http://www.nyxdata.com/Data-Products/Daily-TAQ

However, doing things like producing a histogram or tracking the individual numbers for a given fund require that you be able to identify individual traders, which is generally not possible. Funds will route their orders through brokers specifically to anonymize them, often slicing a given order up into many sub-orders so that it can be spread across multiple brokers. The golden rule of the financial industry is to never ever let your competition know what you're doing; if it gets out that, say, Goldman Sachs is buying a whole lot of Google stock, you can bet that the price of Google will rise and it'll cost Goldman a whole lot more.

When I was still tangentially involved in financial software, there was a lot of interest in algorithms that could heuristically identify possible actors based on patterns in the data, much like how Google identifies webspam or PayPal identifies fraud. I'm not sure how far this got, though, and obviously if some company succeeded in it, they wouldn't tell anyone. Actually, I kinda wonder if this is the sort of stuff Palantir is doing.

"In recent years, what is known as high-frequency trading — rapid automated buying and selling — has taken off and now accounts for 50 to 75 percent of daily trading volume."

From what I've read, its gotten pretty ridiculous overall.