Here's something i've been working on over the last week - how to predict Draws. Thinking about it, the matches that are likely to end in a Draw are:
Now we can't do anything to predict number 4 - matches like PSG draws against a relegation battler but we ought to be able to do something to predict the first 3.
For this post, i'm going to examine how betting on a draw involving 2 evenly matched teams turn out. To identify evenly matched teams, let's take the bookie odds as being a reliable indicator. If both the Home and Away odds to win are between 2 and 3, let's assume that these two teams are evenly matched.
Let's look at the Premier League first. In 354 matches (when i retrieved the data) there were 95 Draws (26.8%). If you had bet a unit on Draws for all 354 matches, your return would have been 370.26 units. A 16.26 unit profit (4.6%).
Filtering for evenly matched teams - there were 44 such matches with 15 ending in a Draw (34.1%). If you had bet a unit on a Draw for all 44 matches, your return would have been 49.58 units. A 5.58 unit profit (12.7%).
La Liga's notorious for the lack of Draws. In 348 matches (when i retrieved the data) there were 81 Draws (23.2%). If you had bet a unit on Draws for all 348 matches, your return would have been 333.91 units. A 14.09 unit loss (4.0%).
Filtering for evenly matched teams - there were 47 such matches with 16 ending in a Draw (34.0%). If you had bet a unit on a Draw for all 47 matches, your return would have been 52.3 units. A 5.3 unit profit (11.2%).
Serie A's an interesting one. In 380 matches, there were 83 Draws (21.8%). If you had bet a unit on Draws for all 380 matches, your return would have been 314.2 units. A 65.8 unit loss (17.3%).
Filtering for evenly matched teams - there were 45 such matches with 15 ending in a Draw (33.3%). If you had bet a unit on a Draw for all 45 matches, your return would have been 49.24 units. A 4.24 unit profit (9.4%).
When two evenly matched teams go head to head, they do tend to draw more often than the general population of matches. That being said, there remains a lot of improvement that needs to be done to better predict draws.
For one, the number of draws was only a small fraction (around 20%) of the total number of draws over the season. Meaning there were a lot of draws that didn't involved two closely matched teams.
For methods two and three, i thought it would be easier to look at forward looking data rather than historical data to assess team form and how they're setup.
