De URS rating, alternatief voor de FIDE rating

In november deden drie Nederlandse jeugdspelers (waaronder natuurlijk ook Bram ten Dam!) mee aan de jeugd WK’s, helaas met ietwat teleurstellende resultaten. Ik zette daar begin deze maand al een stukje over op Schaaksite: https://www.schaaksite.nl/2024/12/01/de-urs-rating-een-alternatief-voor-de-fide-rating/.

Afgelopen week deed ik enige verdere research, met als conclusie: de URS ratings voorspellen de scores van spelers in toernooien beter dan de FIDE ratings, zeker ook voor jeugdspelers.

URS ratings outperform FIDE ratings

Introduction

In November 2024 three Dutch youth players participated in the World Youth Chess Championships in Brazil (U14-U18) and Italy (U08-U12). Based on their November 2024 FIDE ratings, the three players underperformed, and lost on average 105 FIDE rating points per player. As ‘… certain federations typically gain rating points while others lose them at the event …’ (https://en.chessbase.com/post/why-do-some-countries-always-gain-and-other-always-lose-rating-points) it is interesting to try to better understand ‘what could be the possible reasons behind it’.

In an interesting recent article (https://vladchess.substack.com/p/fide-rating-changes-are-they-working) Vlad Ghita discusses the impact of the recent FIDE rating changes, and also discusses ‘Geographical disparities’ and the Universal Rating System (URS, http://universalrating.com).

The URS is a relatively new rating system, supported by the Grand Chess Tour, the Saint Louis Chess Club and the Kasparov Chess Foundation. More details on the theoretical background and the performed calculations can be found on the URS website. One of the goals of the Universal Rating System is to calculate ‘ratings that are significantly more accurate (better at predicting upcoming results) than any of the three individual ratings that are produced by the Elo system’ (http://universalrating.com/about-us.php, under 6. The URS Explained).

This article will compare predicted FIDE/URS upcoming results (ie the expected total number of points per player) in some recent tournaments: the 6 World Youth Chess Championships (U08 – U18), the Olympiad in Budapest (Open section), and the European Individual Chess Championship in Petrovac.

Differences between the URS rating and the FIDE rating

The graph below shows the average difference per player (calculated as URS rating – FIDE rating) for all FIDE federations with over 500 active chess players in the November 2024 FIDE ra­ting list. This difference varies per federation, from Denmark (- 162) to Sri Lanka (+ 259). There seems to be a (linear) dependency of this difference on the percentage of players in a federation with a k-factor = 40.

The table below shows the average differences for some federations:

		Number of	Act players	Percentage	Average
FED	Federation	act players	K=40	K=40	URS – FIDE
CRO	Croatia	2.381	917	39%	-84
NED	Netherlands	3.877	1.616	42%	-125
SUI	Switzerland	1.597	671	42%	-105
…	…	…	…	…	…
IND	India	13.212	9.932	75%	228
AZE	Azerbaijan	844	640	76%	133
CHN	China	1.142	900	79%	98
KAZ	Kazakhstan	1.810	1.492	82%	167
PER	Peru	1.831	1.517	83%	142
VIE	Vietnam	744	624	84%	208

Please note the average URS rating – FIDE rating for the Netherlands is -125. The three Dutch youth players lost on average 105 FIDE rating points.

The average difference between the URS rating and FIDE rating can also be shown per k–factor:

And per FIDE rating:

URS and FIDE rating: comparing expected results for U08 – U18

By using both the November 2024 FIDE rating list and the November 2024 URS rating list, it is possible to calculate the expected results (predicted total number of points) for each individual player, and to compare these two expected results (one based on FIDE ratings, and one based on URS ratings), with the actual number of points each player has scored during the tournament. The results from players without a FIDE and/or a URS rating are removed from the dataset, as are the results from players that played in less than 50% of the total number of rounds of the tournament. Bye’s and reglementary results were also removed from the dataset (source: https://chess-results.com).

The table below shows the results for the 6 World Youth Chess Championships combined, for the same federations as in the table above:

	Number of Players	Average URS – FIDE	FIDE Points +-	FIDE Error	URS Error	Difference Error (%)
Netherlands	3	-109	-105	3,32	1,13	-66%
Switzerland	6	-64	-75	2,18	0,85	-61%
Croatia	4	-43	-73	1,86	0,69	-63%
…	…	…	…	…	…	…
Peru	9	140	41	1,58	1,01	-36%
Vietnam	12	269	57	1,52	1,26	-17%
China	17	138	58	2,16	1,06	-51%
India	16	228	59	1,91	0,95	-50%
Kazakhstan	34	160	65	1,97	0,83	-58%
Azerbaijan	5	217	73	2,61	1,23	-53%
	651	75	2	1,41	1,03	-27%

Average URS – FIDE	the average difference between the URS rating and the FIDE rating per player, calculated as URS – FIDE
FIDE Points +-	the average number of FIDE rating points won or lost per player
FIDE/URS Error	the average absolute value of the difference between the predicted number of points and the actual number of points per player
Difference Error (%)	the % difference between the URS Error and the FIDE Error, calculated as (URS Error – FIDE Error) / FIDE Error * 100%

Some conclusions:

• Players from federations with on average a lower URS rating than FIDE rating (average URS – FIDE is negative) on average lose FIDE rating points
• Players from federations with on average a higher URS rating than FIDE rating (average URS – FIDE is positive) on average win FIDE rating points
• The average ‘URS prediction error’ is smaller than the average ‘FIDE prediction error’ for all federations
• The average ‘URS prediction error’ is 27% smaller than the average ‘FIDE prediction error’

The results for the same set of players/games can also be shown per k-factor:

	Number of	Average	FIDE	FIDE	URS	Difference
K-factor	Players	URS – FIDE	Points +-	Error	Error	Error (%)
K = 40	584	81	2	1,44	1,03	-28%
K = 20	41	19	-7	1,24	1,08	-13%
K = 10	26	32	-2	0,96	0,88	-8%
	651	75	2	1,41	1,03	-27%

Some conclusions:

• Players with a k-factor = 40 on average win FIDE rating points
• Players with a k-factor = 10 or k-factor = 20 on average lose FIDE rating points
• The average ‘URS prediction error’ is smaller than the average ‘FIDE prediction error’ for all k-factors
• The average ‘FIDE/URS prediction error’ is smaller for lower k-factors
• The difference between the average ‘URS prediction error’ and the average ‘FIDE prediction error’ is larger for players with a k-factor = 40 (28%), than for players with a k-factor = 20 (13%) or a k-factor = 10 (8%).

And the results for the same set of players/games can also be shown per FIDE rating:

Rating	Number of	Average	FIDE	FIDE	URS	Difference
FIDE	Players	URS – FIDE	Points +-	Error	Error	Error (%)
– 1500	33	117	43	1,44	1,12	-22%
1501 – 1600	61	107	15	1,29	1,03	-20%
1601 – 1700	96	103	6	1,26	1,03	-18%
1701 – 1800	91	96	0	1,57	1,09	-31%
1801 – 1900	84	104	10	1,43	1,00	-30%
1901 – 2000	65	59	0	1,40	1,07	-24%
2001 – 2100	72	59	-10	1,65	1,05	-36%
2101 – 2200	54	21	-16	1,45	0,96	-34%
2201 – 2300	45	12	-22	1,48	1,01	-32%
2301 – 2400	27	29	-5	1,11	0,95	-14%
2401 –	23	27	-2	0,98	0,89	-9%
	651	75	2	1,41	1,03	-27%

Some conclusions:

• Players with a FIDE rating <1700 on average win FIDE rating points
• Players with a FIDE rating >2000 on average lose FIDE rating points
• The average ‘URS prediction error’ is smaller than the average ‘FIDE prediction error’ for all FIDE ratings
• The average ‘FIDE/URS prediction error’ is smaller for players with a higher FIDE rating 

As the results shown thus far only include youth players, it is interesting to apply the same methodology also to the Olympiad in Budapest (Open section) and the European Individual Chess Championship.

Olympiad in Budapest (Open section)

	Number of	Average	FIDE	FIDE	URS	Difference
K-factor	Players	URS – FIDE	Points +-	Error	Error	Error (%)
K = 40	79	113	22	0,97	0,74	-24%
K = 20	324	89	0	0,89	0,90	1%
K = 10	341	20	0	0,89	0,87	-2%
	744	60	2	0,90	0,87	-3%

Some conclusions:

• The difference between the URS rating and the FIDE rating is larger for higher k – factors
• Players with a k–factor = 40 on average win FIDE rating points
• The average ‘URS prediction error’ is 3% smaller than the average ‘FIDE prediction error’
• The average ‘URS prediction error’ for players with a k–factor = 40 is 24% smaller than the average ‘FIDE prediction error’

Rating	Number of	Average	FIDE	FIDE	URS	Difference
FIDE	Players	URS – FIDE	Points +-	Error	Error	Error (%)
– 1700	12	136	39	1,07	0,85	-20%
1701 – 1800	30	140	23	0,95	0,90	-6%
1801 – 1900	55	110	-1	0,90	0,87	-3%
1901 – 2000	80	104	-2	0,96	0,90	-6%
2001 – 2100	62	100	4	0,80	0,68	-15%
2101 – 2200	79	87	2	0,84	0,84	0%
2201 – 2300	65	68	6	0,92	0,99	8%
2301 – 2400	85	36	3	0,99	1,03	4%
2401 – 2500	83	22	-2	0,92	0,85	-8%
2501 – 2600	104	15	0	0,87	0,86	-1%
2601 –	89	4	0	0,83	0,78	-6%
	744	60	2	0,90	0,87	-3%

Some conclusions:

• The difference between the URS rating and the FIDE rating is larger for lower FIDE ratings
• Players with a FIDE rating <1800 on average win FIDE rating points
• The average ‘URS prediction error’ for players with a FIDE rating <2100 is 3%-20% smaller than the average ‘FIDE prediction error’

European Individual Chess Championship

	Number of	Average	FIDE	FIDE	URS	Difference
K-factor	Players	URS – FIDE	Points +-	Error	Error	Error (%)
K = 40	46	34	14	1,37	1,06	-23%
K = 20	111	35	3	1,25	1,06	-15%
K = 10	218	17	-1	1,00	0,99	-1%
	375	24	2	1,12	1,02	-9%

Some conclusions:

• Players with a k–factor = 40 on average win FIDE rating points
• The average ‘URS prediction error’ is 9% smaller than the average ‘FIDE prediction error’; for players with a k–factor = 40 this is 23%

Rating	Number of	Average	FIDE	FIDE	URS	Difference
FIDE	Players	URS – FIDE	Points +-	Error	Error	Error (%)
– 1900	13	76	15	1,37	0,95	-31%
1901 – 2000	17	46	15	1,08	0,79	-27%
2001 – 2100	27	40	3	1,27	1,13	-11%
2101 – 2200	47	39	14	1,33	1,05	-21%
2201 – 2300	47	29	-3	1,30	1,09	-16%
2301 – 2400	46	1	-3	1,20	1,21	1%
2401 – 2500	74	19	0	1,01	1,01	0%
2501 – 2600	61	21	-1	0,92	0,95	3%
2601 –	43	8	-2	0,92	0,86	-7%
	375	24	2	1,12	1,02	-9%

Some conclusions:

• Players with a FIDE rating <2200 on average win FIDE rating points
• The average ‘URS prediction error’ for players with a FIDE rating <2300 is 11%-31% smaller than the average ‘FIDE prediction error’

Summary and conclusion

• The URS rating indeed is ‘better at predicting upcoming results’
• In particular this applies to players with a k–factor = 40 (including youth players) and players with a lower FIDE rating
• Players with these characteristics on average win FIDE rating points during tournaments

For the three Dutch youth players the conclusion is clear: probably they did underperform a bit; in predicting their expected results the FIDE ratings underperformed a bit more.

Robert Leenes

robertleenes@gmail.com