http://www.cs.ucf.edu/~kstanley/
http://gar.eecs.ucf.edu/
http://nerogame.org/
These three websites explain to an extent a new application of genetic networks. I'm currently doing some research in conjunction with University of Texas and Columbia University to develop a testing model for forex. Here are the two models we are testing:
Model 1
Phase 1: Initial Seeding
During this phase, I just threw four different trading systems:
- Trending
- Ranging
- Scalping
- Backpropagation network.
I then had all of them run against one year of data. The system then picks the system with the highest success score (see my signature "How I Rate Trading Systems").
Phase 2: Evolution
During this phase, the system looks at the past performance of Phase 1, and then moves forward in one-week increments. During every week:
- Backtest to get the success score
- Pick system evolution with the highest success score
- Mutate/Evolve the systems which lost.
- Go to step 1
I have gone through 77 weeks and generations so far. There are a total 43 offspring in this current generation. Average success score is about 160, ranging from 50 to 200.
Model 2
Phase 1: Initial seeding
Straight coin toss is used. Heads long, tails, short, which trains the network. Network is trained to try and predict whether the market will first hit 10 pips long, or 10 pips short.
Phase 2: Evolution
After the intial seeding, every tick will be given a prediction, and a likelihood score.
Example:
Current bid: 1.23456
Long Score: 80 (out of 100)
Short Score: 20 (out of 100)
The success score of this model is around 90.
Obviously, the first project is starting to show more promise, and I would love for there to be a competing project hosted here on ForexFactory (Oh Twee-ee...).
Anyway, have a look around. I would love to see how the FF community evolves this project =D