You are on page 1of 3

Introduction

The controlling logic of the market analysis component within the majority of the Quadrant Capital
models is produced by the process of Genetic Programming.
Genetic Programming (GP) models the same evolutionary process that occurs in nature, with the
purpose of breeding strategies that are capable of profiting from financial market activity. It uses
analogues of naturally occurring structures and processes in order to create software programs
which are capable of prospering within their environment, which is market price activity. The
process is non-deterministic in terms of the output produced; there is no guarantee that the fittest
will be found, merely that the found solutions will meet predefined criteria which define fitness.
This search is constructed by combining the output of a number of technical indicators or
analytical routines (ARs). ARs differ from classic technical indicators in that each AR is configured so
that its return value is a state, which is descriptive of some aspect of market behaviour or pattern.
Typically, each AR has multiple possible states, from a minimum of two or three up to roughly
fifteen or twenty. There is no technical limit on the number of possible states, however the process
works better when there are fewer. Multiple versions of the same AR can be developed to describe
different aspects of the same pattern, each having its own collection of differing states.
When combined, the output of these ARs provides a multi-dimensioned description of the overall
state of the market. The size of the search space created by these ARs is defined by the
combinatorial product of the number of states produced by all the ARs used as components within
the process, and the result is therefore usually a very large number, unsearchable by brute force.
This is to be expected, since the entire market problem, i.e. designing and building profitable
trading strategies, represents a truly vast and possibly undefinable search space. However, the
essential properties of GP as a practical discipline, make it a highly suitable approach to finding
solutions from within such a large search space.
There are many subtleties and nuances to this process, not least of which is the validity of
information being extracted by the ARs. The concept of information is a very important component
to the entire GP process; information is one of the fundamental components of reality, and
information is present in market price action.

Why Genetic Programming ?

Genes carry information. Genes within any living organism comprise the instruction manual for the
organism to replicate itself, both physically and behaviourally, according to the pattern defined
within. Thus the link between genetic material and information transmission should be obvious.
Information Theory is the branch of science concerned with the quantification of information.
Originally formulated to address the problem of encoding and transmitting information to a remote
receiver across a noisy channel, and published as A Mathematical Theory of Communication in
1948 by Claude Shannon, the theory holds that :
Information has predictability, based on the probability of any given outcome. Pattern and
information are deeply entwined.
The existence of pattern allows compression. Meaning can be deduced without the need for the
actual transmission of actual bits (binary digits, the smallest component of data in computer
science).
Pure randomness cannot be compressed because it contains no pattern and therefore no
information.
The inverse of information is entropy, which is uncertainty or unpredictability. The more
uncertain or unpredictable, i.e. the higher the number of possible outcomes, the greater the
amount of information is required to surmount that uncertainty.
Market price action is composed of repeating patterns in both the price domain and the time
domain (cycles). If pattern did not exist then it would not be possible to profit from financial markets
consistently, since, according to the theory, the absence of pattern is indication of randomness.
Conversely, the existence of pattern in market price action implies that information is present and
thus that prior to the patterns completion, its outcome can be predicted. This is the source of the
opportunity for profit in pure outright trading of financial markets.
One further component of Information Theory is the fact that boolean algebra provides the most
efficient means to encode and/or transmit information. Boolean algebra, named after 19
th
century
mathematician George Boole, is the formulation of an algebraic logic system, which demonstrates
that, given any proposition, given any number of terms, it is possible to deduce conclusions based on
the premises through the reduction of the contents of the original hypothesis to a collection of true
or false statements.
This truth has become the basis of modern computing. In computer science, boolean variables, i.e.
variables that store the value true or false, are the fundamental building blocks of all programming.
A logic gate can be either on or off, 1 or 0, and the fact that only two states are needed to describe a
basic decision process leads to the creation of tiers of logic founded on binary, i.e. base 2
mathematics.
In his Information Theory thesis, Shannon uses boolean switching as the process to sort through the
finite set of possible messages, i.e. to overcome the entropy. It is demonstrably the quickest way to
sort a large set of possible outcomes, since at each decision point the program only has to switch
between 2 possible outcomes. The product of successive switching is a population of possible
outcomes whose size is 2 raised to the power of the number of switches. The magnitude of the
efficiency of boolean switching is such that a population of 32,768 possible outcomes is created by
raising two to the power of 15. In other words only 15 boolean switching operations are required to
sift the entire population, as opposed to traversing every single one of the 32,768 possible outcomes
sequentially.
One of the greatest challenges facing developers of systematic strategies is the ability to describe
the market behaviour to the trading logic component sufficiently accurately. Taking, as a very
simplistic example, the idea of a moving average crossover system, there are very few states to be
described; fast average above slow, fast average below, etc. Thus the results of a system of such
simplicity suffer from extreme volatility, as very little information is actually being extracted from
the market by the indicators. The application of an approach to market analysis founded on the
principles established by Boole and Shannon however, allows us to provide a very rich description of
market state or behaviour to the trading model, and this information can be transmitted to the
strategy very efficiently.

You might also like