There's More Than One Efficient Market Paradox

Plus! Reciprocal Tariffs; AI Purgatory; Payments and Credit; LinkedInFluencers; Efficient Markets

Byrne Hobart Byrne Hobart

In this issue:

audio-thumbnail
The Diff March 24th 2025
0:00
/797.126531

There's More Than One Efficient Market Paradox

The debate about efficient markets rests on two pillars:

  1. In 1991, Bill Sharpe pointed out that mathematically, all of us in the aggregate deliver exactly the return of the market, less transaction costs, so on average, active management has to lose compared to buy-and-hold-forever.
  2. Perfectly efficient markets at the level of asset prices involve massive market inefficiencies in terms of how people allocate time and capital. If the market has reached almost perfect pricing, it would take a titanic investment in time and resources to identify and execute The Last Arbitrage to collect one last penny of discrepancy between where an asset trades and what it's worth. That would be a terrible return on investment. At some point, broad market efficiency—rational behavior in light of incentives—requires that some assets be mispriced, but only to a degree that isn't worth exploring or exploiting by the people equipped to do so.

We can extend this further. So much further.

When a market anomaly shows up, the worst possible question to ask is "what's the fastest way for me to exploit this?" Instead, the first thing to do is to steelman it as aggressively as possible, and try to find any way you can to rationalize that such an anomaly would exist. Do stocks rise on Mondays? Well, maybe that means savvy investors have learned through long experience that it's a good idea to take off risk before the weekend, and even if this approach loses money on average, maybe the one or two Mondays a decade where the market plummets at the open make it a winning strategy because the savvy hedgers are better-positioned to make the right trades within that set.[1] Sometimes, a perceived inefficiency is just measurement error: heavily-shorted stocks reliably underperform the market—until you account for borrow costs (and especially if you account for the fact that if you're shorting them, there's a good chance that your shorts will all rally on the same day your longs are underperforming). There's even meta-efficiency at work in otherwise ridiculous things like gambling on 0DTE options or flipping meme stocks: converting money into fun is a legitimate economic activity, though there are prudent guardrails on it just in case someone finds that getting a steady amount of fun requires burning an excessive number of dollars.

These all flex the notion of efficiency a bit, but it's important to enumerate them because they illustrate something annoying about the question of market efficiency: the more precisely you specify the definition, and the more carefully you enumerate all of the rational explanations for seemingly irrational activities, the more you're describing a model of reality so complicated that it's impossible to say whether it's 50% or 90% or 1-ε efficient.

And even this analysis is playing on easy mode. If you look at products that are traded on a market, you can measure their correlations, and (with some uncertainty) estimate how liquid they are. So, in that kind of market, it makes sense to say things like: Gold has poorer real returns than most asset classes, but when real rates are low and geopolitical uncertainty is high, it'll rise. But what about private markets? It's even harder for them to get efficient, and since there's a smaller set of transactions, it's hard to have statistical confidence that anyone's any good, with the exception of incredibly prolific investors like Y Combinator.[2]

But YC illustrates the problems of market efficiency in another way. Part of what YC does is the same thing any early investor does: connect companies with capital and advice in exchange for equity. But it's also great at taking a company from being a small group with an idea to a business with metrics that will get VCs excited. The markup here is hard to predict, but it's very real; they were able to industrialize what used to be a function performed with much more variability by traditional angels.

This is a great business, both in the sense that it's good for the world that YC exists and that it's probably very good to be a partner. But it's also a business, not the sort of asset-light thing you can operate purely with a tiny amount of human capital.[3] The companies with the most valuable intangible assets are the ones that started building those assets long before there was a company—I personally heard about Y Combinator because Paul Graham's essays frequently showed up on Slashdot, and this seems like a not-uncommon experience. And YC was started as an experiment, which happened to work. This is one reason it's been so hard to build another one: empirically, the way you build a YC is to do something else and iterate.

YC has a big enough sample size that it's fair to say that they have skill at some combination of selecting the right founders and getting access to the best deals, the two things early-stage venture investors need to excel at. But the time lag in venture means that statements about skill are mostly retrospective. You can use the present tense for short-term strategies, and the model and franchise mean that YC still has an edge, but a definitive answer about whether or not they're good is always something you get years after you could use it.

And that's even more true for other investors pursuing the same kinds of companies, with the key difference that these other investors have an even smaller sample size and thus an even wider confidence interval.[4] Since it's easier to put up good numbers on a small base of capital than a big one, investment managers will generally put up the best parts of their track record at the point in their career when they didn't have much of a track record, and then, due to luck's regression to the mean and the difficulty of compounding larger amounts of money, results will recede.

But, suppose that after all that, they've still demonstrated skill. What now? Meta-efficiency strikes again: if someone is good at raising capital, and prudent enough not to raise all the money that's available to them, they're choosing who their limited partners are. And one of the best ways for an LP to differentiate themselves in a fund that can choose its LPs is to have been an LP back when that wasn't the case. The fortunate investors who are able to put money into recent Benchmark and USV funds are partly earning an implicit residual on the first investments they made in those funds back when they were more capital- than capacity-constrained.[5] The market in investing talent for identifying inefficiencies at scale is more efficient than the market that gives rise to those inefficiencies, because the managers who exploit them are in the business of making their strategy legible to outside investors without losing their edge.

Yet another problem with measuring efficiency is that there are feedback loops that delay markets from getting efficient. In some fields, crowding just slowly bleeds away returns, but if there's a lag, crowding can actually lead to negative returns even in sensible strategies, or even in strategies that basically have to work out in the long run, like futures basis trades. Sometimes, the reason there's still some edge in something is because the people best positioned to exploit it are also plugged in enough to know that there's already a lot of money flowing into it. The lag problem shows up at the level of investments in operating companies, too, at many different timescales: during the early AI wrapper boom, there were lots of times where you could accurately think "I wish I'd had that idea a week ago." And at the far end of the scale, Intel can look back at their capital expenditures decisions of the last decade and wish that they'd made different strategic/company culture decisions a decade before that. In the very long run, efficiency is the right bet, but in some cases we don't even know what the relevant scale is.

So you might end up thinking of market efficiency as less of a quantifiable claim, and more of a directional tendency. It has to be the case that, as long as it's possible to make better or worse decisions, and as long as the payoff from better decisions includes more money and influence, resources are being continuously allocated towards people who will use them well. But that's an incredibly noisy process, with cycles, epicycles, and plenty of measurement error along the way. It's a good premise to assume, in the face of a perceived inefficiency, that smart people have spent more time than you looking into it, and that it exists for a reason, but that's not an iron law. Sometimes, nobody asked the right question, and sometimes, everyone was too aggressive in applying exactly that heuristic. At every level where you can describe a process by which the market gets more or less efficient, there's a meta-level where exactly the opposite is happening.

  1. This is one reason it's long term return maximizing for the best investors to operate with slightly lower volatility than they can usually tolerate, to keep a bigger liquidity buffer than they generally need, and sometimes to just pay for tail-risk insurance and spend more time than usual worrying about the credibility of very large counterparties: if they get all of that right, then on a handful of occasions they'll be the only buyer in town, and will be able to fully exploit those rare occasions. Opportunities like John Malone's takeover of SiriusXM just days before the company ran out of money ($, WSJ) come along only so often. ↩︎

  2. Other firms better-known for their later investments still have a strong track record in this, like Sequoia,which has made ~1,500 seed investments (these days often through their Arc Program) and ~2,800 total investments. Therefore, they may be the highest signal early stage investor in the sense that they make lots of winning seed investments, but also make the correct decision to double, or triple down the winners in later rounds. The total’s still pale in comparison to YC though, which has made 7,600+ investments to date. ↩︎

  3. In fact, in Sam Altman’s recent Stratechery interview, he claims the moment Paul Graham convinced him to join as President was when he said, “If you run YC, it’s closer to running a company than being an investor.” Sam said that turned out to be true. That may have been an understatement, because in many ways, YC served as a training ground for the CEO that now runs arguably what may be the most consequential company in the world. ↩︎

  4. Annoyingly enough, of the two big axes of VC skill, access and selection, access is the one that can be underwritten upfront: just see if they got into competitive deals! But the more you over-index on that, the more what you're really doing is selecting for someone who's consensus and charismatic. And if they know what's good for them, what they're optimizing for is management fees rather than expected carry. ↩︎

  5. In public markets, this reaches an extreme where the higher a firm's sharpe ratio, the faster it tends to return outside capital (or the more reluctant it is to raise money in the first place: 1.5 & 15 isn't necessarily worth the cost of disclosing enough information about strategy and returns to make investors feel comfortable). This seems to happen less often in private company investing, though there are some privately-held conglomerates that could accurately be described as a PE firm without any LPs. One reason it might be less common in venture than in fields like short-term quantitative strategies or private equity is that it's harder to control the downside; "fully-invested" in early-stage venture might mean a portfolio that allocates about 20% to actual venture, with most of the remaining 80% in treasury notes. You'd basically want to set up a portfolio like this such that a decade of underperformance wouldn't materially alter the strategy, because with a small sample size and a long lag before seeing returns, that's likely to happen at some point. ↩︎

Diff Jobs

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

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.

Find a Role

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.

Find Talent with Diff Jobs

Elsewhere

Reciprocal Tariffs

Big Tech remains an easy target for retaliatory tariffs; if a foreign politician is worried that the US will keep electing protectionists, a good way to deal with that is to retaliate against a group of companies that recently shifted slightly in favor of that same protectionist. But that also means they're a good target for preemptive surrender: the UK is considering cutting its digital services tax, which, given the industries it targets, is in effect a tariff on US ad platform exports. Because it's still unclear just how aggressive tariffs will end up being and how long it will take for them to stabilize, the important point is not the direct impact of this, but what kind of impression it makes on Trump.

AI Purgatory

Cloudflare has a novel way to deal with scrapers: when it detects them, it can create an endless maze of AI-generated pages, full of links to other such pages, which the scraper will follow. This serves several purposes at once:

  1. It means they're using more resources.
  2. It's harder for them to control the quality of whatever data they're scraping.
  3. Every additional click adds to the sample of data about scraper behavior, making the next scraper easier to spot.

We should expect to see more stories like this, because while AI has made it possible to scale some kinds of annoying behavior, it also creates adaptable tools for dealing with exactly that problem.

Payments and Credit

Klarna and DoorDash are partnering to allow people to finance their orders with buy-now-pay-later loans. There are two forces moving in opposite directions here:

  1. For the typical customer who should be regularly using DoorDash, i.e. a busy professional who achieves gains from trade by outsourcing the task of making or picking up lunch, it's presumably quite common to finance their purchase with a different form of consumer loan that also has zero interest by default: they'd put the purchase on a credit card and pay for it when they next pay their statement balance. So there's nothing new about borrowing money to finance lunch delivery.
  2. On the other hand, the marginal BNPL borrower for a small-dollar purchase is probably less creditworthy than these customers. And DoorDash will consider additional payment options only if the increase in purchases (less the attendant fees) exceeds lost sales from a more cluttered checkout. A low credit score is a measure of impulsiveness, and delivery apps are one way to monetize that exact trait.

One reason Klarna may like this is that it's an indirect way to expand their market: some higher-rated customers will try this either because it's novel or because they prefer this kind of payemnt plan, and some of the users will be closer to the core BNPL market, but feeling temporarily flush. So Klarna is paying to get a bit more underwriting data and a bit more name recognition, even if it does mean that sometimes they'll end up writing off a sushi-backed loan as unrecoverable.

LinkedInFluencers

There's a form of weak media arbitrage where, within a given serious category, it's always a winning move to be less serious. In education software, for example, one of the biggest successes is Duolingo, which explicitly prioritizes engagement and retention over learning efficiency. On LinkedIn, where the default update is fairly sober and professional, it simply doesn't take that much effort to be the most interesting post on someone's feed, so there are LinkedIn influencers who have developed followings of tens or hundreds of thousands of people, and are earning money from this that's comparable to what they'd get at their day jobs ($, WSJ). It's a tricky business, because LinkedIn will always try to balance between quantity of engagement and quality of engagement. And in general, power users will continuously get better at optimizing for quantity on their own, leaving the platform with the job of periodically changing the algorithm to focus on quality again.

Disclosure: long LinkedIn parent company MSFT.

Efficient Markets

There will probably not be any CEOs paid over $100m in 2024, for the first time in over a decade ($, WSJ). CEO pay metrics can be noisy when they're tied to big one-time grants that actually pay out over several years. But even if those are, in fact, not as extreme as they look, they still make CEOs a target for the media, politicians, and, in the case of Elon Musk, the courts.

One other reason for this is that pay packages for the CEOs of private equity-owned companies are much less visible, at least until the company in question goes public again and discloses how much stock executives have. But it's usually visible retrospectively: when Standard Aero went public last year, for example, their CEO's pay for the previous year was $2.7m. But he also had shares in the company valued at just under $100m at the time of the IPO, which, since he was hired eleven years before, works out to about $9m/year in compensation. From a corporate governance standpoint, and even from a fairness standpoint, it's probably better for CEOs to make most of their money from increasing the value of the business they run, rather than getting what is, in effect, an advance based on the likelihood that they'll do this in the future.