14th – 19th February Bookings Model Results

Week number 4 – as promised, it’s time for a more comprehensive breakdown of the results so far.

If you’re reading this, make sure to get to the bottom. There is a invite waiting for you.

Results this Weekend

Available as a PDF – Bookings Model Tracking – Week 4

39/66 Winners (59.1% win-rate)
50/66 Unders (75.7% unders)

Very few overs predictions, and they performed poorly – 7/16 winners in total which is not a disaster by any means, but not ideal.

On the contrary, big-difference unders performed well, with 8/10 winners.

Once again, results did not go out way when the model prediction was close to the market line. There were 15 games within the -2 to +2 range (difference between model and market), and only 3 resulted in wins! 3/15 – 20% win-rate. I briefly mentioned how these games ought to be ignored, that looks increasingly likely after this gameweek.

From a macro-perspective, this was the models’ worst performance so far, underestimating the total number of cards by 6.7%, but a 59.1% win-rate by no means something to panic about. I was actually subtly hoping for a poor week so I could try and identify some glaring issues.

Lifetime Results (4 Gameweeks)

The model has been very aggressive on the unders so far, so it might come as a surprise that it has actually overestimated the total number of cards shown. However, it has been incredibly close, in fact, the model has been off by just 1.37%. 11580 Cards predicted, 11460 cards shown.

260 games tracked in total so far.

162 winners (62.3%)
184 unders (70.7%)

Breakdown by League

Bundesliga (17/36 winners, 47.2% win-rate)
Championship (14/19 winners, 73.4% win-rate)
Copa del Rey (2/3 winners)
Coppa Italia (1/1 winner)
FA Cup (11/16, 68.8% win-rate)
La Liga (20/40, 50% win-rate)
League Cup, Carabao SF (2/2 winners)
Ligue 1 (23/36 winners, 63.8% win-rate)
Premier League (20/31 winners, 64.5% win-rate)
Scottish Premiership (3/3 winners)
Serie A (26/40 winners, 65% win-rate)
Turkish Super Lig (22/33 winners, 66%) – with 365 not offering a market here, I’m going to stop tracking. I included initially to just get a larger data set.

Breakdown by Difference Range

Range	Wins	Total	WR
20 to 18	1	1	100.00%
18 to 16	1	1	100.00%
16 to 14	2	5	40.00%
14 to 12	5	7	71.43%
12 to 10	8	9	88.89%
10 to 8	16	20	80.00%
8 to 6	18	27	66.67%
6 to 4	29	43	67.44%
4 to 2	15	31	48.39%
2 to 0	23	40	57.50%
0 to -2	14	30	46.67%
-2 to -4	13	17	76.47%
-4 to -6	7	11	63.64%
-6 to -8	6	7	85.71%
-8 to -10	1	6	16.67%
-10 to -12	2	4	50.00%
-12 to -14	1	1	100.00%

Discussion

Starting with the league breakdown, it’s nice to see most leagues performing in and around the same number – mostly in the 55-65% WR zone. Some leagues underperformed, Bundesliga especially which I will pay much closer to attention to over the coming weeks. La Liga too.

Cup competitions and the Championship have done well for us so far, with Serie A and Premier League not too far behind.

Focusing on the graph for the breakdown by difference range, the shape is largely predictable – its a V-shape, with approximately a 50% win-rate where there is a small difference between the model prediction and market line.

Importantly though, the win-rate rises towards the edges, with a few anomalous points. It’s also worth noting that sample size gets very low at these points, meaning more testing time is required.

Invitation

The next step is to test sharing these to a group.

The idea I currently have, is to share predictions in the morning (or night before weekend games), for all proceeding domestic games. It will have the fixture, model prediction, market line and the difference between the two to identify where the highest value opportunity lies.

For games with a big difference, I will suggest a stake (using Kelly Criterion Calculator) for the main line only.

I’ve set up a Telegram Group to share this, you can join here. There is one game today (Tuesday 18th), which I have already shared. Feel free to join, the group will remain open while I test (at least until the March international break). One thing I ask for in return, if you have a suggestion or critique on how to improve the overall process, please don’t hesitate to message.

So far this project has been extremely well-received, and I’m grateful for your support!