A bit after the fact, but while I was sleeping a buy signal was generated:
BTC Buy at [end time: 10/08/2016 00:00:00, close price: 595.121000]
This is interesting. Is the bot profitable or do you think it can be profitable than a buy and hold type of trading strategy? Sometimes I think the more active a person is in trading the more chances of him losing his money. For a bot to do this I think those chances are magnified.
A bot 100% eliminates emotion from the equation (a huge source of bad trades). It also does exactly what you tell it to do, and only that. The biggest problem is, when you see a bot in action (and believe me, they leave traces of their presence) you can learn how the bot reacts to a situation and it becomes game-able and/or predictable. Nothing worse than a bot that can't win and due to predictability, loses money very fast.
At least according to the regression analysis portion, the top strategies theoretically produce some 200% returns over buy and hold (net fees).
Eliminating emotion was one of the big motivations for developing this. The predictability and "action signature" of a bot is also a good point and something I'll be mindful of as I continue to develop the real-time component. The strategies that have bubbled to the top are almost all pretty obscure (although you might be surprised at how profitable buying on Monday and selling on Wednesday has been for ETH!) and not easy for a human to intuit or track in working memory. That said, by leveraging visibility into the exchange order books (among other things) there are several techniques I can employ to obfuscate activity and "blend-in" so to speak.
The real-time portion of the application is the currently the least mature. I have a good amount of experience developing bots in other domains so I'm excited to see what I can do here.