The Norway Tournament began badly for Fabi fans; he lost to Magnus Carlsen. But then Carlsen lost to So and Caruana surpassed him in the end. The rating gap narrowed slightly, so a new forecast is in order:
In other news, Jakovenko won the Poikovsky Karpov tournament. I first became a Jakovenko fan when I was studying the 6.Be2 Najdorf; he won a number of instructive games with the White pieces in that line. Gelfand returned to the 2700 club.
ReplyDeleteIs this the probability of a 6-6 tie after 12 games? If the draw probability for a single game is only 18%, then I don't see how to reproduce such a low number. Your "methodology" document states that the draw probability is a function of the average ELO, ELO difference and year. At this point, the difference is small. The draw frequency for players with an average ELO of >2800 has been >70%, even over the past two decades. Since draws have also become more common over those decades, shouldn't the (single-game) draw probability exceed 70%?
Yes, it's the probability for a 6-6 tie in the match. The estimated draw rate in each game is a bit above 60%. That does seem low - as you pointed out, 70%+ seems more reasonable. A subtle issue in the methodology is that most of the games in the database come from much weaker players. Of course there are elite tournaments in the database as well, but they are drowned out in the multitude of big open Swisses. So the model is extrapolating from open tournaments and using that to fit outliers like elite matches and round robins. I would like to rerun the model again, but with 1 difference: place much more weight on games between super-GMs and less weight on weaker players. Then the draw rate for super-GMs would depend heavily on super-GM games, not big Swisses. That should lead to better estimates.
ReplyDelete