Strategy
The Edge
The system attempts to find mispriced betting markets using a Poisson expected goals model. It does NOT try to predict match outcomes better than bookmakers — instead, it looks for markets where the bookmaker's implied probability diverges from the model's probability by more than 8%.
How It Works
1. Expected Goals (xG)
For each match, we estimate how many goals each team will score:
attack_strength = (team_goals_scored_avg / league_avg) * 0.5 + 0.5
defence_weakness = (opponent_goals_conceded_avg / league_avg) * 0.5 + 0.5
xG = league_avg * attack * defence * home_away_factor
- Uses rolling 10-game window (no lookahead)
- 50% regression toward league average (reduces extreme estimates)
- Home boost: +8%, Away reduction: -7%
2. Poisson Probabilities
From xG, we calculate probabilities for every market:
- Match Odds: P(home goals > away), P(equal), P(away > home)
- Over/Under: P(total goals > 2.5), etc.
- BTTS: P(both score at least 1)
3. Value Assessment
Compare our probability to the bookmaker's implied probability:
edge = my_probability - (1 / bookmaker_odds)
Only bet if edge >= 8% (above typical bookmaker margin of 5-8%).
4. Kelly Staking
kelly_fraction = (prob * (odds-1) - (1-prob)) / (odds-1)
stake = quarter_kelly * bankroll (capped at 3%)
Conservative: quarter Kelly prevents ruin during variance.
Markets
| Market | Status | Notes |
|---|---|---|
| Home Win | Active | Best performing in backtest (+115% ROI) |
| Away Win | Active | Model weaker here, but included |
| Over 2.5 Goals | Active | Decent edge when xG is high |
| Under 2.5 Goals | Active | Works when both teams defensive |
| Over 3.5 Goals | Active | High odds, high variance |
| BTTS Yes/No | Active | Liquid, reasonable model accuracy |
| Draw | Active | Hardest to predict, highest edge threshold |
| Corners | Excluded | Illiquid on Betfair |
| Shots | Excluded | Illiquid, no reliable data |
| Accumulators | Banned | Strategy is singles only |
Risk Management
| Rule | Value |
|---|---|
| Hard stop loss | -120/day |
| Defensive mode | -80/day (reduce stakes) |
| Max exposure | 800 across all active bets |
| Max stake | 3% of bankroll |
| Min edge | 8% |
| Odds range | 1.50 - 6.00 |
Backtest Results (28 bets, 2025 season, real bookmaker odds)
- ROI: +3.9%
- Win rate: 46%
- Average edge: 11.8%
- Best market: Home Win (+115% ROI on 6 bets)
- Caution: 28 bets is NOT statistically significant