Trading Article
Library
Building Trading Systems Using Automatic Code Generation
by Michael R. Bryant
As more and more traders have moved to automated trading,
the interest in systematic trading strategies has increased. While some traders
develop their own trading strategies, the steep learning curve required to
develop and implement a trading system is an impediment to many traders. A
recently developed solution to this problem is the use of computer algorithms to
automatically generate trading system code. The goal of this approach is to
automate many of the steps in the traditional process of developing trading
systems.
Automatic code generation software for building
trading systems is often based on genetic programming (GP), which
belongs to a class of techniques called evolutionary algorithms.
Evolutionary algorithms and GP in particular were developed by
researchers in artificial intelligence based on the biological concepts
of reproduction and evolution. A GP algorithm evolves a
population of trading strategies from an initial population of randomly
generated members. Members of the population compete against each other
based on their fitness. The fitter members are selected as
parents to produce a new member of the population, which replaces a
weaker (less fit) member.
Two parents are combined using a technique called crossover, which mimics
genetic crossover in biological reproduction. In crossover, part of one parent’s
genome is combined with part of the other parent’s genome to produce the child
genome. For trading system generation, genomes can represent different elements
of the trading strategy, including various technical indicators, such as moving
averages, stochastics, and so on; different types of entry and exit orders; and
logical conditions for entering and exiting the market.
Other members of the population are produced via mutation, is which one member
of the population is selected to be modified by randomly changing parts of its
genome. Typically, a majority (e.g., 90%) of new members of the population are
produced via crossover, with the remaining members produced via mutation.
Over successive generations of reproduction, the overall fitness of the
population tends to increase. The fitness is based on a set of build goals that
rank or score each strategy. Examples of build goals include various performance
measures, such as the net profit, drawdown, percentage of winners, profit
factor, and so on. These can be stated as minimum requirements, such as a profit
factor of at least 2.0, or as objectives to maximize, such as maximizing the net
profit. If there are multiple build goals, a weighted average can be used to
form the fitness metric. The process is stopped after some number of generations
or when the fitness stops increasing. The solution is generally taken as the
fittest member of the resulting population, or the entire population might be
sorted by fitness and saved for further review.
Because genetic programming is a type of optimization, over-fitting is a
concern. This is typically addressed using out-of-sample testing, in which data
not used to evaluate the strategies during the build phase is used to test them
afterwards. Essentially, each candidate strategy constructed during the build
process is a hypothesis that is either supported or refuted by the evaluation
and further supported or refuted by the out-of-sample results.
There are several benefits to building trading systems via automatic code
generation. The GP process enables the synthesis of strategies given only a high
level set of performance goals. The algorithm does the rest. This reduces the
need for detailed knowledge of technical indicators and strategy design
principles. Also, the GP process is unbiased. Whereas most traders have
developed biases for or against specific indicators and/or trading logic, GP is
guided only by what works. Moreover, by incorporating proper trading rule
semantics, the GP process can be designed to produce logically correct trading
rules and error-free code. In many cases, the GP process produces results that
are not only unique but non-obvious. These hidden gems would be nearly
impossible to find any other way. Lastly, by automating the build process, the
time required to develop a viable strategy can be reduced from weeks or months
to a matter of minutes in some cases, depending on the length of the input price
data file and other build settings.
To comment on this article, click here.
