From: scerir (scerir@libero.it)
Date: Sat Jun 09 2001 - 09:59:06 MDT
http://www.economist.com/science/displayStory.cfm?Story_ID=638530
There is a widespread assumption that, over three centuries
of stockmarket trading as we know it, any patterns that do
predict the future will already have been identified and
"arbitraged" out of existence by people taking advantage of them.
David Lamper and his colleagues at Oxford University disagree.
As physicists and mathematicians, they frequently find
themselves dealing with complex natural systems, and in a paper
submitted to Physical Review Letters, they apply that
experience to the artificial world of market mechanisms.
[A paper is at
http://xxx.lanl.gov/abs/cond-mat/0105258 ]
They suggest that price patterns do exist that have not been arbitraged out.
These patterns are created by the collective actions of market traders
themselves-and could be used to predict sudden falls in a market.
To explore what that might be, he and his colleagues built a computer
model which attempts to emulate market behaviour by running a large
number of software "agents" at the same time. These agents "trade"
with each other according to certain strategies. (Similar trading strategies
are used in the real world to carry out automatic share-dealing.)
When both real market data and random simulations of such data
are put before traders, they claim to see patterns. This is no surprise.
People see patterns, where there are none, all the time-in clouds,
in lotteries, in mountains on the surface of Mars, and even in root
vegetables. Predicting a number in a lottery based on such a spurious
pattern cannot, of course, have any effect on whether that number is
actually drawn. But the same is not necessarily true of the stockmarket.
When many traders think they see the same pattern, they may respond
in the same way. They may thus, collectively, create a real pattern.
And not only real, but unstable. For, paradoxically, the more orderly a
market appears, the less stable it is. (Think of dominoes up-ended on
a table. If they are distributed at random, knocking one over causes
little damage. If they are in a line, the whole lot will come down.)
It is the changes in strategy that lead to this orderly instability which the
researchers say they can detect. When their system finds many traders
crowding into one type of strategy, they know that a large downward
change may be imminent. The super-strategy to beat this is obviously
to sell first or even, for the brave, to go short.
The first test was against minute-by-minute historical data from a
foreign-exchange market: a London bank's dollar-yen market between
1990 and 1999. The model detected enough selling and buying
opportunities during this nine-year period to yield a 380% profit.
Being canny, the researchers are anyway keeping details of the algorithm
they use to generate their results under wraps until the patent they
have applied for is locked up. Nor are they too worried if the details
of their super-strategy leak out, since they think they could detect its use
in the market. That would allow them to play a "super-super-strategy"
that made use of the knowledge. On that, they are keeping super-quiet.
[In a different paper R. Mansilla used
http://xxx.lanl.gov/abs/cond-mat/0104472
the algorithmic complexity approach to the
understanding of the complex behavior of financial
markets, and their "pseudo-random" patterns.
He offers some support for the idea that predictability
increases just before big changes.]
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:08:02 MST