2024/25 Total Season Recap
This article is well overdue, but I’ve had my summer break, and it’s time to lock in for the start of the new season. I’ve been working away already on the cards code, aiming to improve it’s accuracy closer to 100%, while also making it more time-efficient to run. With those things fixed, my focus has shifted to the current set of data that has been collected since the start of the year, analysing that a bit deeper to try and find anything that I might have missed, while also reinforcing the lessons from last season.
Given that I started collecting data in January, the dataset is incomplete, so that is something to bear in mind. However, I am incredibly excited to track the progress of the cards code for an entire season – I am confident that the implemented changes will have a beneficial effect over a large sample size.
Anyway, lets get into this write-up, starting off as always with the results.
24/25 Overall Results
Since January 2025, we’ve covered 1699 games across 21 different international and domestic league and cup competitions – the model has marginally over-estimated the total number of bookings by 4.3%, while it has been incredibly accurate on splits – 0.5% off on average per game. Accuracy-wise, those results are highly promising.
However, what matters more is how that translates to returns.
The overall results of the model are:
- Total return of 42.93 units
- 13.18% ROI
There has been a bit of a stagnation over the summer with the lack of fixtures and the challenge of navigating the Club World Cup, but again, the results have been good. I have made several changes over the summer, switching to trixies as opposed to ladders as one example (more on that later), which I look forward to implementing on the EU leagues, when volume is higher.
Ultimately, accuracy and return are the most important numbers for me to properly analyse the model, but there are plenty of other metrics which can provide valuable insight. Let’s go through those now.
As expected, unders dominate both in terms of volume and return. I mentioned this a bit in the previous article, but just based on those numbers alone, it seems wise to focus solely on betting unders for the upcoming season.
But again, we can break that down further. Over the course of the season, I’ve tried to identify which range of differences between the market and model yield the best win-rate, and we have an excellent sample of data now to base conclusions off.
The 10-14 range seems optimal, with a win-rate in excess of 70% off a sample size of around 100 games. This is what we would hope for, and expect. If the model is more accurate than bet365, then when the model suggests a much lower line than bet365, we would expect the under to land a lot more often than the market suggests.
It’s nice to see win-rate increase with model difference, as this is what we would expect to happen. For the higher differences however, results have been a bit variable – looking from a difference of 14 or above. Here, the win-rate is 59%, a slight dip, but the sample size is only 44 – there not enough games to accurately make any deductions. Hopefully that percentage will increase over the next few months as those cells populate with higher numbers.
Going forward, I’ll revert to highlighting the 10-14 range as standout value, in gold. Then, 8 to 10 will be in grey as decent value. I will likely also highlight games greater than 14, but probably in grey for now.
But overall, those numbers look good again. I’ll mention overs briefly, just because I have already decided to brush over those, but results are also again much more variable, without a standout pattern. It doesn’t help that overs are less common than unders, so the sample size is also much smaller.
Now, let’s look into which individual leagues have had the best performance.
Cup competitions have been the most lucrative, alongside Serie A and the Premier League. MLS and Championship predictions have also been decent. It’s too early to comment on the European international competitions, as we only started covering those in the knockouts. So going forward, we will focus a bit more on those leagues with a higher, green-highlighted percentage.
Up until now, totals have been getting along nicely, but handicaps have shown to be more accurate. In most cases, bet365 and the model tend to be quite accurate on the handicaps, but again there is a trend between a higher difference between the two, and win percentage. That implies the model is a better predictor than the market, which is ideal.
However, win-rate does hover around the 50% market for 5, 7 and 9 differences, which we would like to be higher. Going forward, when we get a handicap difference of 11 of higher (which are quite uncommon), these will be highlighted.
Longshots
Initially when I started sharing predictions, my longshot option would be ladders. However, after seeing their performance, I wasn’t fully convinced they were the most effective way of creating higher-odds bets. I shifted to using trixies, and I want to analyse their performance.
I’ve shared 8 Trixie bets, and the results are:
- 2.82 units profit
- 17.63% ROI.
That comes despite being yet to register fully winning trixie. For the next bit, a win will be denoted with a W, loss is L and push will be P.
Twice we’ve had WWP, with respective profits just short of 4 units. Four times we’ve had WWL, resulting in a marginal loss. Once we’ve had WPL, and the remaining time we’ve had WLL – so only once have we lost the full 2 units stakes from a 0.5u trixie.
Those results are very positive overall – we have come closer to the WWW trixie on several occasions, so a win seems overdue. Winning all 3 selections would be massive profit – it would be 3 doubles and a treble win. The WWP are 2 double wins and 2 single wins, so those returns are a lot less than they could be had that P been a W instead. Even still, a 17% ROI is very solid, and I would be happy to maintain that.
However, I would like to re-introduce ladders, and potentially doubles as a means of picking out some longer odds plays. For those, I’ll be looking deeper into specific games with exactly 0 cards shown, trying to pick out or identify any particular trends. I don’t think any one strategy is the best for longshots – I think it comes down to picking and choosing the right time to go for whatever seems best. I’ll be combing through as much data as physically possible before the season starts, and setup a criteria for trixie bets, ladders and doubles, and try to be a bit more flexible with selecting those.
As always, I appreciate the support on this project, and will persist with its’ development! As I mentioned at the start, results have been promising, but there is still lots of room for improvement, so work will continue.
I’m incredibly confident for the upcoming season, and have a few more articles in the pipeline until then.
