2011 Round 1 Results: Everything Went Better Than Expected


Six head-to-head bets for five winners and a narrow loss. Seven line bets for seven winners. Seems there's something to be said for a little offseason number-crunching. 

We were though, more than a tad lucky. Carlton covered by half a point, Geelong snatched victory very late in their game, Adelaide rallied from a long way behind, and the Roos scored in the 11th minute of time-on in the final term to prevent the Eagles from covering.

Here's the summary for both Funds:

Since Investors have a 50:50 mix of the Head-to-Head and Line Funds, their current portfolio return is half-way between the +7.4% of the Head-to-Head Fund and the +19.0% of the Line Fund. That means Investors' portfolios are currently 13.2% larger than they were on Thursday morning. Regression to the mean is going to hurt ...

Home teams fared particularly well this weekend, winning six games and drawing a seventh. On its own this would have been good news for the Head-to-Head Fund, but the fact that three of those winners were home team underdogs was even better bettor news. 

The home team success was also great news for the Home Sweet Home heuristic, which now jointly leads on the Head-to-Head Tipping section of the MAFL Tipster Dashboard. Other tipsters were generally dragged down by the relatively poor performance of the favourites this week - they won only four games and drew another - and by the equally poor performance generated by following the Consult the Ladder heuristic.

Here's the detail, with the Head-to-Head Tipping results in the leftmost column:


(EDIT: This is a corrected version of an earlier table that showed the wrong MAPE value for the Win_7 predictor. It also now includes information about each Margin Tipster's Line Betting Performance. The tezt below has been amended to reflect this.)

The middle columns provide performance data for the Margin Tipsters, starting with the Mean Absolute Prediction Error (MAPE) statistic. (I'm dropping the Median APE metric this season because I think it's too soft on mediocre margin predictors). With all but one of the Margin Tipsters returning sub-20 MAPEs you could only conclude that this week's games went very much as the Tipsters, collectively, envisaged. 

(By way of context for those of you following MAFL for the first time this year, a season-long sub-30 MAPE is a solid performance for a Margin Predictor.)

Though it's far too early to be making definitive statements, the relatively poor performance of the two Tipsters that are based on neural networks - Combo_NN_1 and Combo_NN_2 - has failed to alleviate my concerns about the tendency of such models to overfit the data they're given. Combo_NN_1 especially is showing signs of this.

For the Margin Tipsters I've also provided information about their Line Betting performance had you used their margin predictions as the basis on which to select the winning team on line betting.

On the right are the results for the probability tipsters. The value shown under the heading Score is the tipster's aggregate probability score for the Round. Any value greater than zero suggests that the tipster is predicting at better than a chance level, and the higher the aggregate probability score the better. H2H currently leads the four probability tipsters that forecast probabilities for head-to-head outcomes, and the TAB Sportsbet bookmaker is placed fourth, which is not a position I expect he'll retain for long.

Lastly, the probability scores for the probability forecasts of the Line Fund algorithm is shown. Since the Line Fund is making predictions about a different outcome (viz the line result and not the head-to-head result), its probability score is not directly comparable with those of the four head-to-head probability tipsters. The natural benchmark for the Line Fund to beat is zero since, for it too, this implies prediction at a better than chance level.