The (Working) Theory of Weird Markets
Summary
Andrew Walker’s working thesis is that conventional investment strategies are increasingly dominated, leaving small investors to seek alpha in “weird” situations with few or no historical parallels. Pod shops can process credit-card data faster and with more leverage, quants can screen valuation factors systematically, and machine learning can execute rules or trend-following at scale. His conclusion: “If you want to outperform, if you want to generate alpha, if you want to be different, you have to do something weird.”
The market’s extraordinary rewards make finance Walker’s candidate for “the most competitive game in the world.” His signature comparison is competitive Rubik’s Cube: the championship winning time fell from 23 seconds in 1982 to 20 seconds in 2003 and approached five seconds by 2023; in 2019, the fastest competitor solved it with their feet in 17 seconds. If a $36,500 total prize pool can drive that improvement, finance—with Buffett’s net worth cited at $150 billion even after giving away a great deal—should relentlessly exhaust discoverable edges.
At the highest level of competition, winning strategies often look irrational to less-evolved players. Walker moves from the Fosbury Flop and fourth-down analytics to AI chess and poker: he quotes a description of machine chess as producing moves that can look “simply wrong,” while a former poker semipro told him an AI’s $500 bet into a $100 pot could resemble “a drunk uncle” throwing chips around. The investor implication is that an unfamiliar-looking process is not necessarily mistaken; it may reflect a game whose optimal strategy has changed.
Weirdness matters because highly optimized systems can remain fragile outside the conditions they know. Walker recalls chess players making apparently bad openings to take a computer “out of the book,” and StarCraft AI that was “the best player in the world unless there was even a minor surprise.” He argues that fundamental investors therefore need surprises and fat tails—not merely better execution of the same historical-data playbook.
The strongest examples, in Walker’s framing, combine a nonhistorical catalyst with fundamental thinking that still requires judgment. AI-driven electricity demand was not in historical data four years earlier, yet recognizing power as an AI constraint could make a fortune; spin-offs similarly bring forced selling, limited histories, new management teams, and sometimes liquidity constraints for pod shops. These are places where “the future” is not already cleanly modeled.
Walker treats WBD and unusual management incentives as possible specimens, but openly questions whether his evidence is strong enough. He is long Warner Bros. Discovery and sees its rare bidding war—Ellisons personally guaranteeing an equity deal and saying the bid was not best and final, with Netflix, the Trump administration, and the Warner Bros. board involved—as an N-of-one event. Yet repeatable 8-K signals such as a manager taking all of the next five years of equity compensation in the current year may already be getting “picked over,” creating both opportunity and risk. His honest caveat is that difficulty finding examples might mean “the evidence doesn’t back that up.” He is presenting the theory as a rough draft for feedback because it is central to his investing and annual work for 2026.
Deep dive
1. Alpha migrates toward situations the standard playbook cannot classify
Walker’s theory begins with a blunt premise: “The stock market is the most competitive game in the world.” As capital, technology, and specialization compound, traditional strategies are increasingly dominated by firms built to execute them faster, more systematically, and with greater leverage.
Credit-card data belongs to pod shops; simple deep-value screens belong to quantitative models; general rules and trend-following increasingly belong to machine learning. If the underlying quarterly edge is only 2%, a pod shop may lever it into 8%, while an unlevered small investor cannot justify the same work.
The proposed escape route is “weird”: N-of-one situations without useful precedents. Walker’s favorite historical specimen is Twitter, where the world’s richest individual decided on an acquisition on a whim, then tried to back out and claimed an MAE—one of only roughly five publicly traded MAE cases in history. The past cases he names involved the global financial crisis causing banks to fail or an individual company being hit like by a meteor, rather than the world’s richest man getting cold feet over price.
2. Finance’s incentives ensure that ordinary advantages get competed away
Walker uses measurable sports progression to show what sustained competition does. In baseball, only 11 pitchers averaged a 95 mph fastball in 2007; by 2025, 300 did. What once signaled elite velocity became commonplace in under two decades.
Rubik’s Cube supplies the sharper analogy. Championship times improved only from 23 seconds in 1982 to 20 seconds in 2003, then approached five seconds by 2023 once regular competitions organized and intensified the field.
The punch line is almost absurd: in 2019, the final year of the feet category, the fastest competitor solved the cube in 17 seconds—faster with their feet than the 2003 champion had managed with their hands. The total championship prize pool was only $36,500, with $5,000 for the premier 3×3 event.
Finance offers vastly larger rewards and hundreds of years of competition, from the rumored Rothschild racing-pigeon network used to get news from Waterloo faster than peers and Reuters—named for a man who used pigeons to bridge a telegraph gap—to firms spending hundreds of millions for milliseconds or microseconds. Walker’s challenge: given those incentives, “every edge is sought out and taken.”
3. Mature games reward strategies that initially look insane
The Fosbury Flop is Walker’s cleanest physical example. Before Dick Fosbury won Olympic gold in 1968 by clearing the bar backward, telling an elite high jumper to jump that way would have sounded ridiculous; afterward, the once-insane movement became the dominant technique.
Strategy evolves similarly. Football teams once avoided fourth-down attempts because failure could get a coach fired; analytics now favor going for fourth-and-four around the 40 or fourth-and-goal from the three. Walker invokes Moneyball’s example of hiring a first baseman who could not really hit but took many walks, illustrating that on-base percentage mattered more than batting average.
AI pushes the pattern further. Walker quotes an observer describing AI chess as producing moves humans and human-trained machines found counterintuitive, sometimes “simply wrong,” with games that look like “chess from another dimension.” Poker systems also overbet far more often than humans historically did. A former semipro told Walker that, without knowing the source, one might mistake the AI for “a drunk uncle” recklessly throwing chips around.
4. Surprise is the small investor’s remaining terrain
Walker recalls the Deep Blue–Garry Kasparov example and says Kasparov used apparently insane opening moves to take the AI “out of the book,” sacrificing local optimality to create unfamiliarity.
StarCraft AI showed the same limitation under controlled conditions. It could beat top professionals on a particular map and setup, but altered conditions broke its advantage: “The AI was the best player in the world unless there was even a minor surprise.”
AI-related power demand illustrates a tradeable fat tail. Four years earlier, historical data could not reveal the coming load, and often even future demand was not visible; the valuable insight was that widespread AI deployment “is going to consume a ton of power.”
Spin-offs offer a more structural version: unwanted shares create forced selling, standalone histories are sparse, and new management teams may offer human views AI will not have. Pod shops can participate, Walker concedes, but liquidity constraints may leave openings for smaller fundamental investors.
5. The theory is strongest as a framework and unfinished as a market map
Walker identifies WBD as a current N-of-one and discloses that he is long. Few bidding wars exist as precedents, while this one combines Ellisons personally guaranteeing an equity deal and saying their bid was not best and final, Netflix, the Trump administration, the Warner Bros. board, and numerous “soft considerations.”
He is less certain about turning the framework into a repeatable opportunity set. Unusual 8-K compensation changes—such as a manager taking all of the next five years of equity compensation in the current year as a reward—can be interesting, but AI may eventually learn that pattern too. Walker sees both opportunity and risk.
Walker says the first two-thirds—the overview, theory, and sports parallels—are very good, but he has writer’s block when translating the framework into specific market examples. He is airing this rough draft to solicit examples and objections before building his 2026 annual work around it; the difficulty may mean he needs better examples, or that “the evidence doesn’t back that up.”