The Church of Betting

Rating Models: Goal Ratings

Rating Models: Goal Ratings

So you already know that you can make a lot of money by finding an edge in the betting markets and that finding an edge in the betting markets is very hard. How does one go about it? There are several ways to approach this problem but in this article I talk about Rating Models and Goal Rating Models in particular. Rating Models are a favorite topic of mine and I got inspired to dive into it by the website of Joseph Buchdahl. Therefore I will start by linking you to his article on the matter: Rating Models. I also strongly encourage you to give his book “Fixed Odds Sports Betting” a try – rating model basics are discussed in more length there, next to staking plans and risk management. Definitely an interesting read.

The Basics of Rating Models

The concept of Rating Models is to calculate the probability of a certain team winning a game against another based on the two teams’ relative strength (also called Rating). The trick here is to find a good way to measure this strength that incorporates both the fundamental value of the teams and their recent form, as well as assessing correctly what influence the two factors have upon a team’s results.

In the article quoted above you can see that a simple goal form model appears to find an edge in the British football betting market. I have replicated the model using the goals scored for and against a team in its last 6 games, but this time I made the calculations not only for England, but also for Scotland, Italy, Spain, Germany and France. The data used comes from I have formed the equation from all seasons till 2014/15 and used season 2015/2016 to test the profitability of the strategy. The equation was calculated only for home teams, since we already know it has the highest explanatory power there. Here is what I found:


 with model (best odds)
with model (average odds)
all home wins (best odds)
all home wins (average odds)


y = 0.0120944830x + 0.4533594773
R² = 0.9747701268


 with model (best odds)
with model (average odds)
all home wins (best odds)
all home wins (average odds)


y = 0.0150124252x + 0.4321361372
R² = 0.9614325276


 with model (best odds)
with model (average odds)
all home wins (best odds)
all home wins (average odds)


y = 0.0114462462x + 0.4780770646
R² = 0.9271078320


 with model (best odds)
with model (average odds)
all home wins (best odds)
all home wins (average odds)


y = 0.0110592808x + 0.4709960837
R² = 0.9276883121


 with model (best odds)
with model (average odds)
all home wins (best odds)
all home wins (average odds)


y = 0.0116382314x + 0.4807301722
R² = 0.9412016660

Similarly to the original study I have found the strategy to return a premium in England, although it is still not profitable mainly due to poor performance of the home teams. In Scotland the strategy was profitable and had an edge in front of a blind home-backing strategy. However, in the other tested countries the strategy failed to deliver.

How good are Goal Rating Models?

Although the results are not as promising as one would hope for we must not discard the strategy as useless. The differing results in different countries might just indicate a different style of play. Moreover, in England we have a much higher number of games to work with, which can give us more confidence in the results compared to other leagues.

But mostly, we must keep into account that this is a very simplified system so it would probably be unwise to rely on it to deliver superior returns. There are a few obvious issues with it – it does not take into account against what teams the goals were scored and conceded. Another issue is that it does not differentiate between scoring and conceding home and away. Furthermore, the rating system does not weigh results based on how soon they occurred but just takes the average from the last 6 games.

There are more refined rating systems, which take all those factors into account, most famously the ELO-Ratings. Needless to say, it is also more technically challenging to generate such ratings but one could expect that their predictive power would be higher. I have since long wanted to generate an ELO-like model using the above data, however I still haven’t found the time to do so. If I do, I will surely let you know.

Until then, I am planning to continue with some historical odds analysis for my next article, which I believe will be interesting for every bettor. If interested, follow The Church’s Facebook and Twitter accounts to get updated about new content. Also, if you prefer some topics to be covered more extensively, let me know in the comments. See you around!

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.