5th-10th February Bookings Model Results

---

This week was FA Cup weekend in England, with domestic football taking place elsewhere.

I took this opportunity to throw the model outside of it’s comfort zone, and apply it to cup games. There were also some midweek cup games cross the Copa del Rey in Spain, DFB Pokal in Germany, League Cup and Coppa Italia (cup games highlighted in blue below) which I decided to test. The more data, the better.

Results Breakdown

PDF version can be viewed here – Bookings Model Tracking – Week 3

• 46/71 winners (64.7% win-rate)
• 47/71 Unders (66.2%)

Very similar results to the previous two weeks – a win-rate hovering around the 60% mark (although this was the highest win-rate so far), with the majority exploiting the unders markets. 

This week, however, the model underestimated the total number of cards by 3.6%, which is actually a great sign. It’s still a very low number, but crucially the first two weeks of recording were slight over-estimates. 

The Cup Games

Interestingly, the model seemed to actually lean slightly more towards the over in the cup games – 8 / 21 suggestions were over, while 7 / 21 were marginal under shouts (a difference of less than 5 booking points between the prediction and market line). The market seems to correct well for these games, which is something to bear in mind. 

From tracking the lines manually, we got some incredibly low lines in David vs Goliath games – Leyton Orient vs Man City and Plymouth vs Liverpool were two examples, where a 2.5 cards line was offered. Overs was suggested in both of these games, and it landed, which could be an angle to exploit going  forward. The market seems to overcorrect for supremacy here, especially as teams rotate a lot and play the younger lads, which can often make games much closer than they perhaps should be. 

More Projects

Excited with the model performance, I found myself working on some old projects, the first of which is a corners model. As it stands, I’ve currently got a very basic program working, but would like to add a few additional layers of complexity, some predictive variables, which I believe can give the model a real edge over the market. Thinking of what to include has proven a bit tough though, and requires a bit of thought before any testing will commence. My initial ideas are to use pre-match supremacy, or potentially try to incorporate something like an expected shots prediction, or even something to do with the total number of dangerous attacks. Adding that risks over-complicating the model though, so any ideas are welcome. 

Also, some of you might remember the NBA code which I used at the start of the season, which provided some nice winners. One night even included an 8-0 sweep. The reason I stopped posting basketball bets is because it was quite labour-intensive to input the variables for each player, so I attempted to automate the code a bit more. However, I encountered problem after problem, testing tons of different options to no avail, which was unfortunate because I had tested the code for a majority of the previous season and was extremely confident in the predictions. After re-visiting the program though, I was able to find a solution not too long ago – so those could resume very soon, which is good news. 

Plans for next week

Circling back to the bookings code, it has been steady progress so far, and next week the plan is to include a more comprehensive breakdown of the results. That will hopefully include splitting up each league, checking how overs / unders predictions are doing, and aim to identify an optimal range to produce a higher percentage of winners. 

After 4 gameweeks, there should be around 250 data points to go off, which is a decent enough sample size. Work continues, but the signs are positive.