bn 13F headline as though it were assets under management.
The gap between those figures is not sloppiness. It is the story: with roughly four times gross exposure held in total return swaps, where the prime broker holds the physical shares and the client takes the economics, nobody outside the brokers knew the true size — and no single broker saw the whole thing.
Beyond the filing sat a large Nebius stake built after March, SK Hynix in Korea, and software shorts including Adobe.
By the week of 27 July, prime brokers at Bank of America, Goldman Sachs and JPMorgan were marketing holdings from both sides of the book. The fund sought emergency capital on an ad hoc basis and reportedly offered investors direct access to assets, including its private stake in Anthropic. It was not enough. On the morning of 30 July, before the open, the entire public book — longs and shorts together — crossed to Citadel in a single block.
Then the genuinely instructive thing happened.
This cross-section is the evidence. Had 30 July been a fundamental re-rating of AI infrastructure, Vertiv should have flown and Cipher should have moved in line with its own news. Neither happened. Bloom Energy reported record second-quarter earnings that morning and rose 25.6%; IREN and Nebius, with no news whatsoever, rose 26.5% and 27.1%. When names with idiosyncratic catalysts and names with none move identically, the common factor is doing all the work — and the common factor was the removal of a forced seller.
The decision to transfer the book in one block rather than liquidate it piecemeal deserves attention, because it was the most consequential market-structure choice of the month. A piecemeal unwind of 8.2% of a company's float, repeated across a dozen overlapping names, would have meant days of price-insensitive supply, widening skew, exploding single-name volatility and sympathetic de-risking from every fund holding the same exposure. A single-counterparty transfer means the positions never reach the tape as supply. They simply change owner. Citadel — one of the deepest-capitalised multi-strategy platforms in existence — can now warehouse, hedge or work out of the book on its own schedule rather than the market's.
The mechanism echoes LTCM in 1998 and Archegos in 2021: concentrated exposure, synthetic leverage invisible to public filings, prime brokers discovering the aggregate position only under stress. The difference is the resolution. LTCM required a Fed-convened consortium; Archegos left several banks with multi-billion-dollar losses. Here the book was absorbed privately, in one transaction, at a discount, by a willing buyer with the balance sheet to hold it. That is the system working roughly as designed — and it is why the contagion stopped at the cohort rather than reaching the brokers.
It held six positions. Fifteen months later the reportable book was $13.68bn.
Top ten positions as a share of the reportable book.
Held in swaps, so it never appeared in any public filing.
The forced seller is gone. That question is settled. Whether AI infrastructure deserves its multiple is not. Those are different statements, and the market answered only one of them.
Our read on 30 JulyWhy crowding has become genuinely unobservable — and what we can still watch.
This is not a scandal — dark venues exist for a legitimate reason, which is precisely to let institutions move size without broadcasting intent. But the combination of three features has quietly removed the ability to see crowding at all: execution has migrated off-exchange, leverage has migrated into swaps that bypass 13F disclosure, and the regulatory data that does exist arrives late. FINRA improved its ATS transparency lag from four weeks to two in January 2026. Two weeks is still an eternity.
The practical consequence played out on 30 July. The largest risk transfer of the year was negotiated privately, priced against a distressed tape, and settled before the opening auction. Market participants learned about it from a television reporter. No public order book, no pre-trade signal, no filing.
What remains observable is narrow but real: securities borrow rates, single-name implied volatility and skew, and options open interest. These are the only near-real-time telemetry on crowding available, and they moved before the headlines did. We regard closing this gap as an operating priority rather than a research nicety.
Record profits, a record crash and a record rebound — and a rally that foreigners spent six months selling.
Three shocks arrived within days of each other: the roughly $8.6bn IPO of Chinese memory maker CXMT, which held about 7.7% of global DRAM revenue and roughly 11% of wafer capacity in the first quarter; reports of progress on domestic Chinese DUV lithography; and the Nvidia–OpenAI financing headlines. Each is a legitimate long-term question about Korean pricing power. None explains a 38.6% drawdown in six weeks.
The amplifier was structural, and newer than most people realise. The Korea Exchange only introduced single-stock leveraged ETFs on Samsung and SK Hynix in May 2026 — designed explicitly for short-term intraday trading, and available to retail. Leveraged products now account for roughly a fifth of Korea's 1,142 listed ETFs. The largest SK Hynix single-stock leveraged fund fell between 45% and 47% in one session. Daily-reset leverage does exactly what daily-reset leverage does in a gap: it converts a bad day into a permanent impairment.
This is the detail we would put in front of any allocator with Korean exposure. The KOSPI roughly doubled in the first half of 2026 — the largest gain of any major market in the world — and foreign investors were net sellers of a record ₩148.3 trillion, about $96.7bn, across those same six months. The rally was financed almost entirely by domestic money.
Why were foreigners selling a market that was doubling? The consensus among Goldman Sachs, Nomura and the Korea Economic Institute is that it had almost nothing to do with deteriorating fundamentals, and four things to do with mechanics:
Benchmark weights forced it. As Korean stocks surged, their weighting in global and emerging-market indices rose sharply, pushing many active managers beyond their portfolio and risk limits. Nomura's Asia-Pacific equity strategist described the flows plainly as forced selling by clients. The same dynamic played out in India in recent years — a market can be too successful for its own investor base.
Profit-taking concentrated in the two names. Goldman's flow notes repeatedly attributed the outflows to Korean tech and autos, with Samsung and SK Hynix the specific sources.
Global rebalancing. In one week in May alone, foreigners pulled $13.2bn out of Korea, part of roughly $17bn leaving emerging Asia excluding China — the second-largest weekly outflow on record. Korean retail bought $14.1bn that same week.
The won. Currency depreciation against the dollar quietly erased a meaningful part of the local-currency gain for any unhedged foreign holder.
The streaks were extraordinary in their own right. Before flipping to net buying on 12 June, foreigners had sold for 24 consecutive sessions totalling ₩75.6trn — the fourth-longest foreign selling streak in the history of the Korean market, or ₩84.6trn including the Nextrade alternative venue.
Two lessons follow. First, foreign flows are a thermometer rather than a compass: anyone who shorted Korea in 2026 because foreigners were selling missed a doubling, exactly as they did during the 2020–21 retail-led rally. Second — and this is the actionable one — a market can rise for six months on a domestic bid while its most price-sensitive holders leave. That configuration is not fragile because the fundamentals are wrong. It is fragile because the remaining holders are levered, concentrated and behaviourally correlated. July is what that fragility looks like when it resolves.
One further detail argues against panic. The won did not break. USD/KRW traded near ₩1,467 and moved roughly 0.1% on a day the index fell more than 9%. An equity crash without a currency crash is a positioning event, not a solvency event — the difference between a buying opportunity and a country to avoid.
The policy response was swift: an emergency market-monitoring meeting under Finance Minister Koo Yun-cheol, a public apology from regulators, and a cap on single-stock leveraged ETF exposure at 20% of an individual investor's portfolio. The behavioural damage may prove more durable than the price damage. Korean brokers report clients asking not to be shown equities at all — and the belief that dips are for buying, learned through 2025, has been unlearned in a fortnight.
About $96.7bn — a record half-year, into a doubling market.
From the 22 June high, in little over a month.
A product the exchange had only launched in May 2026.
Barely moved. The crash was positioning, not solvency.
The first triple dissent in one direction since 2016 — and the trap it exposes.
Chair Kevin Warsh held the target range at 3.50–3.75% for a fifth consecutive meeting and pointedly declined to describe the decision as a pause, calling it instead a rigorous review of the economic situation. He has repeatedly used the phrase "family fight" to normalise open disagreement on the committee, and has argued elsewhere that inflation is a choice rather than a condition. Governor Waller has voiced inflation concerns publicly but voted with the majority. The June dot plot had already pencilled in one increase before year-end.
Markets moved immediately. CME futures repriced the probability of a September hike from a minority outcome to roughly 57%, against about 42% for another hold. Some strategists now carry two increases as their 2026 base case. Set against the two cuts priced in January, that is close to a hundred basis points of round-trip inside seven months — and it is the single most important explanation for almost everything else in this note.
Highest since October 2023 — and the 63rd consecutive month above the 2% target.
With April and May revised down by 74,000 combined.
Annualised. Growth is not the argument for tightening.
The 30-year sits near its 2007 high. Term premium, not policy.
The bind is genuine and we do not think it has an elegant resolution. Hike, and you tighten into an economy growing at 1.6% annualised that added 57,000 jobs in June. Hold, and you accept a sixth year above target with a war-driven energy impulse feeding through and a fresh tariff round arriving with a lag. June's inflation data was actually encouraging — headline CPI fell 0.4% on the month on gasoline, core was unchanged — but Warsh has emphasised the direction of travel over any single print.
For allocators the second-order effect matters more than the decision. The disappearance of the rate-cut path is what killed gold, silver and bitcoin this year, and what allowed value, equal weight and cyclicals to beat the Nasdaq. The rate path is the factor rotation. Any model treating July's momentum crash as an AI-specific event is missing half the mechanism. The most consequential date in our calendar is not the September FOMC but Jackson Hole on 27–29 August, where Warsh speaks.
The most under-priced macro event of the month, and the one with the longest half-life.
The pressure is structural rather than speculative. Prime Minister Takaichi's fiscal expansion and persistent wage gains have lifted Japanese inflation expectations. The Bank of Japan raised its policy rate to 1.00% in June — the highest since September 1995 — but is widely judged to be lagging the cycle. With the front end of the curve constrained by deliberate gradualism, the entire adjustment burden fell on the long end and the currency. The ten-year JGB reached 2.901%, a level last seen in 1996, up more than 70 basis points on the year; the twenty-year touched 3.901%.
On 30 July, with the yen at 163.73, Bank of Japan data implied Tokyo may have sold as much as $59bn buying yen in New York hours, acting alone. On 31 July the operation became joint. Both governments confirmed it publicly on 3 August, with Finance Minister Satsuki Katayama and Treasury Secretary Bessent each stating they would not hesitate to act again. The yen strengthened as much as 1.4% to 155.20, a near three-month high.
The New York Fed reportedly sold euros, not dollars. Coordinated intervention is conventionally funded with dollar assets. Selling euros to buy yen targets the yen–euro cross while leaving the dollar leg untouched — a deliberate choice indicating the objective was disorderly conditions and Japan's relative position, not dollar weakness.
Japan was given access to the Fed's FIMA repo facility. This allows Japanese authorities to borrow dollars against Treasury collateral rather than selling Treasuries outright. The architecture was designed so Tokyo could defend the yen without liquidating US government debt into a market where the thirty-year already sits near its 2007 high.
The US Treasury Secretary publicly endorsed correcting what he called substantial yen undervaluation. That is a serving Treasury Secretary naming a bilateral exchange rate objective. It is a meaningful precedent, and it will be tested.
Our reading is that intervention buys time rather than direction — UBS's framing, that the yen will be supported by intervention risk rather than by monetary fundamentals while real rates remain negative, is almost certainly right. But the architecture is a durable change. It establishes that Washington will co-finance a Japanese currency defence in order to protect its own bond market. Anyone short JGBs or short yen now faces a reaction function with an American component in it.
There is a constructive side. Hedged back into dollars, ten- and thirty-year JGBs have been offering in the region of 5.6% and 6.8% respectively — comfortably above comparable Treasuries. Several strategists have moved the asset class from "uninvestable" to "investable". For the first time in three decades the Japanese long end is a genuine allocation question rather than a curiosity, and it now comes with a demonstrated policy floor.
Record fundamentals, falling share prices, and a question about who is paying for it.
Nothing in the physical demand data deteriorated in July. That is the point, and it is worth stating with specifics. Third-quarter DRAM contracts settled 20–30% higher during the month. Google and Meta signed five-year contracts locking in memory prices and volumes. High-bandwidth memory is reported sold out through most of 2027, with no meaningful new supply expected before 2028. AWS capacity is reported sold out through 2028. Microsoft grew cloud revenue 43% and passed a $100bn annual run rate at Azure. TSMC beat on both profit and revenue.
And TSMC fell more than 3% on the day, because it raised capital expenditure alongside the beat.
This is not indiscriminate panic. It is a market learning to discriminate. Through 2025 investors read every capex increase as evidence of demand and a maturing revenue backlog. In July they began asking a different question — not whether the spending is justified by demand, but when it converts into free cash flow, and what the return on the incremental dollar looks like. Meta was punished specifically for free cash flow compression. Goldman now models roughly $5.3trn of cumulative capex for the big four between FY2025 and FY2030, up from about $4.5trn a single quarter earlier.
In the last week of July, the Wall Street Journal and Bloomberg reported that Nvidia was weighing a payment guarantee of up to $250bn to backstop OpenAI's lease of a 10-gigawatt campus in Piketon, Ohio — a project developed by SoftBank's SB Energy on a decommissioned uranium enrichment site on federal land, with power from a natural gas plant reportedly pledged by Japan, and a total cost above $500bn. A separate discussion covered up to $350bn of financing for OpenAI's chip purchases. Together with a partnership with SK Group announced on 24 July, reported deal flow exceeded $750bn.
We want to be careful here, because both the bull and the bear readings are lazy. This is not 1999 vendor financing, in which suppliers lent to customers principally to book revenue that would never be collected. Nvidia is guaranteeing an obligation rather than recognising a sale, and the underlying demand for compute is demonstrably real — it is sold out for years. Equally, the dismissal that this is simply how an economy works does not survive the arithmetic. Analysts have identified more than $800bn of circular arrangements across the AI supply chain. Nvidia's own filings cap disclosed lease-guarantee exposure at $3.5bn.
The honest framing is that a cash-rich monopoly is converting balance-sheet strength into a contingent liability in order to sustain a demand curve it also supplies. That is not fraudulent, and it may well be rational. But it is a transfer of tail risk out of the credit market — which explicitly declined to take it — and onto Nvidia's equity holders, who have not been asked to price it. Michael Burry disclosed and expanded a short position on 24 July; Jensen Huang has called the circularity claims preposterous. Both can be posturing and the structural observation can still stand.
The part of this story equity screens cannot see.
This is the section we would emphasise most strongly, because it received the least attention in July and, in our view, matters most beyond the next quarter. Total AI-related debt issuance is projected at roughly $570bn for 2026. A rapidly growing share is being pushed into project-style structures, private placements, asset-backed deals and credit funds where repayment depends on power delivery, construction timetables, tenant commitments and GPU residual values.
The distinction that matters is between a claim on a corporate borrower and a claim on a narrower pool of assets. A direct claim on Amazon is not the same instrument as a claim on one leased building, or on a stack of accelerators that may depreciate faster than the debt amortises. CoreWeave's $7.5bn Blackstone-led facility, collateralised by GPUs and customer contracts, carried an average rate near 11% with repayments beginning in January 2026 — as collateral values were falling. Sub-investment-grade "neocloud" tenants introduce a category of risk conventional hyperscaler underwriting does not contemplate, and lenders are hedging it with parent guarantees, credit wrappers and re-leasing assumptions of varying credibility.
The reassuring datapoint is that the Chicago Fed found bank exposure to AI-adjacent industries averaging just 0.8% of total assets. The same study cautioned that indirect exposure — lending to the lenders — is not captured. PIMCO's secular outlook states plainly that the default cycle is reasserting itself.
You do not need a collapse in AI demand for any of this to bite. A delay is sufficient. And because so much now sits outside the public balance sheets everyone watches, losses will surface slowly and in the wrong places. Our practical conclusion is that data-centre ABS spreads, private-placement pricing and hyperscaler CDS are better leading indicators for AI equity risk than AI equities are.
A war that reopened, a chokepoint that has not, and a bond market tracking both.
OPEC+ has been steadily unwinding its voluntary reductions, completing the restoration of the 2023 cuts in early August and leaving headroom to add supply once hostilities end. The EIA projects a well-supplied second half; J.P. Morgan forecasts Brent averaging $86 in the third quarter, $80 in the fourth and $78 at year end, with 2027 substantially lower.
For allocators the transmission mechanism is what counts. Treasury yields tracked crude almost tick for tick through July — rising with oil on 31 July as Fed officials talked about the need for hikes, falling on 3 August when the Treasury Secretary signalled a possible Hormuz agreement. Duration positioning without a crude view is an unhedged bet on Middle East diplomacy. We would rather own that risk explicitly than implicitly.
The debasement trade was a duration trade wearing a costume.
The cause is one variable. A hawkish new Fed chair, a war-driven inflation impulse, a strong dollar and fading exchange-traded fund demand together removed the rate-cut expectation that powered the 2024–25 run. Non-yielding assets lose when the risk-free rate rises and stays. Gold underperformed the dollar against the rest of the G10 by roughly 2.6 percentage points during the March–June war period — a safe-haven asset failing its most basic test. US-listed gold ETFs saw around $5.3bn of redemptions in a single month. Bitcoin spot ETFs recorded their worst month on record with $4.5bn of June outflows.
What did not break is the sovereign bid. Central banks are still projected to buy 750–850 tonnes in 2026 — below the 2025 record but among the strongest years on record — and Chinese net imports inflected sharply higher. Per Morgan Stanley, gold now accounts for a larger share of central bank reserves than US Treasuries for the first time since 1996.
Two very different buyers are being confused for one. The sovereign accumulation is slow, structural and price-insensitive, and it is continuing. The leveraged financial demand was a bet on falling real rates, and it has left. We would not treat this year's drawdown as evidence against the first, and we would be more constructive on gold into any confirmed dovish pivot than the current price action implies.
Any one of these would have led a normal month.
The forced seller is gone and the liquidation discount has repriced out. That is settled. Whether AI infrastructure deserves its multiple has not been tested at all. We would not read the snapback as an all-clear, and we expect further dispersion within the theme rather than a uniform re-rating.
Roughly $570bn of AI-related debt this year and a projected $800bn more of private-credit data-centre financing. Equity holders are being asked to underwrite contingent liabilities they cannot see, at multiples that assume no financing friction. Credit at least comes with covenants and a defined claim.
Hedged JGB yields well above comparable Treasuries, in a market where Washington has now demonstrated it will help fund a currency defence to avoid forced Treasury sales. The policy floor under this trade got materially firmer on 31 July. It deserves formal work, not a watching brief.
Record equal-weight outperformance, the best small-cap first half since 1991, energy up 34.7%, community banks up 23.9%. Five years of large-cap growth dominance ended in the first half of 2026. July accelerated something already underway rather than starting it.
Korea doubled while its most price-sensitive holders sold a record ₩148.3trn. That configuration — a domestic, levered, behaviourally correlated holder base — is what turned three legitimate competitive questions into a 38.6% drawdown. We should be running this ownership test on every crowded market we hold.
Three shares in five trade off-exchange, swap leverage never reaches a 13F, and regulatory data arrives two weeks late. Borrow rates, single-name skew and open interest are the only live telemetry remaining. This should be a budget line, not an aspiration.
We hold these views with real uncertainty, and it is worth naming the evidence that would change them. If memory contract pricing rolls over in the fourth quarter — the opposite of the 20–30% increases seen in July — then the selloff was early rather than mechanical, and the credit concern becomes immediate rather than a 2027 question. If September's employment and inflation data come in soft enough to restore a cut into the curve, the entire factor rotation reverses and gold and bitcoin are wrongly positioned as losers. If foreign investors return to Korea in size and the ownership base broadens, the fragility argument in section four weakens considerably. If Nvidia formally commits to the OpenAI guarantee at anything near the reported scale, the equity risk premium for the whole complex should widen rather than narrow. And if the AstraZeneca transaction proceeds, it tells you large-cap pharmaceutical management sees less organic growth ahead than their pipelines currently imply.
| Date | Event and why it matters |
|---|---|
| 7 Aug | US July employment report. The first hard labour data since three officials dissented in favour of a hike. A firm print makes September genuinely live. |
| 11 Aug | EIA Short-Term Energy Outlook. The first full supply-and-demand balance since hostilities resumed. |
| 12 Aug | US July CPI. The most important single release before the September meeting, and the first to carry any Section 301 tariff pass-through. |
| Mid-Aug | Second-quarter 13F filings. Will show quarter-end positioning — and will not show a single dollar of swap exposure. |
| Daily | Korea Exchange investor-type flow data. A sustained foreign return would change the ownership picture in section four. A relapse into retail-only buying would not. |
| 27–29 Aug | Jackson Hole. Warsh speaks. The most consequential scheduled event of the quarter for the rate path, and therefore for the factor rotation. |
| Sept | SpaceX lock-up expiry window. Adds genuine float to a listing already trading below its issue price. |
| 15–16 Sept | FOMC. Currently priced at roughly 57% for a 25bp increase. |
| Ongoing | US–Iran talks and the Hormuz reopening. Drives crude, which drives yields, which drives the rest. |
Index-level calm is not the absence of risk. Sometimes it is just two very large forces cancelling out on the way past each other.
Raffles Square Fund · August 2026Sources. Compiled from contemporaneous reporting including Reuters, Bloomberg, the Financial Times, the Wall Street Journal, CNBC, the Korea Times, Yonhap, Al Jazeera, Axios, NPR, the Japan Times and Fortune, alongside primary data from the Korea Exchange, the Bureau of Labor Statistics, the EIA, USTR, SEC EDGAR, FINRA ATS transparency filings and the BIS Quarterly Review, and research from the Korea Economic Institute, Goldman Sachs, Nomura, SpotGamma, Lazard, J.P. Morgan, CreditSights, State Street and PIMCO.
Confidence. Index levels and monthly returns, the FOMC vote and dissents, the coordinated yen intervention, the Section 301 tariff action, hyperscaler capital expenditure guidance and the Korean foreign-flow figures are multiply sourced. Situational Awareness LP's Q1 2026 13F values are taken from the SEC filing and third-party aggregators; reported assets under management vary widely across outlets — Quiver Quantitative estimates roughly $9.3bn from an ADV filing while press coverage around the unwind described a book of $20bn or more, and the $13.68bn 13F headline is notional reportable exposure rather than assets. Fund leverage ratios and the pricing of the 30 July block are as reported and not independently verified. Single-day Korean flow figures for 31 July are as reported from Korea Exchange data. The Nvidia–OpenAI financing structures and the AstraZeneca–Bristol Myers Squibb discussions are early-stage reports that may not proceed. Certain factor statistics originate from market commentary rather than published index documentation. Charts are drawn from the figures cited; where a continuous series was unavailable, intermediate paths are illustrative and end-points are the reported values.
Disclaimer. This article is commentary and general information only. It is not investment advice, not a recommendation to buy or sell any security, and not an offer or solicitation in any jurisdiction. It does not take account of any reader's objectives, financial situation or needs. Past performance is not a guide to future returns. All figures are as at the date of publication and are subject to revision.
