For once I did my homework. At least that’s my assumption.
Lately I have been playing around with the Poisson distribution model to predict the 1X2, Over/Under, BTTS and correct score of the matches in the Jupiler Pro League.
I have been trying out 4 different Poisson models.
- Poisson distribution based on the goals averages of the home and away teams since 1998 in the Belgian Jupiler Pro League.
- Poisson distribution based on the goal averages of the home and away teams of the 2016/2017 season.
- Poisson distribution based on the goal averages on the H2H matches since 1998 of the home and away team.
- Poisson distribution based on the goal averages on the matches between the home and away team (so all H2H matches) since 1998.
I do know that Poisson is not the best prediction method based on statistics and that it would benefit from a Dixon Coles fit. But still working out how the Dixon/Coles adjustment works.
Without this fix, these would have been the result for the 2016/2017 season in Belgium.
For all 4 of the distribution I did calculate the prediction based on the % and the most probable score.
The first thing that I noticed was that there is a huge difference between the prediction of draws based on the % and most probable score.
The most probable score did predict between 53 and 154 draws, the % did only predict max 5 draws during the season. Still both Poissin model based on the H2H did deliver a ROI of above 28%.
I also do know that these models are based on the average goals at the end of the season, which will not happen in real life.
But we do have the make the assumption that the model gets better near the end of the season because then the goal averages are getting close to the ones at the end of the season.
After this I also did some investigation based on the value of the bets.
Only placing a bet if the odds were right.
The problem I did had here that I did had to use the odds based on the % in combination with the prediction of the most probable score. Which is not exact. But we do have to start somewhere.
This gives me the following results.
As you can see, I can minimize my number of bets using the value betting, while not really affecting my total profit.
Based on these schemes I did decide to try this out for the Play-off of the Belgian Jupiler Pro League.
On each match there are 4 possible bets. 1X2, correct score, Over/Under 2.5 and BTTS. The 1X2 and correct score are placed with taking value betting into consideration. I still have no fix for the draws problem based on the Poisson % distribution, and the profit and ROI are almost identical.
The BTTS and O/U bets are only placed when the odds are above the Poisson distribution.
After running my first test for the upcoming matches I’m doubting if the Poisson model can be used during the Belgian Jupiler Pro League playoffs. This due to the fact that the strength of the teams are almost equal.
Still I will give it a try an din the mean time making my model better.
I also include a Yankee bet on 4 correct scores, a accumulator of 4 on 1X2, BTTS and O/U.
For the first round this results in 21 bets with a total ante of 24,1 points. Some matches are not included due to a lack of H2H games or goal averages in the 16/17 season. This is because 3 teams from the second division are also included in the Play-off 2.
To be complete you can find a list of the bets I’m going to place for the next round.