Inventing Metrics For The Speed Chess Championship
Today's Speed Chess Championship match between Hikaru Nakamura and Fabiano Caruana will feature something truly special that I've been working on for a very long time: most recently with the help of IM Danny Rensch and Gerard Le-Marechal who is head of Chess.com Fair Play and Game Statistics. But before I get into that, let me give you a brief history...
On January 13, 2012, IMs Daniel Rensch and David Pruess squared off in the first ever Chess.com Death Match. The format then was the most brutal of any match that had ever taken place online: Three straight hours of bullet chess. That's it. Just 1-minute madness while both players simultaneously commented on their own games.
IMs Danny Rensch and David Pruess playing in the first Death Match courtesy of the wayback machine.
Seeing that for what it was (craziness of an unhealthy level) IMs Danny and David, who at the time shared the duties of Co-Directors of Content and Professional Relations, adjusted the format to allow for both more high-quality games (at least to start) and for the players to focus on the chess, while commenters did the job of live analysis. Over the last five years, the original "Death Match" has evolved into what is today: a combination of 5+2, 3+2, and 1+1 games to determine who the better online blitz and bullet player is. The format originally became the GM Blitz Battle Championship last year and is now known as the 2017 Speed Chess Championship.
Being a long-time member and a big fan of these events, a few interesting statistical questions occurred to me, and I knew I needed to have answers! I began to gather data and track "live results" of the games, tweaking my own algorithm here and there in the hopes of creating something that could help us understand, and perhaps predict, who is not only favored to win one of these matches, with its very specific format, but also when and if the "match favorite" is no longer the favorite (AKA, track live match-odds as you will see in major sporting events such as poker).
Some statistical inspiration
In the last few weeks, I have been working closely with the Chess.com stat team to develop some new metrics to help us better understand the Speed Chess Championship results. Here is a short description of the new metrics we've created and the really cool stuff you'll see on the live show today as Nakamura and Caruana face off...
DeathMatch (DM) Rating:
The concept of the "DM Rating" is to apply an Elo rating to how well players perform specifically in the Chess.com Death Match format. The initial DM rating for each player is currently based on a combination of their classical FIDE rating at the time, blitz FIDE rating and a few other Chess.com metrics we won't get into now .
After each match, a new rating is calculated for the players based on their match performance. The current top DM ratings have all performed well in the online death-match format. Here are the current ratings for the players remaining in the 2017 Speed Chess Championship.
- Magnus Carlsen (2886)
- Sergey Karjakin (2863)
- Hikaru Nakamura (2807)
- Alexander Grischuk (2779)
- Wesley So (2741)
- Fabiano Caruana (2649)
Without getting too technical, or revealing too much at this point about our secret sauce (algorithm), what we can say is that in tests (and in theory), we are able to predict the expected score of the match (the mean), as well as the random variation that happens from match to match (the standard deviation).
Those two variables together allow us to predict a %Chance of each player to win the match! Just as in poker, we are able to show the %Chance of the players winning the match while the action is live! In addition to the %Chance to win, we have predictions for the expected score for each player pre-match and updated live.
Predicted %Chance to win for one of the most competitive speed chess matches! Note that Giri's chances are inflated initially because of his lack of game history on Chess.com prior to the match.
What in the world is a "Fire Factor??? That was Danny's spontaneous name for one of our favorite new statistics, and it stuck! (NOTE: I think Danny is spending too much time with Twitch gamers, and now thinks of everything in terms of whether or not it is "lit.") The Fire Factor is a measure of how much the %Chance of each player winning bounces around during the match. The more exciting the match, the higher the fire factor! The Fire Factor ranges from 0 (dry) to 100 (en fuego!). We are working on how to display this data live during the matches as well (and it may not be featured today). Here are the Fire Factors above 50 in the 2017 Speed Chess Championship matches so far:
- Wesley So vs Anish Giri (97) (a clear #1 as the match went to overtime with dozens of "lead changes")
- Levon Aronian vs Ian Nepomniachtchi (94)
- Alexander Grischuk vs Richard Rapport (59)
- Fabiano Caruana vs Hou Yifan (53)
- Alexander Grischuk vs Maxime Vachier-Lagrave (51)
These things and more are coming your way in future matches like today's! I can't wait to see the reaction of fans to this project that I have worked so hard on. I'm truly grateful and excited to work with Chess.com directly on bringing it into the limelight!