This year I'm starting the process of projecting the Final 8 a little early. Again this year though I'll be employing a very simple model to predict the outcome of the remaining games, using only the teams' current MARS Ratings and the AFL-designated home team status.
The model I'm using is as follows:
Predicted Home Team Margin = 0.76383 x Home Team MARS Rating - 0.75723 x Away Team MARS Rating
Based on current Ratings this yields the predicted margins for the remaining games as shown below.
For Round 16, for example, the prediction in the Kangaroos v Hawthorn game is that the Hawks will win by about 18 points.
Obviously, this far out from the end of the season, projections about results in later rounds are highly speculative, so you should interpret those predictions and, indeed, the end of season projections in general, in full recognition of that fact.
Nonetheless, I convert the predicted margins you see here to a simulated margin for a given replicate by assuming that:
Actual Home Team Margin ~ N(Predicted Home Team Margin, 36).
These seem like reasonable distributional and parametric assumptions given previous empirical work, which I've reported here on MatterOfStats.
On the basis of the implied probabilities shown here, in about 25% of the remaining games the favourite will carry a probability of between 50% and 60%, and in 25% more it'll carry a probability of between 60% and 70%. Further, in about 30% of games the favourite will carry a probability of between 80% and 90%. All told, that's probably not a bad profile of games where the underdog has a reasonable chance.
I've used this formulation to complete 10,000 simulations of each of the remaining contests, projecting the final competition ladder that results. As has been the case in years past, rather than projecting each team's final percentage I've instead projected each team's For and Against Differential. In most cases, percentage and differential will result in the same team ordering, though there have been examples in recent years where this has not been the case.
After deploying the projected margins in the 10,000 simulations, the following team-by-team manhattan emerges.
In most cases, teams have only one or two ladder positions that they're most likely to fill at season's end:
- Adelaide: 9th or 8th
- Brisbane Lions: 17th or 16th
- Carlton: 13th or 14th
- Collingwood: 6th or 7th
- Essendon: 10th or 11th
- Gold Coast: 8th or 9th
- GWS: 16th or 17th
- Hawthorn: 1st
- Kangaroos: 6th or 7th
- Melbourne: 15th or 16th
- Richmond: 12th
- St Kilda: 18th
- West Coast: 11th or 10th
- Western Bulldogs: 13th or 14th
The exceptions are Fremantle, who are almost equally likely, according to the simulations, to finish 2nd, 3rd or 4th; Geelong, who are most likely to finish 5th, but also about equally likely to wind up in 4th or 6th; Port Adelaide who are about equally likely to finish 2nd, 3rd or 4th; and Sydney who are about equally likely to finish 1st, 2nd or 3rd.
A heat map provides another way to review that same data.
This map makes the fight for 8th place a little clearer, showing the main skirmish to be between Adelaide and Gold Coast, with Essendon and West Coast the other teams enjoying non-negligible prospects for this position.
We can also see that Geelong is the only team outside Hawthorn, Sydney, Port Adelaide and Fremantle to have genuine prospects for a Top 4 position. The Pies and the Roos have small but non-zero probabilities in relation to this position.
Overlaying the simulation results with the latest TAB Bookmaker prices suggests that the only "AFL Futures" wager of any significant value amongst those I've reviewed is that for Hawthorn to secure the Minor Premiership, which is currently priced at $3 and assessed by the simulations as being a 47% prospect. That's quite a healthy edge.
Note that I can't assess value in any Spoon market since, currently, the TAB is not framing any market for this outcome, though it is offering odds on the team to record the "Most Losses". Presumably, if two teams record the same number of losses, both are considered losers as far as this bet is concerned, so a bet on this outcome is not the same as a bet on the Spoon.
I'll finish by reporting the most common Top 2s, Top 4s and Top 8s according to the simulations.
The most common Top 2 has the Hawks in 1st and the Swans in 2nd, while the next most common swaps in Port Adelaide for the Swans.
A Hawks, Port, Sydney, Fremantle Top 4 is assessed as being the most likely, though only slightly more likely than a Hawks, Sydney, Port, Fremantle finish.
Finally, the most likely Top 8 has the following ordering: Hawks, Port, Swans, Freo, Cats, Pies, Roos and Gold Coast. This ordering is assessed as having a probability of only about 1 in 300 however, which is about the same probability as the ordering in which the Swans and Port Adelaide change positions relative to the scenario just described.