The Highest Ratings Ever, Adjusted for Inflation
Friday, May 25, 2012 at 9:17AM According to this German site, Bobby Fischer has the record: 2787, with Garry Kasparov #2 at 2759, Anatoly Karpov #3 with 2722, Mikhail Tal in fourth at 2700 and Viswanathan Anand and Vladimir Kramnik are tied for fifth-sixth at 2699. I'm not sure what their method is and am sure that Ken Regan would disagree with it, but it's at least an entertaining list.
Many, probably most of us do believe that there has been some rating inflation, but even so it's hard to believe that even the Viktor Korchnoi of the late 1970s was stronger - measurably stronger, at that! - than Magnus Carlsen. Korchnoi's peak rating was 2695, while Carlsen is 2835, and our understanding of the game has developed since then. (Even Korchnoi now must know a lot more than he did then, even if at the age of 81 he can longer play with the same strength and endurance that he used to.)
Reader Comments (14)
Regan can almost certainly not be right unfortunately. Even if you aren't convinced that the ELO system self-inflates, you simply have to admit that training methods and knowledge are much better. More very strong kids, more very strong adults ... you get a 100% guarantee that ratings have no real choice but to go up. This is simply because they are in fact stronger and they have stronger players to help sustain their ratings.
This is exactly why Sonas focuses primarily on how dominant players are over their contemporaries as the means of comparison.
[DM: He's a big boy and can take care of himself, and I have my skepticism too. But considering that in addition to his academic credentials he's an IM, it's wise to suppose that if he has made a mistake it won't be anything that's ridiculously obvious.]
Thanks for the link to the Regan paper, which I had not seen before. I am not sure that the German site's conclusions are necessarily in conflict with Regan's, for the following reason:
Regan et al are just looking at the quality of the moves on the board, ignoring the fact that today's players have access to better information (strong computers, opening theory, etc.). Regan doesn't try to estimate how good Bobby Fischer would be today, if he had grown up with access to the same knowledge that Magnus Carlsen did.
[DM: I'm not sure I understand your point, as the German site isn't trying to estimate how good Fischer would be now, either, if he had access to more information.]
I mean, in the absolute sense Einstein's physics are a lot more advanced than Isaac Newton's. But there's no way of telling which man would have been the better theoretical physicist, had they been contemporaries. Einstein had the benefit of standing on Newton's shoulders, while Newton had no such advantage. Today there are probably thousands of physicists to whom Newton's physics are trivial. That doesn't mean they could all have invented it themselves, if they'd lived in Newton's time. Learning what is already well known is a lot easier than discovering it for the first time.
Chess, unlike physics, generates a lot of easily measurable data about skill. There is pretty good evidence that at his peak Fischer was the best chess player on earth, by a wide margin. It's silly to suggest that if he had kept playing, he would not have continued to benefit from the ongoing development of theory, as he did throughout his career. (Indeed, Regan et al note that Kasparov did exactly that.)
Anyhow, as I take it, the German site is trying to make ratings comparable across eras, which is certainly a reasonable pursuit. They're hardly the first ones to do so. Jeff Sonas has made a career out of that.
Before I post some hard data, here's a current example leading to a general query on interpretation.
Through 10 games of the present match, Anand is playing at a 2985 clip, Gelfand 2915, according to my "Intrinsic Ratings" model. This is almost 200 Elo points above their respective ratings, but there's a simple explanation: much of their relatively short games has been from prep obtained using computers to depths where they'd be rated 3200+, so 2900s is just an average of 2700s and 3200.
The question is, how do you feel about this? If you only care about the quality of the moves the players have in their heads when they sit down, regardless of where they got them from, then this is fine. If you feel the computer prep is not indicative of skills as a human, then you will find my "Intrinsic Ratings" to be inflated for today's computer-savvied players. Anyway I use the uniform standard of just judging the quality of the moves made on the board, from Turn 9 onward, not caring where from---unless they are from outside sources during the game :-).
This German site doesn't pretend or claim that their results are gospel truth:
"The right column tries [sic] to account for rating inflation. A simple calculation relates the ratings to the very first list."
While I can read German I didn't find the "simple calculation" anywhere on the site - it seems that ratings are normalized to 1984 (when official and inflation-corrected ratings are identical) rather than to the first list from 1971.
Some thoughts on Dennis' second paragraph: Ratings do not and cannot reflect your 'absolute' playing strength but strength relative to your peers.
[DM: Regan's model tries to get around that.]
So even if Korchnoi kept learning things after leaving the world top, his rating went down because others improved more quickly. And ratings obviously cannot reflect 'potential' strength (if only he was younger and fitter and still completely devoted to the game - I guess he currently spends less time on chess than at the peak of his career).
[DM: Maybe we should use a different example? Korchnoi still seems to spend tons of time on the game, but chess isn't just about what you know - you need to have speed and stamina, and even Korchnoi has slowed down somewhat.]
I have read most of the relevant pieces regarding rating inflation but I still have trouble understanding the concept! IMO, what a rating does is give you a relative worth of your chess abilities as compared to your peers - it shouldn't be viewed as an absolute, to be compared across generations. To me, Kasparov's 2851 rating when (almost?) no one else was over 2800 says a lot about how dominant he was at his peak, but little else.
[DM: Recall that in Regan's model he's trying to give an objective measure with his IPRs - intrisic performance ratings.]
I think there's something to be taken from other sports like tennis, where there is a points system but importance is given to the ranking, not the points - who knows (or cares) what Sampras' points peak was compared to Federer's compared to Nadal's?
Botvinnik is placed below Ponomariov - enjoyed the cognitive dissonance.
[DM: It's a bit funny. But the thing is that while Botvinnik was still in the top 10 when he quit the game, that last bit of his career was when the FIDE ratings started. They're not trying to use historic ratings.]
I don't think Viktor would agree with this. I saw a recent interview with him where he was marvelling at Carlsen's talent and said he was envious of Carlsen's ability to play the best move "with his left hand". That being said, I've always felt that if it wasn't for the Karpov camp's shenanigans, Korchnoi would have won in Baguio and who knows what that would have led to.
[DM: As opposed to Korchnoi's complaints about yogurt and having possible criminals in his camp? On the other other hand, his family was stuck in the USSR; on the other other other hand, Karpov completely dominated the match until his collapse near the finish....]
On the other other hand, his family was stuck in the USSR; on the other other other hand, Karpov completely dominated the match until his collapse near the finish....
Dennis, a couple more hands and you'll be ready to become an economist!
That's the main question, isn't it? Do we take the ever increasing chess understanding into account or not?
ELO is relative to the competion and is statistically based so that the bigger the sample (more players) the more extreme the outliers and therefore the higher the best ELO can be.
What adds credence to Fischer having the all time best ELO rating is his path to the Championship
Interzonal 1970 18 1/2 - 4 1/2 against the all the best players except Spassky and Petrosian
Taminov Match 1971 6-0
Larsen Match 6-0
Petrosian 6 1/2 - 2 1/2
Spassy Match 12 1/2 -7 1/2 (not counting the forfeit)
that comes out to 40 1/2 - 14 1/2 against the very top players in the world he scored 74%
That works out to about 180 rating points above the best competion of his era. In today's world that would be around 2880 to 2900 ELO
The comments above are accurate, and MNb puts the question plainly---I take it into account by not attempting to correct for it. Here are my latest results relevant to inflation, in progress. They include all games under standard time controls in round-robin or small-Swiss events in which both players were rated between 2590 and 2610, expanding to 2580--2620 for 1971--1979. The second column is the IPR, the third and fourth are error bars, the fifth is the # of analyzed moves, and the sixth column is a 4-year moving average of the IPRs weighted by the # of moves. The years 1995--2005 and 2010--now are still to do; I am running 1995 and 1996 right now, while 2006--2009 come from the training sets in my previous papers. The IPRs themselves use a linear fit from 2700+-10 to 2200+-10 in the years 2006--2009 only. Thus inflation prior to those years should show up as values above 2600.
M2600in1971 2506 2379 2633 1589
M2600in1972 2778 2627 2930 0411
M2600in1973 2450 2309 2591 1505
M2600in1974 2388 2136 2640 0457 2499
M2600in1975 2667 2520 2814 0687 2534
M2600in1976 2654 2564 2745 2549 2573
M2600in1977 2618 2524 2712 2378 2621
M2600in1978 2633 2548 2719 2700 2638
M2600in1979 2578 2491 2665 3143 2619
M2600in1980 2627 2553 2701 3659 2614
M2600in1981 2740 2597 2883 0681 2621
M2600in1982 2574 2491 2657 3032 2604
M2600in1983 2694 2562 2825 1235 2627
M2600in1984 2633 2502 2763 1197 2628
M2600in1985 2575 2334 2816 0436 2611
M2600in1986 2750 2647 2852 1573 2686
M2600in1987 2589 2475 2702 2227 2644
M2600in1988 2609 2539 2678 4846 2627
M2600in1989 2699 2632 2766 4864 2655
M2600in1990 2620 2541 2699 3651 2637
M2600in1991 2556 2473 2639 3506 2626
M2600in1992 2694 2607 2781 2223 2643
M2600in1993 2606 2536 2676 5373 2611
M2600in1994 2626 2529 2722 2323 2611
M2600in2006 2563 2451 2674 1926
M2600in2007 2577 2472 2683 2008
M2600in2008 2554 2449 2660 2454
M2600in2009 2716 2594 2838 1256 2589
Based on the 2589 for 2006--2009, versus numbers mostly in the 2610--2640 range up to 1994, one can argue for 30--40 points of rating inflation from then to now. However, it is also possible that the above is measuring the difference between the 5-hour time control with adjournments prior to the 1990 Kasparov-Karpov match, and the faster controls without adjournments which started taking effect in the early 1990's. More as the analysis progresses---recall it takes 6-8 hours to analyze a game on each of my 4 available cores. Helpers welcome...
Continuing my table---again last column is the moving average of IPRs in my 2600-milepost trianing set.
M2600in1995 2641 2553 2730 2672 2633
M2600in1996 2492 2390 2594 2774 2593
This prompts me to ask, when did most tournaments move to a faster time control than the previous 40-in-2-1/2, 20/1, G/30? Note that Kirsan Iljumzhinov arrived in late 1995.
IPR's for the Rapid playoff: Anand 2701, Gelfand 2720, combined 2710. Error bars are wide enough that this much closeness is coincidence.
More table---move totals are low and will be lower in 2000--2005, so I may expand range to 2585--2615 or 2580--2620 in the training sets.
M2600in1997 2647 2507 2788 1,289 2592
M2600in1998 2588 2482 2695 2,117 2583
M2600in1999 2644 2516 2772 1,130 2571
Looks like the site you pointed out as source is now gone, any alternative?
This is my "official" all-time-top-100 list that I'm using:
http://chess-db.com/public/top100alltime.jsp