Ok, this is fantastic. Thanks for evolving how we track bitcoin prices and activity. I presume you're doing this to come up with more reliable predictive analytics so we can all determine which direct the future price and activity is heading?
A few questions:
(1) Have you test your information to prove that your genetic algorithm correctly predicts (relatively speaking of course) the future?
(2) What's your track record for getting it right?
(3) How do we us your predictive analytics to help our buying or selling behavior?
Maybe you're not to that point yet, maybe you're at the point of sharing the analysis. All of it's good, just excited to see where this takes us!
This strategy still uses tehnical analysis and its lagging indicators, so it can't predict the future. The algorithm is excelent at spotting patterns, better than we humans are. It can't predict black swan events, it can't predict mtgox event or coinbase going offline... well maybe that last one it can
The point is, it spots events faster than we usually can, but it can't forsee them. At least not yet.
We are developing a more general strategy that should work on any market and should adapt to any market conditions. We are also working on something that is not just using lagging indicators for its strategy, but it is still too early to say anything about it.
You can see the track record from results, it spots uptrends really well and is usually very quick at correcting itself if it's wrong. It won't try to catch falling knives and it won't try to sell the top too often. The goal was to beat buy&hold strategy, which it did.
It is not that hard to find the "correct" strategy in backtesting. The algorithm using GA will evolve some magic formula for the backtest dataset and craft the strategy for it. Here is one example -
http://imgur.com/xnDBnia. The problem is, if you adjust a dataset just a bit, the strategy won't be necessary best anymore. If you shift the candlestick just for a few minutes, you can get very different results -
http://imgur.com/nnE44CiIt was a real challenge to find a strategy that works on more general datasets. This is where GA helps the most.