Sitonce
Country: US
Show exams for United States Hong Kong
Sign in

Loss Distributions: SOA FAM

Updated 8 min read
Key takeaway

A loss distribution describes the random amount an insurer pays or retains, often after a deductible, limit or coinsurance.

  • For exponential ground-up loss with mean $500 and a $200 deductible, expected payment is $335.16, while the expected payment conditional on a positive claim is $500 because of memorylessness.
On this page13 sections
  1. Ground-up loss versus insurer payment
  2. Limits, coinsurance and loss elimination
  3. Frequency, severity and aggregate loss
  4. Risk measures: VaR and TVaR
  5. Original deductible example and checks
  6. Transform the loss before calculating payment
  7. Limits and coinsurance alter the payment function
  8. Frequency and severity create aggregate risk
  9. VaR and TVaR describe tail risk differently
  10. Check behavior at the boundaries
  11. Use survival functions to integrate excess loss
  12. Separate claim-level and portfolio quantities
  13. Calculate tail summaries from a defined loss basis

Ground-up loss versus insurer payment

Let X be ground-up loss. With an ordinary deductible d and no limit or coinsurance, insurer payment is Y=(X−d)+. That transformation creates a point mass at zero for losses at or below d and a positive payment above it. The payment distribution is not simply the ground-up distribution shifted down in every case because small losses are eliminated entirely.

For exponential X with mean θ=$500 and d=$200, P(Y>0)=P(X>200)=e^(−200/500)=e^(−0.4)≈0.6703. Expected payment is E[(X−d)+]=θe^(−d/θ)≈$335.16. Conditional on a positive payment, the exponential memoryless property gives E[Y|Y>0]=θ=$500. The conditional and unconditional values answer different questions.

Limits, coinsurance and loss elimination

A policy limit caps insurer payment; coinsurance scales a covered amount; a deductible removes an initial layer. Write the payment function piecewise before integrating or summing. For a deductible d and limit u above the deductible, payment can be min((X−d)+,u). Clarify whether a policy limit is an amount above the deductible or a cap on total payment.

The loss elimination ratio measures the share of expected losses eliminated by a deductible under a stated distribution. It can be derived from the expected payment or expected retained loss. Inflation changes loss amounts and can alter the proportion above a fixed deductible or limit. The syllabus asks for loss elimination and inflation effects, so identify whether the contract terms are fixed in nominal dollars.

Frequency, severity and aggregate loss

A frequency model N describes claim count; severity Xᵢ describes claim size; aggregate loss S=ΣXᵢ sums the severities. In a collective model, frequency and severity assumptions combine to determine the aggregate distribution. If N and severities are independent with finite moments, E[S]=E[N]E[X] and Var(S)=E[N]Var(X)+Var(N)(E[X])². State independence before using these formulas.

The syllabus includes distribution classes, the (a,b,0) and (a,b,1) frequency classes, individual and collective risk models, normal/lognormal approximations, convolution and stop-loss expected payments. A normal approximation can be poor in a heavy-tailed or low-frequency setting. Check the problem’s assumptions and requested tail quantity before choosing it.

Risk measures: VaR and TVaR

Value at Risk at level α is a quantile of the loss distribution. Tail Value at Risk summarizes losses beyond a tail threshold, with conventions that can differ for distributions with atoms. The syllabus asks candidates to calculate VaR and TVaR and assess desirable risk-measure properties. Do not use “average of losses above VaR” without checking the definition and discrete-mass treatment used in the question.

For an aggregate model, distinguish an individual claim percentile from a portfolio aggregate percentile. The frequency and severity combine; the tail of the sum is not generally the sum of separate quantiles. If a question asks whether a measure is subadditive or has another property, evaluate the stated definition rather than relying on intuition from a single example.

Original deductible example and checks

The exponential deductible example has an unconditional insurer payment with a zero mass of about 32.97%, because P(X≤200)=1−e^(−0.4). The expected payment per ground-up loss is $335.16. Conditional on payment, the expectation is $500. The conditional value is higher because it excludes losses that produce no payment.

If the deductible increases, expected insurer payment cannot rise under an otherwise unchanged contract. If a limit is added, payment is no greater than the unlimited-deductible payment. Use these monotonicity checks to catch a payment-function error. Verify whether the question asks for amount per loss, probability of payment, or conditional severity before substituting.

Transform the loss before calculating payment

Let X be ground-up loss. With an ordinary deductible d and no upper limit, insurer payment per claim is Y=(X-d)_+=max(X-d,0). The payment has a point mass at zero equal to P(X≤d), plus a continuous or discrete positive part. Its expectation can be computed as E[(X-d)_+], not E[X]-d unless additional assumptions justify that shortcut.

For an exponential loss with mean $500 and deductible $200, the survival probability beyond the deductible is exp(-200/500)=exp(-.4), about.6703. By the memoryless property, the conditional mean excess over $200 given X>200 remains $500. The unconditional insurer payment expectation is.6703×$500, about $335.16. The conditional mean among claims with positive payment is $500, illustrating why the requested conditioning matters.

Limits and coinsurance alter the payment function

With deductible d and limit u on insurer payment, a common payment function is min((X-d)_+,u). A limit on total loss rather than payment can be defined differently, so inspect wording. Coinsurance may multiply covered payment by a percentage after applying a deductible or limit; the order matters. Write the piecewise payment function before using a severity formula.

The transformed distribution contains masses at zero and possibly at the maximum payment. Expected payment can be calculated by integrating the survival function over payment layers, or by summing the pieces of the distribution. For continuous nonnegative X, E[(X-d)_+] equals the integral from d to infinity of the survival function. A policy limit truncates that integral at the corresponding attachment and limit points.

Frequency and severity create aggregate risk

Let N be claim count and X_i independent severities with a common distribution. Aggregate loss S is the sum of X_i for i=1,...,N. Under independence between count and severities, E[S]=E[N]E[X]. Variance is not simply E[N]Var(X); it also reflects variability in N and the mean severity. Conditioning on N gives Var(S)=E[N]Var(X)+Var(N)(E[X])² under the standard assumptions.

These formulas rely on the stated independence and identical severity assumptions. If claim sizes depend on the number of claims, or policies have different exposure, the model changes. A compound Poisson model is one special case, not a universal aggregate distribution. State the model before applying a formula and check whether the question asks about one claim, annual total, or a portfolio.

VaR and TVaR describe tail risk differently

Value at Risk at confidence level alpha is a quantile: a threshold that the loss does not exceed with the specified probability under the chosen definition. Tail Value at Risk summarizes severity in the tail beyond a threshold, with definitions that can handle discrete distributions differently. VaR identifies a boundary; TVaR incorporates information about losses beyond it.

For a continuous distribution, TVaR can often be expressed as the expected loss conditional on exceeding the VaR threshold. In a discrete distribution, probability mass at the threshold requires care, and the exact definition should come from the syllabus or problem statement. Do not assume the conditional expectation formula applies unchanged to every distribution.

Check behavior at the boundaries

At deductible zero and no limit, insurer payment should equal the ground-up loss. As deductible increases, expected payment should not increase. As a payment limit increases, expected payment should not decrease. These monotonicity checks catch reversed integration limits and errors in a piecewise function.

When computing a probability, distinguish P(X>d) from P(X≥d) if there is a point mass at d. For a continuous severity model they coincide, but a discrete model can have probability exactly at the threshold. State whether the distribution is continuous before replacing strict inequalities. Boundary details can determine a multiple-choice answer.

Use survival functions to integrate excess loss

For a nonnegative continuous severity X, the expected payment above deductible d can be written as the integral of the survival function from d upward. This area interpretation is often simpler than integrating (x-d)f(x) directly. If payment is capped, stop at the relevant limit and include the layer boundaries carefully.

For a deductible only, the survival-function identity also makes monotonicity clear: increasing d shortens the integration range, so expected insurer payment falls. This gives a conceptual check independent of the exact severity distribution.

Separate claim-level and portfolio quantities

A deductible payment Y is a claim-level random variable. Aggregate portfolio payment sums such variables across a random number of claims. Expected aggregate payment may use the expected count and mean payment under independence, but a portfolio percentile requires the aggregate distribution, not only its mean.

Do not answer a question about annual total risk with the expected payment for one claim. Identify the exposure unit and whether claim count is fixed or random. The distinction is central to frequency-severity modeling.

Calculate tail summaries from a defined loss basis

Before finding VaR or TVaR, specify whether the loss is ground-up, insurer payment per claim, or annual aggregate payment. A deductible and limit transform the distribution, including its point masses. A percentile for ground-up severity can differ greatly from a percentile of capped insurer payment.

State the confidence level and quantile convention. In a discrete distribution, several values can satisfy an informal percentile description, so use the problem’s definition. TVaR requires a tail convention that handles probability mass at the threshold.

Common questions

What does a deductible do to a loss?

It removes an initial layer, often producing zero insurer payment below the deductible.

Why are conditional and unconditional payment means different?

The unconditional mean includes losses with zero payment; the conditional mean excludes them.

Does VaR equal TVaR?

No. VaR is a quantile; TVaR summarizes losses in the tail beyond a threshold under its definition.