Earlier on in the season I posted some league points projections from different models/websites. Today let’s look at how each performed.
6 Gameweek projection League one & League two
https://syzygyanalytics.co.uk/2025/09/06/efl-league-one-2025-26-6-gameweeks-in/ https://syzygyanalytics.co.uk/2025/09/06/efl-league-two-2025-26-6-gameweeks-in/
This was a comparison between 2 prediction methods, my metric and the Spreadex points spreads.
Figure 1
Spreadex’s prediction error was much less. We’d expect this as 6 games is quite a small sample and so it’s difficult for a pure performance metric to match a market line such as spreadex at that point. The error difference is around 1.5 pts per team.
Premier League Projections after 4 weeks
https://syzygyanalytics.co.uk/2025/09/20/english-premier-league-2025-26-projections-after-4-weeks/
Figure 2
After just 4 games in the premier league my predictions returned a surprisingly impressive result.. Projecting Leeds high and Wolves/Burnley quite low helped my projections perform the best. There will be plenty of variance involved with looking at the results from just 1 league so let’s look at a larger sample.
8 League Comparison during the November international break
Figure 3 – Average for Spreadex leagues counts only the leagues where Spreadex also had spreads for every team. This is so spreadex can be fairly used in the comparison
The MIR (Market implied team ratings) and the spreads at Spreadex are the best. This isn’t so surprising as closing lines and betting markets have a lot of knowledge and wisdom baked into the odds/spreads. I am a little disappointed the prediction error of my ratings did not beat Opta’s predictions. La Liga really hurt me here – I had better predictions in 5 out of 7 leagues but the size of the La Liga prediction error offset this. Girona and Espanyol were the main culprits, Girona’s weak early performances and Espanyol’s strong ones created bad predictions. This does not mean I was often betting against Girona and for Espanyol in the first half of the season – there are other factors involved than simply what numbers my spreadsheet spits out.
In my November article I raised a couple of other points to review come the end of the season. The first one referred to combining my rating and the MIRs to create better predictions. After testing, this did not result in better predictions -> the higher the weight on the MIR the better.
Finally, I planned to see if the favourites had been overrated by the models (inspired by https://www.thetransferflow.com/p/outrights-longshot-bias)
Figure 4
There is no sign the models overvalued the favourites, in fact the favourites over-performed model expectation (this could be due to motivation levels for these teams remaining high until the end of the season). This may add weight to the hypothesis that Ted Knutson discussed in the piece above. I.e. the models are correct, and the bookies do undervalue favourites in long term markets.
Conclusion
Betting odds are accurate indicators and a performance-based metric alone can not keep up, at least for sample of 11 games or less (remember the betting odds can also consider injuries, previous seasons, player ratings e.t.c). Opta’s ratings are fairly weak but they use a unique method (sort of an elo system I believe) which will have it’s own niche (for example comparing teams from different leagues).
If you would like to be alerted to new articles just put your email into the white box on the home page syzygyanalytics.co.uk. Next I hope to preview the world cup, I’m just not sure yet in what capacity it will be.
Leave a Reply