Physics ↔ economics ↔ finance
For a century, economists borrowed the mathematics of physics to model markets — random walks, equilibrium, field equations. Some of it works beautifully. Some of it blew up spectacularly. The thread that survives runs through information theory, and it tells you not which way to bet, but how much.
How to read this. Four stages, in order — each opens with a real episode, builds the model under it, lets the strongest objections fight, and hands you to the next. Stage 3 has a betting simulator you can run yourself. The point isn't to learn a trading rule; it's to watch where the physics analogy earns its keep and where it quietly stops being true.
A stock price is a particle in suspension, jostled at random — its future scatter spreading like a diffusing cloud. — the idea behind Louis Bachelier, Théorie de la spéculation, doctoral thesis, 1900
In 1900, a graduate student in Paris named Louis Bachelier handed in a thesis that modeled stock prices as a random walk — the same mathematics that describes a pollen grain shoved about by water molecules. He wrote down Brownian motion as a price process five years before Einstein's celebrated 1905 paper on Brownian motion in physics. Prices, to Bachelier, were diffusing particles. The market was a heat bath, and a share certificate was something floating in it.
Bachelier's picture is the seed of modern quantitative finance. Treat each price change as a small random shock, independent of the last, drawn from a bell curve. Add them up and the price follows a Gaussian random walk; uncertainty grows with the square root of time. Seven decades later, the Black–Scholes–Merton option-pricing formula (1973; Nobel, 1997) was built on exactly this assumption — that prices follow a geometric Brownian motion. It let traders price an option the way a physicist solves a diffusion equation, and it created a trillion-dollar industry.
The analogy ran deeper than mathematics. The founders of neoclassical economics — Léon Walras, William Stanley Jevons, in the late nineteenth century — built their theory of markets explicitly on the template of nineteenth-century rational mechanics. Utility played the role of energy: a conserved potential that agents climb, with prices settling at the point where the forces balance. Equilibrium in a market was modeled on equilibrium in a mechanical system. The economy, in this dream, was a machine that found its rest state the way a ball finds the bottom of a bowl.
The elegance
"The borrowing was astonishingly productive. It gave us the random-walk model, stochastic calculus in finance, and the Black–Scholes formula — tractable, testable, and worth a Nobel. Physics handed economics a working toolkit it could never have invented alone."
The warning (Mirowski, 1989)
"Economics imported the mathematics of energy physics without the conservation law that gave it meaning. Utility was never conserved like energy. More Heat than Light: physics envy dressed up assumptions as if they were laws of nature."
Where this leaves us
The analogy was both productive and dangerous — it delivered real tools while smuggling in a metaphysics no one had checked against the data. A market that diffuses like a particle and settles like a ball in a bowl is a beautiful object. The only question that matters is the empirical one: where does the picture actually hold? The first place it broke was the part everyone trusted most — the bell curve.
A mathematician staring at a century of cotton prices was about to notice that markets do not have the calm, thin tails the Gaussian dream requires.
Benoit Mandelbrot went looking at cotton prices and found the Gaussian model simply wrong. Price changes did not fall off the way a bell curve demands. They followed heavy-tailed distributions — power laws, of the Lévy-stable family — in which extreme moves are far more common than the normal distribution allows. His book The (Mis)Behavior of Markets (2004) made the case to a general audience: the calm, well-behaved randomness of the textbooks is a fiction, and the tails are where the action is.
Here is what "fat tails" means in money. On 19 October 1987 — Black Monday — the US stock market fell about 22% in a single day. Under a Gaussian model fitted to daily moves, that was roughly a 20-sigma event: something that should not occur once in many billions of years of trading. Yet crashes of that order recur every couple of decades. When your model says "impossible" and the world says "Tuesday," the model is broken.
The field that took this seriously calls itself econophysics — physicists turning their statistical-mechanics tools on financial data. Rosario Mantegna and H. Eugene Stanley laid out the program in An Introduction to Econophysics (1999). Measure the tail of the return distribution and you find a power law with an exponent empirically near 3 — the so-called inverse cubic law of returns. Markets show wild, clustered volatility: quiet stretches punctuated by violent bursts that arrive together. The bell curve does not merely miss this; it systematically underprices the risk of ruin.
The Gaussian modelers
"The normal distribution is tractable and, most days, close enough. It powers the formulas the whole industry runs on. You cannot do calculus on a Lévy-stable distribution over your morning coffee — and for ordinary risk, the bell curve is a fine approximation."
Mandelbrot & the econophysicists
"'Most days' is exactly the trap. The tails are where the money and the ruin live. LTCM's 1998 collapse and the 2008 crisis both ran on models that assumed thin tails — and both met the fat ones. Underpricing the rare move is how you go bust."
Where this leaves us
The random-walk dream was half right. Prices are random — but they are not Gaussian-random. The shape of the randomness, not its mere existence, is the whole game. Get the tails wrong and your model is calmest precisely when it is about to kill you. So if the bell curve is the wrong distribution, is there any piece of the physics import that genuinely survives contact with markets?
One thread does — and it comes not from thermodynamics but from a wartime cryptographer who proved how much a bit of information is worth.
In 1948, Claude Shannon founded information theory — a way to measure information in bits and bound how fast it can travel down a noisy channel. It looks like the purest of pure mathematics, far from any trading floor. But it crossed into finance through Shannon's Bell Labs colleague, John L. Kelly Jr., who saw that the same equations describing a communication channel also answer a gambler's oldest question: given an edge, how much of your money should you put on the line?
Kelly's 1956 answer is now called the Kelly criterion. If a bet wins with probability p and pays net odds b (so a winning $1 returns b dollars of profit), the fraction of your capital that maximizes the long-run exponential growth rate of wealth is
This is not "bet as much as you can when you have an edge." It is the precise opposite of greed. Bet more than f* and your growth rate actually falls while your risk of ruin climbs; bet less and you leave compounding on the table. There is a single optimum, and overshooting it is as costly as a coin-flip mistake. Information theory, applied to capital, says: the edge tells you whether to bet; the math tells you how much.
Kelly bet-sizing simulator
Set your edge, then run 200 rounds. A winning game can still go broke if you bet too big — information theory tells you the exact size that doesn't.
Kelly optimum f*
0.20
Your bet f
0.20
Final bankroll
—Bankroll starts at 100. Each round you stake f of it; win with probability p and gain f·b, lose and forfeit f. The y-axis is logarithmic so both compounding and ruin stay legible.
Edward Thorp turned this from theory into a fortune. He used the Kelly criterion to size his bets at blackjack — the basis of Beat the Dealer — proving a card-counter's small edge could be compounded without going broke. Then he carried the same discipline into financial markets and ran it for decades. The lineage is clean: Shannon → Kelly → Thorp, information theory walking out of a communications lab and into capital allocation. (The popular history is William Poundstone's Fortune's Formula.) The bit, it turns out, has a dollar value: it tells you how much to wager.
The long-run growth rate of wealth, betting a fixed fraction f each round, is the expected log-return: g(f) = p · ln(1 + f·b) + (1 − p) · ln(1 − f). Kelly maximizes this, not the expected wealth — because what compounds is the logarithm, and a few catastrophic losses crush a log-average far more than they dent an arithmetic one.
Set the derivative to zero and you recover f* = (p(b+1) − 1)/b. Check it with the simulator's defaults: p = 0.6, b = 1 (even money) gives f* = (0.6·2 − 1)/1 = 0.2. Crucially g(f) is a hump: it rises to the peak at f*, then falls, crossing back through zero at roughly 2f* — bet double the Kelly fraction in a winning game and your long-run growth is zero; bet more and it goes negative. A positive edge is no protection if you size it wrong.
Where this leaves us
This is the one place the physics-adjacent math genuinely works in markets — and the reason it works is telling. Kelly is not about forcing prices into an equilibrium or pretending returns are Gaussian. It is about information and growth: how to convert a measurable edge into compounded wealth without betting yourself into ruin. The import survives exactly where it stays a statement about information, and stops pretending the market is a mechanical system.
Which raises the harder question: why does the mechanical picture fail at all? The answer is that markets, unlike particles, can read the equations written about them.
Atoms don't read the equations describing them. Market participants do. — the difference that breaks the mechanical analogy
A hydrogen atom obeys the Schrödinger equation whether or not anyone writes it down; it cannot change its behavior in response to the theory. A market is the opposite. The moment a regularity becomes known, traders act on it, and the acting erases the regularity. The system you are modeling reorganizes itself around your model.
Several thinkers named this same effect from different doors. George Soros called it reflexivity: beliefs about prices move prices, which then move beliefs, in a feedback loop with no stable fixed point. Economists know a cousin of it as the Lucas critique — the historical relationships in your data shift once policy starts exploiting them. Managers know it as Goodhart's law: when a measure becomes a target, it stops being a good measure. All three say the same physical thing — the particles change when watched.
So a school grew up to model the economy as what it actually is: an out-of-equilibrium, evolving system, never settling into the mechanical rest state the neoclassical dream required. This is complexity economics: W. Brian Arthur (Complexity and the Economy) and the Santa Fe Institute reframed the economy as a process of constant adaptation rather than equilibrium; Eric Beinhocker's The Origin of Wealth recast wealth creation as an evolutionary search. Doyne Farmer's Making Sense of Chaos (2024) pushes the program into large-scale agent-based simulation — modeling markets as populations of interacting, adapting agents instead of solving for a single resting equilibrium.
Is econophysics real insight, or physicists slumming in someone else's field? Many economists are skeptical. Their charge: econophysics rediscovers stylized facts that economics already knew — volatility clustering, heavy tails — and often does so with weaker statistics, while ignoring the institutions, contracts, and incentives that actually structure a market. Power laws, they note, are easy to claim and hard to establish; eyeballing a log-log plot is not a fit.
The defenders answer that the power-law tail findings are robust and replicated across many markets, and that mainstream Gaussian finance failed catastrophically and expensively — in 1998 and 2008 — precisely where the econophysicists said it would. The strongest version of each side stands: critics are right that institutions matter and that physicists sometimes import bad statistical habits; defenders are right that the empirical tail behavior is real and that thin-tailed finance has a body count. Whether econophysics is a genuine science of markets or a borrowed vocabulary remains unsettled.
Where this leaves us
Physics gives markets a real toolkit — stochastic calculus for option pricing, power-law statistics for tail risk, the Kelly growth criterion for sizing bets under an edge. What it cannot hand them is equilibrium mechanics. Markets are reflexive, adaptive, fat-tailed systems whose particles change when watched. The right lesson from a century of physics envy is to borrow the math and leave the metaphysics behind.
The short answer
Markets diffuse like random walks, so stochastic calculus genuinely prices their derivatives. Their extremes follow power laws, so the physics of fat tails beats the bell curve at measuring ruin. And Shannon's information theory, through Kelly, tells you exactly how much to bet to compound an edge without going broke. But markets are not particles. Their participants read the equations and trade against them, so the regularities dissolve and no equilibrium ever sets in. Borrow the math; leave the metaphysics. Physics tells you the shape of the randomness and how much to wager inside it — not which way the next move goes.
L. Bachelier, Théorie de la spéculation (1900) — the doctoral thesis that modeled prices as a random walk, five years before Einstein's paper on Brownian motion. The origin of the physics-of-markets dream.
B. Mandelbrot & R. Hudson, The (Mis)Behavior of Markets (2004) — the case against the bell curve, from cotton prices to crashes; heavy tails as the rule, not the exception. Start here for the fat-tail argument.
R. Mantegna & H. E. Stanley, An Introduction to Econophysics (1999) — the founding text of econophysics; power-law tails and the statistical mechanics of returns.
P. Mirowski, More Heat than Light (1989) — the history of physics envy in economics; the critical antidote to the whole borrowing project.
W. Poundstone, Fortune's Formula (2005) — the popular history of the Shannon → Kelly → Thorp line; information theory as a betting system.
W. B. Arthur, Complexity and the Economy (2014) — collected essays founding complexity economics; the economy as an out-of-equilibrium, adaptive system.
E. Beinhocker, The Origin of Wealth (2006) — wealth creation read as evolutionary search rather than equilibrium; the accessible entry to complexity economics.
J. D. Farmer, Making Sense of Chaos (2024) — agent-based, complexity modeling of markets and economies; the current frontier of the program.