Poisson Model: How to Price Over/Under Markets
Updated on 2026-07-27 · 930 words
What it is
The Poisson model is a probability distribution used to describe the number of events occurring in a fixed interval of time or space, assuming events happen independently at a constant average rate. In sports betting, it is most commonly applied to soccer to model the number of goals scored by each team in a match. By estimating an average goals per game (λ) for a team, you can calculate the probability of any specific goal count, which then allows you to price over/under markets—for example, over 2.5 goals.
Its popularity comes from mathematical simplicity: only one parameter (λ) is needed, yet it captures the essential randomness of low-scoring events. However, the model relies on strong assumptions—independence and constant rate—that real soccer matches violate. That does not make it useless, but it does make it a starting point, not a final answer.
The math
The Poisson probability mass function is written as:
P(X = k) = (λ^k × e^−λ) / k!
where λ is the average number of goals (the rate), k is the number of goals you want the probability for, and e is Euler's number (~2.71828).
Let’s work with a realistic λ of 2.5 goals per match—common in top European leagues. We compute P(0), P(1), and P(2) step by step:
- P(0): λ^0 = 1, 0! = 1, e^−λ = e^−2.5 ≈ 0.082085 → P(0) = 0.082085 = 8.21%
- P(1): λ^1 = 2.5, 1! = 1 → (2.5 × 0.082085) / 1 = 0.2052 = 20.52%
- P(2): λ^2 = 6.25, 2! = 2 → (6.25 × 0.082085) / 2 = 0.2565 = 25.65%
Summing these: 8.21% + 20.52% + 25.65% = 54.38%. That is the probability of 0, 1, or 2 goals. Therefore, the probability of over 2.5 goals (three or more) is:
P(over 2.5) = 1 − 0.5438 = 0.4562 = 45.62%
If the market were efficient and margin-free, the fair decimal odds would be 1 / 0.4562 ≈ 2.19, or American odds +119. In practice, bookmakers add a margin, so the actual offered odds will be lower (e.g., -110 or 1.91 decimal). The difference between the fair odds implied by your λ and the market odds determines whether you have a positive expected value.
Worked example
Suppose you have a match where you estimate both teams together produce an average of 2.8 goals (λ = 2.8). Compute:
e^−2.8 = 0.06081
- P(0) = 0.06081 = 6.08%
- P(1) = 0.06081 × 2.8 = 0.1703 = 17.03%
- P(2) = 0.06081 × 7.84 / 2 = 0.2384 = 23.84% Sum = 6.08 + 17.03 + 23.84 = 46.95%. So P(over 2.5) = 53.05%.
A bookmaker offers over 2.5 at odds of -105 (decimal 1.952). The implied probability of -105 is 1 / 1.952 = 51.22%. Compare to your 53.05%: your estimate is 1.83 percentage points higher. To check value, compute expected value per dollar wagered:
EV = (1.952 × 0.5305) − 1 = 1.0355 − 1 = +$0.0355 per dollar, or +3.55%.
This suggests a positive expectation, but only if your λ estimate is accurate and the Poisson assumptions hold for this match. The example shows how the model turns a parameter estimate into a actionable decision rule—but the next section explains why that rule can break.
When not to use it
-
Poisson assumes independence between goals, but soccer goals are not independent. A goal changes team behavior: the scoring team may defend deeper, the conceding team pushes forward, altering the rate. This serial correlation means that after a goal, the remaining time has a different rate, violating the constant-λ assumption. The model therefore underestimates the probability of low-scoring and high-scoring extremes compared to reality.
-
The constant rate assumption is rarely true across a full match. Teams tire, substitute players, and adjust tactics. The rate in the first 15 minutes is often higher than the last 15 minutes. Poisson treats every minute as identical, which flattens the timing of goals and misprices live markets and half-time bets.
-
λ itself is uncertain and often estimated from small samples. A team may have played only 10 matches in a season. Using those 10 games to estimate λ yields a wide confidence interval. A λ of 1.5 based on small data could be 1.3 or 1.7 in truth, and that error directly distorts all derived probabilities. This is known as parameter uncertainty and is ignored in the basic model.
-
Poisson systematically underestimates the frequency of high-scoring blowouts (fat tails). Real soccer data shows more 5–0, 6–1 results than Poisson predicts. These extreme outcomes come from factors like red cards, weather, or psychological collapse—none of which are captured. As a result, the model tends to underprice over 4.5 goals and similar markets.
-
The model works poorly for low-scoring leagues or cup matches with defensive play. When λ is below 1.5, the probability of zero goals becomes dominant, but Poisson still assigns probability to high counts that nearly never occur. In these contexts, alternative distributions (like zero-inflated Poisson or negative binomial) often fit better.
For any bet based on a Poisson estimate, treat it as a rough guide, not a precise prediction. Combine it with other factors—recent form, head-to-head, injuries, league context. The model is a tool for framing, not a crystal ball.
Use our match analyzer to compute probabilities and compare with market odds.
Gambling is entertainment with real financial risk. Must be 21+ in most US states. If you or someone you know has a gambling problem, call 1-800-GAMBLER. This content is for educational purposes only.
Frequently asked questions
- What is the main assumption of the Poisson model in sports betting?
- The model assumes that goals are scored independently over time at a constant average rate (λ). This means each moment of a match has the same probability of a goal, and past goals do not affect future goal rates—both of which are unrealistic in soccer.
- How do you estimate λ (average goals) for a team?
- The simplest method is to take the team's recent average goals per match over a representative sample—typically 10 to 30 games. More advanced approaches adjust for opponent strength, home advantage, and league average using regression models.
- Can the Poisson model be used for sports other than soccer?
- Yes, but with caution. It works for low-scoring sports like hockey or baseball where events are relatively rare and independent? In high-scoring sports like basketball, the distribution of points is better approximated by normal or other distributions because the event count is large.
- Why does Poisson underestimate the probability of high-scoring games?
- Because it assumes a fixed rate, while in reality a team that falls behind may abandon defense, and a dominant team may keep attacking. These behavioral changes create a 'fat tail' of extreme scores that the model does not capture.