While I was there, I thought about doing a product that'd try to predict the algorithms other people were using and profit off them, but you run into some really thorny epistemological problems. For starters, you generally can't find out which firms are making which trades, unless you're the firm itself (that's how our analysis product worked; it was for firms to evaluate their own performance, not their competitors). Most buy-side firms aggregate their orders through a few brokers, which keep their clientns' identities confidential. So you see "A sell order for 100 shares of MSFT at $33.01 came through Morgan Stanley", but you have no idea which fund placed the order or what other trades came from them. The signal gets lost in the statistical noise of all the other trades.

There's also no absolute standard of value in finance. Sure, you can benchmark values against discounted cash flow (assuming that you can even predict the cash flows, which is far from certain for most companies), but it may take years for a stock to revert to fair value as measured by discounted cash flow. Many of these automated trading systems operate over minutes or seconds, and don't care that the stock will eventually crash in 2-3 years.