New Chess Rating System More Accurate

Associate Professor J. Isaac Miller

A new chess rating system developed by a team including J. Isaac Miller, an associate professor of economics at the University of Missouri, creates a single rating that has proven to be more accurate than the ratings systems currently in use.

Jordan Yount
News Source: 
College of Arts & Science

Chess players who are rated by the World Chess Federation are ranked three different ways—how they fare in a classical game that allows two hours for the first 60 moves, how they fare in a 30-minute rapid chess game, and how well they perform in blitz chess, which is a game played in five minutes or less. But a new chess rating system developed by a team including an associate professor of economics at the University of Missouri creates a single rating that has proven to be more accurate than the ratings systems currently in use.

J. Isaac Miller says he was approached by the Chess Club and Scholastic Center of St. Louis and the Kasparov Foundation to work with the developers of two other chess ratings systems to see if he could help to improve upon their earlier work. Mark Glickman of Harvard University had developed the Glicko rating system, and Jeff Sonas, founder of Sonas Consulting in California, had created the Chessmetrics rating system.

“I was contacted to be a third team member to come up with some ideas the other two hadn’t thought of, and I bring a different skill set—my specialty is statistical time-series research—so I was brought in to bring a new perspective,” Miller says. Maxime Rischard of Harvard later joined the team as a fourth member. Miller says the Chess Club and the Kasparov Foundation wanted a rating system that effectively weighted games played at different time controls—classical, rapid, and blitz.

Racing the Clock

Miller says a person with basic chess skills will be as good as they can possibly be at the game if there are no limits on time, but as time constraints are imposed, mistakes are made. The faster the game, the higher the probability a player will make a mistake.

“This is one of the things we use in the Universal Rating System (URS)—we want to use the information from fast games because they do reflect how well someone plays chess, but we weigh the fast games less because there is a greater probability someone is going to screw up,” Miller says.

Miller says his team compared their URS to the rating system used by the World Chess Federation in a recent blitz chess championship. The federation’s rating system predicted the winner of a given game 44% of the time, while the URS predicted the winner 67% of the time. Miller says he has compared their rating system to other existing systems, and the URS comes out on top every time.

“The reason why is because the URS is based on the statistical principle of maximum likelihood, so it uses more data in a statistically smarter way than the old ratings systems,” he says. Miller says maximum likelihood simply means choosing a set of parameters (in this case, chess ratings) to make the data one observes be the most likely to come from some statistical distribution.

He says another major difference from existing systems is the URS reassesses the ratings of all players in the system at regular intervals, placing less weight on older games in order to choose ratings that best reflect recently observed outcomes.

Miller says his new chess rating system already has been adopted by the Grand Chess Tour, which will publish monthly ratings that will be used for invitations to Grand Chess Tour events in St. Louis, Paris, London, and Leuven, Belgium this year. He says the Grand Chess Tour is an international circuit of high-profile chess tournaments, including the London Chess Classic and the Sinquefield Cup in St. Louis. In 2014, the U.S. Senate designated St. Louis as the National Chess Capital.

Share This

Facebook icon