Bernoulli Trials As Building Blocks For Claim Counts

Representing Claim Occurrence Risk by Risk

A claim count for a portfolio begins with a simple question about each risk: did it produce a qualifying claim during the period being studied? Representing the answer separately for each risk makes the meaning of the portfolio count explicit. It also helps separate the definition of a count from additional assumptions about whether different risks affect one another.

For example, a hypothetical insurer might study claims from a fixed group of vehicle policies over one calendar year. Before counting, it must specify what qualifies as a claim and which year is in view. The indicator for a policy then records whether at least one qualifying claim occurred during that year. It is zero or one, not the dollar amount of the claim and not, by itself, a model for how often claims occur.

Definition: For risk $j$, define the claim indicator $I_j$ over a specified period by setting $I_j=1$ if risk $j$ produces at least one qualifying claim during that period, and $I_j=0$ otherwise. For a fixed set of $m$ risks, the claim count of risks with at least one claim is $$ N=\sum_{j=1}^{m} I_j. $$

The number of risks $m$ is fixed in this representation, while the indicators can vary from one possible outcome to another. A realized vector such as $(0,1,0,1)$ says that the second and fourth risks produced claims; its sum, $2$, is the realized count. The order of the entries matters for identifying which risks claimed, even though only their sum is needed for the total.

A claim indicator describes occurrence during a period, not necessarily claim multiplicity. If a risk can generate two or more claims in that period, its indicator is still only one when the question is whether that risk had any claim. Counting individual claims instead requires recording the number of claims for that risk, not just its zero-one indicator.

From Individual Indicators to a Portfolio Count

Each indicator is a random variable when claim outcomes are uncertain. Write $p_j=P(I_j=1)$ for the probability that risk $j$ produces at least one qualifying claim. Since $I_j$ takes only the values zero and one, its expected value is $p_j$: the value one occurs with probability $p_j$, and the value zero with probability $1-p_j$.

Adding the indicators gives the number of risks with claims. Taking expectations gives a useful result for the average count. This result does not require the indicators to be independent: it follows from the sum representation and linearity of expectation.

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

This calculation gives an expected count, not the probability of a particular count. In general, the probabilities $p_1,\ldots,p_m$ alone do not tell us how likely it is that none, one, or several risks claim. To determine those probabilities, we also need information about how the indicators behave together. A common event affecting several risks, for instance, can make claims cluster even if each risk's individual probability is known.

Specify the scope.
Identify the fixed risks, the time period, and what qualifies as a claim.

Define each indicator.
Set $I_j=1$ for a risk with at least one qualifying claim and $I_j=0$ otherwise.

Form the count.
Add the indicators to obtain the number of risks with at least one qualifying claim.

Use only the information available.
Individual probabilities determine the expected count; a full count distribution may require information about joint outcomes as well.

Worked Examples

Worked Example: Reading an Indicator Vector

A hypothetical insurer monitors five separate property policies over one calendar year. A qualifying claim means at least one claim reported under the policy during that year. The claim probabilities for the five policies, in order, are $0.10$, $0.20$, $0.05$, $0.15$, and $0.30$. For one realized year, the indicator vector is $(0,1,0,1,0)$. Find the realized count and the expected count. State whether independence is needed for either calculation.

Method. Add the realized indicators for the observed count. Add the individual claim probabilities for the expected count.

Realized count. The observed vector has two entries equal to one: $$ N=0+1+0+1+0=2. $$ Thus, two of the five policies produced at least one qualifying claim in that year.

Expected count. The probability for each policy is its indicator's expected value, so $$ E[N]=0.10+0.20+0.05+0.15+0.30=0.80. $$ The expected count is $0.80$ policies with at least one claim per year, for this fixed group and the stated model probabilities.

No independence assumption is needed to add the observed indicators or to find the expected count. The observed count is a direct sum, and the expected-count calculation uses linearity of expectation. Common mistake avoided: interpreting $0.80$ as a possible realized count or as the probability of at least one claim. It is an average count, not either of those quantities.

Check. The realized count, $2$, is an integer between zero and five. The expected count, $0.80$, is between zero and five, as it must be.

Worked Example: Why Individual Probabilities Do Not Give the Count Distribution

Three hypothetical facilities are monitored for one quarter. Their joint claim-indicator outcomes have the following probabilities: $(0,0,0)$ has probability $0.50$; each of $(1,0,0)$, $(0,1,0)$, and $(0,0,1)$ has probability $0.10$; each of $(1,1,0)$, $(1,0,1)$, and $(0,1,1)$ has probability $0.05$. The outcome $(1,1,1)$ has probability zero. Find each facility's marginal claim probability, the distribution of the count $N$, and $E[N]$.

Method. For each facility, add the probabilities of all listed outcomes in which its indicator is one. For the count distribution, group outcomes by the number of ones.

Marginal probabilities. Facility 1 has a claim in $(1,0,0)$, $(1,1,0)$, and $(1,0,1)$, so $$ P(I_1=1)=0.10+0.05+0.05=0.20. $$ Facility 2 has a claim in $(0,1,0)$, $(1,1,0)$, and $(0,1,1)$, giving $P(I_2=1)=0.10+0.05+0.05=0.20$. Facility 3 likewise has $P(I_3=1)=0.10+0.05+0.05=0.20$.

Count distribution. There is one outcome with zero claims, three outcomes with one claim, and three outcomes with two claims. Therefore, $$ P(N=0)=0.50,\qquad P(N=1)=3(0.10)=0.30,\qquad P(N=2)=3(0.05)=0.15. $$ There is no outcome with three claims, so $P(N=3)=0$. These probabilities sum to $0.50+0.30+0.15+0=0.95$, not one. The stated outcome probabilities are incomplete unless the missing $0.05$ is assigned to some outcome. As written, they cannot define a valid joint probability model.

This check matters: calculations from inconsistent probabilities should not be presented as a valid count distribution. To make the example a valid model, change the probability of $(0,0,0)$ from $0.50$ to $0.55$, leaving all other probabilities unchanged. Now the probabilities sum to one, the marginal probabilities are $0.20$, $0.20$, and $0.20$, and $$ P(N=0)=0.55,\qquad P(N=1)=0.30,\qquad P(N=2)=0.15,\qquad P(N=3)=0. $$ Thus $E[N]=0(0.55)+1(0.30)+2(0.15)+3(0)=0.60$, agreeing with $0.20+0.20+0.20=0.60$.

The joint probabilities show how the risks can claim together; the marginal probabilities alone do not provide this information. Common mistake avoided: proceeding with an outcome table before checking that all its probabilities sum to one. Check: the corrected count probabilities sum to $0.55+0.30+0.15+0=1$, and the expected count agrees by both calculations.

Worked Example: Indicators Versus Individual Claim Counts

Three hypothetical delivery vans are observed over one year. Let $Z_j$ be the number of claims from van $j$, and let $I_j=1$ if $Z_j$ is at least one and $I_j=0$ otherwise. For van 1, the probabilities of $Z_1=0,1,2$ are $0.80,0.15,0.05$. For van 2, the probabilities of $Z_2=0,1,2$ are $0.70,0.25,0.05$. For van 3, the probabilities of $Z_3=0,1$ are $0.90,0.10$. In one realized year the claim counts are $(2,0,1)$. Find the realized number of vans with claims, the realized number of individual claims, and the expected value of each measure.

Method. Convert each claim count to an indicator to count affected vans. Keep the original claim counts to count individual claims. For expectations, first find each van's probability of at least one claim; also calculate its expected number of claims.

Realized values. The counts $(2,0,1)$ correspond to indicators $(1,0,1)$. Thus the number of vans with claims is $$ 1+0+1=2, $$ while the number of individual claims is $2+0+1=3$.

Expected number of vans with claims. For van 1, $P(I_1=1)=0.15+0.05=0.20$. For van 2, $P(I_2=1)=0.25+0.05=0.30$. For van 3, $P(I_3=1)=0.10$. Therefore, $$ E[I_1+I_2+I_3]=0.20+0.30+0.10=0.60. $$

Expected number of individual claims. Using each van's stated claim-count probabilities, $$ E[Z_1]=0(0.80)+1(0.15)+2(0.05)=0.25, $$ $$ E[Z_2]=0(0.70)+1(0.25)+2(0.05)=0.35,\qquad E[Z_3]=0(0.90)+1(0.10)=0.10. $$ The expected total number of individual claims is $0.25+0.35+0.10=0.70$.

The expected number of vans with claims, $0.60$, differs from the expected number of individual claims, $0.70$, because a van can produce two claims. Common mistake avoided: using a zero-one indicator as though it preserved multiplicity. Check: the realized number of individual claims, $3$, is at least the number of vans with claims, $2$; the corresponding expected values have the same ordering.

Assumptions and Common Traps

A claim indicator is a representation, not a complete frequency model. It is useful only when its scope is clear. The risk set must be fixed for the count $N=\sum I_j$, the observation period must be specified, and “claim” must have a consistent definition. If a policy can have multiple claims and the question concerns individual claim totals, indicators alone discard information.

The probabilities $p_j$ may differ across risks. There is no need to give every policy the same probability in order to define indicators or add them. Nor is independence required to obtain the mean count. However, a calculation of the probability of a particular total requires enough information about the joint outcomes; treating risks as independent without justification can produce an incorrect count distribution.

Mixing up a risk and a claim. An indicator counts whether a risk had any claim; it does not count how many claims that risk made.
Leaving the period unstated. “Had a claim” is incomplete unless the observation period and claim definition are clear.
Assuming equal probabilities. Each risk can have its own claim probability; the indicator construction still works.
Assuming independence to add expectations. Independence is not needed for the expected count, but joint behavior matters for the full count distribution.
Confusing an expected count with a probability. A mean such as $0.60$ is not automatically the probability of at least one claim.

Key takeaway: Define one zero-one indicator for each risk and period, then add the indicators to count risks with at least one claim. The expected count is the sum of their individual claim probabilities, whether or not the indicators are independent. A distribution for the total requires appropriate joint-outcome information.

Check Your Understanding

For each problem, keep the observation period and the meaning of the count explicit.

A hypothetical pool contains four risks with claim probabilities $0.05$, $0.10$, $0.20$, and $0.25$ for the coming year. The realized indicator vector is $(1,0,1,0)$. Find the realized count and expected count. (Answer: realized count $2$; expected count $0.60$.)
Three independent-or-dependent-as-stated risks have realized indicators $(0,1,1)$ in a quarter. What is the count of risks with claims, and does this realized vector by itself establish independence? (Answer: count $2$; no, one realized vector does not establish independence.)
A hypothetical repair service has two risks. Their joint indicator outcomes are $(0,0)$ with probability $0.60$, $(1,0)$ with probability $0.15$, $(0,1)$ with probability $0.15$, and $(1,1)$ with probability $0.10$. Find both marginal probabilities and the expected count. (Answer: marginal probabilities $0.25$ and $0.25$; expected count $0.50$.)
For one risk, the number of annual claims $Z$ is zero with probability $0.70$, one with probability $0.20$, and two with probability $0.10$. Find the probability its indicator equals one and the expected number of claims. (Answer: $P(I=1)=0.30$; $E[Z]=0.40$.)
For three fixed risks, the individual claim probabilities are $0.10$, $0.20$, and $0.30$, but their joint behavior is not specified. What expected count can be found, and can the probability of exactly one claim be determined from the information given? (Answer: expected count $0.60$; the probability of exactly one claim cannot be determined without joint-outcome information.)

What You'll Learn