Skip to content
Tutorials Center
My learning

Topics

Lesson 29 of 43 · Risk Theory and Loss Models

Risk Theory and Loss Models: Chapter Review

Review claim counts, claim sizes, aggregate loss, deductibles and reinsurance in one place, run a combined example, and test yourself with ten mixed questions.

Risk theory builds one machine in four stages. Choose a distribution for the number of claims, choose one for the size of a claim, combine them into the aggregate loss, and then decide how the loss is shared between policyholder, insurer and reinsurer. This review gathers the formulas, runs one example through all four stages, and checks your understanding.

One-page summary of counts, sizes, aggregate loss and risk sharing
The chapter in one picture: count the claims, size each claim, add them up, share the loss.

What you will learn

  • Recall the formulas for counts, sizes, the compound sum and risk sharing.
  • Carry a single portfolio through all four stages.
  • Test your understanding with ten mixed questions.

The chapter in one table

Stage Key formulas Watch out for
Counts Poisson: E = Var = λ; NB: E = rβ, Var = rβ(1+β) Overdispersion: variance above the mean
Sizes Exponential θ; lognormal e^(μ+σ²/2); Pareto θ/(α−1) Infinite moments for small α
Aggregate E[S] = E[N]E[X]; Var[S] = E[N]Var[X] + Var[N]E[X]² Skew: normal approximation too optimistic
Exact law Panjer: g_s = (λ/s) Σ j f_j g_(s−j) (Poisson) Needs discrete claim sizes
Deductible E[(X−d)⁺] = E[X] − E[X∧d]; LER = E[X∧d]/E[X] Inflation leverage
Reinsurance Quota q·X; XL (X−M)⁺; stop loss (S−a)⁺ Cost of cover versus benefit

A combined worked example

An insurer writes 500 policies, each with Poisson claim rate 0.2 per year, so λ = 100. Claim sizes are exponential with mean 2,000. The insurer applies a per-claim deductible of 500.

  1. Counts: the number of claims that exceed the deductible is Poisson with mean 100 × e^(−500/2000) = 100 × e^(−0.25) ≈ 77.9. (Thinning a Poisson count keeps it Poisson.)
  2. Sizes: by the memoryless property, a payment given that the claim exceeds 500 is again exponential with mean 2,000.
  3. Aggregate payments: E[S] = 77.9 × 2000 ≈ 155,760, which equals 100 × E[(X − 500)⁺] = 100 × 2000 e^(−0.25); the two routes agree. Var[S] = λ' · E[Y²] = 77.9 × 2 × 2000² ≈ 623,000,000, with standard deviation about 24,960.
  4. Sharing: a stop loss at E[S] + 2σ ≈ 205,700 would pay out only in the worst few percent of years.

The code below recomputes each line.

Try it

Python Combined example: from counts to stop loss

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Python Chapter plot: normal approximation versus simulation

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Common misconceptions

"Frequency and severity can be fitted to totals only." Modelling them separately keeps information about how many claims occurred and how large each was, which is exactly what deductibles and reinsurance act on.

"Thinning a Poisson count by a deductible makes it non-Poisson." If each claim independently exceeds the deductible with probability p, the count of exceeding claims is Poisson with mean pλ.

Check your understanding

Quick check · 10 questions

of 10 answered You scored / 10 · perfect!

  1. 1. A Poisson claim count has mean 12. Its variance is

    For the Poisson, mean and variance are equal.
  2. 2. A negative binomial with r = 4 and β = 0.5 has mean

    Mean = rβ = 4 × 0.5 = 2.
  3. 3. The Pareto distribution with α = 0.8 has

    The Pareto mean exists only if α > 1.
  4. 4. For compound Poisson with λ = 10 and E[X²] = 500, Var[S] is

    Var[S] = λ·E[X²] = 10 × 500 = 5,000.
  5. 5. Which is the correct expected payment per loss with deductible d?

    Payment is X − (X ∧ d), whose expectation is E[X] − E[X ∧ d].
  6. 6. A quota share of 30% means the reinsurer pays

    Quota share is proportional: the reinsurer takes a fixed fraction of every claim.
  7. 7. As the number of independent policies n grows, the coefficient of variation of S behaves like

    Relative variability shrinks with the square root of the portfolio size.
  8. 8. A mean excess function that increases linearly indicates

    The Pareto mean excess is (θ + d)/(α − 1), linear in d.
  9. 9. A deductible of 500 is applied to losses that are exponential with mean 2,000. The expected payment per loss is

    E[(X − d)⁺] = θ·e^(−d/θ) = 2000·e^(−0.25).
  10. 10. Which reinsurance form directly caps the insurer's annual total?

    Stop loss pays the part of the aggregate above the attachment point.

Summary

  • Counts (Poisson, negative binomial) and sizes (exponential, lognormal, Pareto) combine into S, whose mean and variance follow from conditioning.
  • Panjer gives the exact law; simulation and the normal approximation give quick checks, with the normal failing in the tail.
  • Deductibles, limits and reinsurance are functions of the loss; their values are limited expected values or aggregate expectations.
  • Pooling shrinks relative risk but never the systematic risk.

Next: Credibility and ruin theory: chapter overview