MoS for the Men's 2024 AFL Season
/The men’s 2024 version of MoS will be much the same as the men’s 2023 edition, especially on the forecasting front.
You will again see somewhere on the site weekly posts predicitng results for the round ahead, reviewing the results of the round just gone, updating Team Ratings and the Team Dashboard, and projecting the remainder of the home and away season and the Finals using a technique I railed against for many years.
The main changes are decribed in the sections that follow.
ALGORITHMS
The algorithms for MoSSBODS, MoSHBODS, and MoSHPlay have all been reoptimised to include the 2023 season, although the changes do not appear to be large. The new version of MoSHPlay would have finished first on MArgin MAE at Squiggle, although that might merely be a sign of overzealous optimisation leading to overfitting.
For MoSSBODS and MoSHBODS, the changes mostly amounted to choosing new parameter values, including a review of where to split home and away seasons of different lengths, and what value of k to use for updating ratings within each section.
For MoSHPlay, there was some minor work done on how best to combine the team and the player components, but more significant work done on the algorithm that provides that player component. It predicts each player’s AFL Player Rating for the game about to be played and now involves:
(1) Creating a weighted average of previous AFL Player Ratings for each player
Looking only at the last 4 years of data
Increasing 2020 Ratings by 25% to adjust for the shorter quarters
Applying smaller weights to historical performances the further back in time they occurred
(2) Regularising these weighted averages
Revising a player’s weighted average rating if he’s played fewer than 19 games by assuming that he scored a weighted average equivalent of 5.8 in each of the “missing” games. This, obviously, tends to regularise the predictions towards 5.8, and produces what is called a Final Forecast
Once the actual date for the next game is known along with each player’s named position, these Final Forecasts are further adjusted
They are firstly decayed based on the number of days between the player’s most-recent game and the date of the game being forecasted (up to a maximum of 23 days, even for games that were played in a previous season)
Once decayed, a ceiling is placed on the final value at 15.9, and a floor at 5.0
Rookies, for whom no value will exist at this point, are assigned a value of 5.1
Lastly, a position-based adjustment is made to arrive at a Pregame Forecast Rating.for each player
For use in MoSHPlay, a team average of these Pregame Forecast Ratings is calculated (excluding the substitute), which is linearly combined with the relevant MoSHBODS margin forecast for the game.
FUNDS
The same three Funds will operate in 2024 with the new weightings being Line Fund 65% / Head-to-Head Fund 25% / Over Under Fund 10%.
Significantly, all three will follow the same staking strategy, which will be fractional Kelly (20%, and applied to the original, not the current, bankroll) with a minimum edge of 5%. Some examples might help to explain what that means:
Head-to-Head Fund (uses MoSHBODS’ probability forecasts)
GWS best price $1.79 / COL best price $2.10 / MoSHBODS estimate Pr(GWS win) = 72%
Estimated GWS edge = 0.72 x 1.79 -1 = +28.9%
Estimated COL edge = 0.28 x 2.1 -1 = -41.2%
The estimated edge for GWS exceeds 5%, so we wager an amount equal to 20% x Edge / (Price - 1), which is about 7.3%
So, the wager would be 7% of the Fund on GWS
Note that the Head-to-Head Fund will be permitted to wager on Home and on Away teams this season
Line Fund (uses MoSSBODS’ margin forecasts and MoSHBODS estimates for the standard deviation of the margin)
MoSSBODS forecasts SYD over MEL by 2 points.
TAB has SYD -9.5 / MEL +9.5 both at $1.90
Sportsbet has SYD -8.5 / MEL +8.5 both at $1.90
Estimated SYD edge at TAB is NORMDIST(2 - 8.5, 0, 34, TRUE) x 1.9 - 1 = -22% (NB the 34 is MoSHBODS’ forecast margin standard deviation for the game)
In a similar fashion, the estimated SYD edge at Sportsbet is -20%
For MEL, we have estimated edges of +12% at the TAB and +10% at Sportsbet
The estimated edge for MEL is maximised at the TAB and exceeds 5%, so we wager an amount equal to 20% x Edge / (Price - 1), which is about 2.7%
So, the wager would be 2.7% of the Fund on MEL +8.5 at $1.90
Over/Under Fund (uses MoSSBODS’ Total forecasts and MoSHBODS estimates for the standard deviation of the Total)
MoSSBODS forecasts a 169 total for the BRI v CAR game
TAB has o164.5 and u164.5 at $1.90
Sportsbet has o164.5 and u164.5 at $1.88
Estimated Overs edge at TAB is NORMDIST(169 - 164.5, 0, 29, TRUE) x 1.9 - 1 = +6% (NB the 29 is MoSHBODS’ forecast Total standard deviation for the game)
In a similar fashion, the estimated Overs edge at Sportsbet is +5%
For Unders, we have estimated edges of -16% at the TAB and -17% at Sportsbet
The estimated edge for Overs is maximised at the TAB and exceeds 5%, so we wager an amount equal to 20% x Edge / (Price - 1), which is about 1.3%
So, the wager would be 1.3% of the Fund on o164.5 at $1.90
The overall effect of these changes will likely see Investors wagering on more games and with higher average wagers. As a rough estimate, I expect the money in each Fund to turn about 3 times over the course of the season, which is about twice what has happened in previous years. That will, obviously, multiply the ROI by a larger amount, which will be fantastic if it’s positive …
Wagering based on a minimum estimated edge of 5% is nothing new for the Head-to-Head Fund, but is quite foreign to the Line and Over/Under Funds, which have previously had minimum overlay requirements of 8 and 6 points respectively. As a rough estimate, the 5% edge requirement is equivalent to about 4 or 3 point overlays respectively.
It’s entirely possible that these changes will result in Fund behaviour that is far too profligate, but you’ll at least have the fun of watching it all unravel from a safe distance.
