2014 - Round 18b Results : One Streak Finishes and Another (We Hope) Begins

The Hawks' victory on Saturday night - which, by game time, even the bookmakers thought more likely than not - simultaneously snapped the Swans' 12-game winning streak and ended Investors' two-round losing streak.

But for another Swans score of either kind, the return for Investors from that Hawks triumph would have been better still, the SuperMargin bet on a 1 to 9 point victory margin missing out by just a single point.

Failure on SuperMargin wagering was a feature of the round - well, let's be honest, it's been a feature of the season - the lack of success of the two SuperMargin wagers on the Hawks v Swans game being replicated on six other occasions during the round. The resultant 8c loss by the Margin Fund left it down by over 65c on the season and almost certainly means that it will be looking for alternative means of employment next season.

That 8c loss was comprehensively swamped by the Head-to-Head Fund's 1.5c gain and the Line Fund's 13c gain, which meant that the Recommended Portfolio grew by just over 5c across the round. It's now up by almost 13c on the season.

The Line Fund, up by close to 50c for the year, as foreshadowed in the previous post on this blog, halves its bet size per wager next week to just 2.5%. It's been wagering at 5% of the Fund per bet since the start of Round 12 and a review of its performance during the period since then suggests that we probably got the timing about right this season. If we might wish for anything different from the Fund it would be that it'd ramped up its bet size earlier in the season, say from Round 8.

View fullsize

While we're on the topic of hindsight-induced regret, let's spend a moment reflecting on the choice of Combo_NN2 and Bookie_9 as the guides for our SuperMargin wagering. A quick review of the latest data (see below) reveals that, across all of MatterOfStats' Margin Predictors, Combo_NN2 has the poorest ROI on SuperMargin wagering, and Bookie_9 is in the lower half.

View fullsize

Focussing solely on Margin Predictor opinions when they've indicated a home team win or draw, which is the strategy we adopt in our SuperMargin wagering using Combo_NN2 and Bookie_9, eight of the MoS Predictors have recorded a profit so far this season: both ProPred-based Predictors, both WinPred-based Predictors, and all four H2H-based Predictors.

All of which says no more, I guess, than that profitability in this form of wagering has been possible this season, just not with the tour guides we've selected for the journey. 

TIPS AND PREDICTIONS

Being right was also a challenge for the Head-to-Head Tipsters during Round 18, the three from nine performance of as-at-Wednesday pre-game favourites resulting in an all-Tipster average score of just 3.9. Best amongst them was Easily Impressed I who, in its cluelessness, managed a six from nine performance.

Combo_NN1 was the sole Tipster correctly predicting the Adelaide victory, on the strength of which it drew level with Combo_7 in second place on the MoS Leaderboard, just two tips behind Bookie_9.

View fullsize

Margin prediction also proved problematic in Round 18, the all-Predictor average Mean Absolute Prediction Error (MAPE) for the round coming in at 38.9 points per game. The round's best MAPE was ProPred_7's 35.4 points per game, ahead of Win_3's 37.0 and C_Marg's 37.4 points per game.

Combo_7's MAPE of 39.0 points per game, though almost 2 points per game worse than Win_3's, was good enough to leave it atop the MoS Leaderboard, a position it's now held for the past two rounds. 

View fullsize

Win_3's performance was, however, good enough to leapfrog it over Bookie_LPSO and into second place. Two other notable climbers over recent weeks have been the RSMP Predictors, which last year finished in first and second place on the Leaderboard. They're up by about a half-dozen places each over the past three rounds.

Twelve Margin Predictors now have profitable line betting performances for the season and only two of them, Bookie_3 and Combo_NN1, have predicted at lower-than-chance levels.

The week's best - or least worst - head-to-head probability predictions belonged to ProPred, though C_Prob's was virtually identical. That result for C_Prob ensured that it retained second place on the MoS Leaderboard. 

(Despite the accuracy of C_Prob's probability assessments this year, wagering based on them has proven profoundly unprofitable so far this season. Kelly-staking when C_Prob has recommended a wager on the home team has resulted in an ROI of -21.5%. I plan to investigate the reasons for this in a future blog.)

Finally, the Line Fund algorithm recorded another negative probability score this round, its third negative score in succession.