6th November Bookings Model Update
October was an absolute nightmare month in terms of results, simply nothing went right. But instead of dwelling on the negativity, I’m going to be positive and optimistic that things will improve in November, so in this article, as always, I’ll run through the things which I think went wrong, and methods of improvement going forward.
Results
Our lowest win-rate on main lines since tracking started – the aim is to hit the 60% mark, and we’re around the 40% mark, which is just not good enough.
While I am happy with the volume and would like to maintain that, I definitely need that win-rate to improve.
Longshots were not so lucrative this month, but the 0.25u play was profitable, despite only 1/12 bets winning. I might look to push these bets a bit more going forward, seeing how we only need a small percentage of winners to generate a profit.
Our unders strategy just didn’t work at all – we landed one full ladder in the entire month, which is just not ideal.
I mentioned the start of an overs strategy last month, and I think that has gone well. Targeting specific leagues and referees is a strategy I’m commited to giving more time to.
Bad Variance or Bad Bets?
I last evaluated accuracy in September, and we were at the 103.1% mark.
Bet365 were at 108.1%.
Now, model accuracy is at 102%, bet365 at 105.2%.
I touched on the idea that bet365 lines had become a bit sharper, and that data seems to confirm it. While we are still operating at a higher accuracy than bet365, they have improved significantly, which might act as one explanation to explain the low win-rate this season.
I’ve gone back to just sharing predictions rather than bets, as it was for the first few months of the project, just until I get a bit more confidence in the predictions I’m putting out.
Then again, I started tracking bets in January, so the dataset is missing the early months of the European season – this period has been quite useful in a sense, and I will learn a lot of lessons from it.
It’s so strange how the overall process of selecting bets has changed very little this season, yet we’ve had two end-member months in September and October – really high, then really low. But that is just the challenge of betting, and I can only work on trying to improve the accuracy and betting selection process.
Adjusting Referee Weightings
I was noticing that a lot of the gold games were based on referee data, but our trusted refs didn’t work well for us this month.
So I’ve decided to be a bit more risk-averse, and lower the weighting for referees, so the model is now less aggressive – I imagine that will lead to a lower volume, but the hope is to miss those games which provide a false sense of good value.
A New Data Source
The data source for the model is fbref – Opta data, free to scrape (albeit within limits) – the perfect data source.
But in the last week or so, they have introduced a new Cloudfare system which has affected bot traffic. This has basically broken the model, which is obviously not ideal in the short-term.
But the long-term plan was always to move away from using any external websites anyway – I intended to use the data I’ve collected since January to base my predictions off. So, here’s the plan.
I do have a contingency plan in case data sources fail, but it requires a bit more manual labour and is more time-consuming – not ideal. However, we do have an upcoming international break, and I usually use these quieter periods to do some work on the models. So this weekend, I’m going to do manual inputs, and then work on using the previous dataset as the new source data, so everything is self-sufficient.
This is quite a big project, so it’s going to take some time. While I prioritise working on that, I’ll continue to share predictions, but keep bets private.
That’s all for now – work is going to continue on the project, I can only hope October is an anomaly, and things begin to pick up again soon. I view this as a long-term project, so the key thing is to take the lessons and utilise them to improve the model.
I appreciate those that have stuck with me over this rough month, but I understand those that haven’t. Regardless, we will keep moving forward!
