Regular readers will know I've been testing the Poisson distribution method based upon data from the J League. Originally I set out to see if the model could help me with pricing odds. I answered that pretty quickly and no it couldn't, but I was still curious to see if it held any merit as a key indicator of predictions based upon goals scored and XG as the base points. I've run it for four rounds now and whilst the model based on goals scored went well for a couple of weeks, it has nose dived off a cliff the last couple of rounds. So this potentially is the last round of the experiment. Like any betting strategy it has to be understood that it would take full commitment and moreover that it comes with potential losses. For new readers as a marker I've included a very simple model, that of merely betting the favourites in the market noting that the margin for bookmakers naturally sits on a 1X2 market with a home or away win and never the draw. As a reminder this isn't a tipping service, the model and data is purely shown here as a representation of a well known system peddled on the larger majority of videos you'd find on a major platform such as YouTube. As you'd imagine it takes a large degree of time to set up and to input the data after every game to ensure your model remains up to date. For new readers I'm interested in statistics, modelling, patterns and data in general so here I'm doing the hard work for anyone tempted to use the same system thinking it might give them an edge. For an absolute clear picture you'd need to run it over the course of an entire season and I'd suggest not only that but run it for a series of different leagues. It's your time so be my guest if you want to try it. If you do, I'll happily look at the data and analyse it and publish the findings. If you do stumble upon a league where it does work my advice would be to keep that to yourself. Good betting opportunities are always few and far between and if you ever did find a market with an edge, once money flows into it that edge would vanish pretty quickly. I think it's safe to also suggest that the edge you're looking for isn't going to come from this model but that being said I'm wrong about a lot of things.
What I can suggest on this round in particular is that the model used for the starting odds would be heavily weighted in favour of XG in comparison to goals scored because the XG predictions are exactly the same as those that the bookmaker (in this case Bet365) have down as the favourites in the opening markets. The model based on goals scored has just 2 of 10 rounds the same. Goals scored has it down for 7 of 10 rounds to be away wins. Of course it should be noted that the favourites and XG being the same prediction could merely be a coincidence. It probably should also be noted that last round they didn't marry and the XG model got 1 correct result out of 10. The Bookmakers had their best week with 6 priced favourites winning though as I noted in the blog post being right was probably not a good thing for them in that instance. The XG Model is performing worst of all so if GS can say just 3 of the 7 away wins right and 1 home right it should actually return a profit. However it picked 3 the last round out. Still this is an excel spreadsheet and I sound like I'm trying to talk myself into doing something stupid like betting on the ten rounds so I'll stop and just post the predictions.
For the full experiment breakdown of monies placed et al use the following link;
https://anegitsguidetofootballbetting.blogspot.com/2024/07/week-4-pre-game-j-league-modelling.html
Predictions are shown as the side favourite with the bookmaker - Goals scored model - XG Model
Kashima v FC Tokyo 1 - 2 - 1
Machida v Yokohama 1 - 2 - 1
Avispa v Tokyo 1 - 2 - 1
Cerezo v Albirex 1 - 2 - 1
Gamba v Shonan 1 - 2 - 1
Jubilo v Kyoto 1 - 2 - 1
Kashiwa v Kawasaki 1 - 1 - 1
Urawa v Hokkaido 1 - 2 - 1
Vissel v Nagoya 1 - 1 - 1
Sagan v Sanfrecce 2 - 1 - 2
Results will be updated after the weekend. As always if you're of a distrusting nature screenshot above and check after to see I've not altered what's shown above. I'll repeat I'm not a tipping service and I'd not encourage anyone to bet on the above and it's used only for modelling purposes and as a learning and testing tool.
20th July update
One game to go in the round. I could only laugh when I saw the results. I said I nearly talked myself into something stupid. I didn't, but it's a good measure of not betting something and how to deal with the results.
Favourites with the bookmakers
Game 1 @ 1.75
Goals scored model
Correct predictions; 2,3,4,5,6 and 8 @ 4.2, 3.6, 3.5, 4, 3.10, 5.5 so for now outlay £9 and returns of £23.90 (+£14.90).
XG Model
As Bookmakers results
No comments:
Post a Comment