Chess Fashion at Gibraltar 2018 : statistics and a study

The Gibraltar tournament has just ended and its almost 1200 games give us an excellent opportunity to look at chess opening fashion. Why is fashion interesting ? After all, Raetsky & Chetverik claim to have “No Passion for Chess Fashion” ! A line being often played is not a general indicator of soundness, especially in a big open tournament where 2700+ and 1800+ players meet. Personal preferences and emulation of a few champions play a bigger role than the objective assessment of the merits of the frequently played openings. What is more interesting is the relative success of various openings compared with reference points. Then I will turn to the analysis of high-profile games (2650+ for both players) in three openings for which a peculiar success rate was identified.

So let's start with the reference points, namely big averages. My success indicator is the difference between average performance and average Elo. Transpositions are taken into account, so the statistics for the first moves must not be interpreted literally. “1.d4” means all openings that are often reached by playing d4 early. I'm not quoting full statistics if there are less than 20 games in the considered line, except once (the QID).

******************
* White's first moves *
******************
I will stick to the following format all along :
Opening / sample size / %points for White / %draws / average Elo (W/B)/ success
The average Elo is the one of the player who made the last move in the definition of the line, but I'll indicate which player it is (White or Black) anyway to make things clear.
1.e4, 505 games, 54% , 33% draws, Elo 2346 W, +27.
1.d4, 446 games, 57%, 31% draws, Elo 2378 W, +32.
1.Nf3, 125 games, 57%, 32% draws, Elo 2378 W, +29.
1.c4, 66 games, 61%, 39% draws, Elo 2429 W, +47.
Everything else is below 10 games. The English is certainly a bit more drawish and a bit more popular among stronger players, and the latter are slightly more successful with it than with other openings, but all in all the averages are similar enough. The first move advantage makes White's performance superior by 30 points to his average Elo, which is expected.

***************
* Flank openings *
***************

I can't say much about Black's choices against 1.c4 or 1.Nf3, as they are very diverse and each line falls below the threshold of 20 games, with one exception.
1.Nf3 d5 2.g3, 45 games, 66%, 29% draws, Elo 2440 W, +63.
If 14 more games from 1.Nf3 Nf6 2.g3 are added to these 45, the other figures don't change. It is quite surprising to see this line score very well (66% over 45+14 games), with fewer draws than the average. It's not just a surprise weapon to defeat weaker players, as is shown by the performance above 2500.

*******
* 1.e4 *
*******

There are four popular defenses against 1.e4. Let's review them all.
1...c5, 196 games, 52%, 26% draws, Elo 2322 B, -32.
1...e5, 136 games, 53%, 43% draws, Elo 2404 B, -21.
1...e6, 76 games, 59%, 29% draws, Elo 2309 B, -26.
1...c6, 49 games, 59%, 38% draws, Elo 2382 B, -62.
Ironically, the French is now as drawish as the Sicilian... The Open Game is indeed the stronger player's choice and it is unusually drawish (10% more than the average). The Caro-Kann, even though it is also played by relatively stronger players, is suffering right now. Maybe the latest novelties tend to favor White.

Variations of the Sicilian perform very differently.
Najdorf, 39 games, 59%, 26% draws, Elo 2375 B, -78.
Open e6, 34 games, 49%, 27% draws, Elo 2352 B, +25.
3.Bb5(+), 29 games, 52%, 42% draws, Elo 2424 W, -7.
The Najdorf suffers in two lines (6.Be2, 4/5 for White, and 6.h3, 3/4 for White) but earns 50% against the main line 6.Bg5 and other lines. The Taimanov (2.Nf3 e6 3.d4 cxd4 4.Nxd4 Nc6 5.Nc3 Qc7) is especially successful with 11/18 for Black. That explains why Black has a performance superior to average Elo (“Black is OK !”) in the Open Sicilian with 2...e6. Finally, the Bb5(+) systems are very drawish nowadays.

In the Open Games, we also see a contrasted picture.
Spanish (all), 61 games, 57%, 47% draws, Elo 2470 W, +37.
Spanish (closed), 22 games, 57%, 51% draws, Elo 2478 B, -2.
Italian+2Knights, 34 games, 51%, 45% draws, Elo 2355 W, +53.
Open Games are drawish, but the closed Ruy Lopez sets a record with 51% draws. Breyer (9...Nb8) and Zaitsev (9...Bb7) systems are durable battlegrounds. The move 3.Bc4 is, somewhat surprisingly, the statistical path to a draw against a stronger opponent ; that's why it performs well (+53) with a low score (51%).

*******
* 1.d4 *
*******

Against 1.d4, I'll stick with just one tier of categories. Please notice that the “Slav” category excludes the semi-Slav, which could not be isolated with a sufficient number of games.

KID with e4, 26 games, 67%, 28% draws, Elo 2221 B, -54.
Nimzo, 46 games, 53%, 38% draws, Elo 2471 B, -69.
QID, 17 games, 65%, 36% draws, Elo 2439 B, -96.
Slav, 27 games, 72%, 27% draws, Elo 2257 B, -57.
Without c4 , 27 games 57%, 27% draws, Elo 2330 W, +44
QGD (no Nf3), 23 games, 65%, 36% draws, Elo 2413 W, +110.
QGD (with Nf3), 45 games, 49%, 41% draws, Elo 2439 W, =0
Catalan, 40 games, 64%, 33% draws, Elo 2460 W, +39.

The KID and Slav defense don't score well for Black, because relatively weaker players play them against relatively stronger players. The Queen's Indian Defense seems under pressure, that's why it's my only exception to the 20-games threshold. The success record is set by the QGD with no Nf3 with 110 performance points above average Elo for White. Black is well inspired to choose a move order that makes White play Nf3 (either the Indian move order or 3...Be7 instead of 3...Nf6). I thought this was well-known already for quite some time.

******************************
* Study of some high-profile games *
******************************

I've identified 13 high-profile games during the tournament. These are games between players both rated above 2650. Had I set the threshold to 2600, I would have had to consider 35 games and that was too much. I've looked into the opening of these 13 games to see which ones would overlap with an usual statistical result. Four games were selected for a complete analysis.

As the Bb5 Sicilian seem to score particularly poorly for White (performance not superior to average Elo despite being White), I've chosen to study a game from this opening. It's the 10th round game between Liem Le Quang and Daniil Dubov. The Caro-Kann played in the same round between Maxime Vachier-Lagrave and Richard Rapport was also selected because the Caro-Kann was less successful in Gibraltar than its direct alternative. I also wanted to explore why Nf3/g3 was so successful but numerous transpositions made me keep only two games : David Howell vs Nikita Vitiugov from round 9 and the QID transposition between Hikaru Nakamura and Richard Rapport in round 9 too (the QID has a special record too in Gibraltar). No QGD declined without Nf3 was found in Gibraltar between players above 2650 Elo, and for good reasons.

Here is the link to the study :


Enjoy ! Liem Le Quang vs Dubov has a fascinating position to analyze and demonstrates a wonderful breakthrough in a locked position. Amateurs of the King's Indian Attack will be delighted by seeing Howell-Vitiugov revive a line from 30 years ago, with recent echoes. The Nakamura vs Rapport game might be in the next book on the Queen's Indian. As for Vachier-Lagrave vs Rapport, several strange decisions and transpositions make it real fun !
On a side note, while I was analyzing those games with Komodo, I saw a self-reportedly Norwegian account, named “Pacificrabbit”, who had already gone through them at a considerable depth. I guess those games didn't wait for me to attract some relevant attention !

As always, comments are welcome and a little click on the heart below the study board will be appreciated.

Interesting study, good work.
Two comments.
An open tournament is not fit for opening analysis, as top players often play unusual openings to avoid drawing against weaker players. I suggest the Tata Steel tournament may be more suitable for analysis.
Why do not you use the ECO codes A00-E99? Your present descriptions are ambiguous.

I think that an open tournament is more representative of fashion than a supertournament, because the former is not limited to top-20 players who know each other very well and select openings "ad hominem". Players in the ranges 2300-2500 and 2500-2650 also contribute to opening fashion, some even develop strong novelties that you will see in books published next year.

Furthermore, a supertournament like Wijk-aan-Zee doesn't have enough games to compute meaningful frequencies. This could be circumvented by stacking all supertournaments of the last five years. So, I take the point that it would be worthwhile to identify and survey openings played at the very top. This would serve a different purpose, for example it would be an interesting reading key of the opening choices in the coming Candidates tournament. It's an idea for my next study , thank you :) .

About ECO codes, I feel that transpositions make them less and less suitable to classify contemporary openings. The ECO classification was based on tournament practice during the 60's and the early 70's. The fact that it lasts for more than 40 years is already a miracle (and a testimony of the serious work put at that time into developing a durable classification), but I feel that it is really obsolete right now. For example, the closed Grunfeld often occured in Gibraltar after 1.Nf3 Nf6 2.g3 g6 3.Bg2 Bg7 4.0-0 d5 5.d4 0-0 6.c4 c6 (with 7.Qb3 Qb6 8.Nc3 Rd8 to follow), but Nakamura-Howell (round 8) started with 1.d4 Nf6 2.c4 g6 3.Nf3 Bg7 4.g3 c6 5.Nc3 d5 6.Qb3 0-0 7.Bg2 Qb6 8.0-0 Rd8, transposing into the same line. Depending on the source and on the game, it is classified as A07, D78 or E60.
That's why I use an admittedly ambiguous "verbal" description. For example, the "Open Sicilian with e6" covers the moves 1.e4 c5 2.Nf3 e6 3.d4 cxd4 4.Nxd4 Nc6 and 4...a6, and several move orders that transpose to the same position or subsequent positions a couple of moves later (e.g. 1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 e6 5.Nc3 Qc7 6.Be3 a6). Likewise, there are multiple move orders that fall under the general heading "KID with e4" (starting with 1.Nf3, 1.c4 or 1.d4 and converging on move 4, 5 or 6).

I agree on the fact that separating all the move orders would be ideal in theory, but it would require a much bigger sample to extract meaningful frequencies and performances. I chose to sacrifice a little objectivity to increase readability and tractability, hopefully not to the point where readability itself is endangered by the remaining ambiguity. That's a hard choice, like the depth of the transposition that I was willing to take into account.

Thanks for the interesting ideas and for pointing out at the needed clarification.

Videos of the players's interviews were added to the study.


Some of their comments did answer two of my questions : Why MVL played 8.Qe5 instead of 8.Ne5 against Rapport and why Nakamura played to force f4-f5 against Rapport. There was also the unexpected possibility 15.a4 in the MVL-Rapport game.
One question remains : why Nakamura innovated with 10.Nd2 (provoking exchanges and easing Black's problems) when 10.Nb5 worked so well previously.

"Players in the ranges 2300-2500 and 2500-2650 also contribute to opening fashion" I do not agree. Players in the lower ranges have neither the time (they have a real job) nor the capacity to create fashion. They follow fashion. The top players create fashion. They have the time, resources including seconds, and the capability to do that and they need to do that as people prepare against them. Think how Kasparov resurrected the Scotch, Kramnik the Berlin Defence in the Ruy Lopez, Carlsen the London, Fischer the Ruy Lopez Exchange, Karpov the Petrov Defence and the Caro-Kann, Korchnoi the Open Ruy Lopez and the French Winawer. Fashion goes top down, not bottom up. For the same reason you will not be able to predict the opening fashion at the Candidates' Tournament: it is created in the studies of the candidates, not in the preceding tournaments.

The idea behind the ECO codes is that it is a partition: no voids, no overlaps. Transitions are handled in the ECO. Usually the higher ECO code prevails. I find it more useful than the NIC code. You are right, that some games are mislabeled.

You could start with a course analysis A-B-C-D-E. Then you could refine and e.g. split B into B0, B1,...B9.
Then you could refine more and e.g. split B9 into B90, B91... B99. If a class contains many games, then you split, else you do not split.

This would be more systematic and less ambiguous and independant of language. It is also easier to do, as the games are already labeled in ECO code in their pgn file. Only a check for mislabelling is needed.

I've added the Wang Hao - Nakamura game from round 8 in the study. Both players have 2700+ Elo and the opening is the very fashionable Berlin defense, so both criteria were met to include this game. Furthermore I believe that this game has attracted some attention (see my comments in the study).
There is no post-game interview for this game, but Nakamura mentioned it in his interview the next day.

More is coming soon.

Three more games were added.
Enjoy the Najdorf with Be3 between Nakamura and Grandelius, both very dynamic players : it's a wild game with pieces hanging all around. For once we have a post-game interview for each player.
Aronian's crucial victory against Sethuraman is another success story for a Nf3/g3 line (by transposition). A fascinating position could be obtained by Black at move 24 and it seems that Aronian's attack could be survived after all ! Aronian's interview is a blend of revelations and mysteries, as is certainly appropriate with the coming Candidates' tournament !
Wang Hao vs Howell is another occasion to explore the side variations of the Berlin defense. A probable time scramble has led to a phase of "gift exchanges". White's advantage grew to winning proportions and then dissipated. Complications in the Berlin defense make it less drawish than its reputation, or at least it leads to a fighting draw !

There are still some Gibraltar games (typical examples of fashionable openings) that I will add to the study.

Thanks for the study. If I were to give a highlight, despite being under 20 turns, I found the Wang-Nakamura game analysis instructive. The round 4 breakthrough was interesting from the perspective of attempting to draw off stronger players as well.

I also like the game Wang Hao vs Nakamura, it was a crucial game that could have changed the tournament (with an extra victory, Nakamura would have been clear first, so other players would have pressed harder and maybe lost during the last two rounds ; not sure whether pairings would have been affected drastically).

I've added Gupta-Ivanchuk and Oparin-Aronian from round 8. In both cases, we see a relatively younger player with White facing an older superGM. Ivanchuk lures Gupta into a topical line of the Queen's Indian (Kasparov's first win in a World Championship game) but Gupta is well prepared and plays h4/Bg5 that works well in other lines. Oparin knows Aronian's pet lines in the Berlin Defense and leads him into a forced draw that could only be avoided by going into a risky line, two pawns down for what Aronian assesses as insufficient compensation.

A reminder of the link to the study : lichess.org/study/HlsEqsF8

I've looked into more Sicilians with Bb5 (e.g. Oparin-Gelfand), but those games just restate what I've said earlier in the statistical part : this line has become very drawish nowadays. So I will add probably only one last game to the study (Aronian-Short) when I'm done analyzing it.

Nice effort!
Have my like good sir