Why Raw Odds Don't Sum to 100%

This was one of those "aha" moments when I first started working with odds data. I'd convert all outcomes to implied probabilities, add them up, and get... 104%. Then 106%. Sometimes 110%.

That extra percentage is called the margin (or overround, vig, juice). Understanding it is crucial for anyone doing serious sports data analysis.

---

The Math Behind Margins

In a theoretical "fair" market, implied probabilities would sum to exactly 100%:

But real markets look like this: That 5% excess is the margin.

---

Why This Matters for ML Models

When building prediction models, you have two choices for using odds as features:

1. Use raw implied probabilities
Simple, but includes noise from margins that vary by market and source.

2. Normalize to remove the margin
``
True Probability = Raw Implied Prob / Sum of All Probs
`

At OddsFlow, we typically normalize when using odds as calibration targets, but keep raw values when tracking market movement (since margin changes themselves can be informative).

---

Margin Variations by Source

Different data sources have different typical margins:

Source Type

Typical Margin

Sharp markets (Pinnacle)	2-3%
Major operators	4-6%
Smaller operators	7-10%+

This variation is important for multi-source data aggregation. Lower-margin sources generally provide cleaner probability signals.

---

Using Margins as a Feature

Here's something we discovered: margin changes over time can be predictive. When margins tighten (move toward 100%), it often indicates increased market certainty. When they widen, there may be information asymmetry.

We track margin alongside raw odds in our preprocessing pipeline.

---

Practical Calculation

`python
def calculate_margin(decimal_odds: list) -> float:
implied_probs = [1/odds for odds in decimal_odds]
return sum(implied_probs) - 1

Example: 1X2 market

odds = [2.38, 3.17, 3.17] margin = calculate_margin(odds) # Returns 0.05 (5%) ``

---

📖 Related reading: Implied Probability Explained • Odds Movement Analysis

*OddsFlow provides AI-powered sports analysis for educational and informational purposes.*