2015 - Round 5 : A Cautious Advance

/

The Line Fund remains circumspect this week, venturing just a single wager on the Roos with 19.5 points start against the Hawks, but the Head-to-Head Fund sees value to varying degrees in four home teams, one a favourite and the three others all underdogs. 

Not that favouritism counts for much this week, since in only three of the games is the underdog priced above $3.25. Put another way, six of the underdogs are assessed by the TAB Bookmaker as being at least about 30% chances of defeating the favourite.

The Line Fund's solo wager brings its overall activity level for the season roughly into line with the level at the same time last season. By the end of the 5th round in 2014 the Fund had made 21 bets; this year it's made 22. Last year, however, the Fund was down by just over 1% entering the round, but this year it's down by almost 10%.

Whilst the Line Fund has been wagering, if not winning, at about the same overall rate this year as last, activity levels for the Head-to-Head Fund have been relatively higher. It's made 18 wagers representing more than one-quarter of the total Fund over the course of the five rounds in 2015 compared with only 12 wagers representing less than 8% of the Fund to the same point in 2014. Where it has been similar to the Line Fund though is in its relative profitability, this year compared to last. Currently the Fund is down by 4.5%. It was up by 1.5% at the same time last year.

Combined, the Line and Head-to-Head Funds' five bets this week sum to 3.6% of the Portfolio, which is more than last week's total, but still the second-smallest total of the season.

The Kangaroos, despite being the only team on which both a Head-to-Head and a Line bet has been wagered, don't carry the majority of Investor upside and downside. That responsibility falls to Richmond, who have the power to create or destroy about 1.1% of the Overall Portfolio's value based solely on whether they win or lose.

The Roos are capable of producing almost as much profit as the Tigers, which they'd achieve were they to win, and of producing almost as great a loss were they to lose by 20 points or more. The other teams with fiscal responsibilities this week are GWS, who'll knock 1c off the Overall Portfolio if they lose, and add 0.6c if they win, and Adelaide, who'll knock 0.6c off the Portfolio if they lose, and add 0.8c if they win.

In aggregate, a dream wagering weekend would add 3.5c back to the Portfolio, while a nightmarish one would lop another 3.6c off it.

Head-to-head tipsters

With the TAB Bookmaker assessing five of the week's contests as being most likely settled by about two goals or less we might expect the levels of disagreement amongst the Head-to-Head Tipsters to be relatively high this week. We'd be right in that expectation.

Easily Impressed I (EI I) has set what I'm almost sure is a contrarian record this week by selecting the minority-supported team in eight of the nine contests, missing a full house only by choosing the Dons to beat the Saints on Sunday. EI I's 70% Disagreement Index is a figure I honestly didn't think was possible.

Home Sweet Home (49%), Short Term Memory I (40%) and Follow The Streak (40%) are other Tipsters adding more than their share to the Overall Disagreement Index, which this week stands at 25%, the second-highest value its taken on for an entire round this season.

This relatively high level of disagreement has come about not through the Tipsters being highly divided in a handful of games, but instead through 5 to 11 Tipsters supporting the minority team in two-thirds of the matchups. The Adelaide v Port Adelaide derby has been the most divisive, the Tipsters eventually anointing the underdog Crows as their preferred candidate by 19 votes to 11. In other contentious games, nine Tipsters have plumped for a Giants upset win over the Eagles, and a different set of nine for a Blues upset win over the Pies. 

Amongst the leading group of Head-to-Head Tipsters, C_Marg has the most-divergent tips, opting for two underdogs in Richmond and Adelaide. Bookie_9 is also tipping a Tigers victory in its only contrarian tip, while Combo_NN2 has tipped the Lions to pull off the upset win over the Suns in its lone underdog selection. In every other game, all six of the Tipsters currently tied for the lead on the MoS Leaderboard have predicted a win for the favourites.

MARGIN PREDICTORS

Victory margins, evidence suggests, are no easier nor harder to predict when games are expected to be close. There's no reason then to believe that the MoS Margin Predictors will show elevated levels of disagreement just because the Head-to-Head Tipsters do.

And, this week, they don't, the collective MAD for the Margin Predictors coming in at just 5.8 points per game, only the third-highest level for the season so far, higher than last week's 5.0 points per game, but below the figures for each of the two preceding rounds.

Combo_NN1 and Combo_NN2 have the round's largest MADs of around 11 points per game, more than 2 points per game higher than the next-highest Predictor, Bookie_3, at just over 8 points per game. RSMP_Simple has the round's lowest MAD of 2.3 points per game, slightly smaller than Bookie_9's 2.4 points per game.

Only three games have MADs above 6 points per Predictor: Gold Coast v Brisbane Lions (8.8), Richmond v Geelong (8.0), and Adelaide v Port Adelaide (7.6). Two have remarkably low MADs of 3.6 points per Tipster: Roos v Hawks, and Eagles v Giants.

In the context of the MatterOfStats Tipster Leaderboard, the Gold Coast v Brisbane Lions game has the potential to most significantly narrow or widen the gap between C_Marg and Combo_7, the Predictors at the top of that Leaderboard. ENS_Linear, who sits third, might also peg back C_Marg if the outcome of this game is favourable for it (ie a Suns win by 6 points or fewer) but such a result wouldn't do much for it relative to Combo_7 as both have very similar margin predictions for this game.

PROBABILITY PREDICTORS

The three games that have the largest MADs amongst the Margin Predictors also have the largest MADs amongst the Head-to-Head Probability Predictors: Richmond v Geelong (12% points per Predictor), Gold Coast v Brisbane (11% points), and Adelaide v Port Adelaide (9% points). Unlike the Margin Predictors, however, the Head-to-Head Probability Predictors have the smallest MADs for the Sydney v Western Bulldogs (2.3% points) and St Kilda v Essendon (2.4%) games.

At 6.4% points per game, average MAD levels across the Head-to-Head Probability Predictors are only slightly higher this week than they've been for the last two weeks. H2H_Unadjusted has the highest MAD at 8.7% points, just ahead of ProPred at 8.6% points, while Bookie-LPSO has set the low-water mark this week at 4.1% points per game. With the adjustment rule again being invoked for one of Head-to-Head's probability assessments (for the Richmond v Geelong game), H2H_Unadjusted's assessments differ from H2H_Adjusted's for the fourth successive round.

So far, those adjustments have been to H2H_Adjusted's benefit, its Log Probability Score for the season so far currently moderately superior to H2H_Unadjusted's.

Only one game this week has been assessed by the Line Fund algorithm as a 60% chance or better for one of the teams. It's made that assessment about the Giants' chances over the Eagles based on the former's roughly 2-goal positive handicap and its assessment of the relative abilities of the two teams, given the venue. The Giants are, however, the away team, so the Line Fund has no financial stake in the soundness of this opinion.

It does though have a stake in the team that the algorithm's assessed as having the round's second-highest probability of collecting in the line betting market, the Kangaroos, which the algorithm rates as 59% chances. The two teams with the next-best chances are the Dogs and the Lions, each assigned 56% probabilities, but each is playing as the away team in their respective fixtures.

DISAGREEMENT LEVELS AND PERFORMANCE

This week, along with the usual summary of Disagreement levels across the various Tipster and Predictor groups for the season so far, I've included the average performance levels in each week, which for the Head-to-Head Predictors is measured by Accuracy, for the Margin Predictors by Mean Absolute Error (MAE), and for the Head-to-Head Probability Predictors by Log Probability Score.

Though the samples are only very small (four in every case), it's interesting to note that:

  • The correlation between the Combined Disagreement Index and Accuracy is -0.62 for the Head-to-Head Predictors
  • The correlation between the Mean MAD and MAE is +0.64 for the Margin Predictors
  • The correlation between the Mean MAD and Log Probability Score is -0.62 for the Head-to-Head Probability Predictors

In short, so far this season, the greater the levels of disagreement amongst a Tipster or Predictor group in a round, the more poorly they've performed collectively in that round. 

Logically, it's impossible for a group of forecasters to have high levels of disagreement yet perform well collectively since the accuracy of some forecasters necessitates the inaccuracy of those who disagreed with them. It is possible, however, for a group of forecasters with low levels of disagreement to perform well or badly as a collective.

The high correlations above between levels of agreement and collective performance suggest that low levels of disagreement have, so far this season, signalled collectively strong forecasting, while high levels of disagreement have, for the reason just explained, portended a weekend of collectively weak forecasting.

On that basis we might expect a relatively strong collective performance of the Margin and Head-to-Head Probability Predictors this week, and a relatively weaker collective performance by the Head-to-Head Tipsters.

Predicting the aggregate performance of forecasters on the basis of the historical relationship between their levels of agreement and their subsequent performance - what next?