Following our popular article on how to build a betting model, here is a practical example of creating, applying and testing a betting model that predicts wins, losses and draws in the Premier League. Read on to find out.
In this previous article we explained how to build a betting model. Now take a step further and give you an example of how to do it.
For this example we use an approach similar to the Actuarial Control Cycle – a quantitative risk assessment employed by insurance companies. There are five main features in buidling a betting model:
- Defining the problem
- Building the solution
- Monitoring results
- External forces
Below we subdivide these features into additional steps, giving an example to explain each stage.
Please note our example model is basic and we enocurage you to use is as the building blocks behind the thought process of each stage.
Step 1: Specify the aim of your betting model
Our aim is to calculate the outcome of English Premier League games to see if we can predict results more accurately than the bookmaker.
Step 2: Select the metric
Given our aim is to calculate the outcome of EPL games, the metric we will look at is the probability of a home team win, away team win and a draw.
Step 3: Collect, group and modify data
We’ve taken the decision to only consider league games for data purposes and make no modifications.
The data collected would be this season’s scores and subsequent results.
Step 4: Choosing the form of your model
For our example of calculating the probability of a match outcome, we use a simple model that looks back at the past three games of each team.
The outcomes can be calculated using a simple ratio. Let’s say the home team won the past three matches, while the away team lost, won and drew one.
The ‘home win: draw: away win’ ratio would be 4:1:1 with the probability 4/6 =2/3 = 66.66% for a home win and 16.66% (1/6) for the other two outcomes.
This is a crude model but the intention here is to focus on the steps, not the actual model. Let’s call it the ‘3 ratio model’.
Step 5: Dealing with assumptions
Our ‘3 ratio model’ has a number of assumptions which would all need to be tested separately:
- The scale of the goal difference or the goals scored has no effect on the chance of winning
- There is no difference in outcomes between home and away (we know this is not true, as we’ve discussed in the article How to use standard deviation for betting.)
- There are no external factors that affect results – such as cup matches
- There have been no significant team changes since the last three games
Step 6: Build the sports betting model
Once we have colleced the data, we build the model in an Excel file.
Step 7: Test the model
We can back-test the ‘3 Ratio Model’ to Leicester’s 2014 Premier League games. Given they were promoted last season, we exclude the first three games. In testing we start to uncover issues:
In some cases there are no draws. For example when playing Hull away, Leicester had lost their previous three matches, while Hull had won one and lost two. Should we assume that the probability of a draw be zero in this case? Or should we make an adjustment? This means that we have to revisit steps 4 to 6.
If we had used only home matches for the home team and away matches for the away team, would the model results be significantly different? What if we use 2, 5 or 10 matches rather than 3? What if we included cup matches as well?
These results would need to be tested using different assumptions and see how sensitive our output is to each. The more the results vary, the more rigorous our testing should be (back to step 5).
Step 8: Monitor results
Let’s assume this model was accurate, it then needs to be maintained as time progresses. This would lead us back to the starting point of the model.
Now that you know how to build a betting model, go straight to the article Why your staking method is importantto find out the five most popular staking strategies used by professional bettros across the world.