- Wigan to win. (loss)
- Birmingham to win (or draw). (loss)
The odds compilers had these at ~15% chance each, whereas I had them at ~25%. However, both teams got beat badly. Very badly. I also rated the odds of a draw for Birmingham at 30% versus 8% for the odds compilers. That part looks especially bad for my model.
It will be important to focus on matches like these in the remaining weeks because these get at the heart of "how random" these outcomes really are. And of course, that issue is at the heart of modeling them mathematically. - Bolton to win. (loss)
This match was also a surprise. The odds compilers were expecting a draw, whereas my models showed that as unlikely. However, I had this as a fairly even match, whereas it was not in practice. Even the commentators I listened to were surprised by the way this matched leaned so heavily toward Chelsea.
I'm not sure what we can learn from this one. - West Ham to win. (loss)
To be fair to myself, this match was very close. West Ham was ahead for most of the match, and the final goal for Everton came in injury time. I generally prefer to try to learn from the results of matches (rather than arguing what might have been), but I do think it is fair to say that this was a much more even match than the odds compilers expected. My models showed it much closer to even (though still an advantage for Everton), so that part at least looks good. - Newcastle to win. (loss)
Like that above, this match changed outcome in the last few minutes. Newcastle was ahead until an injury time equalizer from Spurs. The odds makers had Tottenham as more likely to win, at ~45%, whereas I had Newcastle to win at ~45%. So one could try to argue that my model was more accurate, as this ended up fluctuating between a Newcastle win and a draw, but I'm not going to pursue that further.
Overall, a bad week for my models in terms of outcomes, though there may be some positives in the details. Still too early to learn anything with statistical significance.
No comments:
Post a Comment