The Binomial Model For Claim Counts

Why Use a Binomial Claim Count Model?

Some insured units can produce at most one claim during a defined period. For example, a covered device might either have a qualifying breakdown claim during the year or have none. If the portfolio contains a fixed number of such units, and each unit has the same claim probability independent of the others, a binomial model gives the distribution of the total number of claims.

The assumptions matter as much as the calculation. A binomial model is not simply a formula to apply whenever a claim count is observed: the number of opportunities must be fixed, each opportunity must have a binary outcome, and the opportunities must share a common claim probability and be independent. The model is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: Let $N$ be the number of claims from $n$ fixed opportunities. If each opportunity produces either one claim or no claim, and the outcomes are independent with the same claim probability $p$, then $N$ has a binomial distribution with parameters $n$ and $p$, written $N\sim\operatorname{Binomial}(n,p)$. Here $n$ is a nonnegative integer and $0\le p\le1$.

For example, the opportunities might be a fixed group of covered items over one specified year, under a hypothetical product for which each item can generate at most one claim in that year. The model counts how many items have a claim; it does not describe claim severity or the amount paid for a claim.

Count Probabilities and Moments

For $N\sim\operatorname{Binomial}(n,p)$, the possible counts are $0,1,\ldots,n$. The probability of exactly $k$ claims is

P(N=k)=\binom{n}{k}p^k(1-p)^{n-k},\qquad k=0,1,\ldots,n.

A particular set of $k$ opportunities has probability $p^k(1-p)^{n-k}$ of producing claims at exactly those $k$ opportunities and no others. There are $\binom{n}{k}$ ways to select which $k$ opportunities claim. The possible selected sets are mutually exclusive, so multiplying the probability for one set by the number of sets gives the formula.

Probabilities for ranges of counts are sums of the relevant exact probabilities. For instance, $P(N\le r)=\sum_{k=0}^{r}P(N=k)$. It is often simpler to use the complement when the requested event is “at least one”:

P(N\ge1)=1-P(N=0)=1-(1-p)^n.

Theorem: Under the binomial assumptions, the expected claim count and its variance are $$ E[N]=np,\qquad \operatorname{Var}(N)=np(1-p). $$ The standard deviation is $\sqrt{np(1-p)}$. These describe the model’s count distribution; they do not say that the realized count must equal its expected value.

The mean is also the expected number of claiming opportunities: each of the $n$ opportunities has probability $p$ of contributing one claim. The variance depends on both $p$ and $1-p$, and is not generally equal to the mean.

Why the Mean and Variance Have These Forms

Define $I_j=1$ if opportunity $j$ has a claim and $I_j=0$ otherwise, for $j=1,\ldots,n$. Each $I_j$ is a Bernoulli indicator with $P(I_j=1)=p$, and the total count is $N=\sum_{j=1}^n I_j$. Since $E[I_j]=p$, linearity of expectation gives

E[N]=E\left[\sum_{j=1}^n I_j\right] =\sum_{j=1}^n E[I_j] =np.

For a zero-one indicator, $I_j^2=I_j$. Thus

\operatorname{Var}(I_j)=E[I_j^2]-E[I_j]^2 =p-p^2=p(1-p).

Variance of a sum equals the sum of variances plus twice the sum of pairwise covariances. Independence makes every covariance between distinct indicators zero. Therefore

\operatorname{Var}(N) =\sum_{j=1}^n\operatorname{Var}(I_j) =np(1-p).

This justification shows why both the common probability and independence assumptions are relevant: if probabilities differ, the common-$p$ binomial formula does not describe the count; if indicators are dependent, the variance can include nonzero covariance terms.

Worked Example: Exact and Cumulative Probabilities

A hypothetical program covers five laboratory instruments for one year. Each instrument can have at most one qualifying claim during the year. Assume independently that each instrument has a $0.20$ probability of a claim. Find the expected count, variance, probability of exactly two claims, and probability of at most one claim.

Method and assumptions. The fixed number of instruments is $n=5$, each has a binary claim outcome, and the outcomes are independent with common $p=0.20$. Thus $N\sim\operatorname{Binomial}(5,0.20)$.

Moments. The mean is $np=5(0.20)=1.00$ claim. The variance is $np(1-p)=5(0.20)(0.80)=0.80$ claims squared.

Exactly two claims. There are $\binom{5}{2}=10$ ways to choose the two claiming instruments. Therefore

P(N=2)=\binom{5}{2}(0.20)^2(0.80)^3 =10(0.04)(0.512)=0.2048.

At most one claim. Add the zero-claim and one-claim probabilities:

P(N\le1)=P(N=0)+P(N=1) =(0.80)^5+\binom{5}{1}(0.20)(0.80)^4 =0.32768+0.40960=0.73728.

The results are $E[N]=1.00$, $\operatorname{Var}(N)=0.80$, $P(N=2)=0.2048$, and $P(N\le1)=0.73728$. As a check, the probabilities for counts $0$ through $5$ are $0.32768, 0.40960, 0.20480, 0.05120, 0.00640,$ and $0.00032$; they sum to $1.00000$. The common mistake avoided here is treating “at most one” as only “exactly one,” omitting the zero-claim case.

Using Complements and Checking the Model

The complement rule can make cumulative probabilities much shorter. For example, the chance of at least one claim is not the same quantity as the expected number of claims. The former is a probability bounded by one; the latter is an average count and can exceed one when there are many opportunities.

Before calculating, check the model’s scope. A fixed set of independent units with one possible claim outcome per unit can fit the binomial setup. The setup does not directly fit a situation where a unit can generate several claims, the number of opportunities is itself random, claim probabilities differ by opportunity, or outcomes are dependent. Nor does choosing an exposure denominator by itself establish the binomial assumptions: an exposure measure and a count model answer different questions.

Worked Example: At Least One Claim Across a Small Portfolio

A hypothetical warranty portfolio contains eight distinct components, each observed for the same one-year period. Each component can generate at most one covered claim during the period. Assume independent outcomes and a common claim probability of $0.10$. Calculate the probability of no claims, exactly one claim, and at least two claims. Also calculate the mean and variance.

Method and assumptions. There are $n=8$ fixed binary opportunities with common $p=0.10$, so $N\sim\operatorname{Binomial}(8,0.10)$.

No claims. All eight components must have no claim:

$$ P(N=0)=(1-0.10)^8=(0.90)^8=0.43046721. $$

Exactly one claim. Choose one of the eight components to claim, then require no claim from the other seven:

P(N=1)=\binom{8}{1}(0.10)(0.90)^7 =8(0.10)(0.4782969)=0.38263752.

At least two claims. Subtract the zero- and one-claim cases from one:

P(N\ge2)=1-0.43046721-0.38263752=0.18689527.

The mean is $8(0.10)=0.80$ claims, and the variance is $8(0.10)(0.90)=0.72$ claims squared. The three probabilities sum to $1$ when the “at least two” event is combined with the zero- and one-claim events. The method avoids calculating six separate probabilities for counts two through eight.

Worked Example: A Threshold Probability and a Model Trap

A hypothetical rental operator tracks ten machines over a fixed season. Each machine can have at most one qualifying claim in that season, and assume independent claim outcomes with probability $p=0.30$ per machine. Find the probability of at least three claims, as well as the mean and variance.

Method and assumptions. The fixed number of opportunities is $n=10$, each with a binary outcome and the same probability $0.30$. Thus $N\sim\operatorname{Binomial}(10,0.30)$. Use the complement $P(N\ge3)=1-P(N\le2)$.

The three probabilities to subtract are

\begin{aligned} P(N=0)&=(0.70)^{10}=0.0282475249,\\ P(N=1)&=\binom{10}{1}(0.30)(0.70)^9 =3(0.70)^9=0.1210608210,\\ P(N=2)&=\binom{10}{2}(0.30)^2(0.70)^8 =45(0.09)(0.05764801)=0.2334744405. \end{aligned}

Therefore

P(N\ge3)=1-(0.0282475249+0.1210608210+0.2334744405) =0.6172172136.

The expected count is $10(0.30)=3.00$, and the variance is $10(0.30)(0.70)=2.10$. As a check, the exact-three probability is $\binom{10}{3}(0.30)^3(0.70)^7=120(0.027)(0.0823543)=0.266827932$, which is included in the event of at least three claims.

Common mistake and correction. A tempting but incorrect calculation is to use $np=3.00$ as the probability of at least one claim. The value $3.00$ is an expected count, not a probability. The correct probability of at least one is $1-(0.70)^{10}=0.9717524751$, which lies between zero and one. The threshold probability above is different again: $P(N\ge3)=0.6172172136$.

Common Mistakes and Practical Interpretation

Using the formula without checking the opportunity structure. If a single policy can generate multiple claims, it is not a one-trial, zero-or-one outcome under this model. The event being counted must be defined precisely.
Assuming equal probabilities from a shared label. A group of policies or items does not automatically have one common $p$. Differences in coverage, period, or risk characteristics can make the probabilities unequal.
Ignoring dependence. A shared event can affect several insured units at once, so their claim indicators may not be independent. The binomial variance then need not apply.
Confusing a count, a mean, and a probability. $np$ is the expected count. $P(N\ge1)=1-(1-p)^n$ is the probability of any claim; in general these are not equal.
Using the wrong complement. “At least two” is $1-P(N=0)-P(N=1)$, not $1-P(N=0)$. State which counts are excluded before subtracting.

Key takeaway: A binomial claim count requires a fixed number of independent, binary opportunities with a common probability of claim. Once those assumptions are justified, exact probabilities follow from the binomial formula, while the mean and variance are $np$ and $np(1-p)$.

Check Your Understanding

For each hypothetical scenario, use the stated fixed-opportunity model and report the requested quantity. Probabilities are rounded to four decimal places.

Four independent covered units each have a $0.25$ probability of one claim and can have no more than one claim. Find the probability of exactly one claim. (Answer: $0.4219$)
Six independent opportunities each have a $0.10$ claim probability. Find the mean, variance, and probability of exactly two claims. (Answer: mean $0.6000$, variance $0.5400$, probability $0.0984$)
Eight independent units each have a $0.20$ probability of a claim. Find the probability of at most one claim. (Answer: $0.5033$)
Twelve independent devices each have a $0.05$ probability of a claim during a specified period. Find the probability of at least one claim, the mean, and the variance. (Answer: probability $0.4596$, mean $0.6000$, variance $0.5700$)
Nine independent opportunities each have a $0.15$ probability of a claim. Find the probability of exactly two claims, the mean, and the variance. (Answer: probability $0.2597$, mean $1.3500$, variance $1.1475$)

What You'll Learn