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

The Prediction Company 
•
155
imitation, and the more firms there are implementing a strategy, the 
less profitable it is for anyone. there are some indications, however, 
that the Prediction company has been wildly successful. As one for-
mer board member I spoke with pointed out, it is still an active subsid-
iary of UBS, after more than a decade. Another knowledgeable source 
told me that, over the firm’s first fifteen years, its risk-adjusted return 
was almost one hundred times larger than the S&P 500 return over the 
same period.
farmer stayed with the firm for about a decade before his passion 
for research lured him back to academia. He took a position at the 
Santa fe Institute as a full-time researcher in 1999. Packard stayed with 
the company for a few more years, serving as ceo until 2003, when 
he left to start a new company, called ProtoLife. By the time they left, 
they had made their point: a firm grasp of statistics and a little creative 
reappropriation of tools from physics were enough to beat the Man. It 
was time to tackle a new set of problems.
Black box models, and more generally “algorithmic trading,” have 
taken much of the backlash against quantitative finance in the period 
since the 2007–2008 financial crisis. the negative press is not unde-
served. Black box models often work, but by definition it is impossible 
to pinpoint why they work, or to fully predict when they are going to 
fail. this means that black box modelers don’t have the luxury of being 
able to guess when the assumptions that have gone into their models 
are going to turn bad. In place of this sort of theoretical backing, the 
reliability of black box models has to be constantly tested by statistical 
methods, to determine the extent to which they continue to do what 
they are intended to do. this can make them seem risky, and in some 
cases, if used injudiciously, they really are risky. they are easy to abuse, 
since one can convince oneself that a model that has worked before is 
a kind of magical device that will continue to work, come what may.
In the end, though, data outclass theory. this means that no matter 
how good the theoretical backing for your (non–black box) model, 
you ultimately need to evaluate it on the basis of how well it performs. 
even the most transparent models need to be constantly tested by just 

---

the same kinds of statistical methods that are used to evaluate black 
box models. the clearest example of why this is so can be found by 
looking at the failure of the Black-Scholes model to account for the 
volatility smile in the aftermath of the 1987 crash. theoretical back-
ing for a model can be a double-edged sword: on the one hand, it can 
help guide practitioners who are trying to understand the limits of 
the model; conversely, it can lull you into a sense of false confidence 
that, because you have some theoretical justification for a model, the 
model must be right. Unfortunately science doesn’t work this way. 
And from this latter point of view, black box models have an advan-
tage over other, more theoretically transparent models, because one 
is effectively forced to evaluate their effectiveness on the basis of their 
actual success, not on one’s beliefs about what ought to be successful.
there’s another worry about black box models, above and beyond 
their opaqueness. All of the physicists whose work I have discussed 
thus far, from Bachelier to Black, have argued that markets are unpre-
dictable. Purely random. the only disputes concern the nature of the 
randomness, and whether they are well enough behaved to be treated 
by normal distributions. In the years since it was first observed by 
Bachelier and osborne, the idea that markets are unpredictable has 
been elevated to a central tenet of mainstream financial theory, under 
the umbrella of the efficient market hypothesis.
And yet, the Prediction company, and dozens of other black box 
trading groups that have sprung up subsequently, purports to predict 
how the market will behave, over short periods of time and under spe-
cial circumstances. the Prediction company, at least, never worked 
with derivatives — its models attempted to predict how markets would 
behave directly, in just the way that many economists (and plenty of 
investors) would have supposed was impossible. nonetheless, it was 
successful.
It’s reasonable to be skeptical about the company’s success. Invest-
ing can often come down to luck. that markets are random is not just 
conventional wisdom in economics departments. there’s an enor-
mous amount of statistical evidence to support it. then again, perhaps 
the idea that markets are random because they are efficient — in the 
sense that market prices quickly change to account for all available 
156 
•
t h e p h y s i c s o f wa l l s t r e e t

---