The Poisson distribution is the mathematical engine behind most serious football prediction models. Here's exactly how it works, why it outperforms gut-feel betting, and how to use it to find value in scoreline, BTTS, and over/under markets.
The Poisson distribution is a probability model that predicts how likely it is to observe a specific count of events (e.g. goals) given an average rate of occurrence. It assumes events happen independently β each goal attempt is unaffected by the previous one β and at a constant average rate.
The formula is:
P(k goals) = (Ξ»α΅ Γ eβ»Ξ») / k!
Where:
Football is a low-scoring sport where goals arrive infrequently and at roughly independent intervals β making it a natural fit for the Poisson model. Research consistently shows that the distribution of goals per team per match closely follows a Poisson pattern across all major leagues.
Say Arsenal have an expected goal rate (Ξ») of 1.80 for this match, and Chelsea have a rate of 1.10. We can calculate the probability of each team scoring exactly 0, 1, 2, 3+ goals:
| Goals (k) | Arsenal P(k) β Ξ»=1.80 | Chelsea P(k) β Ξ»=1.10 |
|---|---|---|
| 0 | 16.5% | 33.3% |
| 1 | 29.6% | 36.6% |
| 2 | 26.7% | 20.1% |
| 3 | 16.0% | 7.4% |
| 4 | 7.2% | 2.0% |
| 5+ | 4.0% | 0.6% |
To get the probability of a specific scoreline, multiply the two independent probabilities. For example:
Arsenal 2β1 Chelsea probability
P(Ars=2) Γ P(Che=1) = 26.7% Γ 36.6% = 9.8%
Implied fair odds: 1/0.098 β 10.2 (or +920 in American odds)
By repeating this for every scoreline combination (0-0, 0-1, 1-0, 1-1, ... up to 5-5 or beyond) you build a full probability matrix. From this matrix you can derive implied probabilities for any market.
Try it yourself
Free Poisson Calculator
Enter xG values for any match and instantly see scoreline probabilities, match outcome odds, and comparisons to bookmaker lines.
Open Poisson Calculator βOnce you have a scoreline probability matrix, aggregating into standard betting markets is straightforward:
Match outcome (1X2)
Sum all scorelines where home team score > away for Home Win; where scores equal for Draw; where away > home for Away Win.
Over 2.5 Goals
Sum all scorelines where total goals β₯ 3. In the Arsenal/Chelsea example: 2-1 (9.8%) + 1-2 (7.3%) + 3-0 (5.3%) + 2-2 (5.4%) + 3-1 (3.9%) β¦ until all scorelines with 3+ goals are summed.
Both Teams to Score (BTTS)
Sum all scorelines where both teams have scored at least 1 goal. Exclude all clean-sheet scorelines (0-0, 1-0, 2-0, etc.).
Asian Handicap
Sum scorelines where the favoured team wins by more than the handicap margin. E.g. for Arsenal -1.5 handicap: sum all scorelines where Arsenal win by 2 or more goals.
In the Arsenal (1.80) vs Chelsea (1.10) example, the Poisson model gives roughly: Arsenal win 52%, Draw 25%, Chelsea win 23%. BTTS: ~47%. Over 2.5 goals: ~52%.
The quality of your Poisson model depends entirely on the quality of your lambda inputs. Using raw goals-per-game introduces noise β a team on five consecutive wins may have been fortunate finishers against low-quality opposition. The best approach:
Use xG, not goals
Expected goals strips out luck and reflects the true quality of chances created and allowed. A team averaging 2.1 goals per game on 1.3 xG is likely to regress.
Apply home/away adjustments
Home advantage in football is real and measurable β home teams score roughly 40% more goals than away teams on average across European leagues. Scale xG accordingly.
Weight recent games more heavily
A team's form over the last 6β8 matches is more predictive than their full-season average. Use a rolling weighted average (e.g. most recent game weighted 3Γ, next 2Γ, earlier games 1Γ).
Adjust for opponent quality
Facing the league's best or worst defence significantly affects expected output. Divide the team's attacking xG rate by a defensive quality factor based on the opponent's average xGA allowed.
Account for missing players
Key absences β a first-choice striker, a creative midfielder, an impenetrable centre-back β can shift lambda by 15β30%. Build in contextual adjustments for significant injury news.
No model is perfect. Understanding where Poisson fails helps you use it more effectively:
β Independence assumption
Poisson assumes goals are independent events. In reality, a team that scores first may switch to a more defensive shape, reducing subsequent goal rates. Models that adjust for score-state are more accurate.
β Red cards and injuries
A sending-off dramatically changes goal probabilities. Basic Poisson models are static β they cannot dynamically adjust for in-game events, making them less suited for live betting.
β Overdispersion
High-scoring matches (4-3, 5-2) occur slightly more often in practice than pure Poisson predicts. The Dixon-Coles correction handles this by weighting low-scoring outcomes (0-0, 1-0, 0-1, 1-1) more accurately.
β Cup motivations & context
Squad rotation in domestic cups, dead-rubber league fixtures, and relegation or title pressure are not captured in xG data alone. Contextual weighting is needed for these matches.
Despite these limitations, Poisson-based models remain a reliable baseline. Academic research consistently shows that well-calibrated Poisson models produce closing-line-accurate probabilities across the Premier League, La Liga, Bundesliga, and Serie A β leagues with sufficient xG data.
The real use of a Poisson model is identifying when your probability estimate differs meaningfully from bookmaker odds. This is the basis of value betting.
Value bet identification example
Poisson model β Over 2.5 Goals: 57% implied probability
Bookmaker odds β 1.85 β implied probability = 54.1%
Edge = 57% β 54.1% = +2.9% β Positive expected value
If your model is calibrated, bets with positive edge are profitable long-term. Apply Kelly staking to size positions appropriately.
Not every edge leads to a bet. Bookmaker margins compress edge quickly, and sharp bookmakers close or limit winning accounts. Focus on markets and bookmakers where your model has the most information advantage β typically early prices before the market sharpens.
In 1997, statisticians Mark Dixon and Stuart Coles published a paper showing that the basic Poisson model slightly underestimates the frequency of very low-scoring draws (0-0) and underdog wins (1-0 home, 0-1 away). Their correction applies a correlation factor Ο (rho) to low-scoring scorelines only (0-0, 1-0, 0-1, 1-1), improving calibration in tight matches.
The practical impact of the Dixon-Coles correction is modest β a few percentage points on 0-0 probabilities β but it matters in markets where correct score or BTTS No has high bookmaker margins. Most commercial prediction APIs, including those powering football data providers, apply a version of this correction.
The Poisson distribution is a mathematical formula that models the probability of scoring a specific number of goals given an expected average rate (lambda). By calculating each team's goal distribution independently and multiplying them together, you can derive exact scoreline probabilities and aggregate them into any betting market.
Over large samples across top European leagues, Poisson-based models using xG inputs produce probabilities that closely match the closing lines of sharp bookmakers β widely considered the gold standard of prediction accuracy. Individual matches remain highly random, but the model's edge accumulates over many bets.
Lambda is the expected goals rate for a specific team in a specific match. Use each team's recent xG rate (weighted toward recent games), adjusted for home/away advantage, opponent defensive quality, and missing players. Avoid using raw goals per game β xG is more predictive.
Yes. Sum the probabilities of all scorelines where both teams scored (for BTTS), or all scorelines with 3+ total goals (for Over 2.5). These aggregated probabilities are your implied fair odds for those markets.
What is xG?
How expected goals are calculated and why they power Poisson models
Value Betting Explained
How to identify positive expected value using model vs bookmaker odds
How Football Predictions Work
The full pipeline from xG data to published predictions
Implied Probability
How to convert odds to probabilities and identify value
Poisson Calculator β
Free interactive tool β enter xG and see full scoreline probabilities
BTTS Calculator β
Model both-teams-to-score probability from individual goal rates
