Dr. Wallace's Favorite Mental Models

The Keynesian Beauty Contest

Why winning so often means guessing what everyone else will guess — not picking what you personally believe is best — and why that single twist, first described in a 1936 economics textbook, explains bubbles, trends, and crowds that look irrational from the outside but are being perfectly rational about the wrong target.

Behavioral finance Game theory & coordination Market psychology ~10 min read

1Fast Overview

The idea in one line
In a Keynesian beauty contest, the winning move isn't to pick the option you personally think is best — it's to predict what the average of everyone else will pick. Since everyone else is running that same calculation about you, the "correct" answer can spiral away from anyone's genuine opinion into layers of guesses about other people's guesses.
Where it comes from
John Maynard Keynes introduced the idea in Chapter 12 of his 1936 book The General Theory of Employment, Interest and Money, comparing professional investing to a newspaper competition of the era in which entrants picked the six prettiest faces from about 100 photographs. The prize went not to whoever picked objectively the prettiest faces, but to whoever's picks most closely matched the average picks of every other entrant — so the smart strategy was never "pick who I find prettiest," but "guess who everyone else will pick."
The twist most people miss
Keynes pushed the logic further than most retellings do. He described investors who don't stop at guessing what average opinion will be, but try to anticipate what average opinion expects average opinion to be — reaching what he called the "third degree" — and noted that some practitioners go even further, into "fourth, fifth and higher degrees."
Why it matters at work
Any setting where value or correctness is set by collective perception rather than fixed fact — pricing a speculative asset, betting on a trend, picking a candidate a committee will approve of, predicting what content will go viral — runs on this same beauty-contest logic, whether or not anyone in the room ever names it that.

2Mental Model Compass

Field of originEconomics & behavioral finance (John Maynard Keynes, 1936)
Core mechanismSuccess is defined by matching what others will choose, not by matching your own honest assessment — so rational players reason about other players' reasoning, often several layers deep
Best used forSpeculative markets, trend forecasting, content virality, and any committee or crowd decision where "will others agree with this?" matters more than "is this objectively right?"
Watch out forConfusing it with simply copying the crowd (social proof) or with a stable mutual best response (Nash equilibrium) — a beauty contest is specifically about chasing a moving, self-referential target

3Learn It: Step by Step

Step 1 — The newspaper competition Keynes actually described+
In Chapter 12 of The General Theory of Employment, Interest and Money (1936), Keynes compared professional investment to a competition that ran in British newspapers of the time: entrants were shown roughly 100 photographs of faces and asked to pick the six prettiest. The prize went not to whoever picked the objectively prettiest faces, but to whoever's six choices came closest to matching the average choices made by every other entrant combined. Keynes used this to argue that professional investing often works the same way — the reward doesn't go to whoever best judges an asset's true underlying value, but to whoever best anticipates what other investors are about to believe and buy.
Step 2 — Why "pick your own favorite" is the losing strategy+
Picking the faces you personally find prettiest is what Keynes would call a naive, first-level strategy — and in this particular contest, it's usually a losing one, because the prize is tied to the group's average, not to any individual's taste. A smarter entrant picks the faces they expect the average entrant to personally find prettiest — a second-level guess about other people's first-level opinions. The contest rewards shifting from "what do I think?" to "what will they think?"
Step 3 — The "third degree," and beyond+
Keynes didn't stop at two levels. In his own words: "We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. And there are some, I believe, who practise the fourth, fifth and higher degrees." In plain terms: Level 0 is "what do I think is pretty," Level 1 is "what will other people think is pretty," Level 2 is "what will other people think other people will think is pretty," and so on — a nested chain of "I think that you think that I think..." that can run as many layers deep as the players are willing to reason through.
Step 4 — From a parlor-game analogy to a theory of markets+
Keynes wasn't just describing a cute party game — he used it to explain why stock prices can detach from a company's actual fundamentals (its real earnings, assets, and prospects) for long stretches. If enough market participants are pricing an asset based on anticipated average opinion rather than fundamentals, the price reflects a chain of guesses about other people's guesses. That chain can inflate a long way above — or crash a long way below — any reasonable estimate of underlying value, and it can unwind with startling speed the moment opinions about opinions shift.
Step 5 — Tested in the lab: the "guess two-thirds of the average" game+
Economists have tested this reasoning directly with what's called a "p-beauty contest" game, formalized by economist Rosemarie Nagel in the 1990s: each participant privately picks a number from 0 to 100, and whoever picks closest to some fraction — commonly two-thirds — of the average number picked by the whole group wins. It's a controlled laboratory version of exactly the reasoning Keynes described. If everyone reasoned through infinite levels of "they'll guess that I'll guess…", the group should converge on zero. In practice, real participants typically reason only a few levels deep, so winning numbers in these experiments usually land well above zero, clustering wherever a few rounds of "guessing the guessers" settles — direct evidence that higher-order, beauty-contest-style reasoning is something real people actually do, and that most of them stop a handful of levels in.
Step 6 — The practical test: are you rating the thing, or rating the raters?+
Before concluding "this is good," "this is valuable," or "this is the right call," ask a sharper question: am I evaluating this based on my own independent judgment of its merits, or based on what I expect other people's judgments will be? Both are legitimate strategies depending on what you're actually trying to win — but quietly sliding from the first into the second, while believing you're still doing the first, is exactly where this model gets missed in the moment it matters most.

4Examples Across Different Worlds

Stock market speculation & meme stocks
A retail trader buys into a rapidly rising stock not because their own valuation model supports the price, but because they expect enough other traders to keep buying that the price will rise further before it falls — betting on the crowd's next move rather than on the company's fundamentals. (Illustrative, composite scenario.)
Fashion & trend forecasting
A retail buyer decides which color and silhouette to stock heavily for next season not based on personal taste, but based on a prediction of which trend other buyers across the industry will also bet on — because a trend only becomes profitable if enough other buyers independently guess the same thing. (Illustrative, composite scenario.)
Hiring-committee groupthink
A panelist who privately liked a candidate best quietly shifts their stated ranking toward whichever candidate they predict the rest of the committee — and the hiring manager's manager — will ultimately favor, voting on anticipated consensus rather than on their own independent assessment of who interviewed best. (Illustrative, composite scenario.)
Social media virality
A creator chooses a headline and thumbnail by predicting what a broad audience will click and share, not by what they personally find most compelling — and once a platform's algorithm detects early engagement, it amplifies the post further based on a prediction of what will keep performing, compounding one layer of guessing on top of another. (Illustrative, composite scenario.)
Art & NFT valuation
A collector in a thin, illiquid market buys a piece anticipating what future buyers will be willing to pay for it, rather than buying purely for personal aesthetic enjoyment — so the price reflects a chain of guesses about other collectors' future guesses about resale demand. (Illustrative, composite scenario.)
Political polling & strategic voting
A primary voter who personally prefers one candidate instead votes for whichever candidate they believe is "most electable" — their best guess at who other voters will support — in order to avoid what feels like a wasted vote on their genuine first choice. (Illustrative, composite scenario.)

5Worked Application

Situation: A five-person hiring panel finishes final interviews for a product manager role and splits 3-2 between two candidates. Before the debrief meeting, the panel chair privately asks each member for their honest recommendation.

Applying the model: One panelist, who personally preferred Candidate B, catches themselves starting to write "Candidate A" instead — not because their assessment of the interviews changed, but because they expect the hiring manager to favor Candidate A, and they'd rather be on the side everyone else lands on. Recognizing this as beauty-contest reasoning, they separate two questions cleanly: what does my own independent read of the interview evidence actually support, and — as a separate, labeled question — what do I think the group's consensus will end up being?

ApproachWhat it actually deliversWhere it goes wrong
Quietly vote for whichever candidate you predict the rest of the panel and the hiring manager will favorA recommendation that's unlikely to stand out or create friction in the debriefThe panel's final decision reflects a chain of mutual guessing about group opinion, not five independent reads of the same interviews — exactly the failure mode the model warns about
Submit your honest, independent recommendation, labeled clearly as your own assessment, separate from any guess about where the group will landA debrief built on genuinely independent signal, which is the entire point of having five separate interviewersRequires a norm where disagreement is safe to state plainly, which not every panel or culture actually provides
Submit your honest recommendation, and separately flag your prediction of where you think the group will land and whyBoth pieces of information reach the chair: the independent signal, and a useful early read on where consensus is formingOnly works if the chair (or whoever runs the debrief) actually treats the two as different things, rather than quietly collapsing them back into one number

This is a composite, illustrative scenario built to show the mechanism, not a specific published case study. Actual hiring processes vary widely by organization, role, and panel structure.

6Common Mistakes

Treating "popular" and "good" as the same thing. Winning a beauty contest means matching what others will pick, which can be entirely disconnected from whether the thing is actually good, true, or valuable by any independent standard.
Confusing it with plain social proof. Copying what other people are visibly already doing is one level of reasoning. A beauty contest is specifically about guessing what others will decide before you can observe their behavior, based on nested guesses about their guesses about you.
Stopping at one level deep without noticing everyone else is doing the same thing. Guessing "others will like X" is only level one. Other participants are very often reasoning about your guess at the same time, which is exactly how the recursive spiral gets started.
Assuming more levels of reasoning always wins. Lab experiments with the "guess two-thirds of the average" game show most real participants reason only a few levels deep. Someone who reasons five levels deep in a group that's only reasoning two levels deep can actually guess worse than someone who stopped earlier.
Using the model as an excuse to never form an independent judgment. Recognizing that a situation has beauty-contest dynamics is a reason to separate your own assessment from your prediction of the group — not a reason to stop forming your own assessment at all.
Mistaking a self-referential guessing game for a stable equilibrium. A Nash equilibrium is a settled state where no one benefits from switching. A beauty contest describes the live, ongoing process of guessing a target that keeps moving as everyone else keeps guessing too — they're related but not the same thing.

7When Not to Use It

When there's an objective, verifiable standard that doesn't depend on others' beliefs. For a factual claim, a safety-critical calculation, or a compliance requirement, anticipating group opinion is the wrong tool entirely — the answer doesn't change based on what anyone expects anyone else to think.
When the group is small and communicates openly. If you can simply ask people what they think and get an honest answer, there's often no need to infer layered beliefs about beliefs — direct communication is faster and more reliable than guessing.
When it becomes an excuse to never commit to a view. The model explains why collective-perception markets behave the way they do; it isn't a license to abandon independent judgment or hide behind "I'm just predicting what others will think."
When you're the one meant to set the standard, not chase it. A judge, auditor, regulator, or fact-checker's job is often specifically to resist anticipated consensus and hold an independent line, even when it's unpopular.

9Try It Yourself: Prompt Builder

Answer a few questions and build a ready-to-use prompt for checking whether you're judging something on its own merits, or quietly forecasting what other people will think of it.

Your generated prompt will appear here.
Copied!

10Save as Skill

Preview
Apply to My Problem

This is the portable skill version of this model — written so you can hand it to any AI assistant and get the same "am I judging this, or am I forecasting what others will judge?" check applied to whatever situation you're facing.


      
      
      Copied!
    

    

      
      
      
      
Your generated prompt will appear here.

      
      Copied!