As we have created a thoroughly detailed collection of Premier League data, which includes odds and results, we can look into recreating our work on the World Cup, and create a relationship to predict odds in the Premier League.
When the World Cup odds were created, we used the bookmakers' odds for each team to win the World Cup as a measure of strength. Past form would have been complicated to use when each team qualified from groups containing different teams, and with the level of the different confederations varying hugely. World rankings are misleading for similar reasons, as well as factors such as Switzerland using clever arrangement of fixtures to secure a high ranking, whilst the only competitive football Brazil played since the last World Cup is the 2011 Copa América and 2013 Confederations Cup (a total of nine matches).
For the Premier League we have been blessed with data going back ten seasons, complete with the odds for each match. We do not have, however, the odds prior to each match for each team to win the league (it would also be a poor choice, as later in the season the majority of teams will have no available odds for this eventuality).
Instead, we can just use past form, since we have the data available to mine. The best form record which predicts odds seems to be simply taking the average points accrued from the previous year (38 matches), which is pleasingly simple.
Finding equations
For each of our 3800 matches, we can look at each side's form, and divide it by their opponent's in order to get a quantitative measure of strength between the teams. Below is a graph showing the odds given versus the measure of strength for home teams in our database:
- Home team victory: 1.0905(x^-1.927)
- Away team victory: 2.5414(x^-1.915)
- Draw (if home team is favourite): 2.1418(x^0.9444)
- Draw (if home team is underdog): 2.167(x^-0.309)
Checking the predictions
Below is a graph comparing our predicted odds to the actual odds given:
This means that our method of modelling the Premier League seems to have merit, and it means we can continue to build a simulator with confidence.