Been building a football value betting model for 4 months -Just looking for honest feedback
I started this without big expectations. I wanted to understand if I could build something that finds real edge in football markets — not copy someone else's strategy, but actually understand from scratch what makes a bet valuable.
Four months later the system is functional and I use it daily. But I'm aware there's a lot I don't know yet, and that's why I'm here.
How it works:
The foundation is Poisson — I estimate expected goals for each team separately and combine them to get outcome probabilities. To remove the bookmaker margin I use the Shin method via numerical bisection. I went with Shin over basic multiplicative devig because it handles the favourite-longshot bias better — but honestly I'm not sure how much difference it makes in practice, and that's one of the things I'd like to hear opinions on. EV is always calculated on the post-devig probability. Kelly Criterion works as the final filter — negative Kelly, bet discarded, regardless of EV sign.
League selection:
I focused on less-followed leagues — Allsvenskan, Eliteserien, Ekstraklasa, Brasileirão, a few others. The logic is that markets with less sharp attention close with more noise and theoretically leave more edge available. I estimate somewhere between +3% and +7% in these leagues, but ROI tracking is still manual and I don't have enough data to confirm that rigorously.
Image:
The image is today's output. 20 bets with positive EV above the Kelly threshold. Sharing it as a concrete example of what the system produces — not as proof that it works.
What's still unresolved:
Over 1.5 and BTTS markets use a simpler Poisson layer that I'm still refining. CLV tracking isn't automated yet. Those are the next priorities.
What I want to know:
Does anyone here use Shin? Is it worth it over simpler methods or am I overcomplicating things for no real gain? And how do you validate a model like this before trusting the numbers — is there an approach you'd recommend?
Thanks to anyone who reads and replies.
P.S. — The interface looks like it's from 1995, I know. It started as a Telegram bot and I never bothered to make it prettier. I just want the picks.
