Skip to content

Figuring Footy’s New 2016 Predictor – FFSS Ranking System Meet the Figuring Footy Scoring Shot Team Ranking System

Posted in Uncategorized

Late last year I introduced SimpElo, a (very) simple Elo-based ranking system for the AFL. During the summer I have been working hard to improve SimpElo, taking in depth looks at a range of different indicators and assessing their predictability for AFL matches. The end goal being to create a ranking system and set of algorithms that can predict the implied probabilities and chances of each team winning each game. This still remains very much a work in progress, but as the 2016 season sneaks up upon us, now is as good a time as it gets to put the system to the test.

 


What is this? Where am I?


If you have no experience with Elo-based rating systems I highly recommend reading through the introduction to SimpElo. In short, an Elo rating system is one where teams gain rating points for winning matches and drop rating points for losing matches. The amount of points they gain or drop is directly proportional to the rating of their opponent compared to their own rating. Beat a good team, you’ve proved your chops. Beat a poor team, you might be already be accurately rated and don’t really deserve a ratings boost.

I’m dubbing the new and improved system the FFSS (Figuring Footy Scoring Shots) Rating System (V.1). The most immediate change from SimpElo is that FFSS also takes into account the margin of victory of each match. More points are awarded for bigger wins, but this tapers off quickly as to not reward teams too heavily for running up the scoreboard in matches they have already won. For example, a team expected to lose by a small margin who actually wins by 20pts would receive a much bigger ratings boost then a team we expect to win by 50pts actually winning by 70.

Secondly, and perhaps most interestingly, the FFSS system uses the amount of and quality of scoring shots a team creates to make inference about the innate ability of each team. We have seen before how Goal-Kicking Accuracy is a highly unrepeatable task and the best teams are often not the ones who kick the straightest. What is certainly more clear is which teams manage to consistently create quality chances in their forward 50s while limiting their opponents chances. Teams that can create these chances deserve to have their ratings reflect this, regardless of any (short-term) unlucky conversion rates. FFSS looks at each scoring shot created by a team in a match and assigns an “Expected Points” number to each of them. This number is calculated by looking at every player that has taken the same type of shot (set-shot or open-play) from anywhere on the ground within 5 meters of that shot since 2012. The “Expected Points” value is essentially an average of these conversion rates multiplied by 5 and plus 1 for the behind.

FFSS sums the “Expected Points” values for each team over the whole match and determines whether each team converted better, worse or just as well as we would have expected them to convert in the long term. If FFSS sees that a winning team has scored many more points then we would usually expect of them given the chances they created and their opponents have scored fewer, then we consider them to have gotten slightly “lucky” in that match and not increase their rating by as much as would have otherwise.

Aside from these two major changes, the FFSS Rating System (V.1) retains distance traveled/ground experience definition of home-ground-advantage, starting rating reversion to the mean and the lower multiplier for late-season unimportant games that SimpElo has always had.

Perhaps most important to note is that FFSS is essentially dumb to player lists, considering a team as an entity of their own. This usually does not make a big difference as ratings update quickly to account for any changes in team quality, but this year, with Essendon essentially a new side, we may see a few weird predictions in the first half of the season.


What will the FFSS ratings be used for?


 

Knowing an accurate rating of each team immediately allows us to calculate the implied probabilities of each and every match before it takes place. Throughout the 2016 season I plan to post FFSS derived tips and predictions weekly as well as brief discussions about why the implied probabilities of matches may differ from the general held beliefs of the public. This will help me recognise shortcomings in the system as well as perhaps find areas where it is, in fact, a better tipper than the everyday fan.

I may also post post-match analyses of interesting games and will of course continue to update the system as the year rolls on. To keep up-to-date with everything, follow me on Twitter.

2 Comments

  1. Phil Ter
    Phil Ter

    Any way we can get the formulas for this? Or is this a “secret” not to get out?

    April 4, 2016
    |Reply
    • Keep your eyes peeled for more in depth discussion on the model throughout the year as I improve bits and pieces. As for files/formulas, I won’t post them explicitly but if you put the effort in you can probably create something similar (or better) based on my notes.

      April 4, 2016
      |Reply

Leave a Reply to Phil Ter Cancel reply

Your email address will not be published. Required fields are marked *