2011 Round 5 Results: Can Tip, Can't Punt

It's just not happening for the Funds at the moment. When the Dons almost drew level with the Pies in the 3rd term on Anzac Day it looked, however briefly, that the Head-to-Head Fund might be about to drag itself back into profitability, but the Pies soon made that unlikely, then improbable, and ultimately impossible as they kicked clear to eventually win by 30 points. Thankfully that meant the Pies covered the spread - by just a point and a half - so all was not entirely lost for Investors as this ensured that the other, smaller Line Fund wager was successful. In total though, the Dons v Pies game represented a net loss for Investors.

Three other games yielded profits. Freo did the right thing by Investors in limiting their victory margin to just 7 points, allowing the Dogs to get up by half a point on Line betting; Gold Coast did a lot more than just protect their 63.5 points start, winning outright; and Carlton were victorious, though not by as large a margin as we'd hoped.

The weekend's big loss - not it that was all that large - came in the first game of the round when the Saints eventually proved too strong for the Lions despite trailing by 5 points at the last change.

All of which is summarised below:

So, with the 50:50 Head-to-Head and Line weightings, that leaves all Portfolios down just over 10% for the season to date.

The picture remains far rosier on tipping. Combo_NN_2 is proving to be quite the tipster, bagging 5 from 7 this week to lead by 1 tip on 27.5 from 38, which is a remarkable 72% accuracy level. Two of the Heuristic Tipsters, STM I and II, occupy 2nd and 3rd spots on 26.5, and BKB is 1 tip further back on 25.5.

Several tipsters managed 6 from 7 tips for the weekend, STM I and II amongst them, and worst performance of the Round went to the Win trio of WinPred, Win_3 and Win_7, which all scored just 2 from 7. This is particularly embarrassing for WinPred given that it was solely created to predict, well winners - hence the name.

Combo_NN_2 also continues to perform particularly well as a Margin Predictor, with an MAPE of 25 points per game and with a record of having predicted the final margin to within a goal in 29% of games so far this season. Bookie_3 has a better MAPE but has only been within a goal of the final margin in 18% of games.

Five Margin Predictors still have sub-30 MAPEs. Combo_NN_1, which has the highest MAPE of the five, also has the smallest bias of all Margin Predictors and the second-smallest standard deviation and RMSE. So, depending on your preferred measure, it's either the fifth- or the second-best Margin Predictor. It also retains the strangest profile of absolute prediction errors amongst all Margin Predictors, specialising in predicitng victory margins that are wrong by between 18 and 42 points. Granted, this is not a particularly bankable characteristic, but it is an interesting one.

To add to the list of Combo_NN_2's virtues, it's also the Margin Predictor with the best Line betting record, having correctly predicted the line winner in 61% of all contests. (Combo_NN_1 could make a somewhat tortured claim to joint-ownership of the best Line betting record. For the claim to hold you'd need to be willing to use it in a purely contrarian manner so that your line prediction would be the opposite of whatever it suggested. Sort of a "so wrong, I'm right" claim.)  

Unlike the situation for Combo_NN_2, across all Margin Predictors MAPE and Line betting performances are only very weakly correlated - the correlation coefficient is -0.17, meaning that MAPE performance explains less than 3% of the variability in Line betting performance. Success on Line betting doesn't require a Predictor's predictions be near the eventual victory margin, only that they fall on the correct side of the spread. Tipping the Cats to win by 100 if they're giving 6.5 start counts as a correct line prediction if they win by 7 points; tipping the Cats to win by 6 in this same game counts as an incorrect line prediction, despite being 92 points closer to the actual margin.

Finally, turning to the Probability Predictors, the ordering amongst the head-to-head probability predictors remains unchanged, and all continue to predict at better-than-chance levels. The probability predictions of the Line Fund algorithm, in contrast, remain sub-chance, which explains why the Line Fund is making a loss despite a 16 and 12 record. If only I'd Level-Staked the Line Fund as I have in every other year ...