In this issue:
- The Surprisingly Weak IPO-Housing Effect—Whenever there's a big IPO, there are news stories about how the money gets spent, particularly how it affects real estate prices. But if you look at historical IPOs, the effect looks like it's close to zero. There are a few plausible reasons why.
- Bottlenecks—Elon Musk is partly diversifying in a prudent way, and partly underwriting ambitious capex.
- More Bottlenecks—The weird political realignment of the last decade continues. The latest symptom: unions siding with big tech.
- Disjoint Models—The harder it is for models to copy each other, the stronger the incentive for different use cases.
- Aligning Incentives—OpenAI wants to get paid for success, partly so its customers will tell them, for free, exactly what success looks like.
- Momentum—The case for momentum as a wonky barometer of market efficiency.
Talk to this post on Read.Haus.
The Surprisingly Weak IPO-Housing Effect
There's a tradition, ahead of big IPOs, to do a news story on how all that money will get spent. This story got written about the Bay Area in 2012, Huangzhou in 2014 LA in 2017, Boston in 2021, etc. It turns out that, if you need to write such an article, it is incredibly easy to get realtors to give you critical insights into their business, such as: if you don't work at the local company that's about to go public, you should buy a house immediately, and if you do, you and your coworkers are all of course going to buy houses.[1]
It makes intuitive sense that real estate prices would go up after a big IPO. Americans like to own houses, but are a little more cautious about letting anyone build any, and someone who worked for a company that recently IPOed has a very wonky portfolio, with most of their net worth in a single volatile asset whose price is tightly correlated to their compensation. Even if they don't get all that much diversification from buying a house—there aren't many worlds where Anthropic shares drop 80% in their first year as a public company and people are glad they bought a condo in San Francisco beforehand—but homeowners aren't usually starting with mean-variance optimization when they buy a house. They're just doing what people normally do when they're flush.
It's a schlep to look at IPOs and acquisitions, try to figure out where most of the stock-based compensation was accrued, and then use that to track local real estate. Fortunately, for problems like this Codex is basically Schlep-B-Gone, and was able to tear through a few hundred S-1s and find cases where lots of people were suddenly richer relative to the size of their real estate market (on the supply side). This includes the obvious candidates mentioned above, but also cases that are more driven by the denominator: ExactTarget was not a big tech acquisition generally, but was a huge Indianapolis tech deal. Pluralsight was not a huge IPO, but was a big deal for the city of Draper, Utah.
And if you do this, you find... basically zero effect! If you look at big liquidity events with geographically-concentrated beneficiaries, and compare home price changes in the relevant zip codes to zip codes with similar economic characteristics, the most expensive third of housing as tracked by Zillow rises 0.26 percentage points more than it did in the year before, with a confidence interval that overlaps with zero. The confidence intervals are also wide enough that there isn't a meaningful difference between the impacts at different pricing tiers.
One reason for this is mere scale. When Facebook went public, they recognized a $10.5bn charge to account for the value of restricted stock units that were unlocked due to the IPO. The effective marginal tax rate for these earners was about 45%, so that's $5.8bn. But the Bay Area's GDP that year was around $600bn, and home sales were about $50bn.[2] So even with pretty aggressive assumptions about how much of their windfall people spent on homes, it's hard for this liquidity event to do more than slightly nudge prices. And Facebook was a big deal on these terms; other Bay Area IPOs were smaller in dollar terms, and the later ones had a higher denominator. Companies outside of the Bay tend to have less broadly-distributed equity; outside of tech, high stock-based compensation tends to mean options for executives, not for line employees. This kind of consideration also shows up in cases like Qualtrics: enterprise companies have headcount that skews to salespeople, whose variable compensation tilts to commissions rather than equity.
The Facebook IPO did coincide with an increase in Bay Area housing prices: they rose ~4 points more than matched zip codes in the year before, and 14 points in the year after. But that's the most extreme case, and they weren't the only source of demand for housing in that area at that time.
There are two other factors to consider in measuring the impact of IPOs on housing:
- Both sides of the transaction were aware of the IPO, which smooths out the price impact. If you wanted to take advantage of the IPO as a home seller, one thing you'd do is delay putting a home on the market until you had more buyers, so some supply would shift back—in other words, though an IPO increases demand for homes, the prospect of an IPO shifts the supply of available homes to line up with that demand.
- Some workers were cash-constrained, but cash salaries for pre-IPO tech companies had already started rising by this point, and workers could do some consumption-smoothing: live a little beyond their cash means for a while, with the expectation that they'd be able to afford it later. So this, too, spreads out the impact. Granted, it's a lot harder to make a down payment than to pay the deposit for an apartment, but someone who switches from renting to owning is also removing someone from the pool of renters, so they're shifting how real estate demand gets expressed rather than actually creating new demand.[3]
Land is a classic factor of production, and for Bay Area tech companies, housing near the office is also a factor. Which means that their employees are running a kind of optimization process: they might take a hit to first-year consumption by moving somewhere expensive in order to take a job whose compensation skews towards equity, but they're also redirecting some local housing resources to create the value that that equity represents. For example, in a way, the Facebook IPO's impact on Bay Area housing started back in the summer of 2005, when the early team rented a house so they could spend the summer working on their company. And the collective impact of employing tens of thousands of people in the Bay Area today is bigger than suddenly minting a few thousand millionaires at the IPO.
Still, it's a little suspicious that the impact of IPOs on housing prices is, in the aggregate, statistically hard to distinguish from zero, especially when one of the big ones is pushing the number upward. There's an answer that rescues the causal story but makes it harder to apply: companies have discretion about IPO timing. Their bankers will also tell them that if they don't want to go public now, they'd better be willing to wait a few years. IPOs are one of the mechanisms by which the market balances the supply and demand for risky equities.
So, when the realtors talk about how everyone's buying houses and you don't want to get left behind, they're right about one factor but wrong about beta. People do respond to IPOs by rebalancing out of their employer's stock into housing, in addition to other assets. But those IPOs happen in good years, and retrospectively, big years for IPOs are followed by bad years for everything that correlates with those companies' fortunes.
Disclosure: long META.
These articles talk about other kinds of spending, too: nice cars, big vacations, art, charitable donations, etc. But the local supply of those categories is elastic, in the sense that if there's a big spike in one city's demand for Ferarris or Cezannes, they'll get shipped there. (If there's a scarce luxury good that windfall recipients are likely to want to buy—maybe no Anthropic employee's home is complete without a Calder sculpture or an Altair 8800, prices for those will respond. But that's a collection of anecdotes that will all sample from different distributions.) So home prices are where you'd expect to see a wealth effect. ↩︎
We have to handwave a bit here, because we have some numbers on total sales at different price levels, but not a mean sale price. ↩︎
In general, many housing debates work better if you think in terms of occupant-months rather than strictly categorizing people into renters and buyers. One thing this tells you, for example, is that effective yield management that keeps properties occupied more of the time is economically equivalent to adding inventory. For example, the big institutional single-family landlords tend to have a higher occupancy rate than single-family homes in the US, albeit by a small margin. (And they're not directly comparable to national averages, since those landlords liked to buy in markets they expected to tighten over time.) If a large-scale buyer adds a point of occupancy to their portfolio compared to a less optimal buyer, this means we get the equivalent of one new home for every hundred they buy. It's an abstract home made out of adroit management of lead-generation systems and frequently-updated pricing algorithms, but that doesn't make a big difference to someone who's looking for a place to live. ↩︎
You're on the free list for The Diff! Last week, paying subscribers read about how non-AI companies want to get control of the flow of data, and to steer towards places that generate more training data ($); Meta's effort to rewrite the rules of teen social media use ($); and AI agents as eusocial organisms subject to evolutionary pressure ($), featuring an extended digression about whether or not evolution has risk premia. Upgrade today for full access.
Diff Jobs
Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:
- Startup building the one API to power every strategy and capital allocation decision in the economy is looking for a co-founder / CTO that’s excited to solve the series of technical grand challenges (and inevitable schleps) involved in making this a reality. If you’re looking to build the economic sensor infrastructure underpinning the next leg of productivity growth, please reach out. (London, NYC)
- A top prop trading firm is looking for an intellectually curious, mathy generalist to work on projects spanning business strategy, technology, and markets. If you’ve got the quantitative chops of a prop shop trader or ML researcher, appreciate that operational alpha decays more slowly than trading alpha, and want to work on entrepreneurial projects that drive the former, this one is for you. (NYC)
- A premier prop trading firm is looking for someone to turn its prime brokerage partnerships into something of a solved, alpha-generating optimization problem. If you’d enjoy thinking through how to price the optionality of having an extra $1bn available when a liquidity-constrained opportunity shows up and continuously mapping the term structure of financing costs across PBs to ensure that $1B shows up at a low cost of capital, this one is for you. No formal background in finance is necessary, if you have a track record of bringing a systems thinking / analytical mindset to situations where success is traditionally categorized by relationship management, please reach out. (NYC)
- High-growth startup building dev tools for wrangling and debugging complex codebases is looking for someone who can personally execute the SaaS bear case: review the third-party software they use and figure out what to keep, what to drop, and what to implement in-house. (SF, DC)
- Series A startup that powers 2 of the 3 frontier labs’ coding agents with the highest quality SFT and RLVR data pipelines is looking for growth/ops folks to help customers improve the underlying intelligence and usefulness of their models by scaling data quality and quantity. If you read axRiv, but also love playing strategy games, this one is for you. (SF)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.
If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.
And: we're now actively deploying capital into early-stage companies through Anomaly. Our focus is on defense, logistics, robotics, and energy. If you'd like to chat, please reach out.
Elsewhere
Bottlenecks
Elon Musk is one of three people, including Ilya Sutskever and Mustafa Suleyman, who have co-founded more than one frontier lab. (For whatever reason, this seems more common in general-purpose technologies: both autos and semiconductors had cases where one person co-founded multiple significant companies in the industry.) Since the early OpenAI days, as AI has gotten bigger, it's been more bottlenecked by infrastructure. And Musk's biggest comparative advantage is finishing ambitious hardware projects early. So SpaceX is manufacturing blades for gas turbines ($, The Information), the most bottlenecked part of the supply chain. Elon telegraphed that he may do this in his interview with Dwarkesh and John Collison six months ago: this is partly the kind of problem SpaceX is operationally excellent at, and thus a normal capital allocation decision for an industrial conglomerate. But in another sense, it's a way for Musk to reiterate that he really does intend to be a material part of the AI capex buildout, and is thinking many tens of billions of dollars ahead when he decides which links in the supply chain need the most investment.
More Bottlenecks
Datacenters are getting less politically popular, but that just makes them more political, and it turns out that unions are pretty gung-ho about all of the construction jobs they create ($, WSJ). Unions' economics are exactly what PE firms love to own: a monopoly on something that's a small component of overall costs but essential to the final product. Which is a good metaphor for politics in a world where more and more of the output is produced by AI: various groups will find themselves in a good negotiating position with AI companies, partly because of regulations that didn't conceive anything like AI when they were written. As AI's share of GDP grows, those bottlenecks experience a kind of Baumol Effect where they capture more of the resulting growth over time. AI is redistributing some wealth, but in a completely unpredictable way.
Disjoint Models
OpenAI plans to drop support for Cursor, saying that Elon Musk has previously failed to adhere to contracts so they can't be confident . They don't explicitly say that xAI is distilling their models, but Cursor would be a great way to do that, and also a great way to deploy the results. As more datasets get locked down by labs who have different levels of distribution for different use cases, models should start to differentiate, too. For now, the data they've been trained on has had enough overlap that there aren't huge differences in model capabilities across domains—the smart ones are typically a little smarter at everything. But over time, model specialization creates a feedback loop, and the less distillation happens, the more that stays contained in particular models.
Aligning Incentives
OpenAI is considerin goutcome-based pricing ($, The Information), where customers pay for results rather than tokens. This kind of setup has intuitive appeal, but it's rare because historically it's been hard to evaluate results in a way that both sides agree to. Now, though, it's possible to build AI oracles that each side agrees to trust (at least, they agree upfront that they'll trust them). For OpenAI, part of the value of this is in the negotiation, not the outcome: the better a sense they have of what outcomes their customers want, the more they can build towards them.
Momentum
The momentum factor is on track for its worst quarter in a quarter-century ($, WSJ), after the violent unwind of the memory trade last month. It's interesting to consider momentum as a proxy for a certain kind of market efficiency: if stocks' performance in the recent past is correlated to their performance in the near future, one thing it implies is that there's easy money in chasing winners. But it also implies that those winners were well-chosen, and that the stocks people have been bidding up are the right ones to bet on. Whereas if betting against momentum worked, it would imply that market participants are systematically miscalibrated. The easier it is to make money with momentum signals, the noisier those signals get—which is true whether the people pushing momentum are doing it in a systematic way or just trading the same fundamental theses that other people found a few months or quarters earlier. So it's a very annoying metric to track: it's good for the market's health when it works, but a warning sign if it's the main trade that works.