I stumbled over some stats that the Thoroughbred Daily News had done a while back where they had asked four companies that offer predictive analytic services at the 2YO in training sales to create a 'fantasy stable' from the spring 2005 2YO Sales. Here is a link to the original data, but the four were asked to pick a stable of 20 individual horses which they tracked to see how they went on the racetrack.
I noticed that the data was incomplete as while they had run the data two years later (May 2007), there were a few horses that were still racing when they did the data pull who after that went on to do well with the most notable being Diabolical, an EQB selection, who went on to compete in multiple Group One events in Europe and is now a leading sire in New Mexico. So I ran the figures again and tallied the results.
The updated data certainly paints an interesting picture. There were 71 individual horses that the four of them selected together (Master of Disaster as an example was selected by three of them) and from that group of 71, 6 of them became Graded Stakes winners. 6 from 71 becoming graded stakes winners is 8.4% and when you consider that Fasig Tipton's Saratoga yearling sale has about 5.3% and it is the best yearling sale as far as that metric is concerned, 8.4% collectively is reasonably good without being exceptional.
The shining light of the group was Jay Kilgore's DataTrack International. Of the 6 Graded stakes winners they had four of them - Henny Hughes, Noonmark, Praying for Cash and Smokey Stover. Sure Noonmark was expensive (he RNA'd at $1.15m) but Praying for Cash ($150,000) and Smokey Stover ($140,000) were buyable horses. I'd hate to think how much in service fees, etc you would have made out of Henny Hughes.
Lesson's learned from the data is that prediction is possible at two year old sales, probably better than most other sales, but it is not easy and you really need to wait until the end of a horses four year old year before making any judgement on their racing class.