2014 - Round 3 Results: Giving a Little Back

Not a great weekend for Investors, mostly because of eight unsuccessful SuperMargin wagers, though the story could have been a little different had the Giants kicked a behind rather than a goal in the dying stages of their clash with the Dees.

That goal propelled them into a 32-point lead, nudging them out of the 20-29 point range that we'd wagered on. Still, there'll always be the narrow losses and the narrow wins; we just need to have more of the latter than the former.

The Line Fund also had a sub-stellar weekend, landing just 3 from 6 - with 1 of the 3 collects from the Hawks at the bargain-basement line price of $1.68 - which knocked 0.7c off the Fund's price. Only the Head-to-Head Fund made a profit, its 2 from 2 lifting its price by 0.6c.

All told, the Recommended Portfolio dropped 2.1c leaving it up by 4.2c on the season. All three Funds remain in profit.

TIPS AND PREDICTIONS

Bookie_9 led the way amongst the Head-to-Head Tipsters this week, correctly identifying the winner in 8 of the 9 contests, elevating it into a seven-way tie for the lead with six of the Heuristic Tipsters. These Tipsters all now have 20 and 7 season-long records (74%).

With seven favourites running out winners it was generally a good week for all the Tipsters. None fared worse than 5 from 9, and only a single Tipster, Easily Impressed I, performed even that poorly.

View fullsize

Amongst the Margin Predictors it was H2H_Adj_7 that recorded the round's low Mean Absolute Prediction Error (MAPE) of 32.30, though six other Predictors managed MAPEs within 1 point of this score.

The two Really Simple Margin Predictors now head the MAFL table, stalked by Bookie_9 which lies only 3.7 points behind the second of the pair. C-Marg has dragged itself from the bottom of the table, its 32.89 MAPE being equal or superior to the performances of both RSMP Predictors this weekend. So, there's still hope for a bookmaker-free Predictor.

There's a lot more than hope as far C-Prob is concerned. It currently tops the Leaderboard amongst the Head-to-Head Probability Predictors having relegated two of the directly bookmaker-derived Predictors into 2nd and 3rd positions. I'll write more later in the week about the results so far of wagering on the basis of C-Prob's opinions.

Lastly, the Line Fund algorithm remains clinging to a tiny net positive probability score for the season to date.

SUPERMARGIN PERFORMANCES OF MARGIN PREDICTORS

In the table that follows I've summarised the SuperMargin performances of Margin Predictors, tallying the number of times each Predictor's margin prediction has been in the correct SuperMargin "bucket" (ie represented an Absolute Bucket Error, or ABE, of 0), in either of the neighbouring buckets (an ABE of 1), and so on. I've also calculated the average Absolute Bucket Error across all games for each Predictor and the ROI that a bettor would have enjoyed had he or she level-staked each of the Margin Predictor's predictions in the SuperMargin market.

The top half of the table provides this information for all games, while the lower half provides it only for those contests in which the Predictor tipped a home team win or draw.

View fullsize

Looking firstly at the top table we find that only four Predictors have delivered a positive ROI to level-staking: Bookmaker Actual (which assumes a margin prediction equal to the negative of the handicap in the line market), Combo_7, Combo_NN1, and Combo_NN2. All have selected the correct bucket in at least four games. Combo_7 and Bookie_9, jointly, have the lowest average absolute bucket error per game at just 3.07 buckets.

Turning to the lower half we find that the same four Predictors are the only ones to have positive ROIs and that all of these ROIs are higher here than the equivalent figure in the top table. Combo_NN1 shines brightest in this section with an ROI of +115% generated through 6 correct bucket predictions from 20 attempts. No other Predictor has been anywhere near as keen to predict home team victories or draws as has Combo_NN1. It's done this in 74% of games, over 10% points more often than the next most home team supporting Predictor, RSMP_Simple (RS), which has tipped a home team win or draw 63% of the time.

Combo_NN1's contrarianism, though profitable, has come at the price of inflating its average absolute bucket error. Its average currently stands at 3.35 buckets per game, a full half a bucket per game worse that the leading Predictor on this metric, Bookie_9.