Wednesday, September 9, 2009

Super 14 Rugby Model Analysis

I managed to get hold of past Super 14 results so decided to build a quick-and-dirty model to see what we could find.  The 14-team competition started in 2006, so I've used data from 2006 through to 2008 to create the model parameters, and tested on the 2009 season.  Well, the results so far are extremely impressive - below are a couple of graphs to show how profitable this model looks.


In the 94 games of the 2009 season the model correctly tipped 70.5 winners, which is right on 75%.  Out of the 39,000 Super 14 tippers on footytips this would have placed us somewhere between 29th and 70th overall, which is outstanding.  As with all these models though, the real fun is to see how profitable it can be.


Head to Head Betting
Firstly we'll look at the stanndard head to head betting - simply betting on the team we think will win.  As with all my betting, I've used the Kelly Criteria.  To keep things simple we'll analyse all potential bets, provided that the probability implies an advantage over the bookmaker - regardless of the size of the advantage, or the probability of the team to win.

Return on Investment
  • All games: 21%
  • Home Team Only: 9%
  • Away Team Only: 32%
After an up and down start to the season (potentially due to the ratings taking a couple of weeks to stabilise),  the model showed consistent profits week after week for the rest of the season, particularly when betting on the away team to win.  More detailed analysis will more than likely show why the away teams are more profitable, but like I said this is just a quick analysis for now.

Margin Betting
A different type of betting option typically available is betting on the margin - the bookmaker will set what they expect the margin for the games to be, and then the price either side of that margin is even money.  As an example, if New South Wales host Queensland then the bookie might set the line at -6.5 points, which means NSW need to win by at least 7 points for that bet to win, otherewise if Queensland win, or even lose by 6 points or fewer then they win the bet. I tweaked the model to predict the probability of or going over and under any given margin, and from this we can bet on either side of the margin if we find an advantage over the bookie exists.

Again, we'll analyse all bets provided there is an overlay - regardless of the size of the overlay or the probability of beating the margin.

Return on Investment
  • All games: 23%
  • Home Team Only: -3%
  • Away Team Only: 42%
A bit disappointing that betting on the home team to win the margin showed a slight negative profit, but that is tempered by the fact that betting on the away team to beat the margin shows over 40% ROI which is amazing.  Not sure why there is such a different profile between home and away margin betting, this will require some more time to analyse - my early thoughts are that the bookies line favour the home team more than they should, hence the bets on the away team provide more value

Like I mentioned, this model was a quick and dirty model - I literally grabbed the data from the net and had a model built in less than hour.  As such it is a pretty simple model, with the main point is that, all other things being equal, the home ground advantage is standard throughout the comp.  That is, the ACT Brumbies face the same home ground advantage by travelling a few hours up the highway to Sydney as they would by jumping on a plane and flying across the other side of the world to play in South Africa.  The first improvemnt to make is set up separate home ground advantages based on whether the opponent is from the same country or from overseas - and then perhaps more complex home ground advtanges could be applied, for example a separate one for each team.

Secondly, in the analysis shown here no filters were put in place - we bet on all teams provided there was an overlay.  Given more time, I'll apply filters based on the size of the overlay, and the probability of winning (or beating the margin).  There's no guarantees that these improvements will results in increased profitability, but if not I can just return to the model as it appears in this report and feel fairly comfortable about its profitability in the future,.  Stay tuned.

No comments:

Post a Comment