The real problem is not in finding an algorithmic strategy that seem to work on historical data - the problem is how to validate sure it will work in the future and how to be confident it's still working even if it's loosing money for some time ?
Still it’s a really fascinating field to explore.