Any genetic programming resources available for developing software that evolves trading systems from historical data? I would like to develop free software for doing that.
that is my idea also. I don't trade with computers, but I asked if anybody was working on evolving systems here a while back. I still get ridiculed for that question from time to time. good luck keep me posted if I was doing it today, I would include more information than just historical price data. Like fundamental info that any econ 101 student could understand. GDP, QE, money supply, etc.
I'm just curious what you have in mind in a little more detail. I'm somewhat familiar with genetic algorithms for optimizing trade system parameters. It sounds like you're looking to apply GA in some other way.
Check this book out: http://www.amazon.com/A-Field-Guide...356700474&sr=8-2&keywords=genetic+programming This is really really good starting guide, highly recommended. It has some kind of framework example in java code and also very extensive appendix with links for further research. I think even some curious layman could read this book. Good value for the money.
http://www.amazon.com/gp/search?index=books&linkCode=qs&keywords=0792376013 I change parameters depending on market conditions but this might be better and will poke around.
GA has never been proven to work for building trading systems that work well over time. They tend to "curve fit" the backtested data. I don't think it's possible to program GA not to curve-fit.
I think there are solid statistical measures to do that. Biggest problem is: What time-frame ? Next 5 minutes, 5 hours, 5 days, 5 weeks, 5 months, 5 years ? If your average time-in-trade is 2 hours, knowing the 5 day or 5 week trend is somewhat worthless, no ? Likewise, if your average time-in-trade is 5 weeks, knowing the 5 hour trend is also worthless. First thing you want to determine: entropy http://www.investopedia.com/terms/e/entropy.asp#axzz2GMU6Zlnv
That is a good application for GAs, called a 'classifier' role. Train bots to identify uptrend, downtrend and chop. Problems with GAs, what does the chromosome look like? An array of bytes or a b-tree structure? How do you splice 'crossover' the b-tree at evolution time? How to avoid overfitting with a GA? If your chromosome is large enough to solve the problem it is large enough to 'remember' key positive or negative specific in your sample data.... Where can you get enough sample data to completely test insample/outsample? How do you insure that a system that is 'random number' based (evolution) is bug free? What is the 'fitness' value in a trading system you are going to evolve the system to find? Net profit? (too risky, large drawdowns but big profits) Sharpe ratio? Develop your system and then put in on cheap GPU cards for parallel processing in CUDA. GAs are much more fun than NN as you can see them learn. NN are more for predicting, GA are more about responding to events. Good trading...