2016 Round 19 : Still Trying

/

As I've mentioned previously, I get nervous about wagering at this time of year because the motivations of teams become less clear. In modelling terms, this can manifest as a change in a team's mean ability, as an increase in the variability of its performance around that mean, or as both. All you can really do is abstain from wagering or hope that, on balance, the motivation-induced error and variance terms work in your favour.

One good example of a motivational conundrum this week is the Lions v Port Adelaide game where Port will be playing to keep their slim finals hopes alive, and the Lions will be potentially imperiling a superior draft pick should they spring the upset. In that scenario, is $6.75 a fair price for the Lions? The Head-to-Head Fund thinks it is - and the Line Fund thinks that 38.5 points start looks attractive too.

The Head-to-Head Fund likes two other underdogs as well, Fremantle at $4.50 taking on a Sydney side hoping to press for, at least, a Top 4 spot, and Collingwood at $2.40 playing an Eagles team with similar aspirations to the Swans' but with lesser claims, statistically speaking, to a Top 4 berth. The Fund's sole bet on a favourite is on Melbourne at $1.25 facing the Suns.

As is usual, the Line Fund has wagered on the same teams at the Head-to-Head Fund, though this week it has gone it alone with an additional, small wager on the Cats in Friday's key game against the Dogs.

That wager, however, is of little consequence to the wagering fortunes of the round, which will be far more shaped by the outcome in the Lions v Power game where as much as 6.8c can be won and as much as 2.2c lost.

Almost the same amount could be squandered on the Pies v Eagles game should the Pies lose by 11 points or more, but the upside in that game is less than one-third as large at just 2.2c.

Fremantle controls the fate of the second-largest potential return for the round, a win by them promising a 3.7c increase in the Overall Portfolio. Should they, instead, lose by 28 points or more, just over 1c will be lopped off the value of that Portfolio.

In aggregate, 6.5% of the original Overall Portfolio has been put at risk by the two Funds, with a maximum upside of just over 13%.

TIPS AND PREDICTIONS

Home Sweet Home aside, there's no disagreement at all amongst the Head-to-Head Tipsters this week except in the Collingwood v West Coast game where MoSSBODS_Marg and the two ensemble forecasters based on it have joined Home Sweet Home in tipping the upset Pies victory.

That makes for an incredibly low 10% Overall Disagreement Index this week.

The Margin Predictors' Overall mean absolute deviation (MAD) is also quite low at just 5.1 points per game per Predictor, bumped higher by the 8.4 MAD in the Crows v Dons game, the 7.2 MAD in the Lions v Port game, and the 7.0 MAD in the Hawks v Blues game.

At a Predictor level, ENS_Linear_MoSS has the round's highest MAD of 7.6 points per game, just ahead of MoSSBODS_Marg's 7.3 points per game. ENS_Linear_MoSS is Predictor Most Extreme in three games this week,  MoSSBODS_Marg in only two.

C_Marg, like ENS_Linear_MoSS, is Predictor Most Extreme in three contests, but has still managed to return a MAD that is only the 5th-highest in the league. Bookie_LPSO, the competition leader, has the round's lowest MAD of just 2.1 points per game.

Details of how MoSSBODS arrived at its margin predictions are provided in the table below where we see, perhaps most interestingly, that the Net Venue Effect is decisive in the Pies v Eagles game. There it turns what would otherwise be a 1.4 SS deficit for the Pies based purely on comparative ratings, into a 1.2 SS surplus, which has been enough to have MoSSBODS tipping the upset.

The Net Venue Effect has also played a major part in determining the size of MoSSBODS' projected margin in the Cats v Dogs game, in which the Effect contributes about two-thirds of the Cats' estimated 6.5 SS superiority.

MoS' Head-to-Head Probability Predictors have mostly made assessments in quite a narrow range this week, the largest span of estimates coming in the Pies v Eagles game where Bookie_OE's 40% defines the low end and MoSSBODS_Prob's 56% the high end, and in the Lions v Port game where Bookie_RE's low of 13% meets MoSSBODS_Prob's high of 29%. The overall MAD across the nine games of 2.9% points per game per Predictor is very low.

Looking at the individual predictors we find MoSSBODS_Prob recording the round's highest MAD of 4.2% points per game, C_Prob the second-highest of 3.7% points per game and Bookie_OE the lowest MAD of 1.8% points per game. At least C_Prob and MoSSBODS_Prob have, by virtue of their diverging opinions, given themselves a chance to narrow the lead currently enjoyed by the three bookie-based Predictors.

TIPSTER AND PREDICTOR PERFORMANCES

This week I thought it might be interesting to review which teams have given the MoS Tipsters and Predictors most difficulty this season so far.

Let's start with the Head-to-Head Tipsters, whose performance we'll assess using the accuracy metric.

(Please click the image to access a larger view.)

Overall, the MoS Head-to-Head Tipsters have fared best when tipping games involving West Coast, in which games they've averaged 89% accuracy. Five Tipsters have tipped in those games with 94% accuracy, the overall accuracy dragged down mainly by Consult The Ladder's and Home Sweet Home's 76% accuracy levels.

The Tipsters have also, collectively, tipped at over 80% accuracy in games involving Essendon, the Kangaroos, and the Brisbane Lions.

At the other end of the accuracy scale, games involving Port Adelaide and Collingwood have proven especially difficult to tip, the overall rate in such games falling below 60%.

MoSSBODS_Marg has struggled more than most in tipping the winner in games involving the Lions, Suns and Dockers, but has performed better-than-average in games involving the Crows, Blues and Giants. C_Marg has had particular difficulty with Richmond and Geelong.

The information in the Home Sweet Home column reflects each team's performance when playing as the designated home team and reveals that only the Dogs and Port have lost more than they've won when they've been deemed to be playing at home. West Coast are also shown to be the most successful team at home having won over three-quarters of such games.

Next, the Margin Predictors. For them we'll use the mean absolute error (MAE) metric to assess performance.

We see that the Western Bulldogs have been extremely kind to the MoS Margin Predictors so far this year, with games involving them coming in with a final margin, on average, less than 20 points different from the expectation. MoSSBODS_Marg has been particularly successful in predicting the margin in Dogs' games, its MAE for these games just 18.6 points.

Other teams for which the all-Predictor average MAE is below four goals are the Roos, Crows and Eagles.

Conversely, margins in games involving the Suns, Cats, Saints and Pies have proven most difficult to predict. In these games the all-Predictor average MAEs have been about 33.5 points per game or higher - over 41 points per game, in fact, in Pies' games.

MoSSBODS_Marg has had greatest difficulty relative to the other Margin Predictors in games involving the Lions, Port, Suns and Cats, while C_Marg has been troubled by the Tigers, Lions, Port and the Cats, though to a lesser extent than MoSSBODS_Marg in the case of the last three teams on that list.

Finally, let's review the Head-to-Head Probability Predictors where we find a lot less variability because three of the five Predictors are just variations on a theme. We assess these predictors using the log probability score (which attributes a score of 1+log(P) to a probability estimate of P for the event that occurred, where the log is measured base 2).

Nonetheless, a reasonably consistent ordering of the teams emerges, even when we include the bookmaker-ignoring C_Prob and MoSSBODS_Prob.

Overall, we find that Essendon has been kindest to Predictors, though only slightly moreso than West Coast. Both C-Prob and MoSSBODS_Prob have missed out a little on the bounty that the Dons have provided, but only MoSSBODS_Prob has done the same with West Coast.

MoSSBODS_Prob's most telling deficiency, however, has been revealed with the next four teams - the Lions, Hawks, Roos, Suns - where its probability assessments have proven to be much less well-calibrated than those of the four other Predictors. It has excelled though, relative to the bookie-based Predictors, with the Crows, Dogs, Giants, Blues, Dees and Dockers. That's not been enough to allow it to outperform them overall.

C_Prob has suffered most for its probability assessments of games involving the Tigers, Dogs, Swans and Cats.