Using the Ratings Systems to Make Predictions

MAFL now tracks the relative strengths of all 18 teams using four Rating Systems: 

• MARS
• Colley
• Massey
• Offence-Defence Model

(The ODM System is explained here, and the Colley and Massey Systems are both introduced here. Details about the MARS Rating System are in the earlier Newsletters which are also downloadable from this site.) 

Very simple head-to-head predictions can be derived from these Systems' ratings simply by selecting to win the team with the higher Rating in a game. This approach ignores the proven advantage of playing at home, but imposes that constraint equally on all Systems.

Had we employed this methodology during the home-and-away season last year (and credited all Systems with 4.5 correct tips in Round 1, since at that point all but the MARS System had no way of choosing between opponents), we'd have obtained the following:

For curiosity's sake, I also created separate predictions based on the two components of the ODM System, one of which only rates teams' offensive capabilities and the other of which rates only their defensive prowess.

These two Predictors, unsurprisingly, performed worst of all, but even they, with their demonstrably one-eyed views, predicted 69% (Offence) and 68% (Defence) of games correctly from Round 2 onwards. MARS performed best of all with a 75% record for that same time period, ODM was next best on 73%, and then followed Massey on 72% and Colley on 70%.

The fact that these percentages are so high merely underscores the highly predictable nature of game results last season.

If we apply the same methodology to the games from this season so far we instead get the following:

The most obvious feature of this chart is that the ordering of the Systems is very different. Massey is best-performed, then ODM, then MARS and ODM-Offence, then ODM-Defence, then Colley. What's apparent when you look more closely at the results is that every System except Colley has recorded an even higher level of accuracy so far this season (ignoring Round 1) than it did over the course of 2012.

Massey is tipping at 82%, ODM at 80%, MARS and ODM-Offence at 78%, ODM-Defence at 69%, and Colley at 67%. It's interesting to note that predicting game results so far this year based on teams' relative offensive capabilities as assessed by the ODM is superior to predicting based on their defensive capabilities, which was also true of predicting game results last season, though the gap in performance was small. This result for the current season is consistent with what I reported in the most-recent Team Dashboard blog entry where I found that teams' ladder positions were more strongly correlated with their Own Scoring shot production (ie their offence) than with their Opponents' Scoring shot production (ie their defence).

All of which demonstrates that even a very simple application of the four Ratings Systems - and even the constituent elements of one - can produce creditable head-to-head predictions.

2013 : Round 2 - Results

There was a lot to be happy about in this week's results, not least the breaking of the long drought in the collect history of the Margin Fund. Combo_NN2 was responsible for two successful SuperMargin wagers, and Bookie_9 was the source of another and was within a kick of landing two more. These successful wagers have left the Margin Fund up by 10c on the season to date. Shame I reduced the weighting of this Fund this season ...

The other Funds weren't as successful, the Line Fund landing 2 from 3 to make a small profit - though not enough to fully recoup last round's losses - and the Head-to-Head Fund losing its sole wager. That leaves the Recommended Portfolio up by a little under 2c on the round but down by the same amount on the season.

TIPS AND PREDICTIONS

Pleasing too were the results for a number of the Tipsters and Predictors.

Combo_NN2 had a particularly spectacular round, tipping all 9 victors including the narrow-underdog Hawks and margin-predicting with a 25.6 points per game MAPE, that despite the potentially MAPE-destroying 148-point victory by the Dons. This performance has left Combo_NN2 atop both the Head-to-Head Tipster and the Margin Predictor portions of the Leaderboard.

What I'm also encouraged by is the margin predicting performance of this season's new Margin Predictors in the shape of the two RSMP variants and the Bookie variant based on the LPSO approach to deriving implied probability. (For details on these predictors follows the links in this blog post.) The two RSMP Predictors are also showing promising early talent in predicting line betting results, each having correctly chosen in 78% of games so far.

 In probability prediction, the TAB Bookmaker is proving difficult to beat once again, but what's pleasing there is that the Risk-Equalising approach to deriving Bookmaker probabilities from market prices is proving very-slightly superior to our previous, Overround-Equalising approach. (Details on the distinction are also available in the blog post I linked to earlier.)