After 10 games played in the German Bundesliga, it is a good time to draw a small summary about the current stats of the betting history. If you follow my blog, you should know, I publish every pick at Pyckio. Until now I only publish the picks of the Poisson model. I am still investigating new prediction models.
The betting simulation for the first 10 matchdays in the Bundesliga provides a huge loss of 256,1 units with 113 bets played. This is the biggest loss during the past 5 years. Until matchday 3 everything looked normal. There was just a small loose. But after this, you can see a big loose with every matchday.
Update: I discovered a error in my simulations. The filter for the value was set wrong. With the correct filter the first 10 match days resulted in a loss of 201,1 units.
Looking at the pick history of Pyckio, you can determine a similar trend line. Last season ended with a positive profit of 43,3 units. After matchday 10 the statistic shows a profit of -19,9 units. The difference to the simulation is caused by two facts. During the simulation I use 10 units for each bet. At Pyckio I decided to use 5 units per bet, as 10 units is the maximum stake. But the main difference is the betting markets. My simulation uses the back and lay markets for the home win, draw and the away win. Whereas Pyckio unfortunately only offers the 1/x/2 market and some asian handicaps. So it is not always possible to make a bet for the corresponding lay market.
But what caused such a big loose in comparison to the past years? For this we could look at the results for the different betting markets. Betting on the away team (back away, lay home) by far caused the biggest loss. About 80% of the complete loss was produced by these types of bets.
This means my prediction model overestimated the away teams really often. Going even more into detail, shows, that nearly 50% of the away loss are caused by one team: 1 FC Cologne. Cologne plays one of the worst season in the history of the club. Last season they were able to reach the international competitions, but now they are at the bottom of the table. This fact causes problems with the variables of my predictive model. The attack & defence strength do not react fast enough to such a performance jump. I already mentioned this problem, which could be solved with a weight factor for matches further back in time. So Cologne is an example for a team, which heavily underperforms. There are also counterexamples. Augsburg is such a team, which really overperformed at the first matchdays compared to the last season. This causes also problems.
But I am not the only one, who got some problems with his prediction models. Joseph Buchdahl, who provides the Wisdom of Crowd prediction model, was also very unlucky during the start of the new season. His model was expected to return 4,5%. Currently it returns -4,3%. This shows, that the current season is not easily predictable.
What does this all mean for the rest of the season? I am currently pessimistic, that my model will be able to balance the current loss over the whole season. The loss is just to big. But I expect the over- and underperforming teams align more to their average performance.
If you have further questions, feel free to leave a comment or contact me @Mo_Nbg.