The Poisson distribution is a mathematical model that predicts the probability of specific numbers of events occurring when you know the average rate at which those events happen. In football, it allows you to predict likely scorelines using just the expected goals (xG) for each team.
This might sound complex, but the principle is simple. If Team A has 1.8 xG and Team B has 1.0 xG, the Poisson distribution tells you the probability of every possible scoreline, from 0-0 to 5-3 and beyond.
How the Poisson Distribution Works
The Poisson formula calculates the probability of exactly k events occurring when the average is λ (lambda). In football terms:
- λ = expected goals (xG)
- k = actual goals in the match
The formula is: P(k) = (e^-λ * λ^k) / k!
But you don't need to calculate this yourself. Once you have xG for both teams, you can use online calculators or simple spreadsheets to generate probabilities.
Building a Poisson Model
Here's the practical approach:
Step 1: Get xG estimates Gather xG data for both teams. Use the average from their last 10-15 matches for current form data.
Step 2: Apply Poisson Use these xG figures as your lambda values for each team.
Step 3: Calculate probabilities Generate the probability grid showing likelihood of each scoreline.
Step 4: Derive market probabilities Sum the relevant cells to get overall probabilities for win/draw/loss, over/under goals, correct scores, etc.
Step 5: Compare to odds Check whether bookmaker odds offer value relative to your Poisson-derived probabilities.
Practical Example
Team A (home) has 1.5 xG. Team B (away) has 0.9 xG.
Using Poisson, you generate:
- 0-0: 8.2%
- 1-0: 12.3%
- 1-1: 7.4%
- 2-0: 9.2%
- 2-1: 6.8%
- And so on...
Sum the probabilities for Team A winning (1-0, 2-0, 2-1, 3-0, etc.) and you might get roughly 54%. If bookmakers are offering Team A at 1.9 (implying 53% win probability), this is fairly valued.
But if they offer 1.85 (implying 54%), the edge has disappeared. At 2.0 (50%), there's clear value on Team A.
Poisson's Strengths
Objective methodology: Uses actual underlying metrics (xG) rather than emotion or narrative.
Full probability distribution: Doesn't just predict match outcome, but generates probabilities for every scoreline.
Applicable across markets: Same model works for 1x2 betting, over/under, correct scores, etc.
Transparent: You can see exactly where value appears in specific scorelines.
Poisson's Limitations
Assumes independence: Poisson assumes each goal is independent. In reality, scoring one goal changes team mentality and tactics, potentially affecting further scoring.
Fixed lambda: Uses average xG across a period, not accounting for in-match tactical changes or specific matchups.
Rare events: Extreme scorelines (5-4, 4-5) are mathematically improbable even when xG suggests higher-scoring matches.
Ignores draws tendencies: Poisson doesn't account for specific teams being draw-prone or draw-resistant.
Set piece effects: xG models vary in how they handle set pieces. Poisson amplifies these variations.
Improving the Basic Model
Several adjustments improve basic Poisson accuracy:
Draw adjustment: Some teams are naturally draw-prone. Adjust by reducing win probability for both teams slightly and increasing draw probability.
Home advantage: Add 0.3-0.4 to home team xG and subtract 0.2 from away team xG to account for home advantage beyond basic quality.
Correlation correction: Goals aren't truly independent. If Team A scores, Team B is slightly more likely to score in response (teams become more attacking). Small adjustments can account for this.
Clustering adjustment: Goals tend to come in clusters (a team scores twice in five minutes). Poisson slightly underestimates this. Apply small adjustments to 0-0, 1-0, and 2-0 probabilities.
Poisson for Different Markets
Match Outcome (1x2)
Sum all scorelines where home wins, away wins, or draw to get overall market probabilities. Compare to odds.
Over/Under Goals
Sum probabilities for all scorelines with combined goals above and below your threshold (e.g., over 2.5 means 3+ total goals).
Correct Score
Poisson directly generates correct score probabilities. If 2-1 appears as 6.8% in your model but bookmakers price it at 5.0%, value exists.
Both Teams to Score
Sum probabilities for scorelines where both teams score at least one goal.
Asian Handicaps
Apply handicaps to the probability grid. A 0 handicap bet on home wins becomes calculating home team win probability plus half of all draws.
Building Your Own Poisson Calculator
A spreadsheet with basic formulas can create a functional Poisson calculator:
Use the POISSON function (or POISSON.DIST in newer Excel versions) with your xG values. Create a table showing all likely scorelines (0-0 to 5-5) with associated probabilities. Use this to generate win/draw/loss probabilities, over/under markets, etc.
Limitations and Reality Checks
Poisson models work best when:
- Both teams have stable xG averages (use 10+ match samples)
- The match isn't a complete mismatch (expected double-digit goals)
- Neither team has specific draw-heavy or draw-resistant patterns
Poisson models work worst when:
- One or both teams have high variance (inconsistent xG)
- Tactical changes are imminent (managerial change, major injury)
- Historical trends show team-specific patterns Poisson doesn't capture
Advanced Applications
Some bettors combine Poisson with other models:
Ensemble models: Use Poisson alongside logistic regression or other statistical approaches, averaging their predictions.
Adjustment factors: Calculate how much teams deviate from Poisson expectations, apply these adjustments to future predictions.
League-specific calibration: Poisson models vary by league. Calibrate for your league of choice.
- The Poisson distribution allows you to convert expected goals into specific scoreline probabilities.
- This provides a mathematical approach to valuing markets beyond just win/loss odds.
- The model has clear limitations, but combined with judgment, it provides a framework for identifying value.
- Start by learning the basic model.
- Run Poisson on several matches and compare predictions to results.
- Identify where the model under/overestimates.
- Then apply adjustments that improve accuracy over time.