The Cockpit · July 2026

Edito

Jonathan Castella, CFA, editor of Socrates on Investing
Jonathan Castella, CFA

The July you read about, and the July in your portfolio

If I had to point to one thing about July 2026, it would be the distance between what you read and what happened to your portfolio.

Follow the financial press for the month and you come away with a dramatic picture. It is also a fair one. Oil rose 21.8 per cent. Inflation fears returned within days of a benign June print. The thirty-year Treasury closed at 5.21 per cent on 30 July, a level the press called the highest since 2007. Semiconductor and memory shares fell hard. The median chip stock in our watchlist lost 18.5 per cent in dollars, and the extremes went far past that. SK Hynix halved. At its low on 30 July it was down 50 per cent in won from where it began the month, and it closed July down 35. Sandisk lost 47 per cent. Margin calls followed. First among leveraged retail investors in South Korea, where the finance minister apologised to parliament. Then at a large American hedge fund, whose collapse became the story of the month.

None of that is exaggerated. It all happened.

Now look at the seven model portfolios in the Portfolios tab above. In July they lost money, and the worst of the twenty-one was down 1.2 per cent. Nothing dramatic, and nothing that changes the picture since the start of the year.

Most of them are still up on the year. The equity portfolios, the most aggressive of the seven, are still close to 10 per cent in all three currencies covered: dollars, euros and francs. The balanced portfolios are still up between 3 and 4 per cent, with flat to modestly negative returns for the fixed income portfolios and the permanent portfolio.

So which of the two were you reacting to in July?

I am not arguing that the news was noise. Two things from it will outlast the month. The first is rates. Three members of the FOMC voted for a hike at the July meeting, more than at any meeting this year, and the next one is now the live question. Oil feeds inflation, inflation feeds the argument for tightening, and the Strait of Hormuz is not resolved. Rates that stay here will press on asset prices sooner or later. Every valuation is a discount rate wearing a disguise.

The second is dispersion. Earnings momentum was strong. At the same time, as Wei Li, BlackRock’s Global Chief Investment Strategist, wrote on 2 August, “the gap between single-stock volatility and index volatility has widened dramatically this year”. That gap cuts both ways. It raises the reward for holding the right names and the cost of holding the wrong ones, which is one sentence read twice. Diversification gives up the first to avoid the second.

Which brings me to Situational Awareness. On its own reported numbers the fund returned 439 per cent in the first half, one of the best runs anywhere, under a 25-year-old who had never managed money and had been treated as a prodigy. Press reports put its assets near $45 billion. If that figure was equity rather than exposure, the positions behind it ran well past $100 billion. Then July arrived. The same letter that claimed the 439 per cent reported minus 67 per cent for the month, and said the fund had sold in order to remove all leverage. Whether its view of the AI trade was right is not something any filing can settle. Leverage does not decide whether you are right. It decides whether you are still there when the answer arrives.

This situation rhymes with the collapse of Long-Term Capital Management in 1998. Back then LTCM was regarded as the most prestigious fund of the nineties, with two Nobel laureates on its staff. After four years of exceptional returns it collapsed in a matter of months and triggered a rescue organised by the Federal Reserve.

Both stories lead naturally to a famous line from Charlie Munger, Warren Buffett’s late partner at Berkshire Hathaway. “There are only three ways a smart person can go broke: liquor, ladies and leverage.”

The Cockpit · July 2026

The Seven Model Portfolios

Seven classic portfolios, from Fixed Income through to Equity, read as references rather than recommendations. Each is passive, total-return and rebalanced quarterly, indexed to 100 with a simulated-expense slider. The graph, the performance and risk statistics, the contribution by holding and the full compositions sit below, frozen as of 31 July 2026. They are a risk dial to set a real allocation against, not products.

Graph

Statistics

Compositions

Contribution

The Cockpit · July 2026

Equity Watchlist

One thousand companies, the S&P 500 plus the 500 others we consider the most important in the world, priced to the close of 31 July 2026. Sort any column to see where the month actually moved, filter by sector, region, currency or size, and open a name to read its ten-year history. Three companies that belong here are absent, and the reason is worth stating rather than hiding. International Holding and Phoenix Group cannot be priced from either of our data feeds. ChangXin Memory listed in Shanghai on 27 July and would enter near the thirtieth largest, but it has a ticker and no usable prices yet. An instrument that reports its own blind spot is more use than one that leaves the gauge blank.

The Cockpit · July 2026

Top News

The July developments that mattered most for markets, read to the close of 31 July 2026. Every figure was checked against a filing, an official publication or our own series. Where a claim rests on press reporting, the outlet is named.

Iran closed the Strait of Hormuz, and oil took back the good inflation news. On 14 July the Bureau of Labor Statistics reported June consumer prices down 0.4 per cent on the month. That was the largest monthly fall since April 2020. The annual rate came in at 3.5 per cent, core at 2.6.[1] The print was already stale. By 31 July West Texas Intermediate had risen 21.8 per cent to $84.67 and Brent 23.6 per cent to $90.12. WTI traded as low as $68.55 on 6 July and as high as $92.19 on 23 July. Energy had supplied most of June’s disinflation. It took it back in three weeks.[2] Both numbers are true. Only one of them described the regime.

The Fed held, and the long end tightened anyway. The FOMC left the range at 3.50 to 3.75 per cent on 29 July. Three of the twelve voters dissented. All three wanted a rise.[3] The bigger change was the statement. It runs to about one hundred and twenty words, with no forward guidance and no balance of risks. Kevin Warsh has removed a thirty-year convention in two meetings. His reason: markets had been “reflecting back what we’ve said”.[4] Markets were not reassured. The thirty-year Treasury closed above 5 per cent on eighteen of twenty-one trading days, and at 5.21 on 30 July. Press put the intraday high at 5.244 per cent, the highest since 2007. The Bank of England held 6 to 3, the Bank of Japan 8 to 1, the ECB after a June rise.[5,6,7]

Then Washington and Tokyo bought yen together, for the first time since 1998. The dollar reached 163.86 yen on 29 July and closed the month at 160.18.[2] Japan is reported to have intervened alone on 30 July, then jointly with the United States on 31 July and 1 August, with the US Treasury selling euros to buy yen. The reason connects directly to the paragraph above. A falling yen forces Japanese institutions to sell what is liquid. What they own that is liquid is United States Treasuries. Selling those into a long end already near 5 per cent is the outcome both governments were trying to avoid. No official figure exists for the size. Tokyo publishes monthly, Washington quarterly, and neither had covered this window as at 4 August.

The momentum trade unwound, and the money went to energy and software. We computed this from our own watchlist of a thousand companies, over the calendar month to 31 July. The median energy stock rose 13.7 per cent. Financials rose 6.7 per cent. Information technology fell 1.0 per cent. The split inside technology is the story. Semiconductors and memory fell 18.5 per cent at the median. Software rose 12.0 per cent. Sandisk fell 47 per cent, KLA 39, Marvell 37, Intel 35. Workday rose 31 per cent, HubSpot and Atlassian 30, Microsoft 25. The popular first-half trade was long chips and short the hyperscalers. It reversed inside a month.[2]

That reversal produced margin calls on two continents. In Seoul the damage was retail and political. Finance Minister Koo Yun-cheol apologised to parliament on 29 July after losses in single-stock leveraged ETFs, a product the Financial Services Commission had approved on 30 January.[8] Retail investors had put about 14 trillion won into them since they launched on 27 May. The KOSPI fell 10.84 per cent on 28 July. Margin balances peaked at a record 38.63 trillion won on 24 June and were 33.0 trillion by 29 July.[9] Goldman Sachs estimates that 320,000 to 360,000 leveraged retail accounts had been fully liquidated by 13 July, two weeks before the worst of it. Citi puts total retail losses on these products near $38.7 billion.[10] A regulator approved the product in January and apologised for it in July.

In San Francisco an AI hedge fund came apart in a matter of weeks. Leopold Aschenbrenner is 25. He entered Columbia at 15 and graduated valedictorian at 19. He joined OpenAI’s superalignment team in 2023 and was dismissed a year later. In June 2024 he published an essay arguing that scaling AI would demand an enormous build-out of chips, compute, memory and power, then raised a fund on that thesis. Jane Street backed him, as did Patrick and John Collison, Nat Friedman and Daniel Gross. He had never managed money before.[13]

Then it collapsed, and how far is not independently known. Every performance figure for this fund comes from the fund. In a letter to investors seen by Reuters, it reported 439 per cent for the first half and minus 67 per cent for July. They are unaudited, and the fund was asking investors for fresh capital in the same weeks it published them.[11] Treat them as a claim, not a measurement. No filing shows what it held on the day it sold. Its quarterly holdings report lists long positions only, so it cannot settle what the fund was net of anything.[12]

The scale is the part worth pausing on. Press reports put assets near $45 billion in early July, and near $10 billion after a block sale reportedly to Citadel, on unnamed sources neither party has confirmed. Nobody has said whether $45 billion meant net asset value or gross exposure. If it was net asset value, then at the leverage a fund like this runs, the positions it controlled may well have been over $100 billion. Either way the losses were large and they came fast. The fund’s own letter says it sold in order to remove all leverage, which is the mechanism worth holding on to: leverage sets how long you can afford to be wrong.

The earnings season was strong, and the headline figure is not what it looks like. With 61 per cent of the S&P 500 reported at 31 July, 86 per cent had beaten on earnings against a five-year average of 78. Blended growth was 47.4 per cent, against 23.2 per cent expected on 30 June.[14] Then strip out two companies. Alphabet’s result included a $98 billion gain and Amazon’s $53.4 billion of non-operating income, both unrealised marks on private AI stakes. Without them, growth falls to 28.8 per cent and the aggregate surprise from 31.4 to 9.2. Alphabet reported $9.11 of GAAP earnings against $2.88 expected. Excluding the gain it earned $2.85 and missed. When private marks add eighteen points to index earnings growth, the word is measuring something else.

Microsoft and Amazon were rewarded for spending. Meta and Apple were not. The test was whether revenue arrived alongside the capital expenditure. Microsoft’s Azure grew 43 per cent and it guided to 45, with backlog nearly doubling to $678 billion. It added about $450 billion of market value in a day, close to close, the largest dollar gain on record. Amazon’s AWS grew 37 per cent, its fastest in eighteen quarters. Meta had no such answer. Earnings missed at $6.18. Free cash flow fell to $784 million. It guided capital expenditure to $130 to $145 billion, with no cloud business to sell. Apple beat on both lines and still fell 7.35 per cent. It guided below expectations and blamed component supply. The AI build-out is consuming the parts Apple needs.[15,16]

The rest of the field, and a credit rating that moved. Alphabet grew cloud revenue 82 per cent and fell 7.4 per cent, on higher capital-expenditure guidance. A record profit met with a 7 per cent fall is what started the scrutiny of AI spending. Tesla’s operating margin was 1.4 per cent. Nvidia had not reported as at 31 July and is due in late August. SpaceX reports on 4 August, its first results as a public company, after falling 31.2 per cent in July. S&P cut Oracle to BBB− on 9 July and named OpenAI as a credit risk. The financing of the AI build-out now carries a credit rating of its own.[17,18,19]

Gold barely moved, and the interesting number was who bought it. Gold rose 1.1 per cent to $4,051, its first gain after five monthly falls.[2] Central banks bought 288.9 tonnes in the second quarter, against 177.9 a year earlier. The World Gold Council calls that “a fivefold increase on Q1’s revised estimate of 57t and a record high for a second quarter”. Poland took the most at 51 tonnes, China 33. Now take the half year. Net demand was 345 tonnes, “the lowest for a first half since 2022”, when it was 241. The drag was Turkey, Russia and Azerbaijan selling, and it happened in the first quarter. By the second, Turkey was down to 4 tonnes and Russia was the only sizeable seller left.[20,21] A record quarter sits inside the weakest half in four years.

Europe shut two of crypto’s largest firms out, and a hardware wallet failed. The MiCA transition ended on 1 July. Tether chose not to apply. Binance applied in Greece, then withdrew on 24 June ahead of a refusal. One declined to be regulated. The other sought regulation and did not get it.[22] Separately, Coinkite disclosed on 30 July that seeds generated on older Coldcard wallets used far less randomness than intended. Updating the firmware does not repair an existing seed. CoinDesk reported nearly 600 bitcoin taken, about $38 million, and called that the figure so far.[23,24] Strategy sold 3,588 bitcoin across two weeks to fund a preferred dividend, and said so in a footnote.[25] Bitcoin still rose 7.3 per cent, to $62,814. It remains down 28.2 per cent this year.[2]

China had the loudest month, and almost none of it was priced in New York. On 16 July Moonshot AI released Kimi K3, an open-weight coding model with 2.8 trillion parameters. Press reports credit it with matching leading US models on coding benchmarks, and with triggering a sharp global selloff in semiconductors. Moonshot is now preparing a Hong Kong listing within six months, at a valuation reported between $20 and $30 billion.[26] Eleven days later ChangXin Memory listed in Shanghai, raised 57.92 billion yuan, and closed its first day about 466 per cent higher at roughly $480 billion.[27] On that value it would rank near the thirtieth largest company in our watchlist. It is absent from it, because neither of our price sources can quote the STAR market yet. It waits for the August close.

And a Chinese company began building the machines that make the chips. Reuters reported on 28 July, citing a single source, that Shanghai Aishengna Electronic Technology Group had started producing immersion DUV lithography systems, the tools that print circuits onto wafers. About five are expected this year, going to SMIC, Hua Hong and ChangXin Memory, with roughly twenty planned for 2027. No company has confirmed it. Two things keep the scale honest. ASML has capacity for around 130 immersion units this year, so five is a rounding error in global supply. And these are DUV machines, not EUV, which is the tool China still cannot build and the one that governs the leading edge.[28]

The Cockpit · July 2026

Methodology & sources

This report is a set of instruments, not a set of instructions. What follows is how each figure is built, where the data comes from, and how it is checked, so that any reader can trace a number to its origin and judge it independently.

How the figures are built

The seven model portfolios and the asset-class series are total-return, indexed to 100 at the start of each period, and expressed in US dollars. Where a simulated expense is shown, it is applied as a daily drag on net asset value, and rebalancing follows the frequency selected in each widget. Yields, spreads and inflation are plotted as reported, without adjustment.

Sources

The data in this report comes almost entirely from two public sources: Yahoo Finance for market prices, and FRED, the Federal Reserve Bank of St. Louis, for the macro series. Fund fact sheets are the reference used for cross-checking. This is public, end-of-day data, not an institutional terminal feed.

  1. Market prices and total returns: Yahoo Finance, using dividend-adjusted closing prices for each ETF and index.
  2. US Treasury nominal and real (TIPS) constant-maturity yields, and the effective federal funds rate: FRED (Federal Reserve Bank of St. Louis).
  3. Moody's seasoned Aaa and Baa corporate bond yields, and ICE BofA investment-grade and high-yield option-adjusted spreads: FRED.
  4. US consumer price index (CPI-U): US Bureau of Labor Statistics, via FRED.
  5. Portfolio weights and fund characteristics: the respective ETF issuers' published fact sheets, used for cross-checking.

News and events

Sources for the Top News items, numbered to match the references in that tab.

  1. US Bureau of Labor Statistics: Consumer Price Index, June 2026, released 14.07.2026
  2. Socrates on Investing: Main Assets series and Equity Watchlist, computed to the close of 31.07.2026
  3. US Federal Reserve: FOMC statement, 29.07.2026
  4. US Federal Reserve: FOMC press conference transcript, 17.06.2026
  5. Bank of England: monetary policy summary and minutes, meeting ending 29.07.2026
  6. Bank of Japan: statement on monetary policy, 31.07.2026
  7. European Central Bank: monetary policy decisions, 23.07.2026
  8. CNBC: Korean leveraged ETF losses and the finance minister’s apology, 29.07.2026
  9. Financial Supervisory Service (Korea), consumer risk response council, 06.07.2026, and KOFIA margin-loan and forced-sale data
  10. Goldman Sachs estimate of fully liquidated leveraged retail accounts, as at 13.07.2026, relayed in press; Citi estimate of total retail losses, reported 29.07.2026. Both are sell-side estimates, not official statistics
  11. Situational Awareness: letter to investors, July 2026, published in full; Reuters, 31.07.2026
  12. SEC EDGAR: Situational Awareness LP, Form D/A 10.03.2026 and Form 13F, first quarter 2026
  13. Biographical detail and named backers: CNN Business, 31.07.2026, and contemporaneous reporting; the fund’s own founding essay, published 14.06.2024
  14. FactSet Earnings Insight, John Butters, 31.07.2026
  15. Microsoft: FY26 fourth-quarter results, 29.07.2026; the single-day market-value figure is Reuters and Axios, measured close to close
  16. Amazon, Meta and Apple: quarterly results releases, 29 and 30.07.2026
  17. Alphabet: second-quarter 2026 results, 22.07.2026; Tesla: second-quarter 2026 update, 22.07.2026
  18. S&P Global Ratings: Oracle downgraded to BBB-, 09.07.2026
  19. SpaceX: announcement of second-quarter 2026 results date, reporting 04.08.2026
  20. World Gold Council: Gold Demand Trends, second quarter 2026, published 30.07.2026
  21. World Gold Council: Central Bank Gold Reserves Survey 2026, published 16.06.2026
  22. Regulation (EU) 2023/1114 on markets in crypto-assets (MiCA); CoinDesk on Binance, 26.06.2026
  23. Coinkite: Coldcard security advisory and technical backgrounder, 30.07.2026
  24. CoinDesk: Coldcard exploit, 31.07.2026
  25. SEC EDGAR: Strategy Inc., Forms 8-K, 06.07.2026 and 20.07.2026
  26. Moonshot AI and the Kimi K3 release: press reporting, July 2026, no filing exists
  27. CNBC: CXMT market debut, 27.07.2026, including the Omdia market-share data disclosed in the listing prospectus
  28. Reuters: China starts production of home-grown immersion DUV chipmaking tools, 28.07.2026, single-source and unconfirmed by any company; ASML capacity context from the same reporting

How it is checked

This report is produced with the help of an AI assistant working under the supervision of Jonathan Castella, CFA, who authors it and is responsible for what it says. Before publication, the underlying data passes hygiene checks, and individual figures are spot-checked against issuers' fact sheets and the primary series. These steps reduce error, but they do not eliminate it: mistakes remain possible.

Corrections and liability

This report is educational. It examines frameworks, regimes and history; it is not investment advice, and it recommends no security, fund or allocation. Nothing here should be relied upon as the basis for a financial decision, and the author accepts no liability for any decision taken, or any consequence arising, from its use.

Accuracy matters to the publication. If you find an error, or have a correction, a question or a comment, it is welcome: write to socrates@socrates-on-investing.com.