Chaos Theory And Neural Network Analysis
by John Kean
Do markets have memory? In this new theory by Edgar Peters, an offshoot of chaos theory, John Kean
explains, markets can be regarded as nonlinear dynamic systems, and neural networks can be used to
analyze and gauge behavioral patterns in price change data useful in prediction.
It's convenient that, at a time when the T behavioral natures of commodity and financial markets are
being rethought, a new technique for examining the data has appeared. Chaos theory, in its in fancy,
challenges the statistically based conventional view of markets as unforecastable series of random events
. A recently published work by Edgar Peters presents evidence that markets do have memory, which
results in price change trends, compared with reversion to a mean. To the extent that markets are indeed
chaotic, they are best thought of as nonlinear dynamic systemsócomplicated and difficult to deal with
using normal analytical methods. Neural networking is a powerful new method of dealing with data that
can provide a test of whether market prices are in any way prophetic, and so whether the conventional
view of markets holds.
Neural networks are an advanced type of artificial intelligence in which the system teaches itself to solve
problems. The internal methods by which the system performs its self-teaching had their origins in the
ongoing studies of animal and human intelligence. The process involves furnishing the program with a
paired series of input data that leads up to an output event, which in turn would be used to form a
prediction. The neural network program then takes the series of inputs, computes its own outputs and
compares them to the actual outputs. In the process of trying to get as many of the computed outputs within the given tolerance limits of the actual outputs, the neural program modifies connection strengths
amid the neurons that make up the network and thus trains itself. A succession of runs is made through
the data until the level of accuracy that is desired is reached. At that point, the "trained" network is ready
to be applied to real-world situations.