# Betting history: Bundesliga match days 1 – 10 / 2018

With beginning of the new season in the Bundesliga I started to use my new predictive model “Team Strength MLP“. This model is no longer a statistical model like the Poisson model from last season. It is trained neural network. After the first 10 matchdays it is time to check the current result of my betting history.

My Team Strength MLP is based on the same variables as the old Poisson model. So I am in a first step able to compare the performance of both models. In a second step I can introduce new variables and see whether and how the performance increases. The backtest of the Team Strength MLP showed a yield of 7.5% for the last 5 seasons. In comparison the Poisson model achieved a yield of 6.1% in the same period.

## Simulation betting history

The simulation of the first 10 matchdays looks really good. The achieved yield of 8.01% is slightly higher than the backtest yield. But of course, 10 matchdays and an overall of 55 bets mean nothing. The sample is just too small to draw any conclusions.

The profit graph shows a typical problem of betting longer odds. Even if you have and edge, the variance is high especially in small samples. At the beginning a small winning streak provides a fast growing bank. But the following short losing streak also reduced it very fast.

## Pyckio betting history

My Pyckio history looks a little bit different. The overall profit is just 12.9 units, which equates a yield of 4.7%.

The monthly overview and the profit line show a similar behaviour as the simulation. The season started with a short winning streak. This cause a yield of 52.8% in August. After that the profit regresses and turns into a lose.

But what causes the difference between the simulation and the Pyckio betting history This explanation is fairly simple. The simulation is based on Bet365 odds. Pyckio uses Pinnacle odds. So both approaches could select different bets as the odds differ. The match between Borussia Dortmund and Berlin the 27th October is one example. Pinnacle offered a odd of 1.49 for a home win, which equals a small value. This was not the case for the Bet365 odds in my database and so this match is not part of the simulation betting history. Another difference occurs as I use a different approach for selecting the bets. During the simulation multiple bets can be placed per match. If the home win and draw offers value, 2 bets are placed. At Pyckio something like this is not possible. That’s why I selected the corresponding Asian Handicap bet in such cases.

After the next 10 matchdays I will do the next analysis and take a look, how promoted teams a handled by my model. Until then I am sitting for my laptop and just follow the numbers.

If you have further questions, feel free to leave a comment or contact me @Mo_Nbg.