This week’s reading list is about the gap between a shock and its consequence. Long bond yields are up sharply across six major markets, but the most careful reading available says this is less a verdict on American institutions than the price of clearing a global market in which more people want capital and fewer are willing to hold duration at any price. A war that should have been a windfall for oil producers appears instead to be dragging peak consumption forward, because demand turned out to be more elastic than the textbooks allow. A US$2 trillion repricing of enterprise software happened over a stretch in which enterprise software revenue growth actually accelerated. The most durable source of excess returns turns out to be not the moat everyone is trained to look for, but the absence of anyone looking at all. Humanoid robots are among the most watched technologies in the world, and the watching tells you very little, because the demo video is the least informative artefact the industry produces. Twenty five years on, the direct damage of the 11 September attacks is dwarfed, by more than twenty to one on a conservative accounting, by what the response to them cost. And a coffee shortage in East Germany in 1977 is a large part of why Vietnam is the world’s second largest coffee producer today. The useful lesson for investors is that first order reasoning gets the direction right about as often as a coin does. The best investors are not trying to predict the shock. They are trying to work out what it will actually set in motion, which is usually something other than the obvious thing.

📚 What is driving the global rise in long-term interest rates? Gene Frieda at Bruegel on the question sitting underneath every valuation model right now. Two loud explanations are on offer: that investors are losing faith in US fiscal policy and Federal Reserve independence, or that higher real rates simply signal stronger growth. Frieda finds neither convincing, partly because the move is not uniquely American, with US, German, French, Italian, UK and Japanese 30 year yields all up between 45 and 79 basis points in the six months to late August. His conclusion is less dramatic but more useful: long term capital has become more expensive because investment demand and the supply of long dated debt have both risen while the buyers who used to absorb duration regardless of price, central banks, reserve managers and Dutch pension funds among them, have stepped back. There is Australian data in here too, and it is striking. Real yields accounted for about 97 per cent of the rise in Australian long rates since September 2024, with almost none of it coming from inflation expectations. (Bruegel)

📚 Something is shifting in the inflation picture: Claudia Sahm explaining why she has moved from hold to hike ahead of the September Fed meeting, having moved from cut to hold only in June. Her framework is the genuinely useful part. A central bank should look through shocks that raise the price level temporarily, which is why tariffs and an energy spike were reasonable to ignore. Her argument is that three things have changed: a Middle East conflict now long enough that energy is bleeding into core prices, a Canadian trade war implying tariffs are not finished, and memory chip shortages from the AI buildout. The last is the sharp distinction. Chip scarcity is demand driven, not supply driven, so the logic for looking through it does not apply. Note she is explicit that better data could move her back to hold, which is a more honest way to hold a macro view than most people manage. (Stay-At-Home Macro)

📚 A long war will permanently damage oil demand: Tim McDonnell on the counterintuitive consequence of crude moving back above US$100. Supply really is constrained, and S&P Global has now concluded for the first time that Middle Eastern production will not return to prewar levels by the end of 2027. But the demand side is doing something the standard model does not expect. Oil demand is assumed to be highly inelastic, which is why prices swing so violently. JP Morgan’s Natasha Kaneva notes that demand has been running around five million barrels a day below last year and has absorbed the largest share of the price shock, with next year potentially the weakest since 2019. The implication for anyone holding energy exposure is uncomfortable: the price of alternatives has fallen far enough that the level required to destroy demand is lower than it used to be. (Semafor)

📚 The SaaSpocalypse was more like a RenaiSaaS: Ernie Tedeschi introduces a weekly index built from Stripe payment volumes across roughly 72,000 businesses, and uses it to ask what actually happened to software revenues during the early 2026 sell-off that erased somewhere between US$1 trillion and US$2 trillion of enterprise software value. The answer is that revenue growth accelerated straight through it, with the index now running above 30 per cent year on year and sitting 4 per cent above its pre-sell-off trend. Younger SaaS businesses grew fastest of all. Tedeschi is careful to say this does not make the market wrong, since valuations are prospective and the concern was about AI encroachment over years rather than quarters. Worth reading with the obvious caveat in mind, which is that Stripe has an interest in the health of the businesses it processes payments for, and that pay-in volume is not an income statement. (Stripe Economics)

📚 Invisible Companies: Jay Barney, Haiyang Zhang and Jerry Neumann on a source of durable excess returns that sits outside both of the standard explanations. Not barriers to entry, not hard to copy resources, but competitive neglect: profits persist because nobody thinks to look, and would-be competitors do not know that they do not know. The evidence is hard to argue with. Constellation Software has compounded at roughly 34 per cent a year since its 2006 listing, against about 11 per cent for Berkshire Hathaway over the same period, buying ski lift ticketing and funeral home record keeping software in deals often under US$5 million, and reckons there are still more than 38,000 such businesses available. Waste Management rolled up 133 haulers in two years because nobody else wanted to be in rubbish. This is the most directly useful piece on the list for anyone who fishes in Australian small and mid caps, and the closing question is a good one: if every screening tool is trained on the same data, does invisibility become harder to maintain, or quietly hardcoded? (Colossus)

📚 14 Reasons Robotics is Hard: Steve Newman with a careful catalogue of what stands between the current humanoid robot demos and an actual artificial worker, and a strong argument for discounting the demos entirely. You may be watching the one success in a hundred attempts, filmed in a scenario arranged to avoid everything the machine cannot do. The specifics are what make it: the human hand has around two dozen degrees of freedom and roughly 17,000 tactile sensors, and no robot hand matches the combination of dexterity, sensitivity, strength and durability. A human sized robot generates about twice the heat of a person with none of the cooling, so some models run fifteen minutes and then need ten to cool down. Waymo has driven over 220 million miles and still occasionally drives into floodwater. And scaling matters: Tesla took fourteen years to reach its first million car year, while ChatGPT reached 100 million users in two months. (Second Thoughts)

📚 The Staggering Financial Cost of 9/11: Robert VerBruggen, for the 25th anniversary, attempting to total the dollar cost of the attacks and everything that followed, in 2025 dollars, with each component estimated separately so readers can adjust or discard any part of it. The direct toll, lives, health, property and the short term economic hit, comes to somewhat over US$100 billion. The response comes to roughly US$1.4 trillion of additional homeland security spending, US$1.2 trillion for Afghanistan and another US$1.2 trillion for Iraq and Syria, with Brown University’s broader accounting reaching US$8 trillion. Against a US economy of about US$30 trillion, that is a very large multiple of the original damage. The piece is published by the Manhattan Institute and its closing paragraphs argue that the scale of the response was nonetheless justified, which readers can take or leave. The accounting itself is transparent and the sources are all linked. (City Journal)