The Physics of Wall Street: a brief History of Predicting the Unpredictable

The Prediction Company 
•
137
computer to solve the equations in real time, keeping one step ahead of 
the actual weather to make accurate predictions long into the future.
few of his colleagues were persuaded. As a first step, and as an at-
tempt to show his fellow meteorologists that he wasn’t crazy, Lorenz 
came up with a very simple model for wind. this model drew on the 
behavior of wind in the real world, but it was highly idealized, with 
twelve rules governing the way the wind would blow, and with no ac-
counting for seasons, nightfall, or rain. Lorenz wrote a program using 
a primitive computer — a royal McBee, one of the very first computers 
designed to be placed on a desk and operated by a single user — that 
would solve his model’s equations and spit out a handful of numbers 
corresponding to the magnitude and direction of the prevailing winds 
as they changed over time. It wasn’t a predictive model of the weather; 
it was more like a toy climate that incorporated atmospheric phenom-
ena. But it was enough to convince at least some of his colleagues that 
this was something worth pursuing. Graduate students and junior fac-
ulty would come into his office daily to peer into Lorenz’s imaginary 
world, taking bets on whether the wind would turn north or south, 
strengthen or weaken, on a given day.
At first, it seemed that Lorenz’s model was a neat proof of concept. 
It even had some (limited) predictive power: certain patterns seemed 
to emerge over and over again, with enough regularity that a work-
ing meteorologist might be able to look for similar patterns in actual 
weather data. But the real discovery was an accident. one day, while 
reviewing his data, Lorenz decided he wanted to look at a stretch of 
weather more closely. He started the program, plugging in the val-
ues for the wind that corresponded to the beginning of the period 
he was interested in. If things were working as they should, the com-
puter would run the calculations and come up with the same results 
he had seen before. He set the computer to work and went off for the 
afternoon.
When he returned a few hours later, it was obvious that something 
had gone wrong. the data on his screen looked nothing like the data 
he had seen the first time he had run the simulation with these same 
numbers. He checked the values he had entered — they were correct, 

---

exactly what had appeared on his printout. After poking around for a 
while, he concluded that the computer must be broken.
It was only later that he realized what had really happened. the 
computer contained enough memory to store six digits at a time. the 
state of Lorenz’s mini world was summed up by a decimal with six fig-
ures, something like .452386. But he had set up the program to record 
only three digits, to save space on the printouts and make them more 
readable. So instead of .452386 (say), the computer printout would 
read .452. When he set up the computer to rerun the simulation, he 
had started with the shorter, rounded number instead of the full six-
digit number that had fully described the state of the system during 
the first run-through.
this kind of rounding should not have mattered. Imagine you are 
trying to putt a golf ball. the hole you are aiming for is only slightly 
larger than the ball itself. And yet, if you miscalculate by a fraction of 
an inch, and you hit the ball a little too hard or a little too softly, or you 
aim a little to one side, you would still expect the ball to get close to the 
hole, even if it doesn’t go in. If you are throwing a baseball, you would 
expect it to get pretty close to the catcher even if your arm doesn’t 
extend exactly as you want it to, or even if your fingers slip slightly 
on the ball. this is how the physical world works: if two objects start 
in more or less the same physical state, they are going to do more or 
less the same thing and end up in very similar places. the world is an 
ordered place. or at least that’s what everyone thought before Lorenz 
accidentally discovered chaos.
Lorenz didn’t call it chaos. that word came later, with the work 
of two physicists named James Yorke and tien-Yien Li who wrote a 
paper called “Period three Implies chaos.” Lorenz called his discov-
ery “sensitive dependence on initial conditions,” which, though much 
less sexy, is extremely descriptive, capturing the essence of chaotic be-
havior. despite the fact that Lorenz’s system was entirely determinis-
tic, wholly governed by the laws of Lorenzian weather, extremely small 
differences in the state of the system at a given time would quickly 
explode into large differences later on. this observation, a result of one 
of the very first computer simulations in service of a scientific prob-
138 
•
t h e p h y s i c s o f wa l l s t r e e t

---