The Misbehavior of Markets

To explain what he saw instead, Mandelbrot reached for the idea that would make him famous outside finance: the fractal. A fractal is a shape whose parts echo the whole at every scale — a coastline that looks equally jagged from a plane, a boat, or up close; a fern whose fronds repeat the pattern of the whole plant. He coined the word in 1975, but the intuition came earlier, and the cotton prices were an early testing ground. When he zoomed into a stretch of price history and then out again, the jaggedness did not smooth over. A day looked like a month looked like a year. The roughness was self-similar.

This was not a metaphor but a measurable property. Mandelbrot borrowed from an obscure earlier mathematician, Paul Lévy, whose "stable distributions" allowed for tails far fatter than the bell curve's — distributions in which extreme events, while still uncommon, are dramatically more common than the normal model permits. Cotton, he showed, fit a Lévy-style distribution, not a Gaussian one. The large price jumps that orthodox theory dismissed as impossible outliers were, in the fatter-tailed world, an expected part of the machinery.

Mandelbrot drew a distinction he thought the whole field had missed: not all randomness is the same. Coin flips and dice are what he called mild randomness — the extremes are bounded, the average is meaningful, and more data makes your estimates converge on the truth. Human height works this way; no one is a thousand feet tall, and the biggest person barely moves the average. This is the world the bell curve describes, and for many phenomena it describes it well.

Markets, he insisted, live in a different regime, which he called wild randomness. Here a single event can dwarf everything around it. One trading day can lose more than the previous several years gained. The average is nearly useless, because the outliers do most of the work — a handful of days can account for the bulk of a decade's returns, in either direction. In wild randomness, the very idea of a "typical" move is misleading, and the concentration of consequence into a few violent moments is not an anomaly but the defining feature.

Step back from cotton and turbulence and a larger question surfaces: if the standard models are so demonstrably wrong about extremes, why did an entire industry keep using them? Mandelbrot's answer is uncomfortable, and it is the real argument of the book. The bell-curve framework survived not because it was accurate but because it was usable. It gave clean formulas, single numbers, tractable equations you could put in a spreadsheet and defend to a committee. Fat tails and long memory offer no such comfort. They tell you that risk is real, large, and only roughly measurable — which is precisely what a manager who has to report a number does not want to hear.

There is an institutional logic to the illusion. A risk model that says catastrophe is nearly impossible lets banks hold less capital, lets traders take bigger positions, lets everyone book profits today against a danger the math has conveniently erased. The comforting model is not just an intellectual error; it is commercially convenient, and convenience has a way of hardening into orthodoxy. When the rare event arrives — and in wild randomness it always eventually arrives — the loss lands somewhere the model never looked.

The cotton prices Mandelbrot plotted at IBM never fit the bell curve, and he spent the rest of his life insisting that the misfit mattered. He died in 2010, two years after a financial crisis in which the risk models of the world's largest banks failed in exactly the way he had described: a supposedly once-in-many-lifetimes event arriving, then arriving again the next week. The models had not merely been unlucky. They had been built to be blind to the very events that destroyed them.