2015 - Round 8 : Slim Pickings

Just four wagers dot the weekend's footballing landscape, and none threatens the long-term health of Investors' or the TAB Bookmaker (insert irony emoji here).

Two smallish head-to-head wagers, the first and smaller a speculative waft on the Saints, and the second a bigger but not materially bolder tilt at the Hawks, are joined by a pair of line bets on other teams, one a favourite and the other an underdog.

Combined the bets represent only 4.1% of original Funds, the smallest aggregate since Round 5, and the third-smallest aggregate of the season. Four collects would see the Overall Portfolio climb by 4.2c, the largest of those windfalls dependent on success for either of the line bets.

Those same two bets also make possible the largest single-game losses, a Gold Coast defeat by 19 points or more, or a Fremantle win by 25 points or fewer, both threatening 1.5% of the Overall Portfolio. The two head-to-head wagers will lead to Portfolio gains or losses roughly in the 0.5c to 1c range.


This week continues the trend amongst the Head-to-Head Tipsters for higher and higher levels of agreement, the Overall Disagreement index coming in at a seven-week low of 13%, the second-lowest value it's taken on this season.

Five contests have either unanimous support for the favourites or have Home Sweet Home as the lone dissenting voice, and there are only five to seven minority Tipsters in three more games. The first group of games has as the majority selections Geelong, Fremantle, West Coast, Collingwood and the Western Bulldogs, the second GWS, Hawthorn and Essendon. 

That leaves just one contest with more significant levels of minority dissent, that the Port Adelaide v Richmond game where Port are the 18-12 majority pick.

Home Sweet Home has the week's most dissenting tips, Short Term Memory I and Follow The Streak the equal-second most dissenting. Again this week the three leading Tipsters on the MatterOfStats Leaderboard have produced the same nine predictions, so again they'll not be split by round's end. Combo_NN2, however, who sits one tip adrift of the leaders could draw level if its prediction of an Adelaide win over the Giants proves correct.


Disagreement amongst the Margin Predictors is also low this week, the Overall Disagreement index at a five-week low of 5.4 points per game per Predictor. That's the third-lowest average this season.

Driving that average up most are the predictions for the Cats v Blues (9.5 points per Predictor), Dockers v Roos (7.5), and Dons v Lions (7.0) games, while the lowest MADs have been recorded in the Hawks v Swans (2.4) and Dees v Dogs (2.6) games. It's remarkable when you think about it, that 19 different algorithms can all produce predictions within a 13 point range, which is the case for that Dees v Dogs game.

Combo_NN2 has gone furthest out on a limb this week, its MAD of 10.6 points per game including three predictions where it finds itself the most extreme Predictor (Hawks by 18, Dockers by 44, and Essendon by 39). Combo_NN1 is actually extreme in more games (4) but has recorded only the fifth-highest MAD.

The week's low MAD of 2.5 points per game belongs to Combo_7, who is the extreme Predictor in no contests at all.


Competing the week's story of general harmony and agreement, the Head-to-Head Probability Predictors have made probability assessments so similar that they've recorded a season-low average MAD of 4.8% points per Predictor per game. Even ProPred, the round's Predictor most different only has a 6.1% point MAD and no single probability assessment that is more than 10% points different from the all-Predictor average.

Bookie-OE, the leader on the MatterOfStats Leaderboard, has the round's lowest MAD of 3.3% points.

Not only are the probability assessments tightly clustered across the in eight Predictors this week but, in eight of the nine games all of the Predictors' assessments of the home teams' chances are on the same side of 50%. Only in the Port Adelaide v Richmond game do we find a defiant WinPred assessing Port's chance as being just a tick under 50%.

Across the games, and mostly as a result of that WinPred assessment, its the Port Adelaide v Richmond game that bears the largest MAD of 8.2% points per Predictor. The Geelong v Carlton game has the next-highest MAD of 7.0% points, it the only other game with an above-average MAD this week. The low MAD is in the GWS v Adelaide game where the entire range of probability assessments spans only 7% points.

The Line Fund algorithm rates most highly in their respective line markets the chances of Richmond (61%), the Brisbane Lions (60%), Carlton (59%), and the Western Bulldogs (57%), all of them away teams and so none of them carrying a Line Fund wager.


Some additional detail and perspective on the high levels of agreement this week can be gleaned from the usual table showing each forecaster's level of disagreement with its peers each week. (Note that the colour-coding is done for each group of forecasters and each week separately, so, for example, a dark green in one week might be achieved with a different disagreement index than a dark green in another week.)

The correlations between average levels of disagreement and average forecaster performance remain mildly positive/negative in the direction we'd expect if dissention were associated with weaker performance:

  • Between average Disagreement for the Head-to-Head Tipsters and average accuracy in a round: -0.49
  • Between average MAD for the Margin Predictors and average MAE in a round: +0.38
  • Between average MAD for the Head-to-Head Probability Predictors and average log probability score in a round: -0.46