Rentech, in particular, data mines weak signals out of gigabytes of historical data, and has a team of PhD mathematicians working to find these signals. Each strategy has some small positive expected value, so in aggregate they can be profitable.

Two reasons you can't replicate this - 'you' aren't a team of super top notch mathematicians with decades of data, and even if you had an exact copy of their trading strategies, you don't have the capital to make the transaction costs, etc, worth it.