This will be the 2nd blog about my betting performance in the season 2022/23 using my ML Poisson model. The beginning of the season was a sobering one. My theoraticaly worst model performed the best and the ML Poisson model, which I am using for betting showed a really bad profit. So let’s have a look, how things changed in the 2nd part of the first season half.
In the last post I already mentioned, that I do not expect my ZIP Poisson model to keep the performance of the beginning. And this already happened. The yield shrinked from +13.2% to +4.5%. Predictive models only based on goal data just do not provide the same predictive power as models, which are e.g. based on xG data.
This is also proved, when looking at the latest performance of my Vanilla Poisson and ML Poisson model. For both models there was a turning point in the performance. The yield of the ML Poisson model increased from -5% to -1.5%. For the Vanilla Poisson model the turn around was even better. The Yield turned positiv to +4.4%.
Unfortunately this turning point was just theoretical one. My real-world betting performance does not indicate such a turning point. The yield only slightly improved from -4.15% to -3.86%. But why is there now such a big difference between the theory and the real world?
At first, we should not forget the margin. All bets in the betting simulation are done against Bet365 fair odds. The odds at broker, where I am placing my bets, contain a margin. It’s always something around 1% – 1.2%. That definitely reduces the profit. But it does not explains the difference of more than 3%. So I have to work on beeing able to directly compare the theoretical bets and the placed bets. Otherwise I will not be able to get an explanation for this difference.
If you have further questions, feel free to leave a comment or contact me @Mo_Nbg.