This week’s reading list is about capacity, by which we mean the difference between having a buffer and having one that has ever been tested. Binance kept an emergency reserve called SAFU, which turned out to be worth rather less than the hole left behind it. The primary dealers absorb Treasury issuance much as a bucket absorbs water, which is to say briefly, shedding about 70 per cent of auction inventory inside the week. Ford generated 18,000 candidate names for a new car, narrowed them to ten, and then the chairman asked why they did not just call it Edsel. Damodaran values Anthropic at roughly US$1.475 trillion, with a bear case of negative US$373 billion, which is a wide range even by the standards of the exercise. American pensions have been replaced by something only about half of private sector workers hold at all, and households appear to experience the resulting risk not as policy but as theft. Pleasure reading has fallen from 28 per cent of Americans on a given day to 16, which makes patience cheaper to acquire than it has been in a generation. And three mathematicians have worked out that the paradox at the heart of majority voting is shaped like a Klein bottle, a surface that needs four dimensions to avoid passing through itself. The useful lesson for investors is that most contingency arrangements are quietly sized for the ordinary week. The best investors are not asking whether a buffer exists. They are asking what happens on the day it gets called.
📚 The Anthropic Prospectus Arrives: My (but not the) IPO valuation! Aswath Damodaran works through the 416 page S-1 filed on 26 October 2026, where enterprise customers spending above US$1 million a year have gone from fewer than 100 to about 1,900, the top five customers are roughly a quarter of revenue, and quarterly compute costs have fallen from over 250 per cent of revenue to 48. Against that sit compute commitments exceeding US$600 billion and a BB+ rating. His base case equity value of about US$1.475 trillion lands below the mooted US$1.5 to US$2 trillion range, and these are his estimates rather than facts. (Musings on Markets)
📚 Treasury Issuance and Dealer Balance Sheets: Four Federal Reserve staff measure what actually happens when the Treasury sells bonds, using data from March 2016 to July 2025. Primary dealers take about half of bill issuance and a fifth of coupon issuance, each dollar of which adds roughly 21 cents to dealer balance sheets, and they clear about 70 per cent of auction inventory within three to five days. A US$1 trillion quarterly coupon volume would add some US$210 billion, against more than US$5 trillion of headroom under the leverage ratio, though average value at risk usage already sits at 35 per cent. (FEDS Notes)
📚 North Korea gets away with $3 billion from Binance: Molly White on roughly US$6.4 billion lifted from Binance hot wallets in late October, after which Tether froze at least US$1.567 billion of USDT and Circle about US$800 million of USDC, leaving the attacker with somewhere between US$2.4 and US$3.4 billion. The exchange’s much advertised SAFU insurance fund stands at about US$1.3 billion. TRM Labs attributes the theft to the group variously called TraderTraitor and Lazarus, which would make it comfortably larger than the Bybit raid, and the figures were still moving as she wrote. (Citation Needed)
📚 Insurers Find Greater Use for ETFs: Matt Toledo on a quiet shift in institutional plumbing. US insurance general accounts held US$49 billion in exchange traded funds at the end of 2025, up 25 per cent on 2024, spread across 768 distinct funds against about 500 five years earlier, with roughly 37 per cent of US insurers now using them. The aggregate allocation is still under one per cent of general account assets, so this is a story about the direction of travel rather than the destination. (Chief Investment Officer)
📚 The Market Owes You Nothing: Vishal Khandelwal on the Edsel, for which Ford’s research division generated 18,000 possible names, cut them to 6,000 and then to ten, enlisting the poet Marianne Moore, who offered “Utopian Turtletop” and “Mongoose Civique”, before chairman Ernest Breech asked why they did not just call it Edsel. Around US$250 million went into development and the dealer network on an expectation of at least 200,000 cars in the first year; it launched in September 1957 and was gone by November 1959. His sunk cost test is the keeper: if I did not already own this stock, would I buy it today at this price, knowing what I now know? (Safal Niveshak)
📚 It’s Never Been Easier To Beat This Market: Tony Isola’s argument is that the edge has not disappeared, the competition has simply stopped turning up. The American Time Use Survey has pleasure reading falling from 28 per cent of people on a given day in 2004 to 16 per cent in 2023, while Alix Earle, 25, holds roughly 8.7 million TikTok followers and has said she has never read a book. He also restates Hendrik Bessembinder’s finding that a small minority of stocks account for all net market gains since 1926, which is worth attributing to Bessembinder rather than treating as settled arithmetic. (Tony Isola)
📚 Everyone wants an affordable America. Good luck agreeing on what that means: Hanna Horvath on what happens when people stop believing they can get ahead. Pew has 60 per cent of American adults under 30 struggling to afford at least one basic necessity in the past year, against 26 per cent of those over 65, while about nine per cent of private sector workers now hold a pension and roughly half are in a 401(k) style plan, which Jacob Hacker calls the great risk shift. The behavioural finding she draws on, from Stefanie Stantcheva’s work at Harvard, is the useful one: people credit their own effort for a pay rise but experience a price rise as something taken from them. (Your Brain on Money)
📚 China’s outbound AI strategy starts with models and governance and ends with chips: Alicia García-Herrero and Théo Storella argue the sequencing runs the opposite way to most equity commentary. The World Artificial Intelligence Cooperation Organization went from 29 member countries in July 2026 to 38 by early September, and on 16 October China published draft export rules covering 23 categories of advanced AI technology. Domestically the bill is already visible: data centres consumed about 170 TWh in 2025, some 1.6 per cent of national electricity, and certain provinces have halved power charges for facilities running domestic chips. (Bruegel)
📚 How Voting Is Like a Klein Bottle: Condorcet showed in 1785 that majority preferences can cycle even when every individual ranking is perfectly consistent, and Arrow proved in 1951 that no ranking method satisfies all the obvious democratic criteria. Ori Livson at the University of Sydney, with Mikhail Prokopenko and Siddharth Pritam, went back to cycles an earlier model had discarded and found that gluing its rims with a twist produces a Klein bottle, which needs four dimensions to avoid passing through itself. Oxford’s Iain McLean supplies the deflation, noting that choosing a voting system is a debate about values rather than a problem with an answer. (Quanta Magazine)
📚 In praise of the dilettante: Jay Miller points out that the word is Baroque Italian and originally described someone pursuing an interest for the delight of it, the sneer being a fairly recent addition. Adorno dismissed the dilettante in 1929 as someone unable to establish any standards, then conceded that the expert critic needs the dilettante in order to prove to himself that he is not one. Goethe and Schiller never finished their 1799 essay on the subject, which is either a failure of application or the strongest argument in the piece. (Aeon)
