A seashell, a coastline, and a stock chart are built from the same rule. Once you can see it, a price chart never looks the same again. Four animations, with everything you need to read them, whether this is your first chart or your ten-thousandth.
Every serious financial disaster of the past century shared one quiet assumption. The models believed markets were smoother than they really are.
They were not.
And the clue to how wrong we have been has been sitting inside a seashell the entire time.
▶ Interactive: the nautilus & the market
Visualisation 1: A nautilus shell and a market chart, drawn by one rule. The spiral is logarithmic, growing by a fixed proportion as it turns, so it looks the same at every magnification and zooms inward without a seam. The point is simple: growth by proportion shapes a shell and a price alike.
Look at the shape above. A nautilus winding inward, and a price chart jittering along. They seem to belong to different worlds, one to biology and one to finance. They do not. They are drawn by the same rule: one simple step, repeated at smaller and smaller sizes, without end.
Chaos has an architecture. That is the whole idea of this article, and everything below is its proof.
The word for that architecture is fractal: a shape made of smaller copies of itself, rough at every scale. Benoit Mandelbrot spent a lifetime showing that fractals run through all of nature: coastlines, snowflakes, the branches of a tree, the florets of a head of broccoli. And, he argued, through the wild zigzag of markets. He was not saying that markets resemble nature. He was saying that markets are nature, obeying the same geometry as a river or a shoreline.
Decoder, fractal. Plain: a shape made of smaller copies of itself, with no smooth scale where the bumpiness stops. Technical: a structure that repeats statistically across scales, with a fractional dimension; named by Mandelbrot in 1975 from the Latin fractus, broken. For traders: a price path’s jaggedness never averages out, because variance arrives from every timescale at once.
How to read this. The story runs in plain language and you can follow every word without maths. The Decoder inserts give the exact terms for anyone who wants them, with a note for traders. Read straight through, or stop and unpack them. And as you go, play: drag the dials, replay the scenes.
The story you have been told (and why it breaks)
Here is the textbook version of how a market moves. Every tiny tick of the price is like a coin flip (heads it ticks up, tails it ticks down), and each flip is blind to the one before. Add up millions of these little coin flips and you get a smooth, polite, bell-curve world. Tidy. Comforting. Mostly harmless.
Decoder, the random walk and the bell curve. Plain: a random walk is a path of independent steps, like a drunk wandering home; the bell curve is the humped shape that treats extreme days as vanishingly rare. Technical: the textbook is a Gaussian random walk, proposed for prices by Louis Bachelier in 1900 and built into finance through Brownian motion and the Black-Scholes formula. For traders: it is the assumption hiding inside most off-the-shelf risk numbers, and the one this article dismantles.
The textbook story is wrong in exactly the places that matter. Real markets trend: they drift one way for far longer than coin flips should allow. They go quiet for weeks, then erupt without warning. And every so often they throw a crash so violent that, in the coin-flip world, it should not happen once in the entire age of the universe. Yet those crashes turn up every decade or two, almost on schedule.
To see why, you have to stop staring at the smooth average and start looking at the texture, which is precisely what these animations reveal.
1. Can you tell which one remembers?
▶ Interactive: same face, different memory
Visualisation 2: Two computer-made charts: one a memoryless coin-flip walk, the other carrying memory you can tune with the dial. The memory series is exact fractional Gaussian noise, and the live lag-1 scatter and autocorrelation readouts confirm the memory is really there. It shows that a trend can be nothing but memory, invisible to the eye yet plain in the statistics.
Two charts. Both made by a computer. One is the pure coin-flip market from the textbook, every step blind to the last. The other has a memory: each step is nudged, just slightly, in the direction the last few were already going.
Take a moment and guess which is which. Go on, really look.
You can’t. And that is the whole point. Side by side, a market with memory and a market without are impossible to tell apart by eye. The difference is real, but it hides. It surfaces only when you stop watching the wandering line and ask a sharper question: does today tend to follow yesterday?
Decoder, autocorrelation. Plain: whether a move tends to resemble the one before it. Technical: the correlation of a series with a time-shifted copy of itself; lag-1 compares each day to the previous one. Near zero is no memory, positive is persistence, negative is reversal. For traders: in real markets the autocorrelation of returns is weak, but the autocorrelation of volatility is strong and durable. Hold that distinction; it is the honest core of the subject.
Now slide the memory dial and watch the lower chart. With no memory it is aimless: up, down, up, down, going nowhere. Add a little memory and something is born, the line begins to hold a direction. Ups follow ups. Downs follow downs. In strings. And a long string of up-days has a familiar name: we call it a trend.
Decoder, the Hurst exponent (H). Plain: the memory dial you are dragging. Technical: a number from 0 to 1 that sets how a fractal series scales over time. H = 0.5 is a random walk, above 0.5 is persistent (trending), below 0.5 is mean-reverting; estimated by rescaled-range or detrended fluctuation analysis. For traders: tempting as a regime gauge, but noisy on real returns, so trust no single estimate.
Decoder, long-range dependence (the Joseph effect). Plain: memory that fades slowly, reaching far into the future. Technical: correlation that decays as a slow power law, the mark of fractional Gaussian noise; Mandelbrot named it after the seven fat and seven lean years, echoing Harold Hurst’s study of Nile floods. For traders: the backbone of momentum and of volatility persistence, and why calm and turbulent regimes cluster.
That is the first jolt. A trend does not need a reason, or a story, or a clever signal. It can be nothing more than memory, a faint tendency for each move to echo the last, quietly adding up. The seed of every trend is something this small, and this invisible.
2. Prices do not have memory. People do.
▶ Interactive: the crowd makes the curve
Visualisation 3: A market assembled from the ground up, from one trader, to many on different time horizons, to a herding crowd. The price is a superposition of those horizon waves, and the copying is a percolation model held near its critical point, where copycat clusters follow a power law and a single giant cluster acting at once becomes a crash. Trends and crashes emerge here from the structure of the crowd, with no outside news required.
A market has no mind. So if it behaves as though it remembers, the memory has to live somewhere, and it lives in us. Equations do not trade. People do. This animation builds a market from nothing to show how a crowd turns into a curve.
Start with one trader. A single person, buying and selling on their own little rhythm, answering to no one. The price they make is dull: a thin, meaningless wiggle. Nothing fractal about it.
Now add more traders. And here is the trick: put them on different clocks. A pension fund shifts slowly, over years. An ordinary investor moves over months. A day trader works by the hour. A high-speed algorithm fires in thousandths of a second. Each group is a wave in the price: some slow and broad, some fast and fine. Stack all those waves, from all those clocks, on top of one another, and a rich, jagged, believable market tumbles out of the sum. The grand slow trend and the tiny fast jitter turn out to be the same picture, seen at different speeds.
Decoder, heterogeneous horizons and superposition. Plain: a market is a pile of traders with different attention spans, each adding its own wave to the price. Technical: summing components across a range of scales (spectral synthesis) is one standard way to build fractional Brownian motion; the heterogeneous market hypothesis reads those scales as time horizons. For traders: it is why structure lives at every frequency at once, with no single natural timescale.
Then let people do the most human thing of all: copy each other. One trader sells, their neighbour sells, the impulse spreads. Watch the clusters of copycats form in the animation: usually small, occasionally enormous. When a small cluster acts, the price twitches. When a giant one all stampedes the same way at once, the price leaps, a crash. And because human crowds clump in a particular lopsided way, countless little huddles and a rare colossal mob, the leaps arrive in that same lopsided way: mostly small, but every so often, catastrophic.
Decoder, herding, power laws, and percolation. Plain: people imitate neighbours, imitation spreads in patches, and now and then a patch swallows the room. Technical: a percolation model near its critical point produces clusters whose sizes follow a power law; the Cont-Bouchaud model turns that into heavy-tailed returns. For traders: a source of fat tails that needs no outside shock; the crowd’s own structure makes the extremes.
No one planned the trend. No one planned the crash. The crowd made both, simply by being a crowd. This is the moment the whole subject should make you uneasy, because it means the wildness is not a malfunction to be fixed. It is what a market is.
3. The three fingerprints
▶ Interactive: wild by design
Visualisation 4: The three fingerprints of a fractal market, shown in turn. Trend is built as a Weierstrass-Mandelbrot sum of waves, the endless roughness is self-affine fractional Brownian motion under deep zoom, and the fat-tailed crashes come from a multiplicative volatility cascade, with a sigma odometer comparing the tails to a bell curve. Together they make the case that the wildness is built in, not bolted on.
If chaos has an architecture, it should leave fingerprints. It leaves three, and this last animation lays them out one at a time.
Trends are built, not bolted on. Begin with one slow wave and keep adding faster, smaller ones on top. A convincing price chart simply appears, and the trend was just the slowest wave underneath all along. (You met this already, as the crowd’s slow traders.)
Decoder, superposition and the Weierstrass-Mandelbrot function. Plain: stack waves whose wavelength halves and height shrinks at each step, and a fractal curve appears from pure addition. Technical: with amplitudes falling as 2 to the power minus H, this is the Weierstrass-Mandelbrot construction, a classic self-affine fractal governed by the same Hurst H. For traders: trend and noise are not two things to separate; they are the low and high ends of one spectrum.
Zoom in forever, and it never smooths out. Take any sliver of the chart and magnify it. You would expect it to calm down, the way a bumpy road looks like a smooth ribbon from an aeroplane. It doesn’t. A zoomed-in sliver is every bit as jagged as the whole thing. Zoom into that, and it happens again. There is a famous party trick hiding here: strip the dates off a market chart and not even a professional can tell you whether they are looking at a hundred years or thirty seconds. The roughness has no size. A coastline does the same: a rocky shore looks equally craggy from orbit or from the beach.
Decoder, scale invariance, self-similar, and self-affine. Plain: the chart looks the same at every zoom. Technical: scale invariance means no characteristic scale exists; a snowflake is self-similar (equal zoom in all directions), a price chart is self-affine (time and price stretched by different amounts, since price scales as time to the power H). For traders: risk does not simply shrink by dropping to a faster timeframe.
Calm and catastrophe are the same thing. This is the unsettling one. Picture a stretch of time. Split it in two and hand one half more energy, the other less, at random. Split each half again, the same way. Keep splitting. The energy does not spread out evenly: it pools, concentrating into a few rare, violent bursts inside long stretches of quiet. That pooling is volatility: calm days clump together, wild days clump together, and the occasional monster move is not a fluke, it is built in.
Decoder, fat tails, kurtosis, and the multifractal cascade. Plain: huge days are far more common than the bell curve allows, and the wildness clumps. Technical: the splitting game is a multiplicative cascade, the engine of Mandelbrot’s multifractal model; it yields volatility clustering and a heavy-tailed (leptokurtic) distribution, with tails that often fall off near an inverse cube law. Mandelbrot called the abrupt jumps the Noah effect. For traders: the formal reason Gaussian Value at Risk understates disaster, and why calm and storm arrive in runs rather than scattered at random.
Lay a market’s daily moves against the textbook’s tidy bell curve and the giveaway jumps out: far too many giant days out in the tails, where the bell curve swears there should be almost none.
4. What this costs you if you ignore it
A number makes it land. In the tidy bell-curve world, a six-standard-deviation crash day should happen about once every two million years. Push to ten standard deviations and the wait runs longer than the age of the universe.
Decoder, sigma and Value at Risk. Plain: sigma is a unit of surprise; one sigma is an ordinary day, and the larger the number, the rarer the day should be. Technical: standard deviation measures spread; under a normal distribution, five or six sigma events carry probabilities that are effectively zero. Value at Risk estimates plausible loss on a bad day, usually on that same normal assumption. For traders: the failure is structural; Gaussian VaR is wrong about tails by orders of magnitude, not by a margin.
On 19 October 1987, the US stock market fell about 23% in a single day, a move of roughly twenty standard deviations by the textbook’s own yardstick. Under that model it should not have happened once in the entire history of the universe, on any planet, in any market, ever. It happened on a Monday. Eleven years later a fund run partly by Nobel laureates, Long-Term Capital Management, nearly took the financial system down when markets did something its models rated as essentially impossible. Then 2008. Then the lightning crash of 2020. The impossible turns out to be roughly a once-a-decade appointment.
Here is the deeper thing fractals do. They change what uncertainty means. The textbook imagines a world that is unpredictable but well behaved, random yet tame, a casino with known odds. Fractals describe a world that is equally unpredictable but fundamentally untameable, where the odds themselves shift and the largest event is always still to come. That is a colder kind of uncertainty, and a truer one. The lesson is not that markets can be predicted; fractals tell you no more about tomorrow than a coin does. The lesson is that the worst day you have ever lived through is no promise about the worst one coming. The tail is always heavier than your experience has shown you yet.
5. Two traders
A trader who believes in bell curves sleeps differently from one who believes in fractals.
The first thinks yesterday measured tomorrow. Their worst imaginable day is only a little worse than an ordinary one, so they treat last year’s largest loss as roughly the largest loss that can happen. They are calm. They are also wrong.
The second assumes tomorrow can always be stranger than anything yet seen. They claim no gift for direction, because fractals grant none. But they plan for a tail heavier than the data has shown them, and that one assumption reshapes everything they do.
It makes them distrust any risk number resting on a normal curve, and reach instead for heavier-tailed tools, the Student-t distribution or extreme value theory that models the tail directly rather than wishing it away. Where the first trader fears the big move, the second builds a method around it: cut the loss quickly on the many small moves that fade, and stay with the few that keep going, so that when a fat tail finally arrives it pays for all the small ones that did not. That is trend following, and it does not predict the wild day. It waits for it. And because even a simple fractal toy serves up shocks larger than its past, the second trader stress tests beyond their own history, treating a back-test’s worst drawdown as a floor, never a ceiling.
The fractal trader has no crystal ball. What they have is the right kind of fear.
Decoder, the trader’s toolkit. Extreme value theory: statistics for the rare extremes, used to size tail risk without assuming a bell curve. Cutting losses: capping the damage on the many moves that go nowhere. Letting winners run: staying in the few that trend, so a single fat tail can outweigh many small losses. For everyone: the edge is not in taming the wild day but in surviving the quiet ones cheaply enough to still be there when it comes.
6. The honest part
Everything you have just played with is a deliberately simple toy, not a simulation of any real market. But that is exactly what gives it force. You do not need news, or panic, or greed, or genius traders to manufacture trends, endless roughness, and impossible crashes. A handful of plain rules, repeated across scales, conjure all of it. The wild behaviour the textbook files away as a bag of freak exceptions turns out to be the ordinary behaviour of anything built this way.
Two honesties deserve to be said plainly. First, the memory in section 1 lives in an idealised process; in real markets the strong, durable memory sits in volatility, the size of moves, more than in direction, their sign. Direction is only faintly persistent, which is why trend following is hard-won rather than easy: it is earned by riding the few real trends and cutting the many false starts, not by any simple recipe for prediction. Second, and more important: even these simple toys throw up shocks bigger than any short stretch of their own history would have led you to expect. The real warning is not that the model is too wild. It is that the world might be wilder still, and the map is always rougher than the ground you have walked so far.
Long before there were traders, there were rivers. Long before there were exchanges, there were coastlines. Nature solved complexity first, and markets merely inherited it. Once you see that, a price chart stops looking like a machine that has malfunctioned. It begins to look like weather.
Chaos has an architecture. We did not design it. We are only learning to read it.
A field guide to the words
A quick reference for every term the animations introduced, plain meaning first.
- Autocorrelation. Whether a move resembles the move before it. Weak for market direction, strong and long-lasting for market volatility.
- Bell curve (normal / Gaussian distribution). The humped curve behind the textbook model, which treats large moves as essentially impossible.
- Fat tails (heavy tails). Far more extreme events than the bell curve predicts; the defining feature of real returns. Measured by excess kurtosis.
- Fractal. A shape made of smaller copies of itself, rough at every scale, with a fractional dimension. Coined by Mandelbrot in 1975.
- Herding / percolation. Traders imitating neighbours form clusters whose sizes follow a power law; a giant cluster acting together produces a crash.
- Hurst exponent (H). The memory dial from 0 to 1. H = 0.5 is a random walk, above 0.5 is trending (persistent), below 0.5 is mean-reverting.
- Joseph effect (long-range dependence). Memory that fades slowly, reaching far into the future; the backbone of momentum and volatility persistence.
- Kurtosis. A measure of tail heaviness. The normal distribution sits at 3; excess kurtosis above that signals fat tails.
- Multifractal / multiplicative cascade. The splitting game that pools volatility into rare bursts, the engine of Mandelbrot’s multifractal model of markets.
- Noah effect. Mandelbrot’s name for sudden, discontinuous jumps, after the flood; the partner of the Joseph effect.
- Power law. A relationship where halving one quantity multiplies another by a fixed factor; a straight line on a log-log plot and the signature of scale-free structure.
- Random walk. A path of independent little steps with no memory; the textbook, and largely mistaken, picture of price.
- Scale invariance. No special scale exists; the chart looks statistically the same at every zoom level.
- Self-similar vs self-affine. Self-similar shapes repeat under equal zoom in all directions (a snowflake); self-affine shapes, like price charts, need time and price stretched by different amounts.
- Sigma (standard deviation). A unit of surprise. Under the bell curve, six-sigma days happen once in millions of years; markets serve them far more often.
- Spectral synthesis / superposition. Building a fractal by stacking waves of shrinking wavelength and amplitude; here, the market’s many time horizons.
- Value at Risk (VaR). An estimate of plausible loss on a bad day, usually built on the normal assumption, and therefore usually too optimistic about disasters.
- Volatility clustering. Big moves follow big moves and calm follows calm; wildness arrives in runs rather than at random.
Richard Brennan writes on systematic trading, complex adaptive markets, and the philosophical foundations of trend following at atstradingsolutions.com. His books include The Fractals of Finance and Complex Adaptive Markets. The forthcoming Carved by Impossibility completes the trilogy.
Want the theoretical foundation for why trend following works?
The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets
The book explores the full architecture of feedback, fat tails, and fractal structure in financial markets, and what it means for how we trade, invest, and understand risk.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.