From: John Conover <john@email.johncon.com>
Subject: Re: Electric Money
Date: 4 Oct 2001 15:29:48 -0000
BTW, the tsinvest's -d5 option is based on the Farmer/Packard
concepts.
In the television program, Farmer discusses how to exploit the fact
that the mechanics of the game of roulette does not have infinite
entropy. Although in the long run, the mechanics will be statistically
consistent, (each number coming up 1 / 35'th of the time, for the
numbers 1 - 33, 0, and 00,) there is a slight bias to local
predictability in the short run.
This can be considered a systemic inefficiency in the random number
generator of roulette-the velocity of the ball when spun around the
wheel by the croupier is not perfectly random, (i.e., the number of
times the ball rotates around the rim of the wheel before striking the
wheel is not perfectly random,) and once the ball hits the wheel, the
chances that it will settle close to numbers where it hit is higher
than for the rest of the numbers.
What this means is that the time series of the winning numbers, over
many spins of the roulette wheel, will have a slight bias to contain
patterns. (Its an important concept that an infinite entropy
stochastic system is probably a mathematical abstraction that is not
perfectly obtainable in the real world-all such systems are chaotic
NLDS to some degree.) For example, if red 7 came up as the winning
number for the last spin, there is not exactly a 1 in 35 chance for
any other number to come up the next spin-some numbers have slightly
more of a chance, others less-a fact that is exploitable, (and just
about cancels the house's advantage,) and can be used to optimize
wagering.
A stochastic/Hamiltonian system of sufficient complexity that is
perfectly efficient will be characterized by a Normal/Gaussian
frequency distribution. However, if it is inefficient, the
distribution will have kurtosis.
What this means in arbitrage systems, (like the value of financial
instruments, over time,) is that price movements have persistence,
i.e., the chances of the next movement being like the last is greater
than 50%. Everyone does not react instantaneously to market
information-which is a systemic inefficiency. (The metrics are that
about half of the share holders react to market information on the
same day the information came out, about half of the remainder on the
second day, and so on-an exponential decay scenario.)
So, in a perfectly efficient market system, the chances of the next
movement being like the last is 50%. The chances of the next two
movements being like the last is 25%, and so on.
However, in a slightly inefficient market system, the chances of the
next movement being like the last, (using daily median metric
numbers,) is about 57%, and the chances of the next two movements
being like the last is 32.5%.
There is an added advantage to doing things this way-the wagering can
be optimized, (i.e., a portfolio management scenario can be
implemented,) with F = 2P - 1, where P has a median value of about
0.57.
What tsinvest does internally, with the -d5 option, is to construct a
table of consecutive like price movements for each stock, (for both
positive and negative movements,) and calculates the probability of
the next price movement being an up movement, and how much to wager on
that probability.
John
BTW, note that just recognizing price patterns is not sufficient-one
has to figure out what to do with that information, too.
John Conover writes:
>
> PBS is airing the program "Electric Money" by Robert Cringely:
>
> http://www.pbs.org/opb/electricmoney/
>
> which has a short section on Doyne Farmer, Norman Packard, and
> the Prediction Company:
>
> http://www.predict.com/html/introduction.html
>
> John
>
--
John Conover, john@email.johncon.com, http://www.johncon.com/