There's an interesting competition "Elo versus the rest of the world" on kaggle.com that is apparently some sort of website for oganizing statistical competitions. Anyway, the task is to invent a system that predicts chess results as accurately as possible - ie. that can predict the future game scores of vast pool of players when large number of their past game results are known.

This is one of the main purposes of different chess rating systems and of course the most prominent among these has already for some fifty years been the Elo system invented by Arpad Elo. Several competitors have already devised systems that seem to make better predictions (shown here: http://kaggle.com/chess?viewtype=leaderboard) than the Elo system (marked in the table by "Elo Benchmark" that sits in #53 at the time of writing).

You can read more at http://kaggle.com/chess. Or perhaps you would even like to devise your own system altough then you better have some serious expertise on statistics as most of the current competitors no doubt have.

Looks interesting to me :) I hope the winning algorith actually sees practical use, whether on chess websites or in official national ratings.