Investors might recall that, in the opening paragraph of the blog describing the wagers we had for Round 10 I noted how closely our fortunes were tied to those of the Home Sweet Home heuristic. Well, Home Sweet Home registered just one correct tip for the round, which consigned us to this:
But for the Cats covering the spread by half a point and the Pies doing what needed to be done to the Eagles, the loss for this round would have exceeded even that for the globus horribilus that was Round 2. The Head-to-Head Fund's performance was the more damaging to Portfolios, its near 30c fall attributable mostly to Richmond's upset loss to Port Adelaide, and to the Dees' and Dogs' excessively large losses in their respective matches.
On the upside, the Line Fund managed to eke out a profit from a 3 and 4 result, which means that Kelly-staking was demonstrably a good idea for this round. Across the entire season however, for which the Line Fund is 32 and 30, Level-staking would have produced about a -2% ROI, some 9% better that what Kelly-staking has produced so far.
At least this year we have more games than last year, so the Fund algorithms will be in possession of information about teams' current-season performances for longer. (I think that's a plus.)
Bookie_9 still leads all head-to-head tipsters despite scoring only 4 from 8 for the Round. This performance was actually marginally better than the all-tipster average for the round which, at just 3.9 from 8, was the lowest for the season. The round's best tipping belonged to Easily Impressed II, which registered 6 from 8, and the round's worst tipping came from Home Sweet Home, which as I've already mentioned scored just 1 from 8. Combo_NN_1 continues to perform admirably, as does Combo_NN_2, both on head-to-head tipping and margin prediction, soundly rebutting my early-season pronouncements about their likely overfitting tendencies.
I've been closely watching Silhouette's and Shadow's tipping this season mindful of their impressive record of consistency in seasons past and wondering if the changes to them made necessary by the introduction of the bye this year would negatively affect their performance. So far, they've not seemed to.
What I realised over the weekend was that I'd not this season looked at the profitability of wagering on their and the other heuristic-tipsters' predictions. So let me redress that now.
Wagering on all of an heuristic's tips would have been profitable for 5 of them: Short-Term Memory I and II, Ride Your Luck, and the two heuristics in question, Shadow and Silhouette.
Punting instead - as I've suggested and tracked in the past - only when a tipster predicts a Home team victory would have been profitable for 7 of the 11 heuristics, amongst them again both Shadow and Silhouette. We all could have made healthy profits over the last few years just level-staking the tips of Shadow or Silhouette.
Turning next to the Margin Predictors, Bookie_3 and Combo_NN_2 both produced impressive MAPE numbers for the round: 27.4 points per game for Bookie_3 and 28.1 points per game for Combo_NN_2. These were the only two tipsters to record sub-30 MAPEs for the round. ProPred_7's results were far less impressive, at 44.2 points per game almost 17 points per game worse than Bookie_3's.
Bookie_3's solid performance allowed it to retain first place on the Margin Predictor ladder. Its season-long MAPE is now 27.30 points per game, some 1.09 points per game ahead of Combo_7 which recorded a 30.2 points per game MAPE for the round. Bookie_9 is in third place on 28.8 points per game, its season-long MAPE dropping this week as a result of a 32.0 MAPE for the round. The all-tipster MAPE for the round was 36.5 points per game, the third-highest of the season.
Just as tipping winners and predicting margins proved difficult this weekend, so too did assessing probabilities.
Amongst the head-to-head Probability Predictors, the TAB Sportsbet bookmaker fared best, but even he barely outperformed what a naive predictor would have scored with a 50% probability assignment to the Home team in every game. The four other head-to-head Probability Predictors all returned mildly negative results for the round, ranging from about -0.06 per game to -0.09 per game. This has allowed the TAB Sportsbet bookmaker to open up a small gap on the the other 4 Probability Predictors.
The Line Fund algorithm's probability assessments were, like those of the TAB Sportsbet bookmaker, just marginally better than naive predictions this week, leaving the Line Fund still slightly negative across the entire season.