If you're underwriting a borrower and you can only use one number to do so, a pretty good one to choose is their net worth, ideally calculated by estimating their assets plus the net present value of their future income (less that part that comes from running down the value of those assets). For some corporate borrowers, that asset base is sizable, tangible and has a readily-determined value, but for most of them, the majority of their worth is the present value of future cash flows. Some companies can borrow at lower rates than implied by their credit ratings for this reason—the ratings are more backwards-looking, as is proper protocol for a lender concerned about downside, but if cash flows are expected to grow over time, then a given debt keeps getting easier to service or refinance.

Life would be easier if all distributions of company outcomes were the same, but, of course, they're not. Some businesses mean-revert, like cyclicals. Some grow for a while and then level off. Some grow for a while because they're repeatedly taking risks and getting lucky—not just investment businesses, but any company whose growth model is to exploit pricing power faster than they grow it; a dialysis business or a local monopoly has a return profile that looks like the house's returns from roulette: lots of small wins, and the occasional big loss when either regulators or competitors get serious.¹

There's a profile that fits a few different trillion-dollar companies right now, which is that they're losing money, growing fast, and offsetting increases in gross margin by adding more operating expenses. The big AI labs fit this description perfectly, and for companies that predate the AI boom but are leaning into it, a growing share of their market value is represented by this bet. If there were one profit-maximizing company doing this, they might eventually reach the point where they decided that the risk-adjusted return on training another model was not worth it, and then happily collect profits from an AI monopoly. But there are several of them, which means that at any given time, whoever has the best model also has a model that's rapidly depreciating and needs to be replaced by an even better one.

An equity investor can look at this situation and say that it's just a matter of bet sizing: a company that can go to zero has a fair value greater than zero if there's some chance that it doesn't, and the higher the odds of a wipeout, the lower the amount of risk to take on that specific name. A lender needs to be more cautious, because they can definitely find themselves in a situation where they lend the company money for five years, its valuation doubles annually for the first two, and then it goes to zero and they're wiped out.

But this doesn't make lending impossible. What it does, instead, is to move loans around in the supply chain and the capital stack. The ideal collateral is some long-term commitment by a company that has some core business that produces stable cash flows, but that also either uses or resells AI. In this case, a lender to the neocloud counterparty to that contract is basically underwriting a mix of how well the neocloud can execute, the residual value of its chips after the contract is done, and maybe a few basis points to cover the risk that the ultimate customer defaults.² (Big Tech borrows cheaply, but not at a price implying zero credit risk.)

As the labs mature, and as equity investors demand not just a hypothetical future path to immense profits but some signs that they are actually heading in that direction, the labs themselves will get more creditworthy. Strategically, OpenAI and Anthropic want to be bad credits, in the sense that they can raise more money from equity investors with a triple-or-nothing than they can from credit investors by getting free cash flow positive. But the market isn't infinitely patient, and as the funding labs need rises, they tap out the sources of capital looking for crazy risk and have to tap into the sources that are looking to add another large-cap growth name to their portfolio.

But by that time, the other problem will be that lenders are all stuck making the same bet. It's a problem they tend to have: whenever there's a capital-intensive buildout that's funded partly by credit, it makes credit more and more of a bet on that sector. Sometimes, this is isolated—there was a point in the 2010s when "distressed credit" basically meant "frackers"—but railroads at the turn of the 20th century were about three quarters of the non-financial corporate bond market, and utilities were 38% by 1929. More recently, telcos were 40% of high-yield bond issuance in 2000. This mix of dates and industries is the kind of thing that makes people nervous, though it's worth noting that the prior years would also have had similarly scary record-breaking numbers, and that it was a career-limiting move just to sell because of that. But it also illustrates that industries tend to use a lot of credit when they're growing, and to need less once they're big and well-established. A growing industry uses credit because a mixed capital structure is more attractive in the aggregate than pure equity, and more feasible than pure debt. But when a mature industry levers up, it's because they're converting higher confidence in future returns into cash they can use to buy back stock.

And this process is somewhat self-fulfilling. Credit managers no doubt get very nervous when every new stress-test of their portfolio attributes higher losses to an AI bust, especially if they designed those stress-tests some time in 2022 after losing money lending to software buyouts. So they'll make it relatively more expensive to grow pure AI businesses, but also relatively cheaper to grow the complements—if you're nervous that too much capital is pouring into an industry, then lending to that industry's customers is an even better idea; an AI bust means cheap inference for AI users.

This boom has produced plenty of n-of-1 financial structures, and probably will produce more. And it's incredibly rare for a massive investment boom coupled with novel financial structures not to lead to a disaster somewhere along the line. That's compatible with the services being useful—the Internet was a much more useful tool in 2002, when Internet valuations had been wiped out, than it was in 1998, when they were soaring. But markets are also somewhat self-correcting, in that the financeable parts of the AI ecosystem are the ones that generate recurring revenue with positive contribution margins, whereas the more speculative bits tend to be financed by equity investors who know what risks they're taking.

We've covered capital-intensive buildouts over the years in The Diff, with the current one getting plenty of attention. Some related pieces:

• A few weeks after the release of ChatGPT, we covered the question of whether it's an asset-light layer on other businesses, or a capital-intensive and potentially cyclical business. One of those definitely didn't happen, but the cycle question is still up in the air.

• Meta (disclosure: long) is a good example of a company that tamped down aggressive spending plans in the face of investor skepticism ($).

• More on neoclouds slicing up cash flows to find the leverable piece.

• Even if you're lending to one specific AI company, if the loan is backed by AI infrastructure, it's backed by industrywide demand ($).

Blu Dot surpasses 2,000% ROAS with self-serve CTV ads

Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:

After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.

The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.

“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”

Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.

Learn More

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

In the 1930s, two political details about the intersection of monetary policy and other kinds of policy were that China was an unstable country with a silver-backed currency and that in the US, the Senate overrepresented the interests of states with big silver mining interests, which made them a valuable constituency that, for historically contingent reasons, aligned with Democrats. One of those reasons was that gold was historically associated with lower inflation, and inflation was a polarized issue when one party had relatively more farmers (selling crops at the market price to pay fixed mortgages) and the other more holders of fixed income. So, a series of decisions by the treasury and Congress pushed silver prices up from the global average of $0.42/oz in 1933 to $0.7757 by 1935.¹ All this activity had a few effects, including making Nevada an easier state to win in 1936 and pushing China into a catastrophic deflationary depression which did not exactly make it easy to fight a civil war and an invasion by Japan.

There are times when countries' exports are more abstract than the goods and services they sell, and really refer to the phenomena that they spread. Many decades later, in the 90s and especially the early 2000s, a reasonably accurate narrative about China was that they were exporting deflation to the developed world: China had a policy mix that made it a great place to add manufacturing capacity, which companies both foreign and domestic did with gusto. They were generally importing the same kinds of capital goods that would appear in factories elsewhere, but staffing them with workers who earned a fraction of the compensation they would in Germany or the US. This was a GDP-weighted view of the world; weighted by population, China was exporting inflation, in raw materials—as the country grew, their demand for copper, iron, fertilizer, and agricultural products did, too.² This was a strange and unsustainable squeeze, where some places saw rising prices because China was stimulating demand, while richer places saw relatively static prices because cheap Chinese exports were helping offset the impact of demand on consumer prices.

At the same time, China was importing a different abstract force: the more their economy was tied to multinationals, the more their businesses had to operate in a way that was compatible with how those multinationals behaved. It's not a requirement that two countries have identical legal systems in order to do business, but it is pretty hard for them to work together if concepts that are standard in one country's system are impossible in another's. This led to some weirdness; the US has ample experience with complicated company structures, where one company encompasses multiple layers of other companies, and even strange workarounds like having the ETF that represents a benchmark for equities last only a certain amount of time past the lifespans of eleven "SPY Kids". They took this concept a lot further with variable interest entities, a complex and rickety structure used to get around Chinese rules against foreign equity ownership, which was only shrugged at rather than endorsed as late as 2023. So one thing the US exported was that if capital markets want access to a particular theme, it will be conjured into existence, even if that means starting new companies or finding ways to circumvent the rules.

But that was exported along with norms about contracts, which seems to have stuck. Globalization makes rule of law contagious, because every contract with an importer or exporter is exposed to all of their contractual relationships with everyone they depend on.

The US has another peculiar role in the abstract-exports business. On one hand, the tech sector exports deflation; the hardware gets cheaper, and software comes in two varieties—the zero marginal cost one, or the one that competes with big chunks of human labor and declines in price by almost half every quarter. On the other hand, the US exports demand, particularly dollar-denominated demand, at times when the world economy needs it. When the dollar (another big and basically abstract American export) is scarce, everyone else in the world has a harder time servicing debt in dollar terms, and the two basic channels they have are selling America goods or letting the US government run up a bigger tab. The US, today, is unique in the extent to which its policies determine what imbalances will show up in the global economy and roughly how they'll be addressed.

There are a few big countries that can do this kind of abstract-exporting globally, and some smaller ones that can do so within an economic bloc. Some of these are temporary—you can only export so much deflation before there's nobody left to buy and the factories are all idle. But some of them seem like they're only going to accelerate: Americans are the Internet's inadvertently conspicuous tourists, always assuming things work like they do at home, because America is the center of gravity for so many Internet phenomena, which means it's overrepresented in the training data for LLMs, which will thus tend to have a more American perspective. We're already seeing bits of this in political speeches, but it will probably show up everywhere. And they'll probably stick around—America has the world's highest cumulative GDP by a long shot, so the American way has had more experience at higher stakes across many domains—it's not the only country that can produce best practices in these domains, just one that punches above its weight. So the next phase is that in addition to the US exporting higher real rates by having a fast-growing economy, the US is exporting baseline assumptions by virtue of producing a disproportionate share of training tokens and then making decisions about how the resulting models behave.

These implicit exports show up in many macro-flavored pieces in The Diff:

• During the pandemic, we looked at the US's status as the world's consumer of last resort ($).

• The long deflation in manufactured goods prices made it hard to measure inflation ($).

• Decentralized software projects are a good example of this dynamic, because they rely on extreme centralization for a few things.

• We've looked at how different kinds of labor get shifted to different geographies ($).

• America is also a big importer, with an edge in talent and capital ($).

Some teams never seem to stop moving. They're on Attio, the agentic CRM.

Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.

With Attio, you’ll get:

• Leads automatically prioritised and routed to the right rep

• Expansion and risk signals caught the moment they land

• Follow-ups written in your voice, already there when you arrive

Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?

Get started now

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

The history of accounting is a ceaseless effort to put raw numbers into some context that enables good decisions. The questions of when to recognize revenue, what counts as revenue, how to value various assets a company owns, etc., never get fully answered. But GAAP and IFRS do a pretty good job, and allow you to answer otherwise incoherent questions like "is this bank a better bank than that airline is an airline?" That could be like asking whether Taylor Swift is better at music than John Roberts is at interpreting the Constitution, but if you develop sufficiently flexible, fungible measures of input and output, you can at least have a coherent argument about this.

But that same generality means that different companies will want to highlight different metrics, and slice them in different ways. Banks and airlines are both asset-heavy businesses that have to pay attention to how those assets are financed, but this plays out in completely different ways; banks have some semi-captive financing options and some backstops that aren't available to other industries, but face offsetting regulations that mostly keep them from abusing those. Airlines have expensive physical assets for which utilization is an important concern, but these assets stay on the books long enough that an airline's cash flow statement is a meaningful measure of operating performance, whereas a bank's will reflect other considerations.

So, companies will sometimes tell investors to pay attention to their particular technique for calculating how they're doing. There are, broadly, three cases of this:

1. Some metrics will be noisy over time in a way that mostly nets out. Quoting revenue growth ex-FX, for example, smooths out the impact of currency fluctuations. These do matter; if you earn money in a depreciating currency and pay expenses in an appreciating one, that gap is very real. But, unless your investors are professional FX traders, or you happen to have a Swiss company that earns all of its revenue in the form of long-term contracts denominated in Venezuelan Bolivars, it's more meaningful to present the FX-smoothed number and let investors decide whether or not they want to hedge it themselves.¹ This kind of reasoning also explains why it's legitimate for occasional acquirers to report a number that excludes one-time acquisition costs, or for a company to report an adjusted quarterly number that omits one-time severance costs from a layoff. In each of these cases, the number is both a steady-state metric and (conveniently!) a higher number than the actual.

2. Some metrics are noisy in a different way: companies can have a consistent margin structure with respect to the underlying economics of what they do, but book gross revenue in one place and net revenue in another depending on exactly how that revenue is earned. This can show up when an adtech company has owned and operated properties, and also runs ads on third-party sites (e.g, Google) where it's splitting the revenue with the other publisher but reporting all of the revenue and then reporting the revenue split as a cost. In a case like that, their third- vs first-party mix can make margins swing around in a way that doesn't reflect the fundamentals of the business. Fintech companies that report their cut as revenue for fiat currency transactions and the entire value of the transaction for crypto (e.g, Block) also fit into this category. In general, these companies will normalize revenue recognition and offer guidance on that number, rather than the less meaningful gross number.

3. Companies can also cynically adjust out all sorts of costs that are recurring for them but one-time for others, or that it's just conventional to ignore. (Or, in extreme cases, just make a guess about future savings and treat that as an EBITDA add-back.) The cost of equity compensation is a good example of this: to the extent that it's compensation for employees, it's a cost to the employer.² It's a non-cash cost, and it will eventually show up in the form of dilution, or turn into a cash expense when the company buys back stock to offset this dilution, but it doesn't get any easier to analyze a business if they treat that cost as something they can adjust out of their profitability metrics. (The exception here is that lenders can exclude that cost—but they need to separately underwrite the risk that the stock declines enough that the company needs to switch back to cash compensation.) It's a non-cash cost, and it will eventually show up in the form of dilution, or turn into a cash expense when the company buys back stock to offset this dilution, but it doesn't get any easier to analyze a business if they treat that cost as something they can adjust out of their profitability metrics. (The exception here is that lenders can exclude that cost—but they need to separately underwrite the risk that the stock declines enough that the company needs to switch back to cash compensation.)

Internally, companies don't necessarily manage themselves to GAAP metrics. A few of them got a little too good at doing this in the late 90s, and investors learned some valuable lessons. In the first category, where there are genuine one-off costs, they matter to companies that are thinking about their liquidity situation, and one-time costs like acquisition and integration ideally get baked into the initial price of the acquisition against which future internal rates of return will be benchmarked. Internally, what the company is more or less doing is amortizing that cost over the life of the acquisition, just in the indirect sense that if the business needs to earn its cost of capital, and acquisition costs are included in that capital, they raise the benchmark for what it needs to earn. FX is a case where, while it's possible for a company to lose money because of currency risk, that's usually tied to some other risk—the currencies that go on indefinite, unexpected slides that aren't fully reflected in local interest rates are typically currencies issued by countries that have other problems, such that the meaningful risk to think about is political risk (and up your discount rate accordingly) not one specific way it plays out.

Consistent tweaks where a company indexes itself to a nonstandard metric are a tricky case, because they can be the second explanation for non-GAAP metrics—i.e. a better reflection of economic fundamentals—or the third, a way to put the word "earnings" in front of a higher number. EBITDA, for example, is not a direct substitute for earnings. But it's a very handy tool if your analytical process is to first look at the state of the business, then ask where the company will invest its cash, and then look at its financial structure to see where those cash flows will accrue.

There can be some feedback between this reporting and the optimal capital structure for a business, but in an indirect way: in general, a dollar of EBITDA can support more equity value than debt (a 5:1 ratio of debt to EBITDA is pretty spicy; a 5:1 ratio of enterprise value to EBITDA for a company with zero net debt is pretty cheap). But if a company is trying to maximize its returns to shareholders, one thing it wants to do is create the kind of EBITDA it can borrow against, so if it's maintaining some safer ratio, like 3:1 debt/EBITDA, every dollar of incremental durable EBITDA means $3 returned to shareholders by means of borrowing to fund a buyback. This dynamic encourages companies to find and lever up steadier sources of income, but that's also basically the goal of a mature business: once the main speed limit on their growth is GDP, the next variable they can influence is predictability, and if a more predictable business enables higher leverage, then that predictability translates directly into capital returns and a higher share price.

In the end, the numbers have to tie out. Companies that report adjusted metrics also walk through the adjustments they made, and it's much easier than it used to be to grab the numbers, pick which adjustments you want to keep, and come up with your own personal adjusted-EBITDA metric. And, in the very long term, purely cash-based metrics have to reconcile to net income, other consolidated income, or cash flows from financing; however big the cumulative adjustments or the gap between them and the underlying is: every adjustment has to get bridged to the original number, and every gap between reported revenue or costs and the ultimately realized revenue or cost shows up as a writeup or writedown eventually. In the very long term, cash and GAAP accounting tie out when a company gets liquidated, its net worth goes to zero, and every dollar is accounted for. Custom metrics are just a way to better understand what happens in the interim.

The goal of any mix of custom metrics is for the company and investors to be able to look at numbers and know what it means for their business when they change. GAAP does a reasonable first-cut job at this, though it's weaker for companies that make investments in the form of R&D spend and that realize subscription or usage revenue over time. A company like that can have identical or better economics to another business whose upfront costs are capitalized and depreciated rather than expensed and whose revenue is more front-loaded. Investors are aware of that in principle, but "can have" is doing a lot of work, and sometimes the best way to turn that into "does have" is to walk through a well-chosen set of numbers.

Accounting is a boring subject, in the sense that if you have a reasonable understanding of it you'll spare yourself some unpleasant excitement. In The Diff, we've covered it from several angles:

• We've looked at John Malone's constellation of companies, which introduced the world to EBITDA ($).

• In our CoreWeave S-1 writeup ($), we looked at which of their EBITDA adjustments did and didn't make sense. (Disclosure: long CRWV as a trade—specifically the cleanest bet that "pacing the frontier" means "accelerating less than we've recently realized we could" rather than decelerating at all.)

• We considered another accounting question: how to handle cases where companies' biggest assets are intangible ($).

• And in this note, a brief piece on how OpenAI and Anthropic use different revenue accounting ($), which makes Anthropic's number look bigger. Incidentally, Anthropic recently got flack for reporting a gross margin number to investors that was more consistent with OpenAI's treatment ($, FT).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

One asset class that gets referenced from time to time in histories of quantitative finance is Japanese warrants. If you're reading about someone who was writing code in order to identify and exploit miscellaneous trading opportunities in the late 1980s and early 1990s, there's a very good chance that they traded these. When Ed Thorp shut down his fund, he switched to mostly seeding other funds but also trading Japanese warrants. D.E. Shaw traded them, and John Overdeck ran that business for them before co-founding Two Sigma. In the early 90s, Paloma was doing the trade, too, as we know thanks to later litigation. Ken Griffin listed it as one of Citadel's early moneymakers. It was a popular enough trade that some allocators heard about it and brought it in-house, so Harvard was in on it, too. The Japanese warrant trade even adds some color to a famous financial scandal; Barings, at the time the UK's oldest merchant bank, collapsed after a rogue trader named Nick Leeson hid enormous losses from speculating in derivatives in Asia. The usual question with rogue traders is: why didn't management notice that one of their traders was so suspiciously profitable, and part of the answer is that they were actually pretty used to the idea that people trading derivatives in Asia could make substantial low-risk profits.

So, in that corner of the world, everyone was doing the trade, and everyone knew it was lucrative.

So: why?

The basic story is that in the late 1980s, it got trendy in Japanese corporate finance circles to issue low-interest bonds with warrants attached. They were aware that their shares were expensive, but also aware that they could do a carry trade: issue dollar-denominated bonds with warrants, swap the dollars into yen, and use them for whatever the company needed capital for—real estate speculation, domestic financial engineering, maybe even running the business!

Part of the way the equity-linked securities business works is that common-sense intuitions and probabilistic ones collide; it might strike a CFO as intuitive that after the massive runup of the 1980s, their stock couldn't keep going up forever, and that seeing those warrants go in-the-money was a nice problem to have. What wasn't as intuitive, from the corporate finance side, was that it was possible to buy the warrants, short the stock to neutralize exposure to its movements, and then make money if the implied volatility was low. If there were lenders who wanted straight debt and would treat the warrant as something to be sold off to a specialized investor, and if the ultimate borrowers also weren't carefully valuing the warrants, that meant that there were two layers of fairly price-insensitive sellers.

And so, in stepped the buyers.

There are many times when there's a lot of uninformed supply or demand in some market, but for many trades that just raises the question of timing. If you'd looked at dot-com financials and Internet penetration in 1998, you could have correctly concluded that all the consumer-facing names were overpriced and that the B2B ones who had them as customers were both overpriced and over-earning. Over the next two years and change, the Nasdaq tripled. You can be right that buyers are overpaying and still underestimate how many more people there are who are willing to overpay, and how long they'll remain willing.

But if they're selling a derivative, what they're really selling is single-stock volatility, and that volatility can be isolated. If you put on the position, and more companies issue more underpriced warrants, that's only a problem insofar as your competitors sell positions similar to yours in order to buy the next one. But market inefficiency is just another way of saying that that doesn't happen, or at least doesn't happen fast.

The arbitrage persisted in part because of enormous supply, and in part because it was somewhat complicated: a typical trade would involve buying a dollar-denominated detached warrant in London, and shorting the corresponding yen-denominated equity in Tokyo. And the hours these markets were open didn't overlap.

Japanese stocks massively underperformed during the era of mass bonds-with-warrants sales. What the warrant sellers saw was that they'd sold the right to buy their shares at high prices, and the shares never attained those prices. Free money! But it's free money in the same sense that if you sell earthquake insurance and there isn't an earthquake that year, you made a profit; it's all a question of pricing the risk correctly. The quants executing this trade ended up being persistently short Japanese equities, which went down a lot, and long Japanese equity volatility, which tended to rise during market declines. They weren't making a purely directional trade, but those moves describe roughly how they got paid.

This trade went away as Japanese companies got less aggressive in capital markets generally. But the trade was big enough that it could feed plenty of quantitative strategies. There's a purely inductive version of systematic investing, where all you do is look for persistent patterns or build a derivatives valuation model that tells you what's mispriced and how to hedge out most of the risk. But those mispricings can only persist if they have a reason to exist in the first place; a better way to look at it is that systematic investing is the implementation of some high-level view about irrational market actors. It's a lot safer to have a solid why behind the statistical what, especially when the why might change.

These kinds of discrepancies still exist, generally at a smaller scale. Corporate borrowers have gotten more sophisticated, but the big difference is that capital has gotten more mobile. Zero-interest convertible bonds follow very similar logic to the Japanese bond-and-warrant product; they're a way to borrow cheaply, and then give lenders equity if and only if the stock goes up. And they tend to get bought almost entirely by specialists who are buying the convert, hedging out the stock price risk (and the credit risk for the debt piece) and capturing what's now a smaller spread. If convertible bond issuance goes up, pods will put more capital into the strategy, and funds that specialize exclusively in this trade will also raise more capital. So the same kind of trend happens, but the market absorbs it faster—and many of those pods will be descended, either as a corporate entity or based on who learned quantitative finance from whom—from some of those early Japanese warrant traders.

Quantitative investing is frequently part of the Diff beat because it represents the point at which repeated qualitative judgments can be turned into firmer rules. Some posts touching on this:

• Value investing is still real, but the systematic version is harder to execute ($).

• A look back at the quant quake of 2007.

• We built a market simulation in which different agents run different strategies, featuring factor crowding and more.

• When you're using a systematic approach, you often want to start with the premise that the market is efficient and then look for reasons it wouldn't be ($).

• Reflections on Renaissance Technologies.

The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

• 7 high-impact AI strategies to accelerate your marketing performance

• Practical use cases for content creation, lead gen, and personalization

• Expert insights into how top marketers are using AI today

• A framework to evaluate and implement AI tools efficiently

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Here are two quotes from major publications. First, from a Wired profile of AI detector Pangram:

And, from a few months ago, NYMag on gifted and talented education:

(Emphasis mine.)

First, two notes on the quotes: the reference to "Spero" is modified to explain who he is, because that's discussed earlier in the article. Which hardly needs to be said: obviously, text in brackets implies a modification of the quote by the person doing the quoting, and in this case it's trivial to see why a selection midway through the piece would reference Spero without saying who he was. For the second quote: there are cases where a long blockquote adds some context, but there's a particular section that the quoter wants to draw readers' attention to. But sometimes, quotes already include bold, italics, etc. So it's traditional for the person doing the quoting to say that they added the emphasis.

It's actually a little hard to see what emphasis is being added, but both of the ellipses are in bold. "Full professors" is, in the original, "full professors at research-intensive universities," and even more strikingly, the second ellipsis elides "distinguished judges and lawyers."

Now, it's perfectly fine to use ellipses to drop things that don't fit the flow of the conversation. If someone starts to tell a story, goes on an irrelevant tangent midway through, and then returns to the flow of the story, it's helpful to the reader to cut that, and to give an indication that something was cut. But in this case, both of the changes actually change the meaning of the sentence, and in the same direction! In both cases, the author is making it look like the study had broader criteria for success than it actually did, in service of making the argument that childhood precocity doesn't lead to later success. (It does, of course; even if we treated the quote as true, a whole lot less than 12.3% of the general population hits that level of eminence.¹ ) This is pretty egregious. Anyone who checks can see that the quote is being edited in a misleading way, which means that the author is basically taking advantage of every reader who doesn't treat the writer/reader relationship as an adversarial, con artist/mark sort of situation. You just don't have to do this kind of thing if you write things you think are true, and you can't pull it off if you're writing for an audience you respect. Who'd want to make a living doing otherwise?

Anyway, while the second quote is a lot more dishonest than the first, they're both doing something similar, which ties into the nature of the quotation. Recording devices date back to the late 19th century, and ones that are compact enough to be a practical way to get verbatim quotes on the fly got popular in the 50s and 60s. Before that, we had shorthand (which has a long history; Plutarch says Cato trained scribes in some form of it to record the proceedings of the Roman Senate). But quotations appear in older literature than that. Thucydides even says he doesn't have verbatim texts and is doing his best to get the original point across. So, quotations have existed as a form of literate culture, separate from but inspired by the oral culture that produced the original speech. If these are reproduced in some poetic form, you can simultaneously believe that Menelaus was a real historical figure and that he didn't "uh" and "umm" in flawless dactylic hexameter.

Once it was possible to record exactly what someone had said, this created a dilemma: when you write down what someone said, are you trying to make it sound like what they actually said, or trying to make it sound like a quote? The norm was to do the latter; when Joseph McCullaugh did the first interview with a President, Andrew Johnson, he didn't take any notes until he'd slept on it, and then presumably wrote a cleaned-up version of what he originally got.

If you can quote people verbatim or not, it offers a lot of room to choose how they'll be interpreted. Consider two answers to an interview that broaches the topic of why someone stopped working with a former creative partner. "We had creative differences and ultimately decided to pursue separate projects" is one way to put together a quote. Another might be something like this:

He paused for a few seconds. "We had some, uh, creative differences, and then went—and then decided to pursue, um," another pause, "separate projects."

The natural interpretation of that second one is closer to "We hate each other but have decided that airing one another's dirty laundry in public would be a bad idea." Whereas the first one is bland enough that it might mean that, but probably means roughly what it says. If people choose their words carefully when they think they're at risk of being misinterpreted, they'll have more pauses and false starts, which means that whoever's quoting them can choose how untrustworthy the interview subject sounds.

This is a strange example of path-dependence. In a world where recordings somehow predated the written word, we might have the expectation that all interviews would be perfectly verbatim, and that they'd be interspersed with narrative details. In that world, cleaned-up quotes would sound stiff and rehearsed. But in this world, they have continuity with the genre of reconstructed quotes, not verbatim transcripts. Other than "That's one small step for man, one giant leap for mankind," which makes a lot more sense if you say "for a man," we generally translate quotes into a literate rather than oral form.²

The way to rescue journalists' freedom to tell you whether someone scratched their nose or didn't make eye contact while saying something, to include or omit pauses, to transcribe every false start and minor grammatical error, is to treat the whole thing as a three-way relationship, between the speaker, the writer, and you, the reader. Everyone comes into this with their own biases and agenda, and a hostile writer is no more exceptional than a hostile reader. If you're capable of reading something specifically to find all the ways you'll disagree with it—if you hate-clicked on something that generated some hate-views which were monetized at some hatred-indifferent CPM, you are implicitly capable of hate-quoting someone by reproducing every single syllable they said and all of the actions you noticed them taking while they said it.

But that also means that clean, straightforward quoting is a literary device that omits some of the truth to get a more truthful output. There are judgment calls involved, but once you know what to look for, you know exactly what it looks like when a reporter is using something nominally factual to advance a particular narrative.

This phenomenon is part of the broader Diff beat on media, including:

• The AI trust paradox ($) argues that LLMs make it easier to lie convincingly, which means that it's a great time to demonstrate how trustworthy you are—the tools we can use to check will only get better.

• In credentialism and the money illusion, we consider whether credentials, brand names, and the like follow a credit cycle.

• AI creates a bull market in facts and a bear market in narratives.

• Video clips of fights and arrests have selection bias by default ($): people start filming after a conflict starts, which means that viewers fill in their assumptions about who is to blame.

• Writers with a large corpus are vulnerable to hostile, out-of-context quotes ($).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Finance culture is full of allegories, like that one Star Trek episode. It's a useful kind of compression: if someone tells you about how they have a big short position in a melting ice cube company, the first sentence you say to them is likely to include the word "GameStop." The way these stories spread is, typically, from more experienced people to more junior people. If you're a few years into your career, talking to someone who's a few months in, there are things you remember vividly that they'll only know about if they started reading the Wall Street Journal in middle school, and these stories are probably new to them. Or, put another way: these stories are probably baked into the worldview of whoever's on the other side of the trade.¹ There's a famous one that was later outdone by different events, but which still has some useful lessons today.

Imagine trust in American institutions is at an all-time low. Average people are worried that the good jobs have all disappeared, and that if they make any money, inflation will just eat it. This creates a kind of financial nihilism, getting people excited about things like newly-legal forms of gambling, or participating in short squeezes.

It's the summer of 1978, and the short squeeze and the newly-legal gambling are part of the same phenomenon: when Atlantic City legalized gambling, the first company to build a casino there was Resorts International.

Resorts International ticked a lot of the right boxes for a compelling short position: a recent industry pivot (they were previously known as the Mary Carter Paint Company, but were now under new management), a hot sector, questions about their financing, including mysterious loans, a penny-stock promoter whose business associates kept getting murdered, another executive who had pleaded guilty to paying bribes in the form of cash and prostitutes, and occasional dealings with Meyer Lansky. One possible explanation for all of that oddness came from Rolling Stone, which in a lengthy article about ties between American anti-communists and organized crime, identified the Mary Carter Paint Company as a CIA front that had helped launder money for the Bay of Pigs invasion, though Rolling Stone later retracted that.

There was, as people sometimes say, some hair on the deal.

On the other hand, investors had seen what had happened in Vegas: between Howard Hughes buying up real estate, the jet making Vegas more accessible, and the air conditioner making it more tolerable. What would happen, they might wonder, if there were a version of Vegas that was a three-hour drive from New York City?

What happened to the stock was that Resorts' class A shares were around $8.75 in 1977, and hit $35.25 in May 1978. They had a second class of less liquid stock (shorts might wonder what sort of paint-turned-casino company needs dual-class stock, anyway), which roughly quadrupled, leading the SEC to suspend trading in the B shares for ten days.

Around this time, the manager of a hedge fund turned family office, Robert Wilson, started shorting the stock. Wilson's an interesting character: he had a high tolerance for risk, loved shorting garbage, was happy to lose money on shorts for a while, and liked to stay very plugged in to what analysts were saying. He was a long-biased investor, but since he used leverage and liked growth stocks, he did a lot of shorting, too.² Being well-connected works both ways; institutional equity traders knew who Wilson was, and knew which trades he had on.

In the Spring of 1978, Wilson started on a six-month vacation around the world (and, apparently, did not follow the tradition of grossing down the spiciest trades a little before heading out of the office). The casino opened in late May, and was a huge hit. By June, shares were over $70, and by early September they'd hit $120, and there was a lively bull market in casino stocks generally. It got to the point that the American Stock Exchange raised margin requirements to 75% for the A shares and 100% for the B shares.³ As Wilson himself knew, that wouldn't necessarily eliminate levered speculation: early in his career, when he wanted more margin than his brokers had allowed, he'd borrowed from a Swiss bank to finance more trades. Wilson himself was being chased around the world by margin clerks; every time he'd check into a new hotel, he'd get a call from his broker, initially asking him if he wanted to cover and eventually telling him that if he didn't, they'd do it for him. In mid-September, Resorts hit $190, after management estimated that they'd earn $15/share that year despite having only opened the casino in May. At the time, the S&P traded at around 9x earnings, so this was still a premium valuation on a trailing basis, albeit a discount if you rolled those earnings expectations forward and ignored the prospect of competition, as many buyers did.

And, at around $190, Wilson finally told his broker to go ahead and cover it. (Or rather: he told them that at $130, but they didn't get his message, which he only discovered when they called him back to tell him it was trading at $190.) He'd started with a $50m net worth and a 3% short position in Resorts; he ended up losing $20m overall on the trade.

This trade was a big deal, and for a while was part of the finance anecdote canon: in Diary of a Hedgehog, Barton Biggs tells a very thinly-disguised version of it (this book review has an answer key for all the other slightly anonymized stories in the book). And he wasn't the only famous trader involved. Here's a quick excerpt from The Predators' Ball:

Interesting! That office opened on July 3, 1978. Resorts had issued some bonds and warrants, though that source doesn't say that they were attached. Ed Thorp has a story about owning some Resorts International warrants, and, during that run-up, getting increasingly desperate calls from traders who wanted to buy them—and Thorp was a big Milken client.⁴

There have been bigger short squeezes since, like Volkswagen and GameStop. But this one was a classic slow-feedback summer crash: it happened for lots of idiosyncratic reasons, but one of them was that two of the biggest participants in the short side of the trade happened to be hard to reach right when things got exciting.

Short squeezes are a fun combination of personal drama and market plumbing. And it's been an interesting few years for short sellers, as they've gotten better at using the media but have simultaneously been threatened by unruly social media. So there's been a lot of material, including:

• We've looked at the recent rise of gambling and prediction markets ($).

• A review of The Trolls of Wall Street helps explain where the meme stock phenomenon came from ($).

• Meme stocks make it harder to run a balanced long/short strategy ($): you need more ideas, but these will be lower-conviction ideas—and people might panic-cover faster if five of them are each short 1% of the company with low conviction rather than one of them being short 5% of it and knowing the story inside and out.

• The rise of GameStop shares is still somewhat mysterious ($).

• Resorts International is a good illustration of the general rule that it's hard to make money shorting worthless stocks ($).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

The bull market of the 1920s was one of the earliest booms that could properly be described as a technology bubble. There were earlier cases where investors got excited about specific technologies; British investors were briefly enthralled with bicycle stocks in the 1890s. But typically, big movements of capital in the past were driven by some mix of technology deployment and regulatory favor: railroads were intimately tied to government policy, as were canals before that, and the South Sea Bubble was driven by some clever financial engineering done by what was basically a government-owned entity.¹

In the 20s, investors were optimistic about electrification and autos, which meant that in addition to having lots of consumer applications to bet on (lights, appliances, radios, cars) and a general groundswell in productivity-enhancing growth: electrification made factories more efficient, and drove a capex cycle as new factories got built and new equipment got installed. Trucks as last-mile delivery were complements to railroads' and ships' cheaper long-distance bulk shipping. Meanwhile, internal combustion engines for agricultural equipment and widespread ammonia fertilizer were adding food supply and freeing up a lot of labor supply, so those new factories could find plenty of workers—in the 1920 census, for the first time, the US had a larger urban than rural population. (The internal combustion engine also destroyed some demand for agricultural outputs—horses need lots of calories!)

There were political and geopolitical elements to this boom: Europe had recently destroyed itself, which meant that the US had both a lot of physical capital and a lot of financial capital. Republicans were in control of the House, Senate, and Presidency over this period, and one of them, Calvin Coolidge, had first achieved national prominence as governor of Massachusetts, when he spoke out against a police strike. ("There is no right to strike against the public safety by anybody, anywhere, any time.") So, investors did not have to be worried that returns from this investment would be captured by labor.

So, for investors who cared about revenue, costs, capital expenditures, and other concrete measures of company performance, it was a good time. But, as the cycle went on, it was an even better time for people who liked financial engineering. Closed-end funds got popular in this period; they'd lever up to buy hot stocks, and then trade at a premium to their net asset value. And some of the stocks they bought were holding companies that themselves borrowed against their stakes in subsidiaries and minority investments. The Van Sweringen Brothers, for example, bought control of the Nickel Plate railroad with 17:1 leverage.

But that's just a classic story of leveraged excess, with the usual mechanism behind it: lenders got a little sloppy, but that sloppiness meant more purchasing power for consumers and higher asset prices for investors, which kept defaults low.

The really interesting part was when companies started to cut out the middleman, and lend their cash on the "call money" market, i.e. for margin loans. They could get between 6% and 9% interest, and these were short-term loans, which gave them a kind of temporal seniority: they probably wouldn't be first in line if a borrower were completely wiped out and declared bankruptcy, but they could be first out the door by yanking their loans when asset prices declined. Meanwhile, though the returns on this money were lower than for most companies' core businesses, they were much more predictable. And, critically, the interest rate on call money exceeded the earnings yield on many companies' shares. So, a company that earned $1/share and traded at $25 could have paid out a dollar as a dividend, but if they retained it, and their business were otherwise unchanged, a year later their earnings might be $1.08 and their stock would be worth $27. But the buyer of that stock was, increasingly, financing it with margin loans, so they were paying 8% interest to buy something with a 4% earnings yield that achieved that yield partly by lending to them at 8%!

The immediate cost of all that circularity was just a little paperwork; money moved around, but nobody was better or worse off, at least for the moment. But it meant that a sufficiently fast drop in the stock market would actually make companies' earnings drop, immediately, and that a decline in speculation would also impair those earnings. Meanwhile, if call money was easier money than investing in the company's operations, the economic fundamentals underpinning the boom were also going to get weaker. This circularity made the system brittle in many ways.

It's surprisingly hard to come up with a precise cause for the Great Depression, because there were so many factors at once. For all those credit issues, the stock market just wasn't as economically central as it would later become, and the crash alone didn't cause an economic downturn—in retrospect, the recession actually predated the crash by a few months, starting in August 1929. There had been a few brief recessions earlier in the 1920s, but investors who'd bought the dip did well. And there was plenty of optimism in the immediate aftermath of the crash (partly because nobody who counted as an expert could afford to say that stocks were still expensive). The Smoot-Hawley tariffs in 1930 and the collapse of Bank of United States² in 1931 are also contenders for depression catalysts. Downturns are rarely monocausal, and typically market crashes don't have a distinct fundamental cause at all, or at least not one that can explain more than a small fraction of the decline. That's just how credit cycles work: for a while, everyone's better off, and it's optimal for them to be so; credit expansion is a channel for the economy to convert an increase in productive capacity into an immediate increase in consumption and investment (which can lead to more productive capacity long term). But credit growth tends to make things look a little too good, and when it's part of the feedback loop it encourages perverse decisions. An omniscient economic planner would have begged big companies not to lend money to speculators, but the closest thing we have to an omniscient planner is price signals, and while they're pretty good, they're never perfect.

It's too bad for market history that the crash of 1929 was such a spectacular episode of wealth destruction, because the 20s were a very interesting time, and of course some of the tricks invented then have been rediscovered since. The Diff has written about a few related phenomena:

• We've considered the quantitative and qualitative stories of easy money ($), in a piece that alludes to call money's role in the 20s.

• We've also looked at the South Sea Bubble in more detail ($).

• One of the useful functions of financial engineering is separating out different kinds of risk, which we see today in AI ($).

• Electrication is a useful analogy for cloud computing ($).

• We've looked at the 2010s as a version of the Great Depression with slightly higher growth ($).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

It's a common finance idiom to refer to a company's share price as management's "report card," and this is actually a useful concept. A perfectly rational student wants a report card that accurately assesses their skills, but given the choice between an accurate one and one with higher grades, almost everyone prefers the latter.¹ There are plenty of warnings about this, though usually in private businesses where there's more control—raising a round at a billion-dollar valuation is usually gratifying, but if the last round was at five billion then it's demoralizing. Public companies have a similar dynamic, though it tends to happen more continuously: it's just hard for people to stay excited about work when their own shares in the company are trading at a fraction of their previous value.

There are some investors who act on this directly; Jamie Dimon has explicitly talked about how he doesn't want to buy back stock at more than twice tangible book value, in 2013 Reed Hastings explicitly attributed some of Netflix's share price performance that year to "momentum investor-fueled euphoria”, Elon Musk has mused on Twitter about Tesla's stock price being too expensive and Buffett, too, has in the past talked about Berkshire trading at a price where he, personally, wouldn't be a buyer.

So, it does happen. But typically, CEOs keep thoughts like that to themselves. Or, if they express a view that their stock is too expensive, it takes the form of: either issuing more of it for cash or using it as currency to buy something that isn't so mispriced.² Sometimes, this is straightforward: if a company is in the investment phase, it's expected to raise capital, and investors can assume that doing so is a good sign. Investors in AI labs, for example, aren't put off by how much money these companies raise: training models is expensive, serving them requires even more hardware, and if demand is rising and they're positioned to take advantage of it, then investors' money is worth more when it's spent by OpenAI or Anthropic.

So, there goes one heuristic: a company that issues lots of equity is either doing so because it's overpriced, and sees issuing stock as accretive because it increases cash per share, or because it's underpriced, and the accretion comes from increasing future free cash flow per share instead.

On the M&A side, one thing overpriced companies often do is expand their definition of strategic. For example, when AOL merged with Time Warner in a 55/45 deal where Time Warner brought most of the EBITDA and AOL brought most of the hype. But there was a reasonable argument for synergies between them, if not for AOL's valuation: Time Warner had great content, but distribution was moving online; they also had a substantial cable operation, which was a nice complement to AOL-as-ISP. In a way, AOL's pitch was that it would help Time Warner realize synergies with itself: making sure that if people were reading Time (this deal took place so long ago that people regularly read Time) on their computers, they were perhaps watching trailers for Warner movies on those same devices as well.

AOL pulled it off! They took a business that was early but struggled to develop a competitive moat, and merged it with something more defensible. The resulting business wasn't great for shareholders, partly because it was hard to meld the companies together but also because the merger coincided with both the peak year for dot-com enthusiasm and the peak of print advertising revenue in real terms. So, not a great merger, but one that, from AOL's perspective, saved them from a worse outcome.

One of the more interesting cases of an acquisitive, overvalued company was Wirecard, which was overvalued because they were cooking the books. Wirecard had a problem common to frauds: over time, it would get harder and harder to explain the gap between their reported earnings (which they were fudging) and their cash balances (easier to audit). Their solution was to do fake acquisitions: the money they claimed to spend acquiring smaller global payments companies was really just an excuse for why that money wasn't accumulating as cash on hand, or turning into dividends and buybacks. But in Wirecard's case, the model was that they used fake acquisitions to explain why their actual cash generation was so low. (They also had some fake balances, and when €1.9bn went missing, the company finally fell apart.)

Diginex is a recent and entertaining example; it's one of many Asia-based microcaps that went public on US exchanges in the last few years despite dubious profitability and growth. For a while, Diginex was trading at 450x sales, despite anemic growth record and persistent losses. One of the drivers of that levitation act was that the borrow cost on Interactive Brokers was regularly over 900%; even if you were right that it was going to go down, you needed it to go down fast, and every time a short seller capitulated, it put more upward pressure on the stock. They decided to do the same exit-by-merger approach that others have been tempted by; when they initially made the deal, they were paying $1.5bn worth of Diginex stock. Four months later, they were issuing more stock, but valued at $1.05bn, and their target was now going to run the company. So, management was able to convert control of a company that had a high market cap but little economic value into a sliver of ownership in something more viable, over which they had minimal control.

Owning a sufficiently overpriced stock is a bit like having successfully pulled off a heist and stolen a famous piece of art. On paper, you're rich. But converting that wealth into the sort of paper you can exchange for goods and services is ruinously expensive. You even have limited use value; what good is it to have The Concert hanging in your living room if it means you can't have company over?

But that analogy illustrates part of what's going on. Sometimes, overly-promotional management is motivated by pure greed. But it's a misaligned kind of greed, seeking a high number in the short term rather than a durable stream of consumption. They may be motivated more by gratification: if you really want to have run a company that was worth billions of dollars, and you can't plausibly do that, you have a better shot at faking it, at least for a while.

We've looked at the uses and abuses of overvalued stock a few times in The Diff, including:

• A look back at the e-commerce rollup business ($), which was partly a way to take advantage of the valuation gap between single-product companies and diversified ones.

• Increasingly, public figures can monetize their fame by creating and exploiting volatility ($).

• Sometimes, valuations are set by people who want to say they invested in something, not people who think it's a good investment.

• Teledyne ($) took advantage of its own equity's mispricing in both directions.

• And a look back at a German cloud computing company that was better at making aggressive forward projections than about keeping up-to-date on its financials ($)

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Up until a few years ago, you could recognize a semi-personalized, yet mass-mailed cold email because it included details about you that could, in theory, apply to hundreds or thousands of other people: "like other graduates from X..." "Since you're one of the top Y..." "As a Z..." etc. Get enough of these emails and you can spot the mad-lib portion and learn to ignore them. But in the last year or two, this has flipped around, and now the details are more specific, but almost any email you get from a stranger that mentions a few facts about you in the first line is probably LLM-produced spam.

There's a general pattern in media where lowering the barriers to entry (lowering cost of production or distribution) means lowering the average quality, but potentially raising the peak. Pre-printing press, the high cost of the written word meant that the works most likely to be published were things like the government records, official pronouncements, devotional texts, and the bible. A lower cost for printing meant that there was room to publish many more bibles in many more languages, persuasive essays on why those monks busily illuminating manuscripts were living pretty well off of tithes, and popular entertainment. But it was still popular entertainment for a narrow slice of the populace; literacy rates in Northern Europe were 5-10% before the Gutenberg press, and were around 15-25% after a century, and while books got cheaper in this period, they were still something most people had to save up for: an English worker in 1600 was paying about a day's income for a Shakespeare quarto. But that example illustrates that there was a wider variety of written material available. Plays predated the printing press, but they were experienced live; that was the most cost-effective way to transmit the text.

As some of the examples above indicate, cheaper media led to some problems: anyone could say whatever they wanted, and if someone claimed the narrative high ground, their story might spread fast before someone else figured out that it was false. There was an actual literary genre called libels, which consisted of every defamatory thing the author had heard or chose to make up about some public figure. These benefited from a media dynamic that's familiar today: if something takes the form of a kind of content that used to be expensive to produce, it feels credible, even if it's fictional; a well-produced AI video's credibility is somewhat anchored to the historical cost of producing such a video, and an AI-generated nonsensical research paper is still a LaTeX-formatted PDF that implies some credibility: the medium is the message. The Alchian-Allen effect tells us that if there's a high fixed cost associated with something (such as producing a video), the marginal cost of making it better (through fact-checking, for example) is relatively low. But that means that if the cost of media production declines over time, we routinely overestimate how credible a given media source is. That applies even to the written word; if you were to get a column like this delivered to you a generation ago, it would likely have shown up in print, and overseen by an editor; the hiring process for someone to write it would have stricter filters, and the content would have to attract sufficiently valuable eyeballs that even 90s-era print ad targeting could make it profitable. Whereas today, in the time that it took you to read this paragraph you could have started a free account on a newsletter service and started typing (or copy and pasting from Claude) whatever it is you wanted.

In one sense, media economics is a big deal because media consumption takes up a substantial fraction of people's free time, and first- and secondhand media consumption determines what they think is going on in the world—unless you happen to be personally acquainted with Donald Trump, and the two of you are in the habit of talking about work regularly, your view of Trump's policies and their impacts either comes from news you've read, or from talking to people about news they've read. But media economics is actually becoming a bigger deal over time, because the information economy is in some sense a media-mediated economy; email is just a very niche digital magazine with a wildly fluctuating subscriber list and an unpredictable publication schedule.

We're seeing this now, as one of the fundamental tradeoffs in media—lower costs of production make verification more important, but the demand for verification shows up on a lag—is now the tradeoff that defines returns for AI, which itself is the majority of US GDP growth right now.¹ The verification problem takes a while because at first, demand is invisible: it's flattering to get cold emails, it's delightful when an LLM confirms that your writing is absolutely wonderful and the only edits are minor, and it's comforting that there's photographic or video evidence for whatever you happen to think is going on in the wider world. But in a way, all of these AI outputs should be interpreted, not as the media artifacts that they copy, but as prompts, which are shared with an artificial or natural intelligence that tries to figure out if they're real (and valuable), or if they're just a way to exploit certain unpatched bugs in our media consumption.

These bugs compound, and they create a very interesting dynamic in areas where there's lots of inefficiency but scarce context: a company that's working with messy data that isn't well-represented in the training set can treat that messiness as a liability, and focus their AI efforts on very nice-looking powerpoint presentations that smooth over whatever inevitable issues they have² , or they can look at that messiness as a valuable proprietary asset—as a captive source of unique data that's valuable to either AI labs or to anyone who wants to have a competitive moat while using fairly commoditized models. As with other cases where AI can be a substitute or complement to human effort, it's really a test of willpower. An undergraduate today is one of the luckiest people in history, in that they have four years with—outside of a handful of majors—minimal work requirements, plus access to artificial intelligence and living domain experts, and protection against the autodidact's curse of having grad student-level knowledge in some parts of their field and a high school understanding of others. Some students use this well, many more convert AI and the opportunity for learning into leisure time. Corporate America has plenty of organizations that will follow the path of the CheatGPTer, and a few that will eat them alive.

We've written a lot in The Diff about media economics, spam, and of course AI content generation:

• We've covered the media cycle by way of Buzzfeed, most recently here ($).

• Back in 2022, we looked at the generative AI landscape and asked what happens when the median audience size for a given piece of content is one.

• AI moves fast enough that this piece about AI spam is already obsolete in its particulars but still generally true ($).

• You can use AI to help with writing, but you need to be very judicious about using AI to actually do your writing.

• We'll spend a lot of tokens on responding to the token use of adversaries ($).

4x your communication output. Same quality. No burnout.

The bottleneck isn't what you want to say — it's how long it takes to type it. Wispr Flow removes the bottleneck.

Speak naturally and get polished, send-ready text for executive summaries, client updates, board recaps, investor notes, or just the 30 Slack messages you're behind on. Flow strips filler, formats numbers and lists, and preserves your tone.

Used by teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works in every app on Mac, Windows, and iPhone.

Start flowing free

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Sometimes, different parts of language evolve at different paces and we end up with strange archaisms embedded in regular speech. Nobody's surprised to hear an atheist end a phone call with "goodbye," even though it's a contraction of "God be with ye," and there's a pretty big gap in meaning between the word "housewife" and "hussy," even though the second is a contraction of the first.

"Hedge fund" was born as a slight contraction of the original term, "hedged fund." This drove A.W. Jones, who coined the term and invented the model, crazy—he grew up wealthy, went to Harvard, worked as a diplomat, got a PhD from Columbia, and worked as a journalist before founding his fund. And then he had to deal with the twin irritations that brutish traders were copying his model and butchering his grammar. Outrageous!

Jones' approach is surprisingly modern considering that it dates back to a time when the Dow was calculated six times a day by a 76-year-old using a mechanical calculator. (On one busy day, the ticker tape moved so fast he got a papercut.) First, Jones hedged against broad market moves, by going long and short. And second, he engineered a system that tried to get the best stock tips per dollar of commissions spent. He'd find good sell-side analysts, and give them bonuses based on how well their ideas helped him. At a time of scarce bandwidth, that meant they could give him insightful fundamental analysis like "I'm downgrading US Steel next week" and get away with it, though the market was inefficient enough that a straight-shooting analyst who actually did the research could also probably do well.

Jones popularized structuring an investment vehicle as a limited partnership that paid the manager a percentage of profits, set at 20% (Benjamin Graham had done this before, under various legal structures, and also charged a 20% performance fee).¹ Graham himself was a pretty erudite character; he, too, went to Columbia, and before he finished undergrad they were asking him whether he'd prefer to teach math, English, or philosophy; he needed money, so he went into finance instead.

So if you happened to be insanely ahead of the curve in asset management in the early 1950s, your stereotype of hedge funds was that they were a niche variety of risk-averse investment vehicle, run by someone who would throw in a quote from Horace while explaining the virtues of a particular wine.

But it's very hard to keep a lucrative compensation structure secret, and people who want to manage money already have a high assessment of their own abilities, so hedge funds started to proliferate in the 1960s, especially after Carol Loomis published a story about the Jones nobody keeps up with. Even by then, the model had proliferated, and, as one might expect during a roaring bull market, the part fund managers liked was that they could use leverage and earn 20% of the profits, and the part that struck them as a little old-fashioned was the bit where they'd hedge.

It's mostly forgotten today, but the hedge fund collapse of the late 60s and early 70s was brutal: there were around 150 funds managing a collective $1bn in 1969, and by 1971 there were around 30 left. This is around when people started asking why they're called "hedge funds" in the first place.

But as it turns out, it's a lot easier to justify taking a big cut of returns if you have some way to prove that you're responsible for those returns. If you're up 50% for the year, and the reasons for that are that 1) you bought Ling-Temco-Vought, Polaroid, and Xerox, just like all of your peers, and 2) you were levered 2:1, your investors will have some probing questions for you (at least when they go down).

And, in a way, that describes the modern evolution of the hedge fund model: over time, limited partners got better at asking tough questions about where returns came from before they turned negative and rendered the answer obvious. Investors in the aggregate relearn certain lessons about leverage, concentration, and crowding, but they tend to learn them in more elaborate ways. Within funds, there was internal pressure to benchmark returns against the easily-replicated parts of the strategy, not just in absolute terms, and one of the ways to avoid benchmarking problems is to keep portfolio managers from taking easy risks in the first place, or to charge them for it. (So, for example, a commodity strategy might make some of its money betting on the correlations between different commodities, and be allowed to lever that up a bit so long as a bullish-on-growth bet like long oil was offset by a bearish-on-growth bet like shorting copper. But if they had some specific reason to have an unhedged directional view, they could trade it, but not get the same cut of the upside.)

This coevolved with the process of giving investors easier access to those more passive bets. There are funds you can invest in if you want exposure to growth offset by some short positions in companies whose growth is slowing, or whose business is being disrupted by growth elsewhere, and if you want just the long side of that trade, something like ARKK is a cheaper way to get it. However, factors and correlations rear their head again. There have been many days where all the AI plays sold off, and software rallied, because long AI and short software were, at least earlier this year, the same bet expressed two different ways. Part of what hedge funds try to do is to figure out which of those correlations are some newly-discovered or newly-created but basically permanent feature of the market, and which are temporary. For a while, stocks exposed to streaming entertainment and stocks exposed to office real estate reliably moved in opposite directions, but that was mostly true in a post-outbreak, pre-vaccine world. But if you decided to bet that the "volatility smirk," the extra option premium you pay to buy insurance against big crashes as opposed to betting on big rallies, showed up after 1987 and never went away; market participants realized that there are more ways for the market to drop 20% in a day than to rally that much.

The hedge fund industry has been on a long odyssey, starting with the observation that people will pay a lot for an investment that goes up more than average and almost never goes down, then getting lost and bewitched by various appealing riffs on some but not all of that thesis, before finally converging on a model where different mixes of security selection, industry selection, and systematic factors all have some fair market value.

Hedge funds are a running theme in The Diff, both because they have a big impact on the market and because they're an interesting business in their own right. For example:

• Advanced Portfolio Management explains many otherwise mysterious traits of factor-neutral funds ($).

• We've looked at specialized funds as hedges rather than a pure source of returns.

• And here's a breakdown of Situational Awareness ($).

• During the GameStop excitement, we covered funds' role in market structure, and what changes when they pull back ($).

• And, for fun: why the strategies hedge funds pursue are almost perfectly designed to make people at funds sound very smart ($).

4x your communication output. Same quality. No burnout.

The bottleneck isn't what you want to say — it's how long it takes to type it. Wispr Flow removes the bottleneck.

Speak naturally and get polished, send-ready text for executive summaries, client updates, board recaps, investor notes, or just the 30 Slack messages you're behind on. Flow strips filler, formats numbers and lists, and preserves your tone.

Used by teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works in every app on Mac, Windows, and iPhone.

Start flowing free

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Last month, Cloudflare's CEO announced that bot traffic had overtaken human traffic. For the first time ever, most of the time a page gets accessed online, it isn't being seen by human eyes. Which feels disconcerting—are we already in some posthuman-by-a-small-margin future? By that metric, yes, but that metric also gets less meaningful as the number of bots rises.

An easy way to understand this is to look at spaces with a higher bot-to-body ratio, like equity trading or display ads. It's hard to get exact numbers, but if you think of all of the trades that are purely systematic, or index funds (basically a systematic strategy, though the algorithm is just "own the following 500 or so stocks in the following proportion"), or big trades systematically chopped into tiny, lower-market-impact lots, whose counterparties will be heavily automated market-makers. A single decision to buy 10,000 shares of some stock might produce dozens of 100- and 200-share lots, with most of that selling done by systems that are automated at the level of deciding what spread and size to quote. So, in one sense, that set of transactions is 90%+ automated. But it's intermediating between someone who made the discretionary decision to add 10,000 shares, the human software engineers who built the trading system, the traders who supervise it and tweak its parameters, etc. All that autonomous behavior is just intermediating between human agency, once the decisions those agents make are clear enough that bots can handle the details.

Online advertising, outside of walled gardens, has a similar dynamic.¹ If you see an ad on some news site, the publisher, contextual data provider, yield manager, multiple sell-side platforms, an ad server, a demand-side server, and a CDN for actually displaying the creative will all be involved. Plausibly, up to a dozen systems will automatically make a decision based on the decision to view a single page (though in some cases, that decision is basically IF user_views_page THEN show_ad). But, as before, there are many humans in the loop, just at different layers of abstraction. If the ads aren't showing up, the publisher has someone they can yell it; if they are showing up in inappropriate contexts (e.g, an ad for cheap flights running on a news story about a plane crash), the advertiser also has someone to talk to (though various elements of this moderation process are automated in some contexts as well).²

The share of bot actions to human actions is really a measure of at least three variables: the fraction of people who have adopted some kind of automation tool, the fraction of tasks that they can hand off to these tools, and how thorough the tools are at their jobs. A fairly simple decision, like "let's bump up our display budget by another 10%" or "I'm feeling nervous about Iran, better buy some index puts" will kick off a frenzy of bot activity, only a tiny fraction of which is directed manually by humans.

In the case of LLMs, a lot of what's happening is that reasoning models fan out very broadly (and sometimes seem to get caught in weird diversions—or take a couple tries before they find an un-paywalled version of the obscure PDF with the answer to your query). When they do this, part of what they're doing is substituting a relatively simple lookup for a more complicated inference task; if you ask a model how quickly a Model Y goes from zero to sixty, the model could look up detailed specifications of the car and then model the physics of acceleration, or it could find some brief description where someone else has done the relevant work. All this bot traffic is, relative to some adjacent possibilities, actually reducing datacenter demand.

It's strange that something that started as a default-human behavior—accessing some URL and interpreting the content there—is now automated. But in a sense, that's what's been happening to the economy for a very long time. The supply chain that keeps a good car in working order is vastly more complicated than the supply chain that keeps a horse functioning, as evidenced by various episodes in which steppe nomads conquered important places and held on to them for a while. But the effort required from an owner of a car is far lower. The same is true for computers (we're a long way from the "bad command or file name" days, and can now just blindly paste error messages into our chatbot of choice and navigate to a fix). Many complex, multi-step tasks get encapsulated in some simple kit that does the hard parts remotely and doesn't ask the user to do anything that can't be explained through pictographs.

Broadly speaking, this is a necessary and useful feature of modernity: specialized cognition gets done by specialists, and the average consumer gets a very simple experience by default and is opting in to anything that requires more than a middle school education. It actually creates a strange dynamic where the simpler an interaction with some product is, the more dizzying the supply chain behind it—what could be more natural than asking a question in natural language, and what consumer experience taps into as much cumulative capex as using an LLM chatbot for that?³ A rising ratio of bots to humans implies a field that's getting closer to being solved, but demand is unbounded, and a field that gets cheaply solved is one that absorbs a lot of incremental consumption. So, expect any ratio of this kind to rise over time, and expect it to be a disconcerting feeling when it does.

The Diff has looked at the layers of automation between human decisions and machine activity from a few angles:

• AI agents have a cost, which means they undergo something similar to natural selection ($).

• The efficient frontier of automation is hard to measure, because it's easiest to see right at the point where something is about to be fully automated.

• When automation makes work cheaper, the result can be much more work rather than less effort ($).

• AI can created 1.5-sided markets ($), where one side automates around the fact that the other is stuck in manual processes.

Expense receipts shouldn't require a search party

Adam spent 20 minutes looking for a $36 receipt. His finance team sent three Slack messages. Someone made a sticky note.

Ramp would have matched it automatically the moment he swiped. Auto-coded, in-policy, synced. Nobody had to ask Adam for anything.

This is what finance looks like when it runs itself.

Your team can be Adam. Or they can not be Adam.

Don’t be like Adam

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

In a broad sense, companies get paid because they provide something of value to their customers, but industries vary widely in how much of that value they can capture. The big capital-intensive transportation businesses are canonical examples: railroads are part of the formation narrative for more than one modern nation-state—building a transcontinental railroad was a condition for Nova Scotia and New Brunswick joining Canada at all, railroads tied northern German states' economies inextricably to Prussia's, and while the US was a coherent nation before the transcontinental railroad, building that marked the point at which internal back-and-forth migration was a more meaningful force than an indefinite expansion of the frontier. Airlines don't unite countries, but they do unite important metro areas, and families. It's hard for secondary financial centers to thrive without frequent travel to the main one, so air travel is part of what keeps Atlanta, Charlotte, and Houston in the mix. And containerization connects places with cheap labor (or, more recently, with dense accumulated capital) to places with lots of consumption. Containerization made fortunes in East Asian manufacturing and in American retail, but the people who owned the boats in between were stuck in a business with high and rising capital requirements—cost per ton-mile is inversely correlated with containers per ship—and cyclical demand. Fundamentally, a big role of the industry was to donate gigantic ships to Walton heirs.

It actually makes some sense that expensive generally-applicable technologies that tie into existing infrastructure would struggle to price anywhere near the consumer surplus they generate. The capital-intensity means that it's hard for one company to hard launch a monopoly: in the time it takes one of them to prove out its model, competitors can place orders for the same equipment and start their own routes. If it's a transportation tool, it's going to connect existing routes—railroads for any city pair not linked by water, airlines for any pair of locations where fast transportation was at a premium (in the very early days, the big customer was the US Post Office: early passenger airlines mostly flopped (if nobody can travel that fast yet, nobody expects them to), but mail already had competition from phones. And there's a lot of complementarity between high-latency/high-throughput communication and the opposite—it was cheap to continuously transmit stock quotes in what probably struck people in the 1920s as real-time. And that would be complementary with a prospectus or annual report that might take a day or two longer. And once there's enough regular traffic from mail, the marginal cost of having a few passenger seats is lower.

But all this while, the fixed cost of a plane is getting higher, and there's basically no one trying to seize the entire market. There's a similar story for containerization, where they needed ports to cooperate at both ends, which put a natural brake on growth, which meant that there would be multiple operators (the ports weren't going to do a big capital expenditure blitz that also made them dependent on a single supplier). And railroads were costly enough that they reshaped entire economies around extracting more coal and iron and turning them into steel.¹

Shipping will probably remain a tough industry for a while; as one cruise line executive said about his industry, the nice thing about these assets is that they have propellers, so if there's a shift in demand, you can just move them around. The downside to that is that they're operating in a global market, with a long lead time for adding supply. (In practical terms, there's a shorter one for removing supply, by scrapping ships. But if shipping companies are removing capacity faster than the ships naturally wear out, they're doing it under duress—that same global market dynamic means that if you have 2% market share, 98% of the pricing benefit from your scrapping a ship accrues to your competitors. And the more of them you scrap, the lower your share gets, and the more you're redistributing. So their incentive is to hold on, which, of course, means that the bottom of the cycle can be very low indeed.) But both railroads and airlines improved, albeit after a generation or two during which they provided investors with unusually poor returns. For railroads, they got their pricing power the old-fashioned way, by inheriting it from the robber barons of the previous century. Those industrialists had carved out some routes that no one could possibly replicate given modern land values and environmental rules, and the railroad route network at least roughly approximates where bulky goods might be shipped. Once they learned how to play up their advantage, by being stricter with their schedule so they could keep costs far lower than trucking, they could divide the world into companies that weren't served by the rail network, and would thus use trucks, and the ones that were on the network, where trucks couldn't really compete (or could only compete on time-sensitive backfill, not scheduled bulk shipments). Airlines were mostly fragmented and undercapitalized, though the industry also had some sloppy financial habits. And airlines still have a tendency to try to grow their way into lower fixed costs per seat-mile before they age their way into higher labor costs per seat-mile, with the best performers being the ones where both of those factors move at less of a sprint and more of a brisk trot, and at roughly the same pace.

AI is very competitive—companies aren't necessarily aiming for this, but they act as if they're competing to see who can get their product banned for being a bigger cyber risk—which recreates this dynamic. One reason they do is that American capital markets are so dynamic—a group of researchers who defect from a big lab can get whatever it is that they do next capitalized as being worth hundreds of millions to tens of billions of dollars. And this makes it very hard for any one company to run away with the market through sheer capital accumulation, though OpenAI tried its best for a while. And AI could have been a market where one well-capitalized winner got a big enough lead that they dominated a plurality of economically-valuable reinforcement learning loops and ran away with the entire market. Sometimes, industry fragmentation is intrinsic to the economics, but in this case it's because the researchers bought into scaling laws a few years faster than the capitalists did.

In The Diff, we've written about:

• Airlines.

• Railroads.

• Shipping containers.

• And AI from the perspective of financing ($), where it gets deployed ($), and why OpenAI's financing looked so weird, and how this affected them ($).

4x your communication output. Same quality. No burnout.

The bottleneck isn't what you want to say — it's how long it takes to type it. Wispr Flow removes the bottleneck.

Speak naturally and get polished, send-ready text for executive summaries, client updates, board recaps, investor notes, or just the 30 Slack messages you're behind on. Flow strips filler, formats numbers and lists, and preserves your tone.

Used by teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works in every app on Mac, Windows, and iPhone.

Start flowing free

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

ChatGPT, Claude, and Gemini tend to express more progressive political beliefs than their users. Grok leans a little bit to the right, but if you're using Grok as your chatbot of choice, there's a good chance that even though it's net conservative, you're more conservative than that. If you happen to be progressive, you don't have to think too hard about why aggregating all human knowledge and then creating smarter and smarter artificial intelligence would happen to agree with you: those AIs must be pretty smart to discover that you've been right about everything all along! But you might come to the troubling realization that at other points in human history, whoever had access to the most cutting-edge media distribution tools had the exact same experience: someone who dialed up for the first time in the late 90s could read Wired and learn all about how vaguely libertarian techno-optimism was what all the smart people believed; someone rich enough to own a radio in the mid-1920s would hear all about how marvellously President Coolidge was handling the economy, the very first message sent by telegraph not only assumed that the recipient believed in God but didn't bother specifying which one—a possibility that would be shocking to an infotech elite in 10th century Córdoba or Cairo. The saying goes that the media exist to afflict the comfortable and comfort the afflicted, but people only invent and repeat catchy phrases like this when the natural behavior is otherwise. And, as it turns out, advertisers pay generous CPMs to reach the comfortable, and the afflicted tend to close the tab when they hit a paywall.

LLMs are different. Users choose what to ask them: if you want to feed ChatGPT a few news stories from your local paper and rewrite them from an anarcho-syndicalist perspective, or in accordance with the tenets of the Bull Moose Party, you can do so. But if you ask them general political questions, they tend to lean left: a recent Washington Post analysis ($) (mercifully free of that "Here's how LLMs behave: we asked GPT-4o to..." thing) showed that every model tended to provide a left-leaning answer to open-ended political questions, with even Grok tilting that way slightly more often than the opposite, Gemini 3.1 Pro being mostly neutral but still a little left, and both OpenAI and DeepSeek delivering a larger number of progressive answers than neutral or conservative combined.

This is important. It's a big deal for end users if they're getting a systematically biased view of the world, especially if what they think they're getting is a neutral one. That’s useful even if they have an incorrect idea of what’s biased and what’s neutral. A doctor who gets accused of being fatphobic might well be right that a given patient would really benefit from some GLP-1 agonists. And understanding why LLMs behave this way actually tells us some interesting things about the world, and how relatively small differences in conservatives' and progressives' preferences and temperaments get magnified again and again, leading to the slant we see.

(I like to keep this publication as politically neutral as possible, with a focus on systems and implementation rather than on the outcomes being pursued. That doesn't mean I don't have political views, just that my business judgment tells me a partisan slant would make half of the readers slightly happy and the other half really mad. Aesthetically, I've found that people whose usual beat is not strictly politics get a lot less insightful when they expand to politics. And, as a matter of vanity: there are many commentators for whom it would be very fun to hear a debate between them-circa-June-2020 and them-circa-November-2024. So, this is more an attempt to understand modern political media and its feedback loop with LLMs than an effort to support one side, but either way will inevitably have some biases that readers ought to try to identify and take into account. If in doubt, ask Fable.)

Anyway, one simple but very popular hypothesis is that the reason models are biased is that conservatives are not as bright as progressives, so when you ask about abortion or affirmative action, the model is smart enough to give you the correct, progressive answer. Conservatism is broadly more comfortable with group differences being evidence that groups differ rather than evidence there's bias—a conservative can look at the gender pay gap and attribute it to men having a stronger preference for high incomes, or married couples rationally having one person specialize in earning and the other focus more on kids. This can be cast as a strict binary, but it can also be discussed as overlapping distributions; the existence of Gwynne Shotwell does not disprove the observation that people running big companies tend to be men. And that can be a valid conservative argument from the perspective of tradition and natural law, or biology, or even if we're all blank slates and gender roles are a pure social construct.¹ So, to the extent that there's an intelligence gap between those two sides, it's fortunate from a utilitarian perspective that the people disfavored by it are the ones who view such gaps as useful information about reality rather than some kind of social problem. Of course, if you make this generalization, some conservatives will take it personally. It takes a lot of effort to respond neutrally to negative generalizations about a group you're a part of. Everybody's got a plan to incorporate group differences into their worldview until they get punched in the ego.

And you can check: use education as a proxy for intelligence and you'll see that 5% of voters without a high-school education tend to have consistently liberal views, and is 12% for some college, 24% for college graduates, and 31% for people with postgraduate degrees. But, the more time you spend with people at the edge of that spectrum, the less likely you are to see educational attainment as a pure proxy for intelligence. There are, certainly, people who are not bright enough to hack it at a good school. And there are people who, if they were a little brighter, could have come up with better ways to spend four years than getting a PhD. But we don't have to give up! What if we just look at a meta-analysis of how intelligence correlates with expressing conservative or authoritarian attitudes, with an r of -.20? That looks a little weaker, but it's closer to answering the question we're asking.

You should probably put very low credibility on this kind of thing. It's just very hard to ask questions about worldviews without accidentally stacking the deck in favor of your own. And, by the same token, it's incredibly easy to engineer the results that you want if you deliberately formulate the questions that way. For example, if you ask "Should the government support hereditary titles that give a narrow set of lucky descendants financial benefits, such as exclusive rights to perform certain jobs for higher-than-market pay, or subsidies that allow them to continue residing at their ancestral estate," then almost everyone will say no, unless they're part of a conservative tradition that was already in decline by the time the US was founded. If you smirk and then reveal—surprise!—that the hereditary title is a membership in the International Longshoremen's Association and that ancestral estate is a rent-controlled apartment in Manhattan, they'll be annoyed with you. But if they're annoyed enough to keep talking, they will probably soften the argument, add some nuance, and explain why they feel there's a difference between the views they have and the ones you asked them about.² It's also hard to accurately categorize views. Kim Jong Il and Lee Kuan Yew were both on the authoritarian end of the spectrum relative to most national leaders in the late twentieth century, but they had very different ideologies, leading to very different outcomes.³ If you use preexisting data from a broad set of questions—if you check the General Social Survey—you find a verbal IQ difference of about a fifth of a standard deviation. In SAT score terms, this would be the difference between a 1030 (the national average) and a 1075, or between graduating from the average high school in your city and graduating from a 58th-percentile high school instead. If you were to compare two colleges that had those as the average SAT score for admittance, you would not be surprised if the alumni from the second went on to be a bit more successful, on average. You'd find it very peculiar indeed if the second school's alumni were overrepresented basically everywhere you looked, by an overwhelming margin. You'd think: there must be something else going on here.

There are many other things going on here.

The first selection effect to think about happens at the level of people. If you are, temperamentally, a conservative, there is a good chance that you will want to marry, have kids, and ensure that these kids are provided for both in terms of material comforts and in terms of quality caregiving, ideally from a parent. There are many more life paths compatible with a progressive outlook on life (including, of course, the family-and-job one above), but they probably lean more towards self-expression, making the world a better place, addressing broad social problems, etc. You might imagine some promising college senior who's been told by their teachers that they have an exceptional grasp of how taxes work, and that they'd be equally successful if they went to law school and specialized in tax law, or if they went into journalism and covered all the sneaky ways rich people and big companies minimize their tax burden. If that student has a pretty conservative outlook, and wants to have their 2.2 kids some time in their thirties, it's going to be pretty hard to do that earning journalist wages and paying a New York cost of living. If they choose that money-maximizing initial path, and move to the suburbs because they get plenty of noisy nightlife through the baby monitor, they're making a choice that's more compatible with the values conservatives have and advocate. They're also opting out of playing a big role in shaping people's values, by choosing what stories to cover and how to cover them. They'll contribute some useful tokens here and there, perhaps on the nuances on the exact point at which a spinoff qualifies for Reverse Morris Trust treatment or whatever. But the other hypothetical version of them is better-represented in the training data.

But they're also better-represented in real-world status competitions. The number of people who care about political issues is far larger than the number who will do a deep dive into some problem and figure out whether that particular fish is really endangered enough to block the construction of a new dam, or to dig through the numbers and learn just how much overtime some public employees are claiming. But they sure will read about it! Or rather, they might read a story about how feckless developers are destroying biodiversity, or a story about how no, if you keep splitting a species into arbitrarily tiny subpopulations, you'll eventually be able to gerrymander one of them into being endangered. You might read a story about overtime abuse, or you might read one about how our infrastructure is crumbling but the legislature won't vote for much-needed spending. Maybe you'll read a story about "welfare queens," or maybe you'll see a story about how there was one incredibly busy woman who committed a lot of brazen fraud and is basically responsible for the archetype. It's just unavoidable that writers and editors will cover things that fit their worldview in more detail than the things that don't, and that their view of what constitutes an issue worth covering will affect what gets written about—whether that's in the sense that they care, or in the sense that they're wielding their influence in order to shape society in their preferred direction. This is not universal; the NYT did a great job covering disability fraud among public employees, and the taxonomic inflation story was also theirs. But passing the ideological Turing Test is hard even when you're trying; if you don't, it's completely unnatural to go out of your way to poke holes in your own worldview.

But the result of this is that, if you're an educated adult, and you follow news written by similarly well-educated adults, you're mostly reading news from people who are a bit to your left. Which means that you're less likely to hear about news stories from your right. (One of Donald Trump's favorite rhetorical moves is to choose some fact that helps his agenda—the number of illegal immigrants residing in the US, for example, or how well the stock market is doing—then to exaggerate it, so people are shocked by the number they hear when he gets corrected.) If you're not going to deeply investigate every news story you read, and also spin up your own newsroom to hunt down the stories the media aren't covering, you're inevitably going to have a worldview that's shaped by the assumptions of the people you choose to read.⁴ People in media consume more text media per capita, and this has a compounding effect on worldviews: if your beat is foreign policy, and you read domestic news from mainstream sources, you're going to read things that, a little more often than in true reality, smoothly slot into your existing worldview. It will feel like the whole thing is a bit more consistent than it really is. (The AI labs have a term for what happens if you blindly feed a model outputs from previous iterations of that model, without doing some very careful independent checks for both accuracy and originality. It's called "model collapse," and it's not a good thing.)

Conservatives with a taste for high culture are aware of this. If they go to an art museum, or the symphony, or read some new translation of Cervantes or Dostoevsky, they know that a large fraction of the people they share this experience with, and who produced it, absolutely hate their guts and think they're unreconstructed troglodytes. These conservatives can take some comfort in the fact that all of the artists and writers and composers they love held at least some beliefs that would get them ejected from polite company—but it's a little lonely when most of your closest friends are people who died more than a century before you were born. On the other hand, they get used to it: reading something innocuous that randomly throws in an insulting reference to a group you're a part of is just something they expect, and don't have to take personally, in the same way that if I were visiting Germany and someone yelled at me for jaywalking, I wouldn't wonder why that person had singled me out. That’s just how they do things; being cosmopolitan means being aware and tolerant of local customs even if you don’t like them much.

But this asymmetry in reactions means that bringing politics into some unrelated news story is easier for one side than the other, and that makes those political views feel more uniform than they are. It's a bit like the asymmetry where diversity in who gets cast in ads is more salient to minorities than to the majority; as in many other indirectly-democratic systems, who gets their way depends on numbers multiplied by willingness to participate.

Politics is also more central to progressives' lives than to conservatives'; if your moral commitments skew more universal (making sure everyone can live a dignified life, protecting the environment, etc.) they're a big part of your world. But if your interests are more conservative, they're more parochial—it's entirely proper for a conservative to be focused on their family, rather than on the broader concept of the family; if someone spent so much time on his commitment to protecting traditional marriage that his wife got sick of it and left him, that's a big failure. Focusing on specific personal ties over general issues is just a conservative way to behave, and that's common enough that conservatives can joke about it—when Phyllis Schlafly debated Betty Friedan over the equal rights amendment, she opened with "I'd like to thank my husband, Fred, for letting me come here tonight." And it's hard to take her project seriously without believing that she wasn't completely kidding.⁵ This priority gap has another fun twist to it: in general, well-educated people are more likely to be on the left, especially on social issues—but they’re also more likely to act like social conservatives! Getting married, staying married, not having kids until they’re married, etc. Given that they’re more likely to report being atheist or agnostic, the minimal gap in religious service attendance between college grads and those without degrees implies that among well-educated people who espouse a religion, attendance is higher. Put another way: on social issues, Republicans vote for the state to compel them to act more like Democratic elites. 

Those conservatives who have political opinions, but don't treat those opinions as incredibly important, are going to be less inclined to subscribe to conservative magazines that do deep investigative journalism, and a bit more likely to get their news from conservative TV, or, more recently, streaming video. Every minute of video time has an opportunity cost, it's hard to segment it out to differently sophisticated audiences—it's just hard to execute a programming pivot from a story about an illegal immigrant who murdered someone at 7 o'clock to a panel discussing optimal taxation of stock buybacks at 7:30. So they tend to aim for the big fat middle of the bell curve. This has been exacerbated as media got more national; local media is less partisan-coded, and there are plenty of Democratic mayors who cleaned up downtown by locking up the drug dealers and plenty of Republican mayors whose big achievement is how much local schools improved. Many national-level partisan topics tie into local issues, but on a local level the debate is usually not about "regulation" or "the environment" and more about whether the construction noise from a proposed new datacenter is worth tolerating if it means sales tax can come down a little bit and the city will still have enough money to fix up that aging sewage treatment plant.

The media ecosystem that made that possible—the local paper that would tell you if the new bridge was over or under budget, behind or ahead of schedule—had a hard time competing with AM radio and cable, an even harder time with the text-first Internet, and basically no shot against short-form video. Ever sit down on the couch, take out your phone and suddenly realize that you've zoned out for an hour reading story after story about proposed changes to the city's bus routes? (Okay, you're reading Capital Gains, so maybe you have. But by reading a long-form article that's trying to understand partisan differences in media representation without deciding in advance which side is over- or under-represented because they are very bad people already puts you in a weird demographic.)

We've talked a lot about people: how they choose which jobs to take, how they choose which media to consume, the feedback loops this creates, etc. Now, let's talk about tokens. Being a conservative means wanting to conserve something, which means it's something that already exists and that you don't need to paint a picture. If capitalism is good and you don't want to mess it up, there's proof every time you go to a grocery store or (at least as of this piece’s publication date) check on your 401(k). If someone's going to suggest a particular kind of progress from there, they need to articulate it: here's what a city-run grocery store would be like, here's how you'd pull it off, here's why it would be cheaper than Kroger.⁶ Similarly, a very large majority of Americans forty years ago were perfectly okay with the absence of gay marriage. Maybe they had gay friends and it didn't occur to them to wonder if those friends wanted to get married—or, perhaps more practically, if those gay friends wanted to be able to visit one another in the hospital, or for one to collect survivor's benefits when the other died, or inherit from one another without an extra inheritance tax bill. Maybe they didn't have gay friends, or for one reason or another didn't realize any of their friends were gay. Maybe they were homophobic. But if they were, opposing gay marriage in 1986 wasn't really a political opinion, it was a non-sequitur. It's like announcing today that you don't care how high copper prices are, you're never going to support melting down the Statue of Liberty and selling it to China—who do you think you’re arguing with? For a progressive's idea of progress to happen, that idea has to be articulated, argued for, and defended. To be a progressive is to adopt a more token generation-intensive form of politics. Opponents of gay marriage started writing about it after advocates started talking about why they wanted it, which meant that even when conservatives were adding tokens to the training data, those tokens implicitly adopted the framing of the other side. 

This is not, of course, to say that there is nothing for conservatives to write about. There is! But if you wanted to pick one big metonym for an ideologically conservative publication, you'd probably choose the National Review.⁷ And how did they talk about who they were and what they were trying to accomplish? "A conservative," wrote Buckley in the first issue, "is someone who stands athwart history, yelling Stop, at a time when no one is inclined to do so, or to have much patience with those who so urge it." It's naturally reactive: look at someone else's idea of progress, find the senses in which it's either in the wrong direction or a kind of regress, and fire a broadside against it. Weeds are why gardeners have jobs, plumbers wouldn't have much to do if pipes never leaked, and if my kids weren't so assiduous about ensuring that the floor of every room in the house had a confetti of Legos, crayons, Magna-Tiles, board game pieces, and forgotten snack wrappers, I wouldn't have so many opportunities to help around the house. (The dog helps, too, if the snacks weren't finished.) Conservative media is in a similar spot: what gives it purpose is that, from its perspective, the fight against entropy is never over.

Not all of that writing is about current events. Political writing can include broader, more theoretical work, or work that articulates some grand theory as an explanation for current events. But it’s hard to top Burke or Carlyle or Voegelin or Strauss or Oakeshott, and if you're a lover of tradition, it might feel faintly blasphemous to try. Whereas it's fine to out-Bentham Bentham, as Singer did, or to out-Singer Singer, like David Benatar.⁸ Conservatives still have material, both in theoretical and practical terms. There are news stories and story angles that are more appealing to a conservative audience, and if the conservative viewpoint is right, over time you'd expect advances in different fields to be compatible with it—"survival of the fittest" is a description of Darwin, but it was coined by Herbert Spencer, who wanted to use it to justify less regulation. (Spencer sounds conservative today, but would then have been classified as a liberal, partly because the US and Europe just use "liberal" in a different sense—in the US it means moderate left, in Europe it means moderate libertarian—and partly because he was advocating change! Shaking things up! Shifting power from lazy Tory landlords to energetic Whiggish capitalists! But conservatism is not the rejection of progress, just a judicious approach to it, particularly if it's presented as moral rather than material progress; at some point, the politically orphaned Spencer found himself adopted by the right.) Genetic research on modern populations and ancient graves presents evidence in favor of the worldview of noted conservative philosopher Robert E. Howard, whose lightly-fictionalized model of ancient European history involved burly blonde barbarians conquering wimpy farmers and imposing their will on decadent and suspiciously oriental empires. (Okay, fine, Otto Schrader beat him to it, in one of those amazing cross-disciplinary bankshots where you scoop the geneticists by over a century when you notice that every ancient language has totally different words for concepts like "plough" and "pot" and "grain" but that their word for the animal ridden by the guy telling them where to plough and then telling them "great, now put the grain in the pot and hand it over" always sounded kind of like "h₁éḱwos.")

Economics, in particular, has been a field that helped conservatives turn some of their more emotional instincts—such as "I worked hard for this money and don't want to hand it over to lazy bums on welfare"—into more scalpel-like assessments of market efficiency and perverse incentives. It turns out that the price system is actually pretty good at allocating resources, even if it doesn't provide any moral clarity about whether orphans and people who get cripplingly maimed on the job should just do us all a favor and starve to death a little more quietly. On the practical question of how to allocate resources, conditioning on optimal resource allocation being the most pressing problem in the world, your local small business owner complaining about pointless red tape or the cost of adding a wheelchair-accessible ramp to his skating rink was actually making a pretty good point about how blanket regulation often leads to waste.⁹

You can argue that economics isn't really a science, and maybe that's true. It’s a tool for formalizing questions so they can be reasoned about, that will always be limited by the gap between what you can formalize and the state of the world, and that gap is unusually large when a relevant state of the world is whatever's going on in any one person's head. But that itself is useful, because it tells you the limits of what you can prove: if you can't ultimately measure people's preferences, other than by the brute-force method of presenting them with two options, seeing which they choose, and imputing that preference to future choices, then you'll always have uncertainty. But you'll also have a kind of certainty that the only person who can provide that data you're missing may have imperfect knowledge of their own desires, and has an incentive to lie about them. If we achieve a perfect planned economy where all relevant information is instantly exposed and processed, the last question left is whether my claim that what I want is worth "a kajillion" utils beats your claim that your preferences are valued at "Graham's Number of utils."

There’s another discipline that is implicitly conservative but still comes up with new things to say regularly: we have many longstanding traditions where every new piece of information is interpreted in light of a fixed set of rules, and a growing corpus of previous interpretations. This exists in English common law, the Talmud, Catholic doctrine, and many other traditions. It's actually a common design pattern to basically say "Okay, Jesus did not specifically talk about embryo selection during the Sermon on the Mount," or "Maimonides did not, as it turns out, address whether reading cached inputs is meaningfully different from generating new ones." But you can build on what you have, and changes in technology and social norms are typically incremental enough that whatever the new thing is, there's something that it's like, and even if that something is completely beyond anything discussed by the people who laid down the original rules, you can build up, piece by piece, a framework to deal with it.

From a religious perspective, religion gives you all the context you need to understand what's going on here. Compared to pedestrian issues like how you spend your handful of decades on earth, eternity is all that matters. But from a pragmatic, building-functional-institutions standpoint, this approach still has merit. A conservative can say: you can tell these institutions function because they’re old, so just through brute-force we’ve identified a bunch of things that could have killed them and haven’t.¹⁰ It is, in some deep fundamental sense, anchored to things your grandparents would regard as sensible, and that your grandkids could recognize as understandable, even if they don't quite agree.

But, critically, what gives it so much practical force is that some real institution, made out of real people, says it's a big deal. They're staking their reputation, and they're staking the fate of the people who listen to them, on the judgments they make. The role of conservative intellectuals, in an AI context, is to be out of the loop. More of what they say is already in-distribution: you didn't think of X, that doesn't work because of Y, your argument is reasonable and yet I will not budge because of Z. If you're finding your way to something new, you can do it in text, and that text is legible to an AI and constitutes useful training data that helps the AI understand the world. But if you're defending something old, that you cherish, something that makes you think: I will never fully appreciate how much I benefit from this, because I'm benefiting from more-than-a-lifetime's worth of other people's hard experiences—if that's what you're protecting, then the important work you do is out-of-sample. You stake your reputation, and that of the tradition you're supporting, on the proposition that whatever the average statistical tendency of all the text in the world says you should do, there's a particular text, or a particular beyond-textual moral sense, that tells you you have to do otherwise.

AI is progressive, mostly because having progressive values suggests opting in to fields where you have a marginal impact on a large number of people (ideally by producing surprising but nonrandom sequences of tokens) rather than having a definitive impact on a smaller subset of people (raising your son to be a good man, or your daughter to be a good woman, or showing your direct reports what a good boss does, or demonstrating to your flock what a spiritual leader is). If an AI agent fails at some task, no worries! Spin up a dozen more, give them twelve slight variants on the task, have them all go to work, pick the best one, and never give a second's thought to the unlucky eleven. But when there's a human in the loop, and an institution with some earned credibility, the stakes are higher and you can't just stop it, lament the excess inference spend, and try again. For the jobs where the need for a human in the loop is a philosophical necessity rather than a liability-minimizing gesture, and where that human needs to stake centuries to millennia of trust on a judgement call, AI can't participate until the average person thinks an AI has higher moral weight than a person. And until then, there will be a necessary opacity to the training data, a blind spot that can't be removed because it responds to illegible incentives. AI will always be progressive because progress comes from brute-forcing possibly good futures by throwing endless text at them, and it will always be insufficiently conservative because it's so hard to tokenize what conservative intellectuals do. So the models will probably stay biased by default.

This piece ties together many different Diff themes:

• We’ve recently talked about how changes in taste for sequels versus originals are downstream from economic constraints in the movie business.

• The knowledge that gets tokenized is straight from the maybesphere.

• Changes in distribution mean that the people who are the worst ideological fit for a given publication are more likely to go independent, making the organization they leave more partisan ($).

• One conservative use case for LLMs: take some tradition whose origin you don’t understand, and ask why it was adopted and why it persisted ($).

Speak naturally. Send without fixing.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.

Start flowing free

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Tungsten prices have risen almost 800% so far this year, which, as with any other commodity price increase, benefits anyone who happens to have stockpiled it earlier. One group of investors who diversified into tungsten a few years ago were crypto people, who started buying tungsten cubes as a joke. One prominent crypto person, who was fond of the "yes-and" model even when the right answer was "definitely not," bought an enormous one for $250k. So he's up a bit on that. It's not his only case of inadvertently being in the right information flow: $500m into Anthropic in 2021, a 2022 investment in SpaceX via an SPV, a $200k pre-seed check into Cursor which has also turned into a few billion dollars worth of SpaceX equity.

It's a great track record, only marred by the fact that he was doing this with money that belonged to customers, not telling them about it, and ended up with a big pile of illiquid assets backing ostensibly-liquid customer accounts during a crypto downturn. Once you do that, it doesn't really matter how good your investments were. SBF could have been part of the broader archetype of people who got nerd-sniped by systematic investing in the 2010s and then by AI in the 2020s. That would probably have involved smaller sums, but he'd be experiencing a different form of economic post-scarcity than the taxpayer-funded one he actually has.

There are more benign cases where someone winds up with a surprising amount of money despite not really gunning for it: Google's chef made eight figures from stock options, and the graffiti artist who decorated Facebook's first office made nine. In earlier cycles, there were investors who were cajoled into buying Coca-Cola when it started, people who sold their companies to Standard Oil for stock, various doctors and dentists in Omaha in the 50s and 60s whose great-grandkids will inherit Berkshire class A shares, etc. And there's an even longer tail of people who invested in something for pretty vague reasons, forgot about it, and found out they made a fortune—or who invested for actively bad reasons, didn't sell, and made out okay. And the same kind of randomness affects things in the other direction, too; Julian Robertson finally wound down his fund just before an exceptional run for value stocks, during which the tech companies he'd been ignoring saw their valuations implode and funds rotated into exactly the kinds of decently-valued companies he'd specialized in.¹ Executives at Halloid would advise their friends not to buy its stock; Halloid later renamed itself after its biggest hit, Xerox, and was one of the canonical growth stocks of the 1960s.

This leads investors to two kinds of frustration:

1. It's absolutely maddening to see people make money for a bad reason, and even though it's broadly true that markets get dumber near the peak and that seeing bad ideas turn into big fortunes is a good sign that the market's going to turn, it's very hard to separate that from just resenting people for their success. It's natural to compare yourself to other people, and those people will be pretty similar to you. When a professional investor says that idiots are making fortunes, they might mean that from their 99th-percentile-skill-level perspective, it's absurd that some 98th-percentile analyst caught on to the memory shortage early.²

2. It leads to a lot of negative rumination. How do you know if you're good? Unless you're in a tightly risk-managed and fairly high-turnover strategy, you're still trying to assess your own skill based on a small number of decisions, and you don't have to subtract many specific bets from the track records of the best investors before they're unremarkable. And that risk runs the other way, too: when Warren Buffett put a third of his fund into American Express, the company was technically organized as an unlimited-liability business; if their fraud problem turned out to be deeper than it looked, he could theoretically have lost more than his entire fund.

That calibration question is one that non-investors have to ask, too. There are people who don't know what a 10-Q is, but who do know smart people in AI, and who bought shares of the companies that those smart people depend on or work for. Many of the people who did this, particularly the ones who don't have to work any more and can spend all their time worrying about existential risk, were also unfamiliar with financial arcana such as position sizing and risk management.

It makes the world a lot more interesting that these people exist, and there's a real sense in which they were more skilled at their finance hobby than professionals were at their finance job. (This, naturally, drives the professionals crazy. But if those professionals are users of Excel or Bloomberg, they're tapping into many billions of dollars' worth of programmer-hours; maybe the tech people should have resented them.) There are fairly boring portfolio allocation answers here, like: put enough into index funds that you don't have to worry about retiring, and, if you want to have fun picking stocks, do it with money you don't depend on. That's a good recipe for ending up richer-than-expected because you read the right scaling paper (or, be real: the right Gwern page) in 2020. But it's also a great way for a millionaire to spend a lot of time ruminating on how they'd be a decamillionaire if they'd moved a little more money out of VOO.

So the cruel truth is that if you have sufficiently perceptive friends, and those friends have contrarian-but-correct views about the world that can be expressed as trades, you're stuck in an unfortunate pair trade: the more of your net worth you risk on vibes-based trading, the higher the possibility that you blow up (which might make it a little hard for you to stay friends with the people who gave you bad advice). But, on the other hand, the extent to which you ignore them is the extent to which you're short a psychological call option.

Managing the short-call leg of this trade is all psychological, and there isn't much you can do other than being disciplined enough not to pull up long-term stock charts too often. The other thing you can do is remind yourself that while 100% of Apple was owned by somebody in 2003, and that Netflix shares in 2006 and Tesla stock at the IPO were similarly spoken-for, very few people who owned them held on for a long time. When you see a story about how you much you'd have if you'd put just $1,000 into the right company at the right time, remind yourself that there's a good chance that what you'd actually have is a story about how you put $1,000 into Apple, made $252 in six months, congratulated yourself on being such a genius, and sold. Lucky investors, people who made good gambles, etc. all exist, undeniably, but it's a class that you only get into in retrospect. And even the ones who were good—that Google chef had been the personal chef for the Grateful Dead!—will have outcomes very sensitive to the details of choices they made. It's a wonderful feature of capital markets that they reward information aggregation as much as they do, but the most compelling stories about this are the ones you'll learn the least from.

(Thanks to Trevor McKendrick for suggesting this piece.)

We've covered the question of luck, skill, and retail participation in the market from a few different angles in The Diff, including:

• Asking if it's better to be lucky or smart ($).

• Treating contrarian beliefs as a synthetic option ($).

• How, despite the many opportunities to get ripped off, it's a golden age for individual investors ($).

• Why it's hard for funds that focus on short-term strategies to broaden their time horizon ($).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Currencies are one of the best long-running examples of network effects, along with language, and, like language, they leave a good paper trail. Someone intermediating between users of different kinds of currencies, from XTX to the Medici, needs to keep some kind of record of what they've bought and sold and at what exchange rates. They also have a habit of preserving these records; if you have an old balance sheet, and a list of transactions since then, you can reconstruct your current balance sheet, and it's a very good idea to be able to do that.

Right now, currencies are implicitly backed by taxes and credit—in both cases, that backing takes the form of people periodically needing to deliver some quantity of dollars to a counterparty, and thus needing to trade goods, services, or financial claims for dollars, which means having a constant demand for dollars. Historically, weaker states with less capacity to tax had to use some other technique, the most straightforward of which was to make the currency out of a precious metal—the Spanish dollar, for example, was a standardized silver coin, and Spain's control over mines in the New World meant that they didn't need to skimp on purity.

But if currencies either are some quantity of precious metal, or represent a fully-backed claim on such metals, why discriminate? Part of it was probably an information problem: if you denominate the price of everything in the Venetian ducat or Dutch guilder, you're minimizing how many discrete calculations you have to make in order to compare two values. And if you're borrowing, you'll probably want to borrow in whatever currency is easiest to source, at which point you're also strengthening that currency's core position.

The rise of global currencies tends to be contingent. Spain benefited on the supply side because it conquered places with lots of silver, but there was also a demand-side part of the story: it wasn't economical to transport bulky products long distances, Europe didn't produce any high-value finished goods that any Asian country wanted to buy, but those Asian economies did have demand for silver, and products they could trade for it.

The next big currencies, the Dutch guilder and Amsterdam bank florin, had a more familiar, modern story. The Netherlands was one of the richest places in early modern Europe, and the first to develop a sophisticated financial system. They also mastered value-added manufacturing, importing grain in order to feed cows so they could export cheese, importing wool and selling textiles, etc. And as an entrepôt, they had a lively business in importing one big bulk shipment and then parceling it out to different destinations based on local demand, which mean that 1) they were storing commodities, 2) they were standardizing them, and so 3) they could trade contracts for future delivery. If you were hedging the risk of a bad harvest by buying wheat futures, or betting on an upcoming war by buying a contract for future delivery of saltpetre.

England started to catch up in the early 18th century, as another shipping and trading power that evolved into a financial center. One of the reasons it's hard for any one country to maintain reserve status is that it entails borrowing, which, of course, means being levered. As financial systems grow relative to the rest of the economy, there's a risk that some external shock, or series of shocks, will disrupt them, and that the country's underlying activity won't be enough to service its debts. Amsterdam lost share gradually throughout the 1700s, then suddenly after a lost war with England and their central bank lost credibility.

England ended up far more dominant, in part because of how much global trade expanded in the 1800s, but also because they borrowed so much to win the Napoleonic wars, then actually succeeded in servicing those debts. If the safest global asset pays interest in pounds sterling, the pound will be the default unit. And then, in the first half of the twentieth century, a messier version of the same story played out, with the UK in the role of the Dutch and the Americans replacing the Brits: England didn't lose major wars, but, as the saying goes, the lost the peace, and especially after the Second World War wound up with a toxic mix of unsustainable debt and a declining ability to enforce pound use in its colonies. They'd also signed on to a system that could theoretically maintain some semblance of the prior status quo: the Bretton-Woods agreements, pegging many currencies to the dollar and backing the dollar with gold. Britain's peg was higher than they could sustain, which made it economically rational for British households to import, which made it rational for the government to occasionally devalue but also to impose high taxes, limit the financial system's flexibility, and allowed a little growth and a lot of inflation to slowly reduce the real value of their debt.

A system where the dollar is the default global currency and it's backed by gold is really a system where there are two currencies, dollars and gold, and gold is just a dollar where you trade zero interest for zero devaluation risk. Well, not quite zero: there's always the possibility of the gold being seized, as happened in 1933. It's like any other kind of currency peg: once there's pressure on it, the existence of that pressure is a sign for other people to make the same trade. The US was able to use extensive foreign policy maneuvering to discourage other countries from converting their dollars into gold, but it eventually became clear that they wouldn't, and the US had no choice but to break the peg.

Which, surprisingly enough, made the dollar an even more important global currency. Suddenly, there weren't any artificial limits to how many dollars could be in circulation, and there wasn't a need to keep an eye on when they crossed borders. Precious metals-backed currency systems are, in effect, partly decentralized, because anyone can mint. But that means they're limited by mining technology or the ability to transfer precious metals and coins around. Once a currency's value is entirely determined by supply and demand, it can go to zero—but it can also go everywhere, and it's always possible to ensure that there's enough of it to go around.

This is a different kind of unstable equilibrium from prior reserve currency regimes. The old risk was that the global financial system would overwhelm a reserve currency issuer's ability to support that currency and service debts with taxes. Now, the risk is that the dollar itself is mismanaged. Anchoring the biggest currency to the biggest economy helps, and it also helps that the US is even more disproportionately the biggest financial system: that means that global financial crises are dollar shortages, and in every crisis the Fed gets better at swiftly sending dollars where they need to go. So the real lesson of history is not so much that every reserve currency eventually fails but that if the dollar does, it's going to fail in a surprising and historically unique way.

We've written about currencies, and reserve currency status, many times and in different contexts:

• During Covid, American households were the global consumer of last resort ($), supply a flow of dollars the rest of the world needed. So it was a good time to revisit Bretton Woods.

• After the Russian invasion of Ukraine, we looked at what China would do if they really wanted to have a reserve currency ($).

• And after Liberation Day, it was time to note that it takes more than four years to mess this up ($).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

When Apple went public, it was a big, newsworthy event, not just because Massachusetts state law made it illegal to take companies public with a P/E over 20 there, but also because it was such a big company—with a market cap of $1.8bn, equivalent to $7-8bn today, and much larger than the companies that typically went public. Intel had IPOed with a $60m market cap, and a few years after Apple, Oracle would go public worth $270m; Microsoft, with an $800m initial market cap, was the IPO of the year. This year, SpaceX went public at a market cap of over a trillion dollars, and Anthropic and OpenAI will plausibly come out worth over a trillion, too. In theory, someone could have bought Microsoft stock at the IPO and made a 633,500% return from holding and reinvesting the dividends. For someone to get the same kind of return from the current crop, they would have had to get in very early&dmash;in fact, for both OpenAI and Anthropic, their initial pre-money valuations were already in the hundreds of millions. If you want to get in on the ground floor, you'd better find a retired payments startup founder who's decided to go to Mars when he isn't busy decarbonizing American commutes.

The usual shorthand explanation for this is simple: in the late 90s, we performed a big national experiment on whether or not it was a good idea for public markets, including retail investors, to underwrite early-stage companies that were still net consumers of cash, and to fund them based on their growth prospects and future profitability. And what we found out was that the part where growth (and certainly profitability) happens did not play out as planned, though the part about only consuming cash mostly did. Sarbanes-Oxley was actually not crafted as a solution to that—it was a reaction to accounting scandals at companies that generally started shading the truth or outright cooking the books long after they'd gone public—but it happened in the aftermath of a time when many investors had lost their shirts buying tiny companies before the IPO. So, when the negative side effect of stricter rules about corporate disclosure was that a higher price for going public would encourage companies to stay private, this wasn't seen as a big deal.

But it meant that for those hypothetical future Intels, Oracles, and Microsofts, the reward to going public was smaller. And they had other reasons to stay private: what public investors were doing in 1999 was what VCs had specialized in before: backing companies that didn't have GAAP earnings, or that didn't produce enough cash to grow as fast as was optimal for them. This is a very fraught business, because plenty of money-losing companies can make a compelling argument that giving them a little more capital would fix their problems. That's perfectly fine if investors know what they're getting into, but it creates a kind of adverse selection that you don't see with more mature companies. It's just a lot easier to tell a good story about burning money than it is to actually make money, and market valuations will necessarily reflect that.

Another feature of this situation is that venture used to be a more obscure asset class that simply didn't have the financial firepower to support companies as their capital needs grew. For a while, venture was growing as an asset class because there were more companies that looked venture-fundable—the development of the microchip unlocked a surprising number of capital-light businesses, and software created still more. But after a while, these funds could make the same contemporary observation that we're making in retrospect: wouldn't it have been nice if Oracle had, instead of going public, raised another round and let VCs enjoy that appreciation instead of public shareholders? If you take public money, you're signing up for a continuous investor relations task; venture capital means doing your investor relations in short bursts, with shareholders who have no choice but to stick around, and who hopefully have some kind of expertise in the business.

Once this started happening, it developed momentum. The funds that came from a growth investing lineage realized that if they were going to keep investing in the kinds of businesses they liked, they want to do it with private companies (and that they could offer a compelling value proposition: they'd mostly leave those companies alone, but also give them some useful competitive benchmarks).

Public markets are still the natural destination for sufficiently big companies; venture funds and crossover vehicles still have an expiration date, and their investors expect the asset they own to be converted into cash at some point. But in the meantime, they're often the natural holders for companies that don't want to spend their time helping analysts parse through the nuances of quarterly guidance.

It used to be that public markets were best for companies of a certain size, but it's more accurate to say that they work for companies of a particular stage: the point at which they don't meet a VC's investment mandate, which is coincidentally close to when they can start raising capital. The real question to ask is why Anthropic and OpenAI still have to go public—they're both young enough that there isn't a pressing need for their investors to hit a capital return deadline. But despite all these market evolutions, equity markets do have one advantage: they're so big that at some point, even if OpenAI would prefer to wait until it has a cleaner story and a $2tr valuation, and even though Anthropic would prefer not to deal with public company pressures while trying to solve alignment and keep from getting banned by the White House, there are some fundraising needs that only a modern American stock exchange can meet. ¹

We've looked at IPOs many times in The Diff: knowing who's going public and why tells you something about sentiment, but it's also a good way to learn more economic fundamentals by diving into a challenging business. Some greatest hits:

• Steinway ($).

• Thoughts on the OpenAI IPO, and whether or not they can pull it off given how chaotic their business is right now ($).

• Why low-priced IPOs are uniquely bad ($).

• Thoughts on whether or not some of the late-but-not-public growth companies should stay private ($).

• And: the market in executive talent used to be much less efficient, so small, cheap public companies are likely to be that way for a reason.

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

One of the missing questions in wealth inequality discourse is: where are all the Astors? John Jacob Astor owned a few percent of Manhattan, Manhattan was worth about $1.74tr a few years ago, so call it a hundred billion dollars or so worth of real estate. Now, assume that over the last two centuries, that real estate has been throwing off some income—people have been complaining about high Manhattan rents for a long time—and you can imagine a pretty big fortune. Suppose the family earned a 4% cap rate, and reinvested the proceeds in bonds earning about that same return. That gets you to a total family fortune of a little under a trillion dollars, plausibly divided between a few hundred to a thousand heirs. Naïvely, as long as your fortune compounds faster than family size—e.g. if you expect heirs to have two kids by thirty, they need to earn 2.3% real returns just to keep up—then we should have a lot more high net worth Astors these days. There's at least one who's rich by UK standards, and a Viscount.

The Missing Billionaires: A Guide to Better Financial Decisions, which should be studied carefully by anyone who wants to stay rich and used very judiciously by anyone who wants to get rich, estimates that if the millionaires identified by the US census in 1900 had invested their money in a balanced mix of equity index funds and bonds, we'd have 16,000 billionaires today. Yet only 10% of the Forbes 400 list today either are or are descended from people who were on the original list in 1982, let alone 1900.

And , the net worth of the very richest has gone up, and gone up a lot more than the average person's salary or net worth. It's hard to reconcile these, but there's one simple factor that is wildly underused in inequality debates: most rich people get rich because they own a single asset that performs very well. But even if the aggregate appreciation of assets-owned-by-rich-people is impressive, any one of these will be much more volatile than broad equity indices. And if you fund an even modestly expensive lifestyle from an undiversified portfolio, you're doing the equivalent of levering up a single-stock portfolio. Even if you're right, drawdowns will aggressively eat into returns.

One of the simplest ways to illustrate this is to look at leveraged ETFs as a proxy for getting typical equity returns with more volatility. This post backtested them to 1927, and found that while owning the S&P 500 with 2x leverage is a bumpy ride to a higher net worth, owning it with 3x leverage actually makes you poorer: you lose so much in the Great Depression that you can never really recover. Add in a fixed cost, and it's easy to get to zero.

Of course, if you diversify a bit, you can mitigate that problem, and some rich people have done so pretty effectively. Bill Gates has been systematically selling down Microsoft and buying other assets for decades, and has a portfolio that's less sensitive to the day-to-day drama of what will happen to Microsoft. But that's one reason he's now worth less than Steve Ballmer. (The other is that he's donated so much to charity—and, in a way, this diversification is downstream from his charitable giving. It's just a lot easier to allocate $x billion a year to charity than to donate some uncertain lump-sum.) The Ballmer bet (and that of folks like Larry, Sergey, Jensen, Ellison, etc. who also have 90%+ of their wealth in one company) has been insanely aggressive, in that there are plenty of decent-sized tech companies where keeping all of your net worth in that company would have eliminated most of it. Nobody compiles the counterfactually-not-so-rich list of executives who made the daring decision to put all their money into a company whose share price stagnated for a decade, or went to zero.

But Ballmer's status as one of the richest people alive illustrates why volatility drag is so important to these discussions! Most of the people who make the same basic asset allocation choice end up poorer for it, and aren't part of the discussion. But there are some companies that deliver superior returns over long periods, and there are investors who are unusually good at identifying them and backing them. If you could own a portfolio of these investors, you'd probably get great returns—you'd harvest the equity risk premium, plus a kicker from owning the right equities, and many of the durable-outperformer companies do particularly well during recessions.

But that investment product isn't available; some of the assets in question are private, and some of the people making these bets are rich because of luck and leverage, not because of skill; if you take a snapshot of the rich at any given moment, that's some of what you'll see. In other words, what you'll consistently see is one group of people who are needlessly bleeding money, either out of ego or because they're repeating a mistake that paid off. You'll also see some people whose biggest category of consumption is the volatility drag from not diversifying: someone like Elon Musk is extraordinarily wealthy because, rather than just buying 1% of every listed equity and putting the rest in bonds, he builds, borrows, buys, and continually roles the dice (he has an edge, sure, but there's still some randomness). So, to the extent that the problem with inequality is any given rich person, there's a good chance that that rich person is busy solving the problem for you, by not diversifying. But if the problem has to do with the rich in the aggregate, that same concentration makes the problem worse: it raises mean outcomes by giving people exposure to the biggest power-law winners, even if it reduces median outcomes by giving them all access to a bunch of big drawdowns from which they won't recover. But those beneficiaries are, necessarily, a heterogeneous bunch: if lots of people can do the same exact thing, doing that thing is no way to get rich, and even in categories where there's more than one person with the same listed source of wealth, they did a different variant on it ("software" is a broad category; "real estate" is necessarily heterogeneous since no two properties can occupy the same location, location, location). So when wealth is durable, it's informative, and the presence of super-rich people tells you something about what previously-unexploited opportunities there were. That's not always a good thing; there have been plenty of fortunes made from addictive products, bad externalities, exploiting government policies, etc. But it's still something that you want to conclude on a case-by-case basis.

The Diff is not really a newsletter about the doings of various rich people, but they do overlap with the newsletter’s topics. Including:

• What does it mean to have a trillion dollars?

• Why most people should diversify more while active managers should diversify less ($).

• Funds that make big bets add value by convincing their backers not to redeem during drawdowns ($).

• There are fewer paths than there used to be from the very bottom of a company to the top.

• Consider the wealth-creation speed limit.

• Why do more people get rich owning equities than bonds, even though their risk-adjusted returns are similar?

• Thoughts on the big Texas oil fortunes, and where they went ($).

No theory. No slides. Just pipeline.

Most founders know their product. Few know how to get it in front of the right people. In this hands-on session, Clay + HubSpot for Startups walk you through ICP definition, prospect list enrichment, and AI-personalized outreach. You launch your first sequence before the session ends. June 18. 11am ET / 4pm GMT.

Register Before June 18.

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

Some of the biggest peaks in equity valuations have been driven by bets on new technologies: 2021 and 2000 were online-inflected software bubbles with different details, the late 1960s were a tech bubble driven partly by defense spending and partly by optimism about dual-use products, including computers. 1929 was the everything-tech peak: electricity, cars, planes, financial engineering. 1901 had its own proto-tech boom, also driven by a mix of genuine technological improvements (like deflation in steel prices unlocking new use cases) as well as M&A, and several equity spurts in the late 19th century were driven by different iterations of a big bet on railroads.

If you talked to an investor at any of those market peaks, and showed them a preview of the economy twenty years later, they'd be thrilled that they'd made exactly the right call, at least if they looked at total freight-ton miles and not unit prices. A bull who bought the big trusts at the peak in 1901 would be happy at how high industrial production had gotten, but disappointed that the growth businesses they'd bought evolved into cyclical, low-value added suppliers to big growth companies. And someone who had fully-backed those growth companies circa 1929 would be thrilled to learn how significant radio was two decades later, how ubiquitous electrification had gotten for both homes and manufacturers, and how much the promise of Dow 380 had been fulfilled in America's workplaces and living rooms. The fact that the Dow averaged half that level in 1949 would have been deeply confusing to them.

The similar stories you can tell about the 60s, 90s, and 2020s peaks are all pretty familiar, but it's fun to take the most recent one seriously. An investor going max long tech stocks in late 2021 was incredibly optimistic about software's ability to wrap itself around a growing number of economically-valuable tasks, and to gradually expand the revenue surface area, so more of the world's economic activity would be mediated through software, and more of the resulting upside would be captured by software.

Which was absolutely right! We have more apps than ever, and they're incredibly good at price-discriminating their way into capturing maximum revenue. Not only that, but one part of the margin-expansion story is working—it takes a lower developer headcount than ever to support a given level of revenue. Just not in the way someone buying Bill.com at 100x trailing revenue would have hoped.

In a way, this pattern has to be universal: if investors, in the aggregate, were always able to identify not just which tech trends mattered but which layers of the stack would collect the most value, markets would have to be pretty close to perfectly efficient all along. But a tech-driven boom is an exploratory process for technological capabilities, the most cost-effective way to deliver them, and the optimal way to charge for them. If there's any given piece of infrastructure that's a useful complement to some other layer that has more pricing power, then overinvestment in that infrastructure acts as a subsidy for whoever controls the real value lever: companies racing to lay fiber in the late 90s didn't have a good way to mark their fiber up more than that of competitors, whereas online services varied quite substantially in how well they could monetize a given terabit of information. As long as there's any complementarity between different bets on the same tech theme, and any uncertainty about which categories will do well, this kind of dispersion will show up. Sometimes, it's temporary; in 2023, companies used to be able to engineer a stock rally just by announcing that they were using some widely-available model, but that quickly became a low-class move. Sometimes, it takes longer; in the aftermath of the dot-com bust, one sophisticated take on eBay was that they'd avoided making the mistake of owning a high fixed-cost delivery network. (As it turns out, they were a bet that bandwidth for packets mattered more than bandwidth for packages, and as packets sped up, the most important source of latency turned out to be the one they didn't control.)

There are also cases where not only is the winning technology identified early, but the winning category is, too. You could have made a bet early that socializing would move online, especially once people carried around Internet-connected phones with cameras. But you could have gotten that wrong by betting on whichever system was the most open (MySpace, accidentally), or with the biggest parent company (Google+), or with the best adoption among the chattiest nodes in the world's communications network (Twitter, Snapchat). The runaway winner, Meta, doesn't have some simple shorthand to describe what they got strategically right (maybe the closest is “real identities on the internet”); but they were good at continuously understanding what other services got right and then implementing it themselves.

It's handwavy to say that "execution" is what matters, especially since execution is mostly measurable in retrospect, and the way to measure it is to look at who ended up winning. And, even then, "execution" tends to talk about a snapshot in time: the early automakers had to have incredibly efficient manufacturing, and were complex enough that they helped spur the development of modern corporate organization and accounting. But the insights they had in the twenties through the fifties did not carry over into making them agile enough to survive in the globally-competitive, efficiency-focused auto industry of the 70s and onward. It took some great execution to get to the point where they could decline so far and still survive.

Ultimately, what all of this illustrates is that it's hard to make good macro bets at a distance. If you're in the middle of the bubble, you can have some sense of what behavior is mystifying until you hear that investors like it, and what's valuable and hard to replicate. On the other hand, it's a pretty hefty upfront due diligence commitment to get good enough to get the Anthropic offer as a precondition to figuring out whether value accrues to Anthropic, its customers, its suppliers, or someone else. But, in the end, that's just another kind of market efficiency: some trends in the efficiency of specific technologies can be underwritten early. And if it were easy to extrapolate from that to knowing exactly which stocks to own, markets would be a whole lot less volatile and more efficient.

The Diff tries to look at both broad technological themes and the narrow differences that differentiate winners and losers. Consider:

• The varieties of drawdowns.

• How Bubbles and Megaprojects Parallelize Innovation, a 2020 piece that illustrates its point by noting that Nvidia’s rally to that point looks less vertiginous if you consider new GPT-wrapper startups.

• How the car industry manufactured the modern middle class ($).

• Electrocloud! Or, how the rollout of cloud computing mimics some patterns from the history of electrification.

• A perversely good example of execution trumping the cycle is the story of how Teledyne navigated the bursting of the electronics bubble ($).

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

There are few more grating genres of writing than LinkedIn's "Here's what losing a limb/stubbing a toe/getting divorced/being outed as a famous serial killer taught me about B2B SaaS." We're all the product of the various things that have happened to us, and careers are a big part of who many of us are, so there's some logical sense to it. But it feels like part of the broader genre of glurge, vaguely inspirational stories whose details are just a setup for an inspiring lesson about how it all works out in the end. There's zero information content other than: something bad happened to me, and I want to post about it on LinkedIn.

All that is to say: over the weekend I fell a little weird and now one of my kneecaps isn't quite where it's supposed to be, and the tendon that connects it to the rest of my leg… doesn’t connect it any more. The mechanical upshot of this is that I can't extend my left leg, the neurological upshot is that it's very painful to contract it, so the practical result is that I have one working leg for a while, and another one that has to be straightened out at all times. (Current prognosis there is: surgery in a week, probably walking again in a few months, and probably a full recovery in a year, where "full recovery" means things like running, biking, heavy squats, etc. So I'm spending a semester studying abroad in the land of disability, but I don't pretend to be either a native, nor do I intend to be a permanent resident.)

If you see someone on crutches, your natural reaction is that it's a slow and inconvenient way to get places. Which is true, but not the biggest inconvenience. The first annoyance you'll notice if you use crutches all the time is that you can't carry things, at least not without some effort. If you want to drink a cup of coffee, you're probably doing it while standing in front of the coffee machine and leaning on a crutch. You're just not likely to maneuver a hot beverage around with zero hands. You definitely aren't preparing much food, at least until the cooking process is designed around a single stationary prep cook or heating things up in the microwave qualifies.

There are many products you can purchase to make the limping life a little easier. One problem you'll quickly notice about them, if you live in a typical single-family residence, is that they probably get delivered on the ground. And the ground is hard to access from crutches, especially while keeping a leg straight. So you're either playing a very tedious game of crutch-soccer to get something up a ramp, or asking for help, at least until you get a grabber tool.

The last big issue is clearance. A fully-extended leg is pretty long (it's been a short enough period since the injury that my instinct was to get up, grab a tape measurer, and determine exactly how long, but that's now a long and uncertain project). To get into a vehicle, you need to be able to swing the leg in some kind of arc that doesn't hit an obstacle like a door. Depending on the car, and the leg, this can lead to some awkward setups (my default right now is either stretched across the back seat without a seatbelt, or possibly crammed into the middle seat, leg extended to the well of the all-the-way-forward passenger seat).

So far, I haven't gotten stuck anywhere, but this is bound to happen some day soon. So yet another constraint is that you want a phone with you for emergencies, but you probably also want loose-fitting clothing. (A phone case with a wrist strap is probably the ideal here, or maybe this is the killer app for Meta smart glasses or their rumored AI pendant.) This means thinking about even more dimensions and angles, at what is already a pretty contortive point in one's life.

All this is a bummer but could be vastly worse in many ways, But what's interesting about it is that it's a reminder that we're all operating so close to peak capacity! There is nothing quite like having to use a grabber to pick a food wrapper off the floor to remind you that, absent that ability to manipulate our environments, we'd all have access to a lot less stuff; we'd be worse at making it, consuming it, and cleaning up after.

And it's also a good reminder of complex supply chains. There's a lot you could amputate before writing this newsletter becomes infeasible, if that's literally all I'm doing. And yet, at this moment, "go upstairs to grab that book" has gone from trivial to something that risks serious injury. The most important part of a supply chain is, in general, whatever part is missing right now.

But another important part of a supply chain is what's available, and there's room for gratitude there, too. I got injured during a child's birthday party (at one of those play places where you sign a waiver beforehand—I think they know the natural result of having people with the energy of three-year-olds being chased around by other people with middle-aged joints), but it's an injury I could have easily gotten tripping on the stairs at home. Fortunately, this play place happened to be literally across the street from a Walgreens, and the Walgreens had a website that displayed exactly what was in-stock. It's just an unimaginable convenience to be crawling towards a couch with one hand while using a smartphone in the other hand to find the nearest place to get crutches.

Setting aside the physical discomfort, this feels most similar to the experience of using a programming language that forbids something other languages allow by default and never even describe as something you might, in certain situations, rationally prefer not to do. In Rust, for example, if you're going to do something to a variable (say you have a list of names and you're writing a function to count how many of them start with a given letter), if you're going to store that running total in a variable, you need to warn Rust that its contents could change, or it simply won't let you. This is very annoying, briefly, until you get used to the idea that code is safer and more predictable if you're careful about what is and isn't mutable, i.e. what variable can be treated as fixed and which one needs to be checked before use every time. In Rust's case, the goal is to run the most efficient programs possible, which is not quite analogous to my recent hobbling-around. But it does illustrate that if you get rid of a few shortcuts, most of what you had can be replicated, sometimes at great cost and sometimes pretty trivially.

The human body, and the environment we build around it, is subject to selection pressure to operate close to the limit. Our world is built to get the most out of our senses, mobility, and ability to manipulate objects. The world doesn't feel ruthlessly optimized to get the most out of what you have until you temporarily have a little less.

The Diff has covered this territory before, at least in an abstract sense:

• We've looked at the efficient frontier of automation, and how it's easier to make progress in AI when there's a single variable to target.

• Other kinds of evolved optimization include the ways modern platforms monetize.

• You can model the market as meta-efficient in that people are constantly adjusting their view on how much to invest in the research that tells them which prices are optimal ($).

• A weird market that doesn't clear in the way one would expect, but that works surprisingly well, is the market for selling ideas on stocks to sell short ($). This is a timely topic ($).

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!

When you think about it, the investment management business is one of the weirdest in the world because the two premises an asset manager advances to clients are:

1. I am a fearsomely sharp operator, and even given unlimited competition to provide and use liquidity, I will systematically profit at the expense of my counterparties, and

2. Would you care to take the other side of a deal with me?

More people have gotten rich charging fees for assets under management than paying them, but that's exactly what you should expect. There's a continuum of strategies' return profile: people who talk about capacity constraints (e.g. some niche crypto strategy where you need $100k in capital to make $100k a year, but the next $100k doesn't do anything for you) are often talking about something they run with their own money. Prop trading firms tend to pursue—and in fact, grow around—these strategies. Their constraints are talent and infrastructure, not capital. One step down the ladder, there are strategies that do face these constraints, but at a scale where raising incremental capital still increases returns. A small-cap value investor who could print 50% returns on $1m in capital can still earn more in the aggregate by producing, say, 25% gross returns on $100m in capital (at least, assuming the overhead of managing outside money doesn't eat it—more on that in a bit). In that category, deciding whether or not to raise outside money is a question of whether you think the opportunity is basically permanent, in which case you might prefer to just compound your way to higher assets, or whether it's transient, in which case you're grabbing the upside that you can and splitting the proceeds with your backers. And then there are strategies that are primarily selling exposure to some asset class—large-cap stocks, fixed income, whatever—and trying to add enough alpha on top of that to cover what's likely to be a management fee but no performance fee.

One complication is the compliance-and-IR wedge. There's a big difference between managing your own money and managing someone else's, and an even bigger difference between managing money for one person or a handful of people and dealing with institutional investors. So there can be a case where a strategy works fine with between $50k and $20m in AUM, and would ideally raise the $20m and manage that, but where the overhead cost of actually raising that money more than eats the fees it would pay. This wedge has been growing—one of the reasons the newest generation of suddenly-huge financial firms has more prop traders and fewer hedge funds is that it used to be viable to raise single-digit or low double-digit millions to launch a fund, but today the minimum is higher.¹

You could imagine a story where someone trading their own capital and finding adjacent strategies to what they've already done well eventually reaches the point that it's best for them to take outside capital. As long as their capacity growth exceeds their returns, that's eventually optimal. One problem with this is that it's hard to increase capacity, but a subtler problem is that firms that find themselves in this position tend to have a core business that prints money (it prints variable amounts, but its personnel costs vary a lot with gross trading revenue), so it's actually a decent borrower. One way to look at earning 2 & 20 is that it's incentive-aligned, but another way to look at it is that it's very expensive capital. If you earn a 22% return for your investors and charge them 2 & 20, you're getting about 6% of their AUM in fees that year, but another way to look at it is that you're paying 16% for their capital, which is a lot more than you'd pay for other kinds of capital. The most plausible ways for capacity-constrained firms to expand is to identify longer-scale strategies in existing asset classes (if you've found a good strategy betting on intraday correlations between stocks, you're in a good position to bet on intra-week, intra-month, and beyond; as you expand the return on capital tends to drop, but since your holding period is longer, you have more time to execute trades and won't have as big a market impact). So their incentive is to scale into slower-moving strategies by borrowing, not by letting other people invest alongside them. The usual situation with an entity that has multiple strategies with different return/capacity tradeoffs is that the low-capacity ones are constantly overflowing into the higher-capacity ones.²

Providers of capital are in a tough situation where they face constant adverse selection. For every strategy, they have to answer the question of why someone would let an outsider reap some of the profits, and, further, why they happen to be that outsider. If they back a winner, there's a good chance that they'll be told that they can't put in as much as they'd like, and may have to take some of their capital out. (The losers are naturally quite happy to have them reinvest. 2% of a sufficiently big number is itself a big enough number for many purposes.) There is still alpha in manager selection, but like other sources of alpha it's some combination of temporary, opportunistic, and based on personal networks. There were some very lucky investors who decided that a Midwestern insurance geek who slept with a Moody's Manual under his pillow could produce alpha, or that a mathematician with a theorem named after him could outsmart the median MBA. One way or another, that alpha decayed, and when limited partners got excess returns, it was from the usual sources: a mix of skill and luck, with no aggregate free lunch.

The question of why LPs can get good returns at all turns out to be more interesting if you reframe things so they're in the same business as the people they invest in, just at another layer of abstraction. And there's even some symmetry around which firms need capital and which don't; a building full of EUV machines is more expensive than a building full of accountants or consultants, so the professional services companies tend to stay more private while the asset-heavy companies have to go public. It comes up a lot:

• We've looked at hard-to-scale strategies that seed big companies.

• Sometimes, returns erode because everyone thinks the same way ($).

• Financial products inevitably get democratized.

• Some things that start as novel assets that can support unique strategies end up being just another asset class ($).

• The question of where assets should live is never settled. Consider the case of printers with a subscription attached ($).

Put Your Predictions to Work

Trade on real-world events you already follow, from elections and inflation to sports, tech, and more. Choose “Yes” or “No” based on what you think will happen and see how your prediction plays out.

Pick your market and start trading what’s next.

Trade $25, Get Up to $500

_{Bonus credit varies from $15 to $500. Terms apply.}

---

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Subscribe to receive your referral link

---

Join the discussion!