¿ìè¶ÌÊÓƵ

Doyne Farmer looks with quiet satisfaction on his recent research into
complex physical phenomena. ‘We are certainly seeing results that give us
confidence,’ he says. ‘We could publish papers that would really shake up
the economic community.’ But Farmer and his colleagues won’t be publishing
their results; they won’t be publishing anything, except perhaps profit
and loss accounts. Mainly profits, they hope.

Until recently, Farmer was head of the nonlinear dynamics group at Los
Alamos National Laboratory, New Mexico, probably the most high powered intellectual
crucible of thinking on complex systems. Such systems often look random,
but in fact contain elements of predictability that can be traced. Chaos
theory is just one example – the most widely known – of how to analyse complex
systems. ‘Most systems in the world are nonlinear,’ explains Farmer, ‘and
we are just beginning to understand them.’ Only the development of powerful
computers has enabled scientists to peer into nonlinear phenomena – conventional
mathematical analysis cannot cope. Farmer believes that the study of nonlinear
systems will become as hot commercially as it is intellectually.

A little more than a decade ago, Farmer was one of a group of young
researchers at the University of California, Santa Cruz, who helped to push
the newly emerging science of chaos into the arena of academic respectability.
Now, with Norman Packard, another of the Santa Cruz crew, Farmer is at the
scientific head of the Prediction Company, a formidable assembly of physicists
and computer whizzes who aim to exploit new methods of analysis to penetrate
the uncertainties of financial markets.

‘We hope to demonstrate that there are pockets of predictability in
the markets, and to make some money on the way,’ says Farmer. He is not
talking about small change. The company is looking for a financial partner
– a Wall Street firm that trades in stocks, bonds, and currencies – prepared
to put up, say, $50 million to turn the Prediction Company’s brand of nonlinear
analysis into large profits, some of which would probably find their way
back to support academic science. If the venture succeeds, the company will
be the biggest operation of its sort in the notoriously unpredictable world
of financial markets.

Analysis and prediction of fluctuations in the financial markets is
an elusive goal, but one that holds out the promise of great financial reward.
Some individuals – the successful traders who become legends on Wall Street
and in London’s Stock Exchange – seem able to it. Traders such as John Templeton,
Peter Lynch and John Dreman seem to see clearly enough through the turmoil
of the markets to edge ahead of them, and make huge profits in the process.
But what, precisely, do these financial geniuses do?

To most people, who can do no better than coin-flipping to predict the
movement of markets, the secret of their success remains elusive. Farmer
sees it merely as a challenge: ‘Our goal is to come up with a formula for
trading that can be executed in an automated fashion,’ he says, ‘and doesn’t
depend on some day-by-day brilliant insights by one individual into the
way the markets are behaving.’

MODEST BUT CHIC ROOTS

The Prediction Company is located in a modest-looking single-storey
house, just a few yards from the historic main square of Santa Fe, New Mexico,
the chic, artistic centre of the American southwest. Inside, the building
is white walls and polished wood floors. A framed picture of Einstein, not
yet hung, leans against a wall. Copies of The Wall Street Journal lie scattered
on a low table. Sun Workstations – intellectual status symbols – quietly
process financial data, slowly building quantitative models of financial
worlds that cheat easy description.

‘I trace the roots of our current venture back to the roulette project,’
explains Farmer. ‘From that we learned we could indeed predict things that
people said were impossible.’ The roulette project, an obsession of Farmer’s
and Packard’s, began in 1977, when they were both doing doctorates in physics
at Santa Cruz. They set out to prove that it was possible to predict, to
some significant degree, the fall of a roulette ball, and to make some money
on the way. With many of their closest associates, and now calling themselves
the Dynamical Systems Collective, Packard and Farmer started a project that
would run for more than a decade.

Most adventurers who seek to beat the wheel scientifically hope to spot
imperfections, such as a slight imbalance of the wheel that leads to a tiny
but significant bias in the way the ball falls. Farmer and Packard positively
welcomed a perfectly balanced, perfectly crafted wheel. Working with equations
of motion of a rolling ball on a circular track, the Santa Cruz crew built
computer models that would produce predictions on which bets could be made.
‘You’d look at the ball as it was rolling, look at the wheel and its behaviour,
and feed the information into the computer,’ explains Packard. ‘The output
would be a prediction of the octant of the wheel where the ball might exit,
and this narrowed down the range of numbers on which the ball might finally
±ô²¹²Ô»å.’

The project relied on a rule of roulette, which allows bets to be made
while the wheel is spinning. It also depended on the researchers having
the technical ingenuity to conceal a computer, input and output connections
and all, on an innocent-looking gambler. ‘The computer was built into a
shoe, and the input was by toe-operated switches,’ explains Packard. ‘We
tried all kinds of outputs, like little electrodes that gave shocks on certain
parts of the arm or the leg. But we ended up with a belt arrangement in
which there were three tiny buzzers, each of which buzzed at one of three
´Ú°ù±ð±ç³Ü±ð²Ô³¦¾±±ð²õ.’

CASINO FORAYS

Thus equipped, members of the Dynamical Systems Collective made forays
into casinos, where they pitted a creative combination of high and low technology
against the semi-chaotic behaviour of a spinning and bouncing ball. At the
time, the physics faculty already considered Farmer and his partners to
be misguided in their enthusiasm for studying nonlinear, chaotic systems.
For some, this obsession with the roulette project was confirmation that
the members of the collective were less than committed to respectable academic
studies.

They did win a few bets, though. ‘We made some money,’ says Packard.
‘Not much, it’s true. Certainly not much compared to the effort that went
into the project.’ But he agrees with Farmer that the roulette project was
a harbinger of their current venture. ‘The essential feature that has stretched
across all these years, linking what we were doing then and what we plan
to do now, is using data input to build models, from which we then make
predictions.’ Techniques for building models of financial markets are slightly
different from those they used for the roulette project, but both are based
on building models directly from data. These days, the data are economic,
political, and financial, not about balls and wheels and motion. The output
is – in theory – predictions of the day-by-day movements of prices and currency
values.

At first, prediction was not a compelling goal for the Dynamical Systems
Collective. Its members were simply enthralled with the intellectual novelty
of tackling nonlinear phenomena. Their chief interest was in looking for
structure in these highly complex and seemingly random systems. But even
as early as 1980, Farmer and Packard, in company with their Santa Cruz colleagues
Robert Shaw and James Crutchfield, wrote a paper that hinted at what was
to come. In it they described how, to a limited extent, it should be possible
to take a dynamical system such as a fluid flow, stick a probe into it to
determine its behaviour at a particular point in space, and from this reconstruct
the behaviour of the whole system.

Although Farmer and Packard long nursed an entrepreneurial urge, in
1980 the notion of applying the techniques of nonlinear science to financial
markets was barely formed. By 1985, the Dynamical Systems Collective had
achieved academic respectability, and had dispersed to various high-powered
institutions: Farmer to Los Alamos and Packard to the Institute of Advanced
Study at Princeton. In March 1985, Farmer was visiting Packard at Princeton,
and, prompted by a casual conversation with a financial expert also at the
institute, the two began to talk about interesting ideas. ‘We realised that
it should be possible to extract predictability from financial markets,’
recalls Packard. ‘We tried to figure out how we might be able to use the
ideas we were having about complex systems to get at that predictability.’
Perhaps they would set up a company, they dreamed. And this time, they would
make a lot of money.

The two men had identical intellectual goals: to identify structure
and extract predictability from complex systems. But they were using different,
though complementary, technical tools to reach these goals. The dream of
attacking the markets lingered at the back of each of their minds for another
four years. ‘By 1989 we realised that, with our different approaches, we
were close to being able to extract predictability from arbitrary signals,
including financial signals,’ says Packard. That realisation stemmed not
only from an ‘intuitive feel’ about how their analyses were developing,
but also from models of financial systems. Tentative attempts to forecast
market movement – with modest amounts of money at stake – were encouraging.

WALL STREET BECKONS

Packard’s keenness to put his analytical tools to the test in the real
world soon made him a target of Wall Street head-hunters. Farmer, meanwhile,
after a decade at Los Alamos, was growing weary of the administrative duties
that came with being head of his group. He was also becoming frustrated
with the shrinking research funds available for nonlinear studies, as ‘big
science’ projects sucked money away from the politically impotent small
science. The time was right to do something. The question was: what?

‘We realised that the quickest way to apply our tools, to see some results,
would be to join an established Wall Street firm,’ says Packard. ‘We would
be able to slot into an infrastructure ready to trade with the information
we could provide.’ Packard, now at the University of Illinois, was prepared
to move to New York; Farmer wished to stay near Santa Fe. As well as his
personal reasons for this choice, since 1984 the city had been home to the
Santa Fe Institute, a centre for the study of the emerging science of complexity.
This intellectual frontier was one with which both Farmer and Packard had
become deeply involved. So they decided to take the riskier, but potentially
more rewarding, route: they would establish the Prediction Company. ‘We
felt we’d learned a certain amount from the roulette project, and it seemed
a pity to waste the experience,’ Packard recalls.

This time, the venture would not rely on energetic but naive amateurs.
Farmer and Packard brought in business expertise in the shape of James McGill,
a member of the early 1970s Santa Cruz physics crew with a successful subsequent
record with start-up companies. Venture capitalist James Pelkey, a long-time
friend of McGill’s and a close associate of the Santa Fe Institute, became
lead investor. Half-a-dozen young researchers completed the team.

Starting with huge bodies of financial data, both historical and current,
their task was to build models – algorithms and computer programs – that
would describe the behaviour of the markets. These models should form the
basis for accurate predictions. ‘If we were to achieve a 60 per cent rate
of correct predictions, that would give us an overwhelming advantage,’ says
Farmer. Given that a 50 per cent rate represents chance, an extra 10 per
cent might not seem much, but because of the amounts of money involved,
it would represent an enormous profit. Even a 5 per cent margin would be
impressive. ‘I don’t want to give anything away about the standard of performance
we are achieving already,’ says Farmer, ‘but I’d say that certainly seems
to us to be quite feasible, the 55 per cent rate.’

EASIER THAN THE WEATHER

Economists and financial experts are likely to be sceptical about such
claims until they see hard evidence in the form of successful long-term
predictions. But, as Farmer points out, a 55 per cent success rate for other
complex systems, such as weather, would not be very impressive. ‘From that
point of view financial markets seem a lot easier than the things we’ve
tackled in the past, because the standards of predictability are a lot lower,’
he says.

So why have financial markets defied ready and consistent prediction,
except by a few individuals who clearly possess exceptional skills? ‘Markets
are complex entities that depend on the behaviour of many diverse individuals,’
observes Farmer. ‘Nonetheless, financial analysts read the same papers,
watch the same tickertapes, respond to similar rumours and act according
to normal human emotions such as fear and greed. The market is an aggregate
of their behaviour. When they act as a herd, the market may obey rules that
are quite simple.’ Farmer and Packard aim to discover these rules and use
them to predict the financial future.

Farmer’s description of financial markets as ‘complex entities that
depend on the behaviour of many diverse individuals’ is in fact a general
description of any complex system, the kind of challenge on which nonlinear
science cut its intellectual teeth. That is why Farmer and his colleagues
are convinced they will be able to shift their skills from the laboratory
to the market place.

PICKING POCKETS

The phrase ‘pockets of predictability’ is one Packard uses frequently.
‘What it means is that you are being bombarded with all kinds of financial
data – exchange rates, bond prices, interest rates, and so on,’ he explains.
The Prediction Company treats this information about the markets as they
would data derived from any complex system. At any given moment, all data
for that instant represent just one point in space. Thousands of these points,
representing the data for each of thousands of moments, will distribute
themselves throughout the space, according to the ‘shape’ of the system.
‘Some parts of the system may be empty or sparsely populated,’ explains
Packard. ‘Something from the outside may kick the system, and shoot it to
some other piece of space. For various reasons, therefore, there will be
areas that aren’t predictable. A limitation from external perturbations;
a limitation from limited data; and then some parts are unpredictable because
of chaos, where the dynamics of the system causes it to diverge wildly,
too fast to follow.’ Among all this will be pockets of relative stability
– pockets of predictability. These are the targets of the Prediction Company’s
models. ‘We are confident that we’ll be able to extract close to the limit
of predictability that’s in these signals,’ Packard says.

Economic theory concerning financial markets has changed through the
years, most recently coming to view them in terms of equilibria. For some
physicists, including Farmer, this is unconvincing. ‘To me, markets look
inherently unstable,’ he says. ‘The instabilities you see, the crashes and
so on, in an intuitive way speak against viewing markets in terms of equilibria.’
Whether markets are poised at equilibria or at the edge of instability,
the key question is how predictable they are. Economic theory has vacillated
over this point, sometimes as partly predictable. Those who accept some
predictability have attempted to exploit it mainly through linear tools,
for example trend following, and have had some success.

Nevertheless, financial markets, like most complex phenomena in nature,
are probably governed by nonlinear patterns. Now that nonlinear analysis
has become something of a trendy academic pursuit, it is not surprising
that these tools should be brought to the marketplace. The Prediction Company
is not the first operation of its sort, but when it links up with its financial
partner, it will be the biggest.

There are four or five groups applying nonlinear analysis to financial
markets in the US, and about the same number in Europe. Generically, they
may all be said to be similar, inasmuch as they derive their inspiration
from chaos theory and models based on nonlinear dynamics. But they are all
rather different. The ultimate comparison will be in their performance.
If one group does turn out to be dramatically successful, it will become
part of the market, and so could affect market behaviour. ‘That’s the kind
of problem we’d like to have,’ says Farmer.