Hypothetical Means Across Risks

Why Means Across Risks Matter

Two policyholders can have different expected claim outcomes because their risk characteristics differ. Even for one policyholder, actual outcomes can vary from period to period around that policyholder’s expected outcome. Credibility theory separates these ideas by representing a risk’s underlying characteristics with a risk parameter, then asking for the expected outcome conditional on that parameter.

The resulting conditional expectation is called the hypothetical mean. Considering it across a population of risks gives a distribution of expected outcomes: some risks may have lower means, and others higher means. That distribution describes persistent differences in expected outcomes among risks, not the random variation of realized outcomes within a risk.

Definition: Let $\Theta$ denote a risk parameter and let $X$ be the outcome of interest for a risk. The hypothetical mean for a risk with parameter value $\theta$ is $$ m(\theta)=E[X\mid \Theta=\theta]. $$ If a risk is selected from a population in which $\Theta$ varies, then $M=m(\Theta)$ is the random variable representing the hypothetical mean across risks. Its distribution is induced by the distribution of $\Theta$.

The outcome $X$ is random for a risk, even after its parameter is specified. The hypothetical mean $m(\theta)$, by contrast, is a number once $\theta$ is fixed. It becomes a random variable $M$ only when we consider selecting different risks whose parameter values vary.

From a Risk Parameter to a Distribution of Means

The first step is to identify what outcome is being modeled and what information the risk parameter represents. The hypothetical mean is always conditional on that parameter. It is not automatically the observed average for a particular risk, nor is it necessarily the same for every risk.

For a discrete risk parameter with possible values $\theta_1,\ldots,\theta_k$, first calculate each conditional mean $m(\theta_j)$. Then assign that mean the probability that the risk parameter equals $\theta_j$. If two parameter values produce the same mean, combine their probabilities when reporting the distribution of $M$.

P(M\leq t)=P\bigl(m(\Theta)\leq t\bigr).

This expression also defines the distribution when $\Theta$ is continuous. To find probabilities for $M$, identify which parameter values make $m(\Theta)$ fall in the requested range, then use the distribution of $\Theta$. If $m$ is increasing, for example, $M\leq t$ corresponds to $\Theta\leq m^{-1}(t)$, when that inverse exists. If $m$ is not one-to-one, all parameter values satisfying the inequality must be included.

A useful check is the law of iterated expectation. Provided the relevant expectations exist, the overall expected outcome equals the mean of the hypothetical means:

E[X]=E\bigl[E[X\mid\Theta]\bigr]=E[m(\Theta)]=E[M].

This result checks how the mean across risks relates to the unconditional mean of outcomes. It does not say that individual outcomes equal their hypothetical means; it says that averaging the conditional expectations over the population of risks recovers the overall expectation.

Worked Examples

Worked Example: Discrete Risk Types and Claim Counts

Method: Calculate the conditional expected claim count for each risk type, then use the risk-type probabilities to construct the distribution across risks. In a hypothetical portfolio, 60% of risks are Type A and 40% are Type B. For a one-year period, a Type A risk has claim count $N$ equal to 0 or 2, each with probability 0.5. A Type B risk has $N=0,1,3$ with probabilities 0.2, 0.3, and 0.5, respectively. Find the hypothetical mean for each type, the distribution of hypothetical means, and the overall expected claim count.

Type A conditional mean. Given Type A, multiply each possible count by its conditional probability and add: $$ m(A)=E[N\mid A]=0(0.5)+2(0.5)=0+1=1. $$

Type B conditional mean. $$ m(B)=E[N\mid B]=0(0.2)+1(0.3)+3(0.5)=0+0.3+1.5=1.8. $$

Distribution across risks. A randomly selected risk has hypothetical mean 1 with probability 0.60 and hypothetical mean 1.8 with probability 0.40. Thus $P(M=1)=0.60$ and $P(M=1.8)=0.40$.

Overall expected claim count and check. Weight the type-specific means by the proportions of risks: $$ E[M]=0.60(1)+0.40(1.8)=0.60+0.72=1.32. $$ The unconditional calculation gives the same result: the Type A contribution is $0.60(1)=0.60$, and the Type B contribution is $0.40(1.8)=0.72$, for a total of $1.32$ claims. The common mistake is to report only 1.32 as though it were each risk’s mean; it is the population average, while the two type-specific hypothetical means are 1 and 1.8.

Worked Example: A Continuous Distribution of Hypothetical Means

Method: Transform the risk-parameter distribution and use the transformation to find probabilities. Suppose, hypothetically, that a risk parameter $\Theta$ is uniformly distributed on the interval from 0.5 to 2.5. Conditional on $\Theta=\theta$, the annual claim count $N$ has a Poisson distribution with parameter $2\theta$. Find the distribution of the hypothetical mean $M$, its expected value, and the probability that it exceeds 3 claims.

Conditional mean. A Poisson random variable with parameter $\lambda$ has mean $\lambda$. Here the conditional parameter is $2\theta$, so $$ m(\theta)=E[N\mid\Theta=\theta]=2\theta, \qquad M=2\Theta. $$

Distribution of $M$. Since $0.5\leq\Theta\leq2.5$, multiplying the interval endpoints by 2 gives $1\leq M\leq5$. More precisely, for $1\leq t\leq5$, $$ P(M\leq t)=P(2\Theta\leq t) =P(\Theta\leq t/2) =\frac{t/2-0.5}{2.5-0.5} =\frac{t-1}{4}. $$ Therefore $M$ is uniformly distributed on $[1,5]$.

Expected hypothetical mean. A uniform distribution on $[1,5]$ has mean equal to the midpoint: $$ E[M]=\frac{1+5}{2}=3. $$ Equivalently, $E[\Theta]=(0.5+2.5)/2=1.5$, so $E[2\Theta]=2(1.5)=3$.

Probability of exceeding 3. Because $M$ is uniform on an interval of length $5-1=4$, the portion above 3 has length $5-3=2$. Thus $$ P(M>3)=\frac{2}{4}=0.5. $$ The check is consistent with 3 being the midpoint of the uniform interval. The common mistake is to confuse $M$, a risk’s conditional expected count, with the realized Poisson count $N$: $M$ ranges continuously from 1 to 5, whereas $N$ is an integer-valued random count.

Worked Example: Hypothetical Annual Loss Across Risk Classes

Method: Find the conditional expected loss for each risk class, then form the distribution using the class proportions. Consider a hypothetical product with annual claim count $N$, where claim counts conditional on risk class $R$ are Poisson. The class probabilities are $P(R=0.5)=0.20$, $P(R=1)=0.50$, and $P(R=2)=0.30$. For class $R=r$, the Poisson mean is $0.4r$. Each claim has a fixed payment of $5,000, so annual loss is $X=5{,}000N$. Find the hypothetical annual loss mean for each class and the mean across risks.

Conditional expected loss. For class $r$, $E[N\mid R=r]=0.4r$. Since each claim pays $5,000, $$ m(r)=E[X\mid R=r] =E[5{,}000N\mid R=r] =5{,}000(0.4r)=2{,}000r. $$ Thus the class-specific calculations are: $$ m(0.5)=2{,}000(0.5)=\$1{,}000,\qquad m(1)=2{,}000(1)=\$2{,}000,\qquad m(2)=2{,}000(2)=\$4{,}000. $$

Distribution across risks. A randomly selected risk has mean $1,000 with probability 0.20, $2,000 with probability 0.50, and $4,000 with probability 0.30. These are probabilities over risk classes, not probabilities of observing those exact annual losses.

Mean across risks and check. $$ E[M]=0.20(\$1{,}000)+0.50(\$2{,}000)+0.30(\$4{,}000) =\$200+\$1{,}000+\$1{,}200 =\$2{,}400. $$ The same result follows by first finding $E[R]=0.20(0.5)+0.50(1)+0.30(2)=0.10+0.50+0.60=1.20$, then $E[M]=2{,}000E[R]=\$2{,}400$. The common mistake is to use the Poisson frequency parameter $0.4r$ as the dollar mean; multiplying by the $5,000 payment per claim is necessary to obtain expected loss.

Interpretation, Weighting, and Exam Traps

The distribution of hypothetical means describes how expected outcomes differ across the population of risks being considered. It does not describe how often an individual risk realizes each outcome. For example, in the annual-loss example, the distribution assigns probabilities to risk-specific means of $1,000, $2,000, and $4,000. It does not say that an individual risk’s annual loss must be one of those amounts; the number of claims, and hence annual loss, remains random conditional on its class.

The population used to define the distribution must also be clear. If the probabilities are proportions of policies, $M$ describes the hypothetical mean for a randomly selected policy. If the analysis instead selects a random unit of exposure, risks with more exposure can receive more weight. Those sampling choices can produce different distributions of hypothetical means, even if the risk classes and their conditional means do not change. State what is being selected before interpreting the weights.

A common exam trap is to replace all conditional means with the unconditional mean. In the first example, the overall mean is 1.32 claims, but that does not make 1.32 the hypothetical mean for either Type A or Type B. Another trap is to average the type-specific means with equal weights when the risk proportions are not equal. Using 50% for each type in that example would give $0.5(1)+0.5(1.8)=1.4$, not the correct population mean of $0.6(1)+0.4(1.8)=1.32$.

Specify the outcome and parameter.
State what $X$ measures and what risk characteristic is represented by $\Theta$.

Calculate the conditional expectation.
Find $m(\theta)=E[X\mid\Theta=\theta]$ using the model for outcomes at fixed $\theta$.

Describe the population of risks.
Use the distribution of $\Theta$ and the chosen risk weights to derive the distribution of $M=m(\Theta)$.

Check the overall mean.
When expectations exist, verify that $E[M]=E[X]$ using the law of iterated expectation.

Key takeaway: A hypothetical mean is a risk-specific conditional expectation. Its distribution across risks is obtained by applying the conditional-mean function to the population distribution of risk parameters, using clearly stated weights. It represents persistent differences in expected outcomes, not the distribution of realized outcomes within a risk.

Check Your Understanding

For each problem, identify or calculate the distribution of hypothetical means across the stated population of risks.

A hypothetical population has Type A risks with probability 0.25 and Type B risks with probability 0.75. Their conditional means for annual claim count are 0.8 and 1.6, respectively. Give the distribution of $M$ and $E[M]$. (Answer: $P(M=0.8)=0.25$, $P(M=1.6)=0.75$, and $E[M]=1.4$ claims.)
Suppose $\Theta$ is uniform on $[2,6]$, and $m(\theta)=3\theta-1$. Find the distribution of $M$, its mean, and $P(M>11)$. (Answer: $M$ is uniform on $[5,17]$, $E[M]=11$, and $P(M>11)=0.5000$.)
In a hypothetical portfolio, 40% of risks have conditional Poisson claim count mean 0.5, and 60% have conditional Poisson claim count mean 1.5. Give the distribution of $M$ and its population mean. (Answer: $P(M=0.5)=0.40$, $P(M=1.5)=0.60$, and $E[M]=1.1$ claims.)
A randomly selected policy has a hypothetical annual loss mean of $1,000 with probability 0.20, $2,500 with probability 0.30, and $4,000 with probability 0.50. Find $P(M>\$2{,}000)$ and $E[M]$. (Answer: $P(M>\$2{,}000)=0.8000$ and $E[M]=\$2{,}950$.)
A risk parameter $\Theta$ takes values 1, 2, and 4 with probabilities 0.20, 0.50, and 0.30. Conditional on $\Theta=\theta$, the annual claim count is Poisson with mean $\theta$. Find $P(M\leq2)$ and $E[M]$. (Answer: $P(M\leq2)=0.7000$ and $E[M]=2.4$ claims.)

What You'll Learn