Be rk <= (ri, rj) a directed relation rk between the ordered pair of relata ri and rj, with rk ≠ ri and rk ≠ rj. We will call a relation a terminal iff ri = rj. Furthermore, we call ri (the left-hand-sided element in the ordered pair) the normative relatum and rj (the right-hand-sided element in the ordered pair) the associative relatum.
Definition:
As I have shown in a previous post in GP entry and exit decision rules are encoded in a tree form. The decision rule tree returns a boolean value for every processed bar, which is interpreted as an entry (root node returning true) or exit (root node returning false) signal. Some authors suggest using a single rule tree for entry and exit signals, but I personally prefer evolving dedicated rule trees for both entry and exit rules, as I believe them to produce better signals.
Writing a software program that creates - or to be more exact, evolves - trading strategies with genetic programming (GP) requires a set of design decisions to be taken concerning different aspects. In this article I will presume that the goal is to evolve a trading strategy consisting of technical indicators only, which will return entry and exit signals. Some elements like stop and limit order can be co-evolved with the basic entry and exit signals. Although it is also possible to use GP for stock selection purposes based on fundamental data, we will not look deeper into this possibility.
