Rating changes

[Jonathan Rogers]

I agree with Stewart (in the default thread) - we should really have a separate thread to discuss the proposed changes to calculating FIDE ratings: a k factor of 30 for everyone under 2400, a k factor of 20 for everyone above, effective from next July, as I understand it.

I'm no expert on ratings but I find the proposed changes shocking; as with many reforms in the chess world, they seem quite radical but as far as I am aware with relatively little empirical support.

1 No doubt the main justification is indeed that juniors/other improving players should be able to gain points quickly, and that declining players should similarly see a proportionate decline in their ratings. But what evidence is there to suppose that juniors are always so hideously under-rated? I do not see much; where it is true they seem to gain their points quite quickly as it is. From my own 4NCL club, Samuel Franklin gained just over 200 points in a year, just for example. Similarly I am not sure that there are many old timers who are hanging on to ratings which are really reflecting their glory years. Again, i don't suppose for a moment anyone has offered any evidence on this. The late Bob Wade's rating went down to something which was certainly reflective of his later years rather than his prime. If you see an old timer with a high rating, the chances are that he or she simply does not play much, but the accuracy of the ratings of inactive players is something which should be tackled differently altogether (perhaps everyone should lose 10 points automatically in a year when they play less than 6 FIDE rated games).

2 But apparently old players over 2400 are allowed to slow down their rating decline, whilst the Magnus Carlsens etc who are still rapidly improving even when over 2400 then have their rises checked to some extent by a lower k factor. What is the reason for this different k factor over 2400 ? Isn't this just political - the 2600s won't enter open tournaments anymore if they stand to lose too many points to 2200s (but it's quite Ok for 2300s to risk losing lots of points to 1900s, even though it is more likely to happen because 1900s are more likely to be seriously under-rated)? Whatever the reason for the 2400 difference might be, it tells us that the need to accurately measure rapid improvement and decline is not purely what the new system is all about.

3 I think it is fundamentally unfair to have two opponents in one game playing off different k factors (assuming that both players are reasonably established and not newcomers). Of course it has always been like this, but why? It means that one opponent - the one likely to lose in any event (as in the case of 2350 v 2550) - has more to lose than the other in the event of an equally disappointing result (a disappointing result being a loss for the 2350 or a draw for the 2550). If it is justifiable to have different k factors at all, I think that when you have a game between people on different k factors, then the only fair way is for both players to have the same k factor for that particular game - namely the k factor of the higher rated opponent.

[Maxim Devereaux]

Actually, if you read the Chessbase article by Macieja (report for the ACP on the FIDE congress), he gives a reasonable explanation of this change: it goes hand-in-hand with the proposal to publish FIDE rating lists on a 2-month rather than 3-month cycle from the next year also, and is also attempting to remedy a past failure to account for this when we moved from 2 to 4 rating lists a year.

There isn't really any correct figure for the K-factor, though I do recall seeing a statistical analysis that suggested the ones used by FIDE were perhaps a little on the low side. Whether the K-factors are 'correct' at 20 and 30, with 6 rating lists a year is a moot point - what they are apparently trying to achieve is a close approximation to the system that was in place with K-factors of 10 and 15 and 2 rating lists a year.

You'll always get the situation where players decline more slowly than they should, because these players are generally less active than the average in any case - it almost doesn't matter how high you set the K-factor if they play (almost) no rated games. Equally, any player who plays significantly more games than the 'normal' number in a rating period can in fact cause their actual rating to be higher than the level at which they performed. Mark F, Lorin and I had an interesting discussion of this in Greece, and came to the conclusion that Stefan Macak had performed at around 2470 yet managed to raise his rating from ~2350 to ~2550 in one period by playing 80-something games.

[Richard Bates]

Actually, if you read the Chessbase article by Macieja (report for the ACP on the FIDE congress), he gives a reasonable explanation of this change: it goes hand-in-hand with the proposal to publish FIDE rating lists on a 2-month rather than 3-month cycle from the next year also, and is also attempting to remedy a past failure to account for this when we moved from 2 to 4 rating lists a year.

There isn't really any correct figure for the K-factor, though I do recall seeing a statistical analysis that suggested the ones used by FIDE were perhaps a little on the low side. Whether the K-factors are 'correct' at 20 and 30, with 6 rating lists a year is a moot point - what they are apparently trying to achieve is a close approximation to the system that was in place with K-factors of 10 and 15 and 2 rating lists a year.

You'll always get the situation where players decline more slowly than they should, because these players are generally less active than the average in any case - it almost doesn't matter how high you set the K-factor if they play (almost) no rated games. Equally, any player who plays significantly more games than the 'normal' number in a rating period can in fact cause their actual rating to be higher than the level at which they performed. Mark F, Lorin and I had an interesting discussion of this in Greece, and came to the conclusion that Stefan Macak had performed at around 2470 yet managed to raise his rating from ~2350 to ~2550 in one period by playing 80-something games.

Are there any specific obligations upon organisers to submit results as rapidly as possible? I remember in the old days people who struggled to pass the 2400/2500 ratings barrier were able to go back years in the past to find some "lost" five pts and just add them on, even though any error would have long corrected itself. Obviously the 2/3 month lists are undermined somewhat if organisers can sit on results for months before submission. I notice that one of Macak's results (a gain of 40+ pts) on the Apr-Jul 08 list was submitted from the previous October.

Didn't FIDE introduce a rule a few years ago limiting the amount of points any individual could gain on one list?

They will have to close the above loophole or someone like Macak will pop up in the World's top ten before very long.

[Sean Hewitt]

Are there any specific obligations upon organisers to submit results as rapidly as possible?

You register a tournament for a specific period and I was told that you must submit the results by the deadline for that period (14th of the month before the next list publication date) or the event wouldn't be rated.

Whether FIDE enforce this I have no idea - I always submit my rating report the day after the tournament ends.