The Diff - Longreads + Open Thread

Longreads

• The FT has a piece on the rise and fall of Shanghai as a global financial center (free, but registration required). It's an interesting combination of problems: China's economic growth was slower than what many global financial companies underwrote when they opened offices there, even though China still outgrows the rest of the world. The country doesn't want to relax capital controls, which places limits on how big their financial sector can get. And one of the reasons outside companies were invited in the first place was skill transfer, which is always a factor in foreign direct investment. In manufacturing, it can take a long time for new entrants to have a competitive advantage, but in China the local companies started out better-connected by default, and so well before they were at skill parity with outsiders, they had a better business.
• Neal Stephenson on the philosopher Charles Sanders Peirce, who has thoughts on being wrong. What usually prompts this kind of question is the observation that a lot of people have simply crazy beliefs, and many of them are registered to vote. But exploring it in any detail means implicitly staking out some claim on the object-level issues in question. Whereas reading the same discussion at a time when the ambient mainstream narratives and conspiracy theories have no emotional valence whatsoever is a good way to get the right abstractions.
• A new contribution to the literature on very long-term returns by asset class. Apparently bonds are better than we thought, the US really is an outlier, and the zero-interest era, too, was truly an outlier. The curse of finance is that there's too much data until you're trying to prove any specific claim, at which point there's never enough. 19th century equities are sampling from a different distribution than today's, whether that's by industry, geography, overall capital structure, source of returns, or even investor attitudes. (Those attitude shifts are subtle, but they're real; the story of the really big innovations in quantitative finance is the application of rigor to phenomena that were previously identified through trial and error and ad hoc pattern-matching.)
• In Inference Magazine, the case for building AI datacenters in the UK, the case for supplying them with nuclear power, and claims that this is actually a feasible project. There's a lot in the piece that will be familiar to readers of Situational Awareness or advocates of nuclear power. But it also has specific regulatory moves that could actually make this possible. (Disclosure: the first listed author, Jack Wiseman, previously worked for The Diff.)
• Tim Dellinger argues that the distribution of employee performance is not Gaussian, because even if the underlying traits are normally-distributed, companies don't hire from every part of the bell curve, and a truncated distribution has different characteristics. This is an important point, and a good argument against aiming for a certain level of attrition as opposed to assuming that most hires will work out and deciding which ones won't on a case-by-case basis.
• In Capital Gains this week, we look at what to do when an asset's price follows a self-referential feedback loop, with the fascinating case of MicroStrategy (disclosure: I'm short a little bit) as an example. Turning an overpriced stock into a narrative that drives that price even higher is quite an accomplishment, though not one that produces good long-term outcomes.

Books

Buffett's Early Investments: A new investigation into the decades when Warren Buffett earned his best returns: In dollar terms, most of Warren Buffett's wealth creation happened late in his career. When Berkshire was managing hundreds of billions of dollars and beating the market by a few points, with favorable leverage, it really added up. But Buffett has periodically mused that 1) his best returns in percentage terms happened in the 1950s, when he was running a smaller amount of capital and could find correspondingly more underpriced opportunities, and 2) that such opportunities still exist. This book aims to document some of Buffett's earliest bets, including a mix of winners and losers as well as a mix of by-the-numbers deals and more qualitative ones.

It avoids some of the case studies you've heard about (there are probably more people familiar with the balance sheets of Sanborn Maps in the 50s and Dempster Mill in the 60s than of the median company of a similar inflation-adjusted market cap today, because both of them are straightforward studies of Buffett's early strategies, and have been documented in detail in multiple biographies). There's comparatively less material out there on Buffett's investment in Grief Bros. Cooperage or Philadelphia and Reading. And every time a biographer chooses which case studies to include, they're trying to tie them into a narrative: if the point of your chapter is "Buffett started out looking for balance sheet bargains," you write about the company he bought for less than its net cash on hand, not a growthier name.

So this book helps complicate some stories in useful ways, and stands as a reminder that some of the fundamentals haven't changed even if the specifics have. For example, for most of his time as a publicly prominent investor, Buffett has been good about avoiding major conflicts of interest: his net worth is in Berkshire, and he likes to keep managers aligned (David Sokol resigned in 2011 after suggesting the acquisition of Lubrizol, a company whose shares he'd bought in the days before making the pitch). Back in the 1960s, though, he did a similar maneuver: Philadelphia and Reading had turned into a small conglomerate, and he pitched one of his partnership investments as an acquisition target to them, which they accepted. This could be entirely aboveboard: he found an underpriced company, invested in it, and then suggested that someone else he knew acquire the whole thing. But it's a story that makes it hard to avoid at least the appearance of impropriety. (For the avoidance of doubt: given Buffett's overall track record, I tend to believe that that's what happened, and that Buffett got a good deal both for his partners and for P&R.)

And he was more tolerant of that kind of behavior at investees, too: in his famous investment into Disney, one of the risks that doesn't get much play in other case studies I've read is that Walt Disney (the guy) had an affiliated company whose revenue included money licensing Walt Disney (the brand name) to Walt Disney (the publicly-traded company). This is the kind of arrangement that can represent an indefinite drain on a company's results; if they have a key employee who's also a financial counterparty, it's hard for shareholders to know how much upside they'll retain. But this was, as it turns out, the kind of risk that can be handicapped and incorporated into an overall valuation. So if you've ever felt bad about investing in a good company with shady management, know that you're in very good company.

One of the better stories in the book illustrates how little has changed in stock picking over the decades. If you were to imagine the prototypical great pitch at a modern hedge fund, it might look like this: "We've found a company that's best known for its core business, which has faced structural challenges. They don't break out as much information as we'd like on their subsidiaries, but we found an alt data tool that tracks one of those subsidiaries, and it's clearly a rocket ship. We think earnings will inflect positive as that continues to grow, and we're baking in some multiple expansion as the story gets out." And that's exactly what Buffett found with Studebaker: their automotive business wasn't great, but one of their subsidiaries made an increasingly popular fuel additive. Buffett knew what the main component was, knew which company sold it, and knew where it was being made—so he spent time in Kansas City counting Union Carbide rail cars, and used that to validate that the fuel additive business was growing fast.

This is probably not the first Buffett book anyone should pick up, but it's useful nuance and provides some case studies that I hadn't seen before. Knowing more about someone's process is useful whether or not you copy it: some of the value is from expanding the available stock of analogies, and some of it is from recognizing that the effort and risk involved in this business is only worth if if you're extremely motivated by money or if counting rail cars in Kansas is your idea of a fun vacation.

Open Thread

• Drop in any links or comments of interest to Diff readers.
• What’s everyone’s macro outlook these days? There’s a weird combination of broad optimism and narrow uncertainty—geopolitics in general and tariffs in particular are unknown, and equity markets are pretty concentrated. But arguably they’re less thematically concentrated than at other peaks (S&P earnings were a lot more dependent on the financial sector in 2007 than they are on AI today, but those businesses get different multiples). 

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• A growing pod at a multi-manager platform is looking for new quantitative researchers, no prior finance experience necessary, 250k+ (NYC)
• A well funded startup founded by two SpaceX engineers that’s building the software stack for hardware companies is looking for a staff product manager with 5+ years experience building and managing data-intensive products. (LA, Hybrid)
• Ex-Ramp founder and team are hiring a high energy, junior full-stack engineer to help build the automation layer for the US healthcare payor-provider eco-system. (NYC)
• An AI startup building tools to help automate compliance for companies in highly regulated industries is looking for a director of information security and compliance with 5+ years of info sec related experience at a software company. Experience with HIPAA, FedRAMP a plus. (NYC)
• YC-backed, post-revenue AI company that’s turning body cam footage into complete police reports is looking for a senior founding engineer/tech lead who can build scalable backend systems and maintain best practices for the engineering org. (SF)

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.