Why Over/Under Is My Favorite Market to Model
Among all the markets I've built prediction models for, Over/Under (totals) consistently produces the best results. Here's why: it's a cleaner prediction problem.
Instead of predicting *who* wins (three outcomes, heavily influenced by individual moments), you're predicting *how many goals* will be scored. This is more amenable to statistical analysis.
How O/U Markets Work
The market sets a line (usually 2.5 goals), and you predict whether the total will be over or under that number.
| Line | Total Goals | Over | Under |
|---|
| 2.5 | 0, 1, 2 | Loses | Wins |
| 2.5 | 3+ | Wins | Loses |
| 2.25 | 2 | Half win/Half lose | |
| 2.75 | 3 | Half win/Half lose |
The xG Connection
Expected Goals (xG) data transformed how we model totals. Instead of using actual goals scored (noisy, high variance), xG measures the quality of chances created.
Key insight: xG has much higher predictive power for future goals than actual past goals.
At OddsFlow, our totals model uses:
- Team xG per 90 minutes (home/away splits)
- Team xG against per 90 minutes
- xG trend over recent matches
- Head-to-head xG history
Feature Engineering for Totals
Beyond xG, we've found these features valuable:
Attack indicators:
- Shots per game
- Shot conversion rate
- Big chances created
Defense indicators:
- Shots faced per game
- Save percentage
- Big chances conceded
Context factors:
- Match importance
- Days since last match (fatigue)
- Weather (rain tends to reduce goals)
Why Models Outperform on Totals
Three reasons:
- 1Less randomness: Individual goals are random, but expected totals over 90 minutes are more stable
- 2Better data availability: xG data is widely available and standardized
- 3Market inefficiency: Recreational participants often have stronger opinions about winners than totals
Quick Reference Table
| O/U Line | Typical Scenarios |
|---|
| Under 1.5 | Defensive matchups, important low-stakes draws |
| 2.5 | Standard market, ~50% of matches go over |
| Over 2.5 | Attacking teams, weak defenses |
| Over 3.5 | High-scoring matchups, open play styles |
📖 Related reading: How AI Predicts Football • xG Analysis Techniques
*OddsFlow provides AI-powered sports analysis for educational and informational purposes.*