We've shown previously that it's possible to predict the TAB Sportsbet Bookmaker's head-to-head prices to a high level of accuracy using only MARS Ratings, the Interstate Status of a game and information about the very recent form of the Home team.
That model was built using the data from 1999 to 2010. If we restrict ourselves to the period 2006 to 2011, it's actually possible to construct a model with a higher R-squared that's also slightly more parsimonious.
Home Team Bookie Prob = logistic(0.3718 * Interstate Status + 0.03278 * Home Team MARS Rating - 0.03274 * Away Team MARS Rating)
This model has an R-squared of just over 85%.
We know that there's a link between the TAB head-to-head probabilities and the handicap for the same game. It's no surprise then that we can also build a respectable model that predicts the TAB Bookmaker's handicap for a game given the same variables that we used in the model above. What is perhaps surprising is that I've not attempted this before.
One model that results from such an attempt is the following:
Predicted Start for Home Team = 0.7498 * Away Team MARS - 0.7518 * Home Team MARS - 8.122 * Interstate Status
Note that to build this model I excluded from the training data those games for which the Bookmaker's start was 6.5 points or less because, for some years, the points start for such games did not reflect his true opinion of the likely victory margin. (See this blog for details.)
This model's R-squared - based on all games during the period, not just those where the start was 7.5 points or more - is 84%.
A few things about this model are interesting:
- An appropriate handicap for a game can be derived by taking three-quarters of the difference in the teams' MARS Ratings and then subtracting 8 points if the Away team is playing out of its home state (and, strictly, the Home team isn't).
- It comes up with the same favourite for a game almost 90% of the time. Amongst the 10% of games for which it chooses a different favourite, in 60% of those games the TAB Bookmaker's Home team probability lies between 45% and 55%, and in only 12% of those games is this probability below 40% or above 60%.
- Treating its predicted handicap as a prediction of the final victory margin and then wagering on the Line market accordingly would lead to a 52% success rate - comfortably better than chance, but not quite good enough to overcome the typical overround in this market
- Though it still tends to offer home teams too much start, the average handicap-adjusted margin obtained using it is about +1.5 points per game, which is 0.8 points superior to the TAB Bookmaker's +2.3 points per game
Once again it turns out that we can do quite a bit with our MARS Ratings.