World Cup Recap: Day 15 (Conclusion of quarterfinals)

We finally get a rest day, so let’s start with a deep breath. After 15 days of frantic chess action, a starting field of 128 players from around the world has finally been whittled down to the Final Four. In tie break action the last two players eliminated before the break were Wei Yi and Shakhriyar Mamdyarov. The former, a superprodigy at just 16 years old, had been owning the headlines with his remarkable deep run, but finally it came to an end at the hands of Peter Svidler. The latter, from the host country of Azerbaijan, was also a pleasant story to follow as he carried the flag for local fans, but his run too came to an end with his loss to Sergay Karjakin.

By finalizing who their opponents would be, these two matches also of course had small impacts on Anish Giri and Pavel Eljanov (the players that had already punched their own tickets to the semifinal round the day before). Here are the new odds of reaching the finals for all six players, and how those odds changed based on the tie break results:

Seed Player Rating New Odds of Reaching Finals Change
11  Sergey Karjakin (RUS) 2764.6 55.1% 24.2%
16  Peter Svidler (RUS) 2733.6 29.7% 15.1%
4  Anish Giri (NED) 2795.7 70.3% 0.4%
26  Pavel Eljanov (UKR) 2751.4 44.9% -3.2%
24  Wei Yi (CHN) 2737.0 0.0% -15.5%
19  Shakhriyar Mamedyarov (AZE) 2745.0 0.0% -21.0%

With just four players left, there are only two matches for us to discuss. These are arguably the two most important matches of the entire tournament, however, as the winners of these semifinals will each earn a place in next year’s Candidates Tournament, arguably a bigger prize than can be won in the finals where “only” the title of World Cup Champion is at stake. So let’s take a close look at those two matches:

Anish Giri vs. Peter Svidler

Giri is the top remaining seed, and highest rated player left in the field. Our model of course sees him, therefore, as a strong favorite in this match (70% to win), and the most likely eventual tournament winner (44%). He is also the player with the least to lose, as that strong rating of his will almost certainly earn him a spot in the Candidates Tournament even if he doesn’t advance here. How did his path to this point look?

Giri has perhaps been a little bit underwhelming in the classical games, settling for more draws than his rating would indicate and dropping 2.4 rating points in the process. However this solid approach has worked well, as he has not lost a classical game either. Three times he has won one of two classical games, drawn the other, and advanced without tie breaks. The other two times he drew both classical games, then proceeded to win the first set of tie breaks 2-0. Despite having been made fun of by none less than Magnus Carlsen for his drawish tendencies, they have provided him a remarkably comfortable path to the semifinals with very little danger of elimination at any point so far in the process.

Svidler, on the other hand, comes into this round as an underdog (30% to win the match, and just 12% to win the whole tournament). He got here after a tense back and forth tie break with Wei Yi. Svidler has in fact been more solid even than Giri, with just two wins and eight draws (but no losses) in the classical games so far. Of course one of the wins was against Veselin Topalov, and Svidler entered the event rated much lower, so this performance has been enough to impress the Elo system and add 6.6 points to Svidler’s rating.

Svidler also has not lost a rapid game, but hasn’t been as convincing as Giri in tie break matches. He won his first two tie break matches in the first pair of games with one win and one draw, but got severely tested by Wei Yi drawing both G/25 games and the first G/10 game, before he finally broke through to win and advance in the fourth rapid game.

We have two players who have faced relatively little adversity, and have drawn many of their classical games. Svidler has the experience edge, having won the 2011 World Cup when his opponent was just 17 years old. However Giri is now all grown up and the strong favorite by rating. We estimate that Giri is a 70% favorite to win, and will begin to find out the reality of the match tomorrow.

Sergey Karjakin vs. Pavel Eljanov

This match is much closer by rating. Karjakin has the edge, but our model (which does not look any further than current live ratings in its prognostications) says he is a 55% favorite (with a 25% chance to win the event). Not too far from a coin flip. Karjakin was seeded #11, and is the second highest seed to reach this stage. He survived adversity in round two, losing his first game to Onischuk, but bounced back to force tie breaks, which he won. Overall Karjakin has four wins, but also the one loss, in his classical games, for a net rating gain of 2.6 points, showing less drawish of a nature than our two players on the other side of the bracket, however it has still left him in tie breaks three times so far.

Against Onischuk, he drew both G/25 rapid contests, before winning 2-0 in the G/10 portion. Against Andreikin he won the first G/25 game and held with black to end the tie break match there. Against Mamedyarov he again drew both G/25 games, but then both G/10 games. So his classical loss to Onischuk is his only defeat so far in the tournament, but he has been tested a fair bit.

Eljanov, meanwhile, is the underdog story. Seeded 26th, and entering the tournament rated just 2717, he opened up on an absolute tear winning his first six classical games! While he settled down to draw twice with Jakovenko, he was back to his winning ways against Nakamura taking home the classical match with a win in the first game, and a comfortable draw in the second. This insane +7 score in classical games has meant Eljanov’s rating has risen an astounding 34.4 points, bringing him up to #14 in the live ratings, a gain of 18 spots on the rating chart! It has also meant he has only played one tie break match, giving us less of an idea of what to expect from him in that phase. Against Jakovenko he drew the first three rapid games, before winning the fourth.

We’ve discussed before that we are highly skeptical of the idea of “form”, believing instead that most results that are better or worse than expectations are simply the expression of random variance. However if “good form” is in fact a thing, Eljanov has it right now. The model does give him some credit, in that it uses his massively improved live rating to estimate his chances, rather than judging him on his much lower rating from before the tournament began, but his 45% chance to reach the finals (and 18% chance to win the whole thing) are still based on his current live rating of 2751. He certainly looks to be playing at beyond that level so far, and if that continues then Karjakin may be in far more trouble than the model gives him credit for.

It is worth mentioning that Eljanov isn’t exactly coming out of nowhere. Rather, he’s coming back. Five years ago he was rated 2761 and ranked sixth in the world, so he’s been here before, just not recently.

Tomorrow games will begin again after this brief respite, and we will see if anyone jumps out to a 1-0 lead with a decisive victory in the first classical game. Until then, let’s consider one other question: what merit do our predictions actually have? Many readers may be wondering how accurate our predictions actually are. Well, we’ve provided odds for 124 matches so far in this event that we now know the results of. Let’s divide those up into groups, by the odds we gave the favorite, and see how we’ve done:

Odds Range Matches Favorite Win%
90-100% 21 95%
80-89% 22 91%
70-79% 31 71%
60-69% 23 65%
50-59% 27 59%

Overall our model seems to have done a pretty good job, considering nothing but live ratings. Players in the 80-89% range have won a little more often than expected, and players in the 70-79% range a little less, but one or two matches going differently would true up those numbers, and these sample sizes are small.

Of course it’s always possible to cherry pick cutoff points to imply differently. Statistics don’t lie, but they can be manipulated and selectively presented in ways that will cause people to draw bad conclusions. If you wanted to make the case that our model has underrated strong favorites, you could set a cutoff point at exactly 80.4%. Players given an 80.4% chance, or better, of winning did in fact win 39 matches in this event, with just one loss. That’s a 97.5% edge, when we’d expect a mere 90% edge or so. There should have been four upsets in that range and there was only one!

However five players rated as between a 79.4% and 80.4% favorite, just below the cutoff point, and three of those five lost. Cherry picking data can be fun, but doesn’t help if you want the truth. Overall, if our odds on the 124 matches so far have been accurate, we should have expected an average of 92 favorites to win with 32 underdogs advancing. In actuality, players we’ve deemed to be favorites have won 93 matches and lost 31. Seems like our predictions have been pretty valid to us.

Advertisements

6 thoughts on “World Cup Recap: Day 15 (Conclusion of quarterfinals)

  1. I assume that when you say that your model gives a pretty accurate success rate, you are referring to the tournament as a whole. I reckon that this is generally due to the earlier rounds (Rds. 1 thru 3) where the favourites have much higher ELO than their opponents.

    However, the latter rounds where the ELO gaps are much smaller, playing favs based on ELO did not and neither will hold. For instance, in the Rd-of-16 (Topa-Svidler, Giri-Wojta, So-MVL, Ding-Wei, Naka-Adams, Fabi-Mamed, Karja-Andrei, Jako-Elja) only 3 of 8 favs by ELO won, i.e., as 5 of them lost, the success rate was a dismal 3/8 = 37.5%. If we lumped Rds 4 and 5 (Q’finals where 3/4 favs won), that figure rises to 6/12 = 50% but still below the expected rate for the favs.

    As a side note, Giri (a hungry lion) and Karkain (both wounded and hungry) are still fighting but the other hungry or wounded lions (Fabi, Wesley, Ding and Wei) have been killed (excuse my terminologies). Giri has neither won a World Cup nor a Grand Prix, whilst Karjakin was wounded in the last Grand Prix and insulted by exclusion from the Grand Chess Tour. Note: Eljanov and Svidler do not belong to this “lion” group (elite Top super-GMs) but we cannot discount them.

    Like

    • Well, I’m saying that if I name someone an 80% favorite then they’ll probably win about 80% of the time, and if I call them a 60% favorite, they’ll probably win 60% of the time.

      In later rounds there have been more 60/40 edges and fewer 80/20 edges, as the weaker players mostly got knocked out early and rating gaps shrunk. So certainly the percentage of matches in which the ratings favorite will win drop in later rounds, as the favorites have smaller edges. I don’t think the accuracy dropped, though. I think a 60% favorite in the semifinals would win with about the same frequency as a 60% favorite in a first round match (which is to say, six times out of ten).

      Like

  2. Btw, reaching the semifinals already involves a prize (besides the money and fame) – qualification for the Grand prix series of the next cycle (assuming that there’s stability and the next cycle will include such a series with qualifying criteria similar to the previous ones). Of the four players, that’s particularly important for Eljanov & Svidler, who might not make the cut based on ratings.

    Like

    • That’s a good point. I hadn’t started mentioning the next cycle yet because nothing has been finalized on its qualification criteria, but it is sensible to assume some components (like this one) will probably be the same. So in the super way too early race for the 2018 Candidates Tournament, Svidler and Eljanov did indeed see a bump in their odds from reaching this point.

      Thank you for the input!

      Like

  3. I have doubts why Anish did not play like Giri in game 1 of Semi-finals. Suddenly, he deviated from solid play, refused Peter’s draw offer (?), and later pushed the pawn to G5. The loss reminds me of discussions here in Chessnumbers where some say that Naka or Jako would gain more if they lose a game.

    As you’ve perfectly described it, Giri is playing proxy → his placement in World Cup will determine who amongst the Russians can bag the Candidates’ slot. Losing in SF of World Cup gives the slot to Peter; winning gives it to either Grischuk or Kramnik. Whichever outcome is closer to his chest may win the slot!

    But considering the big $$$ differential between semi-finalist vs. finalist vs. Cup champion, Anish will NOT entertain such idea.

    Like

    • It’s Svidler who refused Giri’s draw offer – following 30.Bxa4? – not the other way around (“For me, I felt that I’ve taken a number of strange draws in this tournament, but they were mainly in second games; in a first game, when suddenly I have this opportunity to play a sharp position for a win with Black – I felt that this is obviously the best position I’ve had in this game yet, and if I don’t see mate [for White] I should continue” – Svidler in the postgame interview).

      Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s