Hello, David,
genetic algorithms are not used in trading directly. They help us to arrive at the best model (model parameters) and variable values (value parameters).
Model parameters determine what concept or model will be used in the strategy in question. Example:
- we are not sure which moving average calculation will be the best one to use with our trading strategy, so we design an input like
iMAType. Let's say that the range for this input parameter will be 1 to 4. According to the selected input value (1, 2, 3 or 4) we calculate the moving average within the strategy either as a simple, exponential, weighted or triangular one.
Value parameters are pretty self-explanatory. Their value is used as an input to already existing model. Example:
- we want to know what moving average length is the best for our strategy, so we design a variable like
iMALength. This variable is used in the strategy code directly as an input for calculating the moving average length.
The two main advantages of GAs are:
- their fitness landscape is shaped gradually, because they don't try all possible combinations. This means that GAs are not easily fooled into local maxima
- the solving speed is tremendous compared to the standard exhaustive methods
I use GAs in my system development every day. My best strategies are designed using GAs. Of course, GAs are no holy grail - the system developer must still know very well what is he doing and put a lot of hard work into the development.
The markets have become more efficient since 1990ies due to the widespread computer analysis software. If you want to gain an edge, it's necessary to use computer aided system development tools like, for example, GA.
Another interesting AI tool used in trading strategy development might include back-propagated neural networks, but that's another topic
Please feel free to suggest additional important links relating to this subject.
