We're starting things a little earlier for Round 1 of this year, partly because the new Fund models don't require me to wait for complete bookmaker data to use as inputs, but mostly because I want to give myself time to adapt to the new rhythms of the season given the non-trivial changes in my approach.
This year I'll aim to post the week's wagers, tips and predictions on a Monday or Tuesday night, and then provide a brief update once the Unders/Overs markets are released later in the week. I realise that means I'll be showing my hand to the TAB Bookmaker before being able to lock in those Unders/Overs bets, but I suspect that'll make a difference statistically insignificantly different from zero ... to us both.
So, firstly to the summary.
As it did last year, the Round Summary provides MARS Rating, Head-to-Head and Line market TAB prices (in today's blog as at around 5:30pm on Saturday), and the MoS Funds' Head-to-Head and Line market wagers at those prices.
Investors have then, six wagers so far for this round, encompassing Head-to-Head and Line wagers on Melbourne, Sydney and the Western Bulldogs. The Head-to-Head bets are all, as they will be all season, equally sized, while the Line bets range from just 0.9% of the Line Fund to 2.4% of the Line Fund.
There's also room in the Summary this year for details of the Overs/Unders market and any wagers that might have been placed on it. These markets, however, will tend to be posted nearer game start times - presumably because of the significant effects of any inclement weather - and so will only appear in the update I'll post later in the week.
For now I'm doing away with the Ready Reckoner that maps results to wagering returns on the basis that, on balance, I don't think it adds much new information and that it would, in any case, be difficult to amend to incorporate Overs/Unders betting. Suggestions are welcomed.
TIPS AND PREDICTIONS
For those of you new to MoS this blog post from a couple of months back will provide you with some background to the Tipsters and Predictors that will be sharing their opinions with us this season.
Since I wrote that post, the only change has been the addition of a couple more Head-to-Head and Margin Predictors, these statistical models taking MoSSBODS Ratings as inputs. (For the technically minded, ENS_Linear_MoSS and ENS_Greedy_MoSS are ensemble forecasters that take MoSSBODS Ratings as their sole inputs. They include twelve base-learners - though not all are used by ENS_Greedy_MoSS - spanning an ordinary least squares regression, some forests, xgboost, a nearest neighbour model, and some other secret herbs and spices. Unlike KFC, however, the recipe is available on request.)
We find that the 11 Head-to-Head Tipsters are unanimous in five contests and nearly so in two more, Home Sweet Home the dissenter in one of those matches, and Consult The Ladder the dissenter in the other.
That leaves just two contests where there's significant debate: the Dees v Giants clash, where the narrow majority supports the Giants, and the Dogs v Dockers clash, where the Dockers are very slightly favoured.
The column headed "Disag" provides a measure of how different are that Head-to-Head Tipster's tips, with larger values representing greater divergence from the group opinion. Technically, this percentage records how likely it is that the Tipster will have a different opinion to another, randomly chosen Tipster in a randomly chosen game. This week Home Sweet Home wears the crown of Tipster Most Different - a crown that it had virtually welded to its head for most of last season.
Directly beneath the Head-to-Head Tipsters are the Margin Predictors, and they find themselves unanimous in their opinions about which team will win in seven of the nine contests. Even in those games, however, there's considerable disagreement about the margin of those victories. In the Gold Coast v Essendon game, for example, the smallest predicted victory margin for the Suns is just 4 points while the largest is 40 points.
You can get an idea of the level of dissent amongst the Margin Predictors by reviewing the last two rows of their data block where you find the range of their predictions (in the row labelled Min / Max) and their Mean Absolute Deviation (in the row labelled MAD). The latter measure is calculated by a simple averaging of the absolute difference between each Predictor's margin and the all-Predictor margin for that game. Larger numbers connote more variability in the predictions.
Given that understanding of the MAD measure, it's clear that the margin for the Gold Coast v Essendon game is the hardest one to tip this week, and that for the Roos v Adelaide game the easiest.
As well as calculating a MAD for each game we can calculate a MAD for each Predictor, which gives us a measure of how different are that Predictor's margins from the all-Predictor averages. This week sees MoSSBODS_Marg as the Predictor with the largest MAD of 7.8 points per game, and ENS_Greedy as the Predictor with the smallest MAD of 2.3 points per game.
The final block of data in the table is for MoS' Probability Predictors, which this season number only five. As the Margin Predictors, the Probability Predictors find themselves on the same side of 50% for the same team in all but two of the contests, though their distances from 50% vary considerably.
Here too the bottom two rows of the data block provide information about the variability of the predictions viewed across all predictors, and it is again the Gold Coast v Essendon game where variability is greatest. The average absolute difference between the individual Predictor's probability assessments and the all-Predictor average of a 74% probability for Gold Coast is 12.4%. The range in those assessments runs from 55% (for MoSSBODS_Prob) to 86% (for Bookie_LPSO).
The smallest range and MAD is seen for the West Coast v Brisbane Lions game where the probability assessments span only a 4% range and the MAD is just 1.3% points per Predictor.
For the Probability Predictors too we have a MAD-based measure of individual Predictor disagreement, which this week has MoSSBODS_Prob as the Predictor Most Different, its probability assessments, on average, being almost 8% points different from the all-Predictor averages.
DETAILED MoSSBODS PREDICTIONS
As foreshadowed in this post from early February, each week I'll include detailed MoSSBODS Predictions that break down the components of MoSSBODS' margin predictions and provide additional predictions about the projected score for each team and their aggregate score.
To recap, briefly, on how to read this table, consider the first row where we see that MoSSBODS (see this post to see how MoSSBODS works) has Richmond as 6.5 Scoring Shot (SS) favourites, they being 4.2 SS better in Defence (+1.8 vs -2.4) and 2.6 SS better in Offence (-0.4 vs -3.0). The game is being played at the MCG, which favours Carlton slightly (by 0.3 SS), which gives us 4.2 + 2.6 - 0.3 = 6.5 as the predicted Richmond margin.
This SS margin is converted to actual points using last season's average Conversion rate for all teams. Richmond is therefore tipped to win by 24 points, specifically 93-69 as the next two columns show. A 6.5 SS expected margin implies, empirically, a 77% victory probability, which is the figure in the fifth-last column. The remaining columns provide information about wagering thresholds.
Looking across the set of predictions, one notable feature is the scarcity of expected scores over 100 points. Only West Coast is predicted to achieve this mark. Last season, about 31% of all scores were over 100 points, on which basis we might expect to see 5 or 6 of this magnitude in this first round. The reason we see fewer amongst the predictions is, I would posit, an artefact of the regression of MoSSBODS Team Ratings towards zero that takes place at the end of each season.
I'll write a brief post sometime in the next couple of days once the Overs/Unders markets have been posted and additional bets, potentially, have been made.
Until then, please go ahead and comment on this post or send me an e-mail if you've any questions or suggestions.