December 1st Links

• Without a meaningful drop in US crude oil production in 2026, the global oil market rebalancing will be stalled. The other way to think about it is that if the expected surplus is ~1.5 million b/d, then either global oil demand has to increase by that much or the supply side has to decrease by that much. In this case, if US crude oil production is flat at $60/bbl in 2026, then the only way to rebalance the market is for demand to increase. If there are any signs that demand does not keep pace and global oil inventory builds pile up, then you will see oil prices crash to push on the supply side. This is the asymmetric situation that we are running into. Inevitably, the supply side has to respond (via a decline). The global economy is not that healthy to pull us out of this incoming glut entirely, so without a material supply reduction, the incoming downturn is going to look very ugly. If history rhymes, then it will punish producers. That’s the situation we are going into now. To make matters worse for energy investors, Canadian oil stocks are priced for perfection today. [HFI Research]
• On the technology S-curve, we’re early. But that tells you almost nothing about the equity S-curve. You have to keep those two curves separate. If anything, history suggests the opposite: the earlier you are on the adoption curve, the more likely it is that markets have already tried to cram 30–40 years of expected impact into a 3–5 year price spike. The stock market is not patiently tracking real-world diffusion; it’s front-running the story. And then you have to ask what actually changes when a bubble peaks. Spoiler: it’s not the code, not the chips, and not even the next two quarters of P&L. What changes is the story in people’s heads, and once that flips, everything else (multiples, flows, “fundamentals”) gets reinterpreted through that new lens. Technologies diffuse on multi-decade S-curves. Stocks move in short, violent spikes. AI can be genuinely early on one and already mature on the other at the same time. [Early in AI, Late in the Trade]
  
• To me, it seems like the single most important quality in an investor is an ability to look for situations that are too good to be true, and then chase those down ruthlessly. It’s a strange combination of traits at the extremes of the explore/exploit dichotomy. Great ideas are rare, so you should turn over lots and lots of rocks. But they do exist, and you need to be able to recognize them and, when you really find one, put everything else aside to make sure you actually are exploiting it. If you optimize too much for the “explore” part of the equation, you will fail to follow up and execute on your best ideas as thoroughly as you should. But if you don’t explore enough, you will spend lots of time researching subpar ideas, eventually convince yourself they are great, and have mediocre performance. [Sardine Trader]
• The problem is simply how does one turn the complicated style of Thucydides into English. The two new translations take different approaches. W. Blanco writes, “I have tried to make a translation of Thucydides’ famously difficult text that would be accessible to students and general readers. To do so, I have relaxed the compressed, often crabbed, syntax of the speeches and have adopted a relatively colloquial vocabulary for them and for the narrative as a whole. I offer no apologies” [Bryn Mawr Classical Review]
• There are thousands of variations you could have used, each resulting in a similar-yet-slightly-different program. And that’s why prompt engineering is needed. There is no a-priori reason why your first, naive program key would result in the optimal program for the task. The LLM is not going to “understand” what you meant and then perform it in the best possible way — it’s merely going to fetch the program that your prompt points to, among many possible locations you could have landed on. Prompt engineering is the process of searching through program space to find the program that empirically seems to perform best on your target task. It's no different than trying different keywords when doing a Google search for a piece of software. If LLMs actually understood what you told them, there would be no need for this search process, since the amount of information conveyed about your target task does not change whether your prompt uses the word “rewrite” instead “rephrase”, or whether you prefix your prompt with “think steps by steps”. [François Chollet]
• In investing in emerging markets, Firebird often tries to divine the truth not from anything a company or brokerage analysts say – which frequently is misleading – but from what they don’t say, as well as other non-information that creates a mosaic. Just one example from hundreds I have: in the early days of the Ukrainian stock market an oleaginous broker came to New York pitching all the main exchange-listed stocks – except one, the one that to us seemed most interesting. He never mentioned the other one at all. Of course, it turned out that omitted stock was the only good one and the broker’s more important clients were already likely looking for shares, while the others were low-quality, with plenty of sellers. In reference to Sherlock Holmes’ Hound of the Baskervilles, I call that situation “The Dog that Didn’t Bark”. [Harvey Sawikin]
• Let's talk about something that will shape our future, but no one is yet considering or predicting: China will eventually open its borders to mass immigration. When it does it will absorb truly astounding numbers of immigrants, if proportional to say what the UK or Canada this will be 200-300 million or so people. This might sound almost physically impossible, but the surprising almost disturbing truth is that in the modern world it is easy move 200 million people. In 2023, 4.4 billion passengers were carried by the world's airlines. 200-300 million people in a graying ageing world of low fertility is enough to raise global labor costs. This will be completely distinct second China shock to Western companies hoping to offshore to places with cheap labor. The cheap labor good at working in factories will move to China. The West will get scraps of misplaced high end human capital and food app deliverymen. East Asian democracies such as Japan and South Korea are experimenting with mass immigration as their ageing population reaches true crisis. There is populist backlash in East Asia, but just as in Europe and America the people don't get a say when business owners speak. Unlike democracies authoritarian governments aren't picking the absolute most unassimilateable and unproductive immigrants they can find. But they are still importing massive numbers of workers. Look at Russia, Belarus, or countries such as Dubai and other Gulf monarchies. Authoritarian governments just want workers; they don't need welfare recipients to vote as instructed. No impactful elections, means there is no need to import political clients to socially engineer elections. This solves some but not all societal problems seen in the West. Neither China being an East Asian country nor it being an authoritarian country are therefore reasons to not expect eventual mass migration into China. When China exhausts its rural labor reserve, likely in the next decade, we can expect pressure to mount for either outsourcing or immigration. In the West we did both. I expect in China the CCP will after some nationalist grandstanding ironically come to the conclusion that mass immigration will be preferable to outsourcing. Marxist elites will prioritize maintaining the overall concentrated industrial base and cdemocracy. Mass immigration China will bitterly disappoint China doomers: China will not collapse due to low birth rates. It will do what bureaucrats do around the world. Human quantitative easing to make line go up. Mass immigration China will bitterly disappoint liberals: It will turn out it is much more economically efficient and politically sustainable to only import workers and not voters. We already saw 200-300 million people move to work in China's factories. It just all happened inside of China. The distances involved in these migrations are no less than say immigration from South or Southeast Asia would be. I also want to remind nationalists who might like China's nation state of the obvious point that the political institution that carried out the cultural revolution doesn't actually have any fundamental commitments to Chinese cultural continuity. I think the same kind of decision-making that interns the Uighurs for reeducation, imposes the one child policy, imposes internal border controls is actually exactly the kind of ruthless people as numbers technocratic decision-making that would import 100 million Indonesians to Shenzhen while policing them heavily to avoid demographic driven economic collapse. [Samo Burja]