From a Poisson Model to an Exact-Count Probability
A claim-count model describes how many claims occur in a specified period or exposure. Once a Poisson model has been selected and its mean count is known, the next question is often precise: what is the probability of observing exactly \(k\) claims? The answer depends on both the mean and the particular count. It also depends on recognizing that a claim count can be zero or any positive whole number; a Poisson model does not impose a fixed upper bound.
In this tutorial, \(N\) denotes the number of claims in the specified exposure and \(\lambda\) denotes its expected count. The model is \(N\sim\operatorname{Poisson}(\lambda)\), where \(\lambda>0\). The mean is not necessarily an integer: an expected count of \(2.4\), for example, is compatible with an actual count that must be a whole number.
The formula gives the probability of one exact value. The quantity \(\lambda\) is the model’s mean number of claims for the specified exposure, while \(k\) is the count being queried. Do not substitute the mean for the observed count unless the question explicitly asks for that count. The factor \(e^{-\lambda}\) is part of every probability, and \(k!\) adjusts for the number of possible orderings represented by a count of \(k\).
The formula applies when the Poisson model is appropriate for the count being modeled. It does not turn a non-count quantity, such as claim size in dollars, into a Poisson variable. Nor does it make a non-integer or negative claim count possible. The Poisson model has no finite maximum count, but every realized count must still be a nonnegative integer.
Why the Formula Has This Form
One useful justification comes from the rare-event limit of the binomial model. Imagine \(n\) independent opportunities for a claim, each with probability \(\lambda/n\), so the expected number of claims is \(n(\lambda/n)=\lambda\). The binomial probability of exactly \(k\) claims is $$ P(N=k)=\binom{n}{k}\left(\frac{\lambda}{n}\right)^k\left(1-\frac{\lambda}{n}\right)^{n-k}. $$ For fixed \(k\) and \(\lambda\), take the limit as \(n\) grows. The combination of terms involving \(n\) can be written as $$ \binom{n}{k}\left(\frac{\lambda}{n}\right)^k =\frac{\lambda^k}{k!}\left(1\right)\left(1-\frac{1}{n}\right)\cdots\left(1-\frac{k-1}{n}\right), $$ which tends to \(\lambda^k/k!\). The remaining factor satisfies $$ \left(1-\frac{\lambda}{n}\right)^{n-k} =\left(1-\frac{\lambda}{n}\right)^n \left(1-\frac{\lambda}{n}\right)^{-k} \longrightarrow e^{-\lambda}\cdot 1. $$ Multiplying the limits gives \(e^{-\lambda}\lambda^k/k!\), the Poisson formula. This argument explains the formula as a limit for many small-probability opportunities; using a Poisson model does not require that every application be described explicitly as a binomial limit.
Worked Calculations
A reliable calculation begins by identifying the model parameter and the requested count separately. Then check that the requested count is in the support, evaluate the factorial and power, substitute into the formula, and round only the final probability. The examples below report probabilities to four decimal places.
Worked Example: No Claims or Two Claims
A hypothetical small equipment policy has a Poisson claim count \(N\) over one specified coverage period, with mean \(\lambda=2.4\). Find the probabilities of exactly zero claims and exactly two claims.
Method and model. Use \(N\sim\operatorname{Poisson}(2.4)\). Both requested counts, \(0\) and \(2\), are nonnegative integers, so both are in the support.
Exactly zero claims. Substitute \(k=0\). Since \(0!=1\) and \(\lambda^0=1\), $$ P(N=0)=\frac{e^{-2.4}(2.4)^0}{0!} =e^{-2.4}\cdot\frac{1}{1} \approx 0.0907. $$
Exactly two claims. Substitute \(k=2\), evaluate \(2!=2\) and \((2.4)^2=5.76\), and then calculate: $$ P(N=2)=\frac{e^{-2.4}(2.4)^2}{2!} =\frac{e^{-2.4}(5.76)}{2} =e^{-2.4}(2.88) \approx 0.2613. $$
Check and common mistake. The zero-count calculation uses \(0!=1\), not \(0\). For two claims, omitting the factorial would give \(e^{-2.4}(5.76)\), twice the correct probability. The correct answers, rounded to four decimal places, are \(0.0907\) and \(0.2613\).
Worked Example: A Count Equal to the Mean
A hypothetical roadside-assistance portfolio is modeled with a Poisson count \(N\) having mean \(\lambda=4\) events during a designated period. Find the probability of exactly four events.
Method. The mean is \(\lambda=4\), and the requested count is \(k=4\). A Poisson count can equal four, so apply the exact-count formula with \(4!=4\cdot3\cdot2\cdot1=24\).
First compute the power and factorial ratio: $$ \frac{\lambda^k}{k!}=\frac{4^4}{4!} =\frac{256}{24} =\frac{32}{3}. $$ Therefore, $$ P(N=4)=\frac{e^{-4}4^4}{4!} =e^{-4}\left(\frac{32}{3}\right) \approx 0.1954. $$
Check and common mistake. The result is a probability, between zero and one. A common error is to treat “the mean is four” as meaning the count is always four, and report probability one. The mean describes the distribution’s average across outcomes; it does not determine a particular outcome. The probability of exactly four is \(0.1954\), rounded to four decimal places.
Worked Example: Checking Whether a Requested Count Is Possible
A hypothetical claims administrator uses \(N\sim\operatorname{Poisson}(1.5)\) for the number of claims in one specified exposure. A report asks for the probability of exactly three claims, while a second calculation proposes using the same formula for exactly \(2.5\) claims. Evaluate the valid probability and assess the second request.
Exactly three claims. The count \(k=3\) is a nonnegative integer, so it is in the support. Compute \(3!=6\) and \((1.5)^3=3.375\): $$ P(N=3)=\frac{e^{-1.5}(1.5)^3}{3!} =e^{-1.5}\left(\frac{3.375}{6}\right) =e^{-1.5}(0.5625) \approx 0.1255. $$
Exactly \(2.5\) claims. A claim count cannot be \(2.5\); the support contains only integers. Thus \(P(N=2.5)=0\), and the Poisson probability formula is not applied with \(k=2.5\). The factorial expression in the stated count formula is defined for nonnegative integer \(k\); substituting a fractional count is not a valid way to obtain a claim-count probability.
Check and common mistake. The mean \(1.5\) is allowed to be fractional, but the outcome \(2.5\) is not. Confusing the mean with a possible realized count is the trap. The valid probability is \(0.1255\), rounded to four decimal places; the probability of exactly \(2.5\) claims is zero.
Support and Interpretation
The support is not merely a technical detail. It tells you which values can be substituted as \(k\), and it distinguishes the Poisson model from a fixed-trial count model with a maximum number of opportunities. A Poisson count can be \(0\), \(1\), \(2\), and so on, with no fixed upper endpoint. In the model, each such value has positive probability when \(\lambda>0\). A negative number, a fraction, or a value such as infinity is not a possible count.
| Proposed value of \(N\) | In the support? | Reason |
|---|---|---|
| 0 | Yes | Zero is a nonnegative integer, and \(0!=1\). |
| 6 | Yes | Every nonnegative integer is included; there is no fixed maximum. |
| -1 | No | A count cannot be negative. |
| 2.5 | No | A count must be an integer. |
A useful distinction is that “no fixed maximum” does not mean that arbitrarily large counts are equally likely or even likely in a particular application. It means only that the mathematical support does not stop at a finite count. For a specified \(\lambda\), the formula assigns a probability to each nonnegative integer, and those probabilities decrease for sufficiently large counts.
Common Errors to Avoid
- Using the mean as the count. The parameter \(\lambda\) need not be an outcome. Set \(k\) from the exact event asked about, even when it differs from \(\lambda\).
- Omitting \(k!\). The denominator is essential, including when \(k=0\), for which the convention is \(0!=1\).
- Applying the formula to a fractional or negative count. First verify that \(k\) is a nonnegative integer. A fractional mean is allowed; a fractional claim count is not.
- Rounding intermediate values. Keep the exponential and other calculations at full calculator precision, then round the final probability to the requested number of decimals.
- Confusing an exact probability with a range probability. \(P(N=k)\) concerns one count only. A request involving several possible counts requires combining the probabilities for those counts, rather than substituting a range into the exact-count formula.
Check Your Understanding
For each problem, first identify the Poisson mean and the requested count. Report numerical probabilities to four decimal places.
- A hypothetical bicycle-repair plan has \(N\sim\operatorname{Poisson}(0.8)\) repair claims in a specified period. Find the probability of no claims. (Answer: \(0.4493\))
- A hypothetical glass-coverage product has a Poisson claim count with mean \(2\). Find the probability of exactly one claim, showing which factorial is used. (Answer: \(0.2707\))
- A hypothetical parcel insurer models the number of damaged-parcel claims in a route as Poisson with mean \(3\). Find the probability of exactly three claims. (Answer: \(0.2240\))
- A hypothetical warranty portfolio has Poisson claim count \(N\) with mean \(1.2\). Find \(P(N=2)\), including the value of \(2!\). (Answer: \(0.2169\))
- A hypothetical equipment pool has Poisson claim count \(N\) with mean \(2.5\). Find the probability of exactly four claims, and state whether exactly \(4.5\) claims is in the support. (Answer: \(P(N=4)=0.1336\); \(4.5\) is not in the support.)