The Coin Isn’t Fair: What Persi Diaconis Teaches Traders About Randomness

At fourteen, Persi Diaconis ran away from home to become a magician. He told no one, walked out of a New York high school, and spent the next decade touring with the sleight-of-hand legend Dai Vernon and working card tables on ships between New York and South America. The writer Martin Gardner, who knew him, said the teenager had a “fantastic second deal and bottom deal.” Translation: he could deal you the second card from the top, or the card off the bottom, and you would never notice. He got paid to make randomness lie.

Then he picked up a probability textbook he could not understand, went back to school at twenty-four, earned a Harvard PhD, won a MacArthur “genius” grant in 1982, and became a professor at Stanford. He spent the rest of his career doing to the universe what he once did to cruise-ship gamblers: proving that the things we assume are random usually are not, and the things we treat as meaningful are usually just noise wearing a costume.

That is the entire job description of a trader. And almost every famous Diaconis result has a twin sitting somewhere in your trading account.

The magician who learned randomness from the inside

Most probability theorists study randomness from the outside, as a clean mathematical object. Diaconis studied it from the inside first, as a professional who was paid to counterfeit it. A card cheat’s entire craft is the manufacture of fake randomness: a shuffle that looks thorough but preserves order, a cut that looks fair but isn’t, a deal that looks square but delivers the card you already decided on. When you have spent ten years faking randomness for money, you develop an unusual instinct for when the real thing is present and when it is being faked to you.

He carried that instinct into mathematics. Casinos later hired him to inspect their automatic card-shuffling machines. He found flaws almost immediately, enough hidden order that a sharp player could exploit them. The executives, the story goes, told him: “We are not pleased with your conclusions but we believe them, and that’s what we hired you for.” He also spent years debunking claims of ESP and exposing psychics, because his real specialty was never cards. It was the gap between what looks meaningful and what actually is. That gap is where traders live and where most of them lose money.

Here are four of his results, and the trade each one is trying to warn you about.

Lesson 1: The coin isn’t fair, but the edge is one percent

“It’s a coin flip” is our standard phrase for pure chance. In 2007, Diaconis, Susan Holmes and Richard Montgomery published a paper showing it isn’t. When a real human flips a real coin and catches it, the coin wobbles as it spins (a motion called precession), and that wobble means it spends slightly more time with its starting face up. Their physics model predicted that a flipped coin lands on the same side it started about 51% of the time.

For years that was just an elegant prediction. Then in 2023 a team recorded 350,757 real coin flips, using coins from dozens of countries to rule out any single coin’s design. The result: coins landed the same way they started 50.8% of the time. The bias Diaconis predicted was real. A coin flip, our purest symbol of fairness, is quietly rigged by about one percent.

Now sit with how small that is. A one percent edge is completely invisible in any single flip. It is invisible across ten flips, or fifty, or a hundred. You could flip a coin all afternoon and never feel it. The researchers spelled out what it’s worth: if you bet a dollar per flip and knew the starting side, across 1,000 flips you’d earn about nineteen dollars. A genuine, mathematically provable edge — and it produces almost nothing until you’ve stacked up a thousand repetitions.

This is exactly why we keep telling traders to run their strategy like a house, not a hero. When you trade prop firms like a casino, you stop caring about the outcome of the next hand and start caring about whether your one-percent coin is still your coin. It’s also why the fight over whether trading is really gambling misses the point: the casino and the disciplined trader are playing the same game Diaconis played, harvesting a tiny known edge over an enormous sample. The gambler at the table is playing the coin from the wrong side.

Lesson 2: Seven shuffles, and the sample size you don’t have

Diaconis’s most famous result, proved with Dave Bayer in 1992, answers a question every card player thinks they already know: how many times must you shuffle a deck to properly mix it? The answer is seven ordinary riffle shuffles. Not three, not four. Seven.

The surprising part is how the deck gets there. For the first few shuffles the deck barely mixes; large chunks of the original order survive, hidden but intact. A computer, or a trained magician, can still reach in and find a card you moved. Then, right around the seventh shuffle, the deck suddenly collapses from “still quite ordered” into “genuinely random” almost all at once. Mathematicians call this a cutoff phenomenon: nothing, nothing, nothing, then everything, in one step.

Two things fall out of this that every trader needs.

First: a small sample still contains hidden order. A deck riffled three times feels shuffled. It looks shuffled to you. It is not shuffled — there is enough structure left that a professional can read it. Diaconis and Bayer even built a card trick on exactly this fact.

Second: the cutoff is a warning about your equity curve. Systems that look smooth and controlled can stay that way for a long time and then flip state suddenly, the way the deck does. A strategy that has “always worked” through one regime can tip into failure in a single step when the regime changes. Order can hide right up until the moment it doesn’t. This is the same reason we push traders to know the probability mindset of thinking in batches rather than judging themselves on single trades: the batch is where the truth lives, and the batch has to be big enough to have actually mixed.

Lesson 3: Why a “genius” trader always exists

In 1989, Diaconis and Frederick Mosteller wrote a paper on how to study coincidences, and gave us a line worth tattooing on the inside of every trader’s eyelids: “With a large enough sample, any outrageous thing is likely to happen.” They called it the law of truly large numbers.

The math is brutal and simple. An event with a one percent chance in a single trial will, across 1,000 independent trials, almost certainly occur — the probability it happens at least once is about 99.99%. Rare events aren’t rare when you run enough trials. And crucially, we misread them, because we personalise the coincidence and forget how many trials were actually running in the background.

Point this lens at the trading world and it explains something you see every day.

This is survivorship bias with a mathematical spine, and it’s the single most expensive mistake in trading education. It’s also why Nassim Taleb’s work sits so close to Diaconis’s; if you want the full version of this idea aimed straight at markets, it’s the whole argument of the best trading books we keep pointing traders toward. Diaconis’s contribution is the reminder underneath all of it: don’t personalise the coincidence. The surprising streak isn’t happening to a genius. It’s happening because the sample is enormous and someone had to be the one it happened to.

Lesson 4: The machine that only looks random

There’s a detail in the coin work that turns the whole thing inside out. To prove his point, Diaconis helped build a mechanical coin-flipping machine that could produce heads every single time. Same launch speed, same angle, same height: same result, on demand. A coin toss, it turns out, is not random at all. It is fully determined by physics from the instant it leaves your thumb. It only looks random because a human can’t control or measure the initial conditions precisely enough.

And the shuffle work has a matching twist. In their experiments, even after many shuffles the deck never became perfectly mixed; a clever method could always guess a moved card slightly more often than one in fifty-two. Real structure almost never fully disappears. It just becomes very, very hard to see, and very small once you do.

The Diaconis discipline, in Mind, Method and Money

Strip away the coins and the cards and Diaconis leaves traders with one coherent discipline, and it maps cleanly onto the three things a complete trader has to get right.

Method. Treat your edge like a coin to be measured, not a belief to be defended. Diaconis never assumed a coin was fair or a deck was mixed; he tested it against a large enough sample to know. Your backtest, your journal, your setup all deserve the same suspicion. Demand seven shuffles of evidence before you trust a number, and keep testing it out of sample, the way he kept testing casinos that swore their machines were random.

Money. Size for the edge you actually have, which is a one-percent coin, not the certainty your best days make you feel. A small, real edge harvested across a huge sample is the entire business. Bet as if every good trade proves you right and one truly large number will eventually clean you out.

Mind. Refuse to personalise the coincidence. Your losing streak probably isn’t a broken system; it may just be a fair coin doing what fair coins do, clustering. The guru’s winning streak probably isn’t genius; it may just be the surviving coin. The trader who internalises the law of truly large numbers stops being the mark at their own table.

The bottom line

Persi Diaconis spent his first career learning to fake randomness and his second career refusing to be fooled by it. He knew, better than almost anyone alive, exactly how a shuffle hides order, how a “fair” coin cheats, and how a large enough crowd guarantees a miracle. That is not trivia. It is the exact skill a trader is trying to build: the ability to look at a chart, a track record, or a leaderboard and ask the only question that matters. Is this a real edge, or is it randomness in a costume?

The traders who last are the ones who keep asking. They size small because the coin’s bias is small. They wait for the sample because three shuffles lie. They stay skeptical of the genius because someone always had to be. Diaconis would recognise every one of them. They’re doing to the market what he did to the deck.