To be frank, we're not all that satisfied with our ELO rating system, for either Australian Rules Football or for the Canadian brand in the CFL so far this season. It's doing an adequate job of forecasting the outcome of games, but no better than what we can do on our own - and in the CFL, that's saying something!
Sidelight: Here's how unpredictable the CFL has been so far this year. For comparison, we'll start with the footy season...
In the tipping contest AFL.com.au is running, the range of scores from the contestants spans from a low right around 50% correct (flipping a coin) to a high of 77% (we are currently running very close to the top, around 75%). An average score is around 60%. Got all that?
For the CFL contest, there have been eight games played so far.
NOBODY has gotten all eight right.
Only TWO people (out of 11,000) have gotten SEVEN right.
We have successfully predicted only three out of the eight games so far, worse than fifty-fifty...yet we are in the top-half of the contest (the average score is about 2 1/2 right, out of eight).
We are still doing better than the professionals, however: BoDog, the official gambling outlet for the CFL, has posted the line for each game and been WRONG 75% choosing the winner.
Impressively, the CFL.ca writer in charge of predicting games for the league, Jamie Nye, is five for eight so far - WAY above average!
The problem has been the quarterbacks: So many have been knocked out of games that the favorites become the underdogs DURING the game! In four of the games already, starters have left the game for the favorites, leaving the underdogs to rise up and win. Two others were severely altered by substitute QB play, one in either direction (we can't WAIT to see Marshall's Rakeem Cato play again for Montreal!).
So, back to our experiment:
One of the difficulties with taking as your starting point the way a team ended the previous season is that they change so much over the off-season, and your rating can't keep up because you can't know how those changes will work. So we're re-calculating the seasons in various different ways, by hand, and placing more emphasis on the first of the season games and then tapering off to the normal rate of adjustment as the teams settle in. So far, though, while we've created versions that match the more conventional ratings and power rankings, we've not been able to make one which more accurately predicts the games themselves, which was the entire point of reworking the model.
So, we're still re-working the model. Until then, you'll continue to see the version we've been running. By the way, speaking of that...
CFL - entering Week 3
Hamilton (41.5) - 1-1
Toronto (35.1) - 2-0
Calgary (32.9) - 1-1
Edmonton (30.4) - 0-1
Montreal (29.8) - 1-1
BC Lions (27.4) - 0-1
Saskatchewan (27.0) 0-2
Ottawa (24.4) - 2-0
Winnipeg (21.5) - 1-1
AFL - entering Round 15
Hawthorn (82.8) - 9-4
West Coast (78.5) - 10-3
Sydney (70.0) - 10-3
Fremantle (67.1) - 12-1
Richmond (61.0) - 8-5
Collingwood (60.0) - 8-5
Geelong (57.7) - 6-6-1
Port Adelaide (56.1) - 5-8
Adelaide (52.8) - 7-5-1
North Melbourne (52.2) - 6-7
Western Bulldogs (44.8) - 8-5
GWS Giants (41.7) - 7-6
St. Kilda (36.4) - 5-8
Gold Coast (31.6) - 2-11
Melbourne (30.8) - 4-9
Essendon (30.1) - 4-9
Carlton (26.5) - 3-10
Brisbane (18.9) - 2-11
No comments:
Post a Comment