Matter of Stats

View Original

2015 - Round 20 : Going Very Quietly Into That Good Night

Investors can have Sunday off this weekend, Round 20's action, such as it is, all being confined to Friday and Saturday. On Friday, we've taken the Swans at -22.5 in the line market, while on Saturday we've a tiny head-to-head wager on the Dons at $4 along with two line bets, one of them on the Dons with 26.5 start, and the other on Port Adelaide giving 14.5 start.

The Dons then, barely, offer the round's greatest upside and downside, a victory by them promising a 1.6c gain in the Overall Portfolio, and a defeat by 27 points or more threatening a Portfolio loss of the same magnitude. In the two other live contests, both the Swans and Port can add 1.4c or subtract 1.5c from the Portfolio. It's very much a "nobody gets hurt" kind of round.

HEAD-TO-HEAD TIPSTERS

There's a little more dissension amongst the Head-to-Head Tipsters this week, with six of the round's games eliciting minority support of at least a half dozen Tipsters.

The game generating the highest level of disagreement is Sunday's Dockers v Eagles derby, where the Tipsters have come down 17-13 in favour of the designated home team. Next most divisive is the Port Adelaide v GWS clash where nine Tipsters have opted for a Giants upset. (I would note though that all nine are Heuristic Tipsters and, Consult The Ladder aside, none of them have exactly shot the lights out this with their prescience this season.)

Two games see unanimity of opinion, the Roos and the Dogs both being selected by all 30 Tipsters.

Overall, the Disagreement Index for the round has come in at 28%, which is at the high end of values for this season (it's the 5th-highest). Mostly, it's been driven up by the Heuristic Tipsters, though even Bookie_3 has turned in one of its most apparently contrarian set of tips this week. When widescale contrarianism contributes to your definition of the norm, even conservativism looks contrarian.

Amongst MoS' leading Head-to-Head Tipsters, it's only the Lions v Blues game generating any disagreement at all, with BKB opting for the Blues (because it follows Consult The Ladder when there are equal-favourites), Bookie_3 tipping a draw, and Combo_7, Bookie_9 and C_Marg all selecting the Lions.

MARGIN PREDICTORS

Much less disagreement is in evidence amongst the Margin Predictors, the round's all-Predictor average Mean Absolute Deviation (MAD) coming in at just 4.7 points per Predictor per game, the 2nd-lowest value we've seen all season, larger only than last week's 3.3 points per Predictor per game.

Saturday's Roos v Saints game has generated the broadest range of opinions, the MAD for that game 8.7 points per Predictor and the range of predictions spanning just over 5 goals from a 14- to a 46-point Roos victory.

The only other game with a MAD greater than a goal is the Tigers v Suns matchup, which actually has an even bigger range of predictions (41 points) than the Roos v Saints clash.

Combo_NN_2, which is Predictor Most Extreme in six of the round's games has easily the highest MAD of just under 10 points per game, Win_3 coming in 2nd-highest at just 6.6 points per game, a result it achieved despite being Predictor Most Extreme in only two games.

MoS' leading Margin Predictor, C_Marg is most at risk of giving up ground to the rest of the field if Port Adelaide, Hawthorn or West Coast fail to win convincingly.

WHICH TEAMS' MARGINS HAVE BEEN HARDEST TO PREDICT THIS SEASON?

This week, a quick analytical diversion into margin prediction before we move to probability prediction.

For the table at right I've calculated the Mean Absolute Errors (MAEs) on a team-by-team and a home v away basis for the predictions of selected Margin Predictors so far this season.

From the table we can see, for example, that C_Marg has, overall, predicted Fremantle's final margins most accurately, on average across their 18 games predicting margins less than 4 goals different from the actual margin. For Dockers' home games C_Marg has been even more accurate, predicting, on average, within about 21 points of the final margin.

Next best has been C_Marg's performance in Adelaide games where it's predicted with an MAE of about 4 goals both for the Crows' home and their away games. C_Marg has done worst in Gold Coast games, where its MAE has been over 6 goals, Suns home games being especially problematic and resulting in a 43 point MAE.

Looking across the other Margin Predictors for whom results have been provided (the TAB figures, by the way, inferred from the TAB's handicap in the line market, adjusted slightly where the handicap prices were not $1.90 for both teams) we see that the pattern of MAEs by team are broadly similar.

All of the Predictors have done well in forecasting Carlton and Essendon away games, and Richmond and St Kilda home games, for example.

Overall though, C_Marg has outpredicted the TAB by about 0.8 points per game, C_Marg being especially superior in predicting:

  • Sydney, West Coast, Brisbane Lions and Western Bulldogs away games
  • West Coast, Carlton, Brisbane Lions, Western Bulldogs, Collingwood and Melbourne home games

Conversely, the TAB has been especially superior to C_Marg in predicting:

  • Port Adelaide and Melbourne away games
  • GWS, Kangaroos and Gold Coast home games

PROBABILITY PREDICTORS

Levels of disagreement are also low amongst the Head-to-Head Probability Predictors this week, the round's 3.8% points per Predictor per Game MAD the 3rd-lowest of the season, behind only last week's 3.6% and Round 14's 3.5%.

C_Prob, for the second consecutive week, and for the sixth time in seven weeks, has the round's highest MAD, though at 5.1% points per game that's nowhere near as MAD as it's been in other weeks. Win_Pred has the 2nd-highest MAD of 5.0% points per game, but is Predictor Most Extreme in five games to C_Prob's four.

Relative to Bookie-OE and Bookie-RE, both of which sit higher than it on the MoS Leaderboard, C_Prob has most to lose if St Kilda, GWS or Fremantle prevail.

Looking from a game-by-game viewpoint, it's that Freo v Eagles game that has generated the most disparate probability predictions, the all-Predictor MAD coming in at 6.9% points per Predictor, and the range of probability assessments spanning 18% points. There's a slightly wider span in the Tigers v Suns game (19%), though there the MAD is only 4.4% points per game as five of the nine Predictors have made assessments within a 5% point range from 79% to 84%.

The Line Fund algorithm has been particularly bold this week in rating the chances of five teams as 59% or higher: Port Adelaide (66%), Hawthorn (66%), West Coast (66%), Sydney (60%), and Melbourne (59%). Three other teams, Essendon, the St Kilda and the Carlton, have all been assessed as 54% chances.

DISAGREEMENT HISTORY

At the end of Round 19, the correlations between Tipster/Predictor overall performance and average by-round disagreement levels were as follows:

  • Head-to-Head Tipsters' average weekly Disagreement Index and overall tipping accuracy: -0.86
  • Margin Predictors' average weekly MAD and overall MAE: +0.39
  • Head-to-Head Probability Predictors' average weekly MAD and overall log probability score: -0.39

So, we still find that conformity to all-forecaster averages has tended to produce more accurate predictions this season. C_Marg and C_Prob remain notable exceptions however.