NFL betting demo
Every game here was predicted by a model trained only on earlier seasons, so this is what betting it would actually have looked like. Choose when to bet and how much, and watch the bankroll play out over 20 NFL seasons.
Final bankroll
–
–
Win rate
–
–
Return on stakes
–
–
Bankroll over time
- Bankroll
- Starting bankroll
Loading games…
Line chart of the bankroll after each bet placed by the chosen strategy, compared with the starting bankroll.
How often the model is right, by confidence
- Break-even
Bar chart of the model's accuracy on out-of-sample games, grouped by how far its probability is from 50%, against the break-even win rate.
Show as a table
| Model confidence | Games | Accuracy |
|---|
| Last 10 bets | Line | Pick | Result |
|---|
How it works
- The model: predicts the chance a game goes over the posted total from Elo team strength, recent scoring form, rest, travel and game-day weather.
- No peeking: the model and its settings were chosen on games before December 2003. Each later stretch of games was predicted by a model trained only on what came before it.
- Try: raise the edge to bet only the games the model is surest about, switch to ½ Kelly to see how sizing changes the risk, or move the start year to see how much the result depends on when you began.
Data: NFL games with closing Over/Under lines, joined with stadium weather. Pushes return the stake. Read the case study or the code.