This post stems from two reasons.
- There have been a few calls over the months for somebody to investigate just how good bookmakers odds are at predicting FPL performance. Lots of people are using this data, but with little to back it up.
- My past analysis found that the Team Planner tool achieved an average accuracy of 95.8% in forecasting bookmaker odds across 5 GWs. The value of this totally correlates to how well bookmakers odds work within FPL.
Fortunately I have access to data which allows me to delve deep into this topic.
Using Bookmaker Odds to Forecast Performance
The first step is to understand how I performed this analysis and also how the algorithm works.
Bookmaker odds can be used to forecast player goals, clean sheets and team goals- though margins need to be carefully removed. FPL Review uses this data for 1 GW ahead predictions and historic xG and match data to predict assists, cards and saves. A bespoke algorithm is used to determine expected bonus points.
The end result is a neat and tidy prediction of points per player for a 90 minute performance for every player in FPL. Historic predictions per player, along with xG data, can be found on the Player Database page.
Using the Season Review tool it is possible to visualise past forecasts from any past GW for any of the 6.1 million managers in FPL. These forecasts are adjusted by the minutes a player was fielded for as FPL Review does not predict a players game-time.
I decided to analyse a sample of 250 managers performance by the end of GW21. (Team Codes 750000-750249 if you want to check the data yourself). For each team the following key pieces of data were collected:
- Overall FPL Points
- Expected Points Based on Bookmaker Odds
- Expected Points Based on xG Performance Data
I also collected this data per player from the
Player Database page.
The plots below show the relationship between Expected Points Based on Bookmaker Odds and overall FPL points for the 250 randomly selected teams and 571 FPL players by the end of GW21.
It’s very clear from the plots above that bookmaker odds when correctly dealt with provide an extremely strong indicator of FPL performance. Across the 250 random team selection there was a correlation factor of 0.881. Among the 571 players that value factor 0.966.
Football itself and FPL points are quite influenced by luck, however it is possible to remove the noise effect caused by luck a bit by using xG data. All xG data in this analysis and on the site is sourced from understat.com and treated to adjust for quality by FPL Review.
Comparing the FPL points expected from bookmaker forecasts to the points expected from xG data yields extremely interesting results.
With the noise element in FPL points removed by using xG data it can clearly be seen how good the bookmakers are. This time for the 250 random team selection there was a correlation factor of 0.950. Among the 571 players that value factor 0.980.
It should be clear the power that bookmaker odds have in terms of FPL- however only if the data is correctly treated. It should be no surprise either, bookmakers are experts in predicting probabilities and have access to data that is beyond the reach of most individuals. By maximising your team forecasts based on bookmaker data you will achieve results.
In my previous post I found that the Team Planner tool achieved an average accuracy of 95.8% in forecasting bookmaker odds across 5 GWs. Based on how powerful using odds is and how accurate that tool has been shown to be, the Team Planner is most certainly one of the most potent tools in FPL- with more features to come.