Are 2600+ players underrated?

[JustinHorton]

A data scientist, Albert Hamood, has produced a report, Rating deflation: today's chess elite are underrated which appears to suggest that for players above 2600 (which is the dataset examined, it makes no comment about lower-rated players) there has been a degree of rating deflation and a given rating now represents a higher playing strength than hitherto.

I don't have the maths or the knowledge of rating systems to make a proper judgment on this report, but I'm guessing other people here may do so.

[Thomas Rendle]

I haven't read the report - but the answer is yes. At least if you're comparing playing strength as what would happen if a 2600 today was transported back 10 years ago to face a 2600 in 2016 (this seems the fairest way to do it). I would estimate the difference at maybe 60-70 points, so 2600 player today would maybe expect to score close to 60% vs a 2600 10 years ago.
I suspect in tactics and endgame play the difference is less clear / non-existent.

It's hardly surprising - information is so easy to come by, and computers are incredibly strong. Everyone's openings are way better now. I'm 50 points below my peak and my opening knowledge is above where it has ever been (and early middlegame knowledge, possibly). For juniors who can learn an opening course in a couple of hours, this is even more pronounced.
Even at 40 or so, if I want to learn a whole new opening it's probably about 10 hours work to get pretty confident. 20 years ago it would've taken much longer, for objectively worse lines / openings.

Even at a club or lower international level, most 2000/2100 players have been forced to do some computer work to make their openings more sound - if you don't you lose 100 points or so!

[JustinHorton]

It's hardly surprising - information is so easy to come by, and computers are incredibly strong. Everyone's openings are way better now.

Although the report (which as I say, is limited to 2600+ players) doesn't consider this to be the most important element.

[Thomas Rendle]

I have skimmed the report. It has it at about 100 points over 15 years as an estimate? Slightly more than I thought, but not ridiculous by any means.

I suspect it's quite related to opening knowledge still though. Everyone knows their openings better, and all the plans and ideas are more clearly written about. Every chess course comes with a set of annotated instructive games to play through, etc. - of course you could do all this 15 years ago, but it would've required a lot more leg work!

[Roger de Coverly]

I don't have the maths or the knowledge of rating systems to make a proper judgment on this report, but I'm guessing other people here may do so.

I don't think the paper has much to say about the mathematics of rating systems. What he has identified is that over time, players ability to match engine suggestions has increased.

Rating systems measure results, so that if the really top players get better, those who expect to score one from four against them have to get better as well to maintain their relative position.

At lower levels there have been a couple of recent shocks to the system. One was the suspension of OTB play during Covid, This meant that players improving through study etc outside OTB chess had ratings well below their strength when play resumed. The other was FIDE's compression of the 1000 to 2000 band into a 1400 to 2000 band. These would offset one another as presumably was intended. It's probable these effects eventually would feed through to GM level.