You Only Live Once, Statistically

Here is a coin worth turning down. Heads, your money grows by half. Tails, it shrinks by forty percent. The up move is bigger than the down move, the coin is fair, so on average every flip pays — about five percent a round. A spreadsheet says take it. A brochure would say take it with both hands. Play it for real, round after round, and it takes almost everything you have.

That sentence should not be able to be true, and the cleanest way to watch it become true is the instrument Nassim Taleb reached for constantly: a Monte Carlo simulation — a machine that rolls the same random situation thousands or millions of times and lays out the spread of histories that could have happened. Taleb, a former options trader, treated it less as mathematics than as a personal tool. "I have two ways of learning from history," he wrote in Fooled by Randomness: "from the past, by reading the elders; and from the future, thanks to my Monte Carlo toy."

Point that toy at our coin and the trick shows itself. Run ten thousand people through fifty rounds and the average fortune in the room climbs past a thousand dollars. Follow any single one of those ten thousand lives down its own timeline and the typical one is left with about seven. Same coin, same rules — two answers pointing in opposite directions. Here is what that looks like.

A hundred people, or one person a hundred times

The gap has a precise shape, and Taleb gives it the cleanest possible statement. "The difference between 100 people going to a casino and one person going to a casino 100 times" — that is the whole thing. In the first case a few gamblers go bust, the rest do fine, and the average across the room is reassuring. In the second case the one gambler carries every result forward; if he busts on round 28, there is no round 29. "Let us call the first set ensemble probability," Taleb writes, "and the second one time probability." The crowd's number is an ensemble average. Your life is a time average. They are not the same figure, and you only ever get the second one.

What separates the two is what erases the randomness. Across a big enough crowd the luck cancels — sample size does the smoothing, and the average is real. Down a single timeline luck does not cancel; it compounds. Each round multiplies whatever the last one left, so time does not average your results, it strings them together — and a string is only as strong as its worst stretch.

Why the two averages split

The arithmetic is almost insultingly simple. Win once and lose once — a heads then a tails — and a hundred dollars goes to a hundred and fifty, then down forty percent to ninety. Not back to a hundred. Ninety. The order does not matter: one good round and one bad round multiply to nine-tenths of what you started with, every time. So the crowd, adding up everyone's dollars, sees plus five percent a round; and you, watching your own pile compound, lose about five percent a round. The brochure quotes the first number. You live the second.

A positive average is not a lie. It is just not a promise to you.

The smaller, truer number is the geometric mean, and it lags for a plain reason: a loss costs more than the same-sized gain repairs. A forty percent fall needs a sixty-seven percent rise to undo it, not a forty percent one. Physicists call the general phenomenon non-ergodicity — a system where the average over a crowd and the average over time genuinely diverge. Ole Peters, who built the modern argument around this exact coin, put the punchline plainly: wealth averaged over many people grows, while wealth averaged over one person's long run shrinks.

The wall you can't climb back over

Multiplying has a second, harder edge. Add and subtract dollars and you can always dig out of a hole; multiply them and there is a floor you cannot leave. Reach zero — or close enough — and every future round multiplies zero, and you stay there. Peters and Gell-Mann state it in a line: "multiplicative dynamics imply an absorbing boundary. Unlike under additive dynamics it is impossible to recover from bankruptcy." Ruin is not just a bad outcome on the list. It is a trapdoor under the list.

This is where Taleb's favorite parable does its work, and it is grimmer than a coin. An eccentric tycoon offers you ten million dollars to play Russian roulette — one bullet, six chambers, pull the trigger. Five of the six histories leave you rich; one ends in an obituary. On a spreadsheet it looks superb: a naive expected value would boast an "83.33% chance of gains, for an 'expected' average return per shot of $833,333." But, Taleb finishes, "if you played Russian roulette more than once, you are deemed to end up in the cemetery." That expected value is computed across the crowd of possible yous. Only one of them is going to be doing the living.

From there Taleb draws the line that makes this more than a curiosity: "If there is a possibility of ruin," he writes, "cost benefit analyses are no longer possible." The average stops being decision-useful the moment one of the outcomes ends the game. And identical winnings turn out not to be identical: "$10 million earned through Russian roulette does not have the same value as $10 million earned through the diligent and artful practice of dentistry." The dentist can do it again tomorrow. The gambler spent his survival to get there, and that bill is still outstanding.

How Taleb actually used the simulator

The Monte Carlo toy was not only for parables. In Fooled by Randomness Taleb runs a working example: an investor who genuinely earns, say, fifteen percent a year over Treasury bills, with the ordinary ten percent or so of volatility that rides along with it. A good bet, honestly held. Then he changes one thing — not the returns, only how often the investor looks. Checked once a year, the portfolio shows a gain about ninety-three percent of the time. Checked second by second, that collapses to roughly fifty-fifty — a coin flip of elation and misery, almost all of it noise. (You can roll that exact kind of simulation in The Monte Carlo Fan: hundreds of futures fanning out from today, with one dial for how wild each year is — turn it up and watch the average future pull away from the typical one.)

The simulation was there to teach the same lesson the coin teaches from the other side: the number you fixate on depends on the frame you read it through, and the wrong frame manufactures feeling out of nothing. A crowd is one frame. A second-by-second ticker is another. Neither is your actual life, and both will happily sell you a story about it. The toy was Taleb's way of stepping outside the single history he happened to be living and asking how much of it was luck — the move his "lucky fool," who reads a fortunate run as pure skill, never makes. (That failure has its own essay.)

The honest doubt

A site called Against Certainty owes its own sources the scrutiny it asks of everyone else's, and this idea has real critics worth hearing. The awkward part is that the core is old. Peters concedes it himself: the geometric-growth rule at the center of the argument "is well known among gamblers as Kelly's criterion of 1956," and his "modest contribution is to frame these observations as a question of ergodicity." Economists have pushed harder — one paper flatly calls ergodicity economics "pseudoscience" on the grounds that it "has not produced falsifiable implications," and another argues it mostly renames results expected-utility theory already had. The statistician Andrew Gelman, sympathetic but unmoved, caught the mood: "I'm with Peters in his disagreement with the textbook model, but, yeah, we know that already."

So hold the grand version loosely. Whether "ergodicity" is a revolution in economics or a sharp relabeling of things Kelly and Bernoulli already knew is genuinely disputed, and this piece takes no side in that fight. What survives it is smaller and sturdier, and it is all you need: an average taken across a crowd is not a forecast for the one life you get to run, and any bet with a real chance of ruin has to be judged by that second number, not the first. The fix is not mystical either — you just don't stake everything. On this very coin, the growth-maximizing move is to risk about a quarter of your pile and hold the rest; the game as posed stakes all of it, four times that, which is exactly why it turns a positive average into a slow-motion wipeout.

How to read an average

None of this says averages lie, or that every tempting bet is a trap. It says an average is the answer to one specific question — what happens to the crowd? — and you have to check whether that is the question you are actually asking. Usually it isn't.

We built the coin so you can run it yourself: The Average Is a Liar. Play your own hundred dollars and watch it sink, then run ten thousand lives at once and see the handful of winners haul the average up over a room full of quietly ruined players. It is the same instrument Taleb kept on his desk, pointed at the one question an average is built to hide — not how did the crowd do? but what happens to me? For the same logic applied to whole careers and companies, its companions are The Genius Trap and Defensible If It Loses.