Fabiano Caruana and Paul Richter: Who had a better week?
You probably think I’m joking. The whole chess world knows about Fabiano Caruana. The 22-year-old Italian/American, playing in the strongest tournament in chess history, the Sinquefield Cup, has scored an unbelievable 6-0 in the first six rounds. He has beaten Topalov twice, Carlsen, Nakamura, Aronian, and Vachier-Lagrave.
The commentators are struggling to find any comparison. The only other person to start a super-GM tournament 6-0 was Karpov at Linares, 1994. Karpov’s opponents were not too shabby: Lautier, Bareev, Illescas Cordoba, Topalov (who thereby becomes the answer to a trivia question: Who was a victim of the two best individual tournament performances ever?), and Polgar. But there aren’t any world champions on that list. Caruana’s performance includes a victory over the world champion, and I think that makes it a little bit more impressive.
On the other hand, the chess world is not buzzing about Paul Richter. Nobody even knows about him except readers of my blog. But I would say he has almost had as good a week as Caruana. While Caruana has gone 6-0 against his rating equals, Richter at this weekend’s CalChess state championship went 4½-1½ against people rated much higher than him.
How do the ratings compare? In six games, Richter had a performance rating of 2577, which was 386 points above his initial rating of 2191.
By comparison, Caruana supposedly has a performance rating of 3596. But in my opinion this is BOGUS. Let me explain.
When I started out in tournament chess, the procedure for computing performance ratings was very simple and clear. You take your opponents’ ratings, add 400 if you win, subtract 400 if you lose, and then average the results for all your games. If you do this for Caruana’s tournament, you get 3196, not 3596.
So why does Caruana’s performance rating show up as 3596? Because some genius decided to tinker with the Elo rating system. When a player goes undefeated, the thinking was, the games don’t really tell you how strong he actually was.
For example, let’s suppose that a 2000 player goes to a tournament and plays against a bunch of 1300 players. Then the maximum performance rating he can possibly get, using the method I just described, is 1700. But really the games don’t tell you that his strength is 1700. They tell you his strength is at least 1700. In reality, it could be anywhere between 1700 and infinity.
S0 how do you interpolate between 1700 and infinity? You can’t give a player an infinite rating, because then it would be infinity forever. So this genius had a bright idea: let’s add 400 more points! So now the player’s estimated strength become 2100, not 1700.
While I understand the argument, I disagree with it. The thing is that 400 points is just an arbitrary number. Adding it once is arbitrary, and adding it again is equally arbitrary. What really matters is that the rating system should be consistent. There’s no reason that somebody with a score of 5½-½ should get a bonus of 400 points per win, while somebody with a score of 6-0 should get 800 points per win.
So if we compute the performance ratings in a consistent way, Caruana’s performance for six games is 3196, which exceeds his pretournament rating by 395 points. That is better than Richter, but only by the slimmest of margins.
Curiously, when I looked up Richter’s rating history I saw another very strange phenomenon that also may indicate a defect in the rating system. His real (not performance) rating rose from 2191 to 2271 after this weekend, and I assumed that was his first time over 2200. But it isn’t! I was surprised to see that after his second tournament, way back in 2008 when he was ten years old, he had a provisional rating of 2211!
How on earth did that happen? Well, after his first, five-game tournament he had a provisional of 1373. Then in his second, four-game tournament he went 4-0. His performance rating for that tournament was 2027 if you use the consistent 400-point method, but 2427 if you use the Fabiano Caruana 800-point method. His provisional after the second tournament should be a weighted average of the two tournaments. Let’s see how that works out.
(5/9) x 1373 + (4/9) x 2027 = 1668: new rating according to the consistent method
(5/9) x 1373 + (4/9) x 2427 = 1841: new rating according to the bogus 800-point method
Neither one of these is even close to the 2211 rating he actually received. So I think that what’s going on here is actually something else: bonus points, or feedback points, which were instituted to enable underrated people (especially young players) to get their ratings up to their true strength faster.
I’m not really sure how bonus points are computed nowadays. In the dark ages when I started out, you got an additional 1½ points for each point you earned above 32. If you add this in, you get:
1668 + 395 = 2063: new rating using the consistent method with bonus points
1841 + 654 = 2495: new rating using the bogus 800-point method with bonus points.
So I still can’t explain how the USCF computers got 2211. But the main point I want to make is that whatever system they used, it way overcompensated. After the spike to 2211, Richter’s rating gradually dropped down into the 1700s, which is probably where it should have been in the first place. I think that the use of bonus points with provisional ratings is rather suspect, because provisionals are quite labile anyway. I’d rather see a system where the bonus points were phased in gradually, although this has the disadvantage of making the system more complex.
Anyway, the messed-up system that we have makes Richter a possible answer to two trivia questions. One is: Who had the largest rating increase from one tournament? His rating went up by 838 points. Maybe someone out there has had a bigger increase, but I’ve never seen anything close to this. The second trivia question is to name a player whose rating skipped over the 1400’s, 1500’s, and 1600’s. I actually wrote a long post about this a while ago, called Skipping a Grade.
I call it “skipping a grade” when a player skips over a 100-point rating range. It’s fairly common for top juniors to do it. (I skipped over the 1700’s myself.) It’s much rarer to skip over two in a row, and I found only one person on the top-100 lists who had skipped over three in a row: Nolan Hendrickson, who skipped the 1300’s, 14o0’s, and 1500’s. So Richter is the only person I know who skipped the 1400’s, 1500’s, and 1600’s.
P.S. Updating my entry from yesterday, GM Ioan Chirila managed to beat GM Mark Paragua in the last round of the CalChess State Championship, so he tied with GM Oliver Barbosa for first. Arun Sharma and Paul Richter therefore tied for third, not second.
P.P.S. Until I get my Comments section repaired, please e-mail any comments to me at qxpch (at) yahoo (dot) com. So far nobody has done that during the four days my comments have been broken, but I just want to offer it as a possibility.
