What you're seeing
A population of agents, each holding some wealth. They all start exactly equal — one unit each. Then they trade, over and over. Each trade is between two of them, chosen at random, and it is scrupulously fair: they stake an amount and flip an even coin for it. There is no cheating, no skill, no advantage — the expected gain of each trade is zero for both parties. And yet, as it runs, the wealth does not stay spread out and it does not jitter around equality. It condenses: almost the entire field goes dark while a handful of cells blaze brighter and brighter, until — if you let it run with no redistribution — essentially one agent ends up holding everything and everyone else is left near zero. The Gini coefficient, a standard measure of inequality where 0 is perfect equality and 1 is one person owning all, marches steadily toward 1.
The rule
This is the yard-sale model (a "kinetic exchange" model from econophysics). One transaction:
- Pick two distinct agents i and j at random.
- The amount at stake is a fraction f of the poorer one's wealth:
Δ = f · min(wi, wj). (Staking the poorer party's share keeps every bet affordable to both — nobody can lose more than they have.) - Flip a fair coin: with probability ½, agent i wins Δ from j; otherwise j wins Δ from i.
That is the entire model. Every trade is symmetric, zero-sum, and fair — the two agents are interchangeable and neither is favored. Total wealth never changes; money only moves between agents. Run many such transactions per tick and watch the distribution evolve.
The redistribution knob adds one more step. Each tick, a small fraction τ of every agent's wealth is collected into a common pot and handed back equally to everyone — a flat wealth tax with a flat rebate. It conserves total wealth (it just moves money around), but it continually nudges the distribution back toward the mean.
Why it matters
This is one of the most counter-intuitive results in econophysics: fairness at the level of each transaction does not produce fairness in the aggregate. The reason is multiplicative compounding. A win and a loss of the same absolute size do not cancel as fractions of your wealth — a 20% gain followed by a 20% loss leaves you below where you started. Across millions of fair trades, these multiplicative shocks compound, and it can be shown that wealth condenses onto a single agent almost surely. No villain is required. Some inequality, the model demonstrates, needs nothing but chance and compounding.
The redistribution knob is the pedagogical heart of the entry. With τ = 0 you get runaway condensation — a winner-take-all oligarchy. But add even a small redistribution and the outcome changes qualitatively: the relentless drain toward one agent is arrested, and the system settles into a stable, realistic spread of wealth — a Pareto-like distribution — instead of total collapse. Fair trade alone tends to oligarchy; a little redistribution buys a stable spread. Bruce Boghosian and colleagues showed that this model maps onto a known mathematical equation (a Fokker–Planck equation, with the tax appearing as a drift term) and that, suitably parameterized, it can be tuned to resemble the broad shape of real wealth distributions.
In the wild
Kinetic exchange models are a real and active strand of econophysics: borrowing the mathematics of colliding gas molecules to model money changing hands. They reproduce a robust empirical regularity — real wealth and income distributions are roughly exponential through the bulk with a Pareto power-law tail at the top — and the yard-sale model with redistribution can be tuned to match that shape.
But it is essential to be honest about what this shows, exactly as the Atlas's Schelling segregation entry is careful to be. The model demonstrates that random fair exchange is sufficient to generate inequality and condensation — not that real-world inequality is only, or even mainly, the product of fair random luck. Real economies are full of things this model omits entirely: wages and labor, production and growth, inheritance, structural advantage, policy, taxation schedules, discrimination, fraud, and a great many transactions that are not fair, not random, and not zero-sum. The yard-sale model is a clean thought-experiment that isolates one mechanism and proves it is powerful. It is a lens that reveals one force — that fairness in the small can fail to add up to fairness in the large — not a complete theory of how actual wealth came to be distributed as it is. Read it that way and it is illuminating; read it as the whole story and it badly misleads.
Try this
- Run with redistribution = 0 and watch the Gini march toward 1 as the field goes dark and one cell takes everything. The sorted-wealth curve below collapses to a flat floor with a single spike at the right edge.
- Reset, then set redistribution to ~0.01: inequality still grows, but it stabilizes into a spread instead of collapsing. The Gini settles at an intermediate value and the wealth curve holds a smooth Pareto-like shape rather than a single spike.
- Raise the stake f toward 0.5 and watch condensation speed up — bigger bets compound faster. Lower it toward 0.05 and the same fate arrives, just more slowly.
- Compare with Schelling's model, the Atlas's other "fair micro, surprising macro" social entry: there, mild individual preference is sufficient for segregation; here, fair individual trades are sufficient for oligarchy.
References
- Dragulescu, A. & Yakovenko, V. M. (2000). "Statistical mechanics of money." The European Physical Journal B 17, 723–729.
- Chakraborti, A. (2002). "Distributions of money in model markets of economy." International Journal of Modern Physics C 13(10), 1315–1321.
- Boghosian, B. M. (2014). "Kinetics of wealth and the Pareto law." Physical Review E 89, 042804.
- Boghosian, B. M. (2019). "The Inescapable Casino." Scientific American 321(5), 70–77.
- Hayes, B. (2002). "Follow the Money." American Scientist 90(5), 400–405.