How much do past seasons affect current Elo ratings?

Teams that are good one year tend to be good the next year. For this reason, ratings systems like Elo don’t reset at the end of each year. In my Elo system, teams regress only a little bit (10%) towards the league-average rating between seasons; this means that good teams in 2016 were still rated above-average before a game had been played in 2017.

Calculating ratings this way gives better results – the system predicts the results of games more accurately than it would if you wiped the slate clean at the start of the year. But carrying teams’ ratings over from the previous year means that teams’ ratings aren’t just a measure of their performance in the current season. This has led several people to ask me how much teams’ performance last season is still affecting their current rating.

To measure that, I’ve calculated what each team’s Elo rating would be after Round 15 if we had reset all ratings at the start of the year, with every club reverting to the league average of 1500.

The biggest difference is Brisbane. Even after beating Essendon at Docklands, the Lions’ actual Elo rating is dismally low – 1280, among the worst in the club’s history. But if we had wiped the slate clean at the start of the year, the Lions would be rated 1367 – that’s a common or garden variety very bad team, not a historically awful one.

Essendon, as well, would be rated higher if we’d reset the ratings at the start of the year. They’re currently rated 1466, a little below the league average of 1500. If we’d ignored the past, they’d be rated 1520, a touch above average.

Sydney are another notable team here. The actual Elo rating currently thinks they’re the second best team in the league, more or less agreeing with other ratings systems that have them at or near the top. But part of the reason for their very high rating is that the Elo system remembers how good they were last year. If we’d reset the ratings in the off season, the Swans would be in a virtual dead heat with the Giants as the fourth best team in the league. It’s pretty notable that even if we completely disregard everything before 2017, the Swans still come out near the top of the ratings just based on the current season, despite having lost their first six games.

The Dogs are also a little flattered by the ratings’ memory of their past accomplishments. They’re currently rated 1525, just a smidge above the league average, but if we disregarded the past they’d be rated 1476, a little below average.

We don’t reset ratings at the end of the year because doing so would reduce the accuracy of the system’s predictions. Teams rarely turn from bad to good overnight, so ratings systems benefit from being a little sceptical about reacting to results. But if you’re just interested in how teams would rate in the Elo system based on their performance so far this year, without history getting in the way, now you have your answer.

Creator of The Arc