1.1 Why Insurance Exists — Risk Pooling and the Law of Large Numbers
Insurance turns a loss you can’t predict into a small fee you can. Imagine a thousand neighbors who each fear a house fire: none knows whose house will burn, but together they can predict roughly how many fires will happen, so each pays a little into a shared pot. The law of large numbers says the more policies in the pool, the steadier the claim rate becomes. But losses that hit many people at once, like a hurricane, don’t average out. Those must be carried by capital, reinsurance or catastrophe bonds.
Why it matters: Pooling works for separate risks, not for risks that strike together.
Summary: Insurance turns an unpredictable individual loss into a small, predictable premium by pooling many similar, independent risks. The law of large numbers makes the pool’s claim rate stable: its standard deviation shrinks with the square root of the number of policies. Correlated losses do not shrink with size, so they must be carried by capital, reinsurance or catastrophe bonds.
- With a 2% claim probability, the claim rate’s standard deviation is 0.14 percentage points for 10,000 policies and 0.014 for 1,000,000.
- A hundredfold larger pool cuts relative uncertainty only tenfold (1 ÷ √n).
- A pairwise correlation of just 0.01 sets a floor of 70% relative uncertainty, however large the pool.
- Capitalizing a correlated book as if it were independent understates the needed buffer a hundredfold in the worked example.
- Hurricane Andrew (1992) caused about $15 billion of insured claims and pushed at least eight insurers into insolvency (Moody’s); the Insurance Information Institute counts eleven.
Insurance exists to convert an individually unpredictable, potentially catastrophic loss into a small, predictable, affordable cost. It works through risk pooling: a large number of people face similar risks (a house fire, a car accident, a death), each individually unlikely, and each pays a small premium into a shared pool; the pool pays out to the smaller number who actually experience a loss in any given period.
The mathematical foundation making this work reliably is the law of large numbers: while any single individual’s chance of a claim is highly uncertain, the average claim rate across a sufficiently large pool of similar risks becomes remarkably stable and predictable — which is precisely what allows an insurer to price a premium with measurable statistical confidence, rather than guessing.
The precision of a pool grows with the square root of its size. If each of n policies has a claim probability p and claims are independent, the standard deviation of the pool’s claim rate is √(p(1 − p) ÷ n). With p = 2% and 10,000 policies: √(0.02 × 0.98 ÷ 10,000) = 0.0014, or 0.14 percentage points, which is 7% of the 2% expected rate.
| Policies in the pool | Standard deviation of the claim rate | As a share of the expected 2% rate |
|---|---|---|
| 100 | 1.40 percentage points | 70% |
| 10,000 | 0.14 percentage points | 7% |
| 1,000,000 | 0.014 percentage points | 0.7% |
Two lessons sit in that table. First, a hundredfold larger pool cuts relative uncertainty only tenfold. Second, the whole result depends on independence. If every pair of policies shares even a small correlation ρ (one hurricane, one pandemic, one court ruling that reprices a whole class of claims), the variance of the claim rate becomes p(1 − p)[ρ + (1 − ρ) ÷ n], which never falls below p(1 − p)ρ however many policies are added. With ρ = 0.01: √(0.02 × 0.98 × 0.01) = 0.014, a floor of 70% of the expected rate — the same relative uncertainty as a 100-policy pool, at any size. Correlated risk cannot be diversified away by selling more policies; it has to be carried by capital or passed on through reinsurance and catastrophe bonds (Part 1.7: Reinsurance — Insuring the Insurers).
Flipping one coin, you cannot predict whether it lands heads or tails. Flip ten thousand coins, and you can predict with very high confidence that close to 50% will land heads — the individual outcome stays random, but the aggregate becomes reliable. An insurer cannot predict which specific policyholder’s house will burn down this year, but across a large enough pool of similar houses, it can predict the aggregate number of fires with genuine statistical precision.
If losses are independent, measurable and individually small relative to capital, then pool them and price on expected loss plus a loading; scale cuts relative volatility by 1 ÷ √n. Once the correlation between policies exceeds roughly 1 ÷ n (the point where the correlated term overtakes the independent one in the formula above), size capital and reinsurance from the modeled catastrophe tail, not from the policy count. For a 100,000-policy regional property book that threshold is a correlation of only 0.00001. Ignore the rule for books with no meaningful common shock in normal years, such as individual life mortality, but stress those separately for pandemics.
Capitalizing a concentrated book as if its risks were independent. For 1,000,000 homes with a 2% claim probability and a $15,000 average claim, expected claims are 1,000,000 × 2% × $15,000 = $300 million. A three-standard-deviation buffer is 3 × 0.014% × 1,000,000 × $15,000 = $6.3 million if claims are independent, but 3 × 1.40% × 1,000,000 × $15,000 ≈ $630 million at a correlation of 0.01, a hundred times more. Hurricane Andrew in 1992 produced about $15 billion of insured claims and pushed at least eight insurers into insolvency, according to Moody’s (2022); the Insurance Information Institute counts eleven property-casualty insolvencies, ten in Florida and one in Louisiana. Aggregate exposure by zone and peril, model the tail, and buy protection for it.
What is the law of large numbers in insurance?
It is the statistical result that the average claim rate of a large pool of similar, independent risks settles close to its expected value. One policyholder’s outcome stays unpredictable, but across 10,000 policies with a 2% claim probability the realized rate typically lands within about 0.14 percentage points of 2%, which lets an insurer price losses it has not yet seen.
Does a bigger insurer always carry less risk?
No. Size reduces only the risk that is independent across policies. A large insurer concentrated on one coastline, or in one liability line exposed to the same legal trend, can be riskier than a small diversified one, because correlated losses arrive together. Regulators and rating agencies therefore examine catastrophe exposure and reinsurance, not just premium volume.
Why won’t insurers cover some risks at any price?
Pooling needs losses that are measurable, accidental and not catastrophic to the whole pool. Losses the buyer controls, losses nobody can estimate, and losses that hit most policyholders at once break the mechanism: the premium needed for the correlated tail exceeds what buyers will pay. Such risks tend to end up with governments, capital markets or nobody.
Insurance pools many similar, independent risks so that the claim rate becomes predictable: its standard deviation falls with the square root of pool size, from 0.14 percentage points at 10,000 policies to 0.014 at 1,000,000 for a 2% claim probability. Correlated losses set a floor that no amount of size removes, so catastrophe exposure must be met with capital, reinsurance or catastrophe bonds rather than more policies.
Four questions on this chapter. Decide on your answer first, then click “Reveal Answer.”
1. An insurer has 2,500 independent policies, each with a 4% annual claim probability. What is the standard deviation of the pool’s claim rate?
- About 0.04 percentage points
- About 3.92 percentage points
- About 0.39 percentage points
- About 1.96 percentage points
Reveal Answer
Answer: C. √(0.04 × 0.96 ÷ 2,500) = 0.0039, or 0.39 percentage points, about 10% of the 4% expected rate.
2. If an insurer quadruples the number of independent, similar policies in its pool, what happens to the standard deviation of its claim rate?
- It falls to a quarter of its former level
- It falls to a sixteenth of its former level
- It stays the same as before
- It falls to half its former level
Reveal Answer
Answer: D. The standard deviation scales with 1 ÷ √n, so four times the policies gives 1 ÷ √4 = one half.
3. A regional insurer’s homeowners book has a small positive correlation between policies because of shared hurricane exposure. What does adding many more policies in the same region do?
- Leaves a floor on volatility that size cannot remove
- Turns the correlated risk into fully independent risk
- Removes all volatility once the pool passes one million
- Raises volatility in proportion to the total policy count
Reveal Answer
Answer: A. With correlation ρ, the variance never falls below p(1 − p)ρ, so correlated risk must be met with capital or risk transfer.
4. Which risk is hardest to insure through ordinary pooling?
- A loss that is accidental and easy to measure
- A loss that hits most policyholders at once
- A loss whose probability is well documented
- A small loss that each buyer suffers independently
Reveal Answer
Answer: B. Pooling needs independent, measurable, accidental losses; a common shock to the whole pool defeats the law of large numbers.
1.2 Types of Insurance
There are three big families of insurance, and each behaves differently. Life insurance makes promises that can last decades and depends on how long people live and whether they keep their policies. Property and casualty insurance, covering things like homes and cars, depends on how often claims happen, how big they are and on catastrophes. Health insurance depends mainly on rising medical costs. Because their risks differ, you need different measures to tell whether each line is healthy.
Why it matters: A number that works for one kind of insurer can mislead for another.
Summary: Life, property-casualty and health insurance differ in how long their promises run and how correlated their losses are. Life books turn on mortality, longevity and lapse; P&C books on frequency, severity and catastrophes; health on medical cost inflation. Those differences decide which metric tells you whether a line is healthy.
- One year of $1,000,000 coverage costs $3,115 in expected claims for a US man of 40 and $16,455 at 65 (SSA 2023 period table).
- Level premiums overcharge early and build a reserve that pays for the costlier later years.
- Two extra years of annuitant life raise an annuity’s value by 7.4% in the worked example.
- Term, whole, universal, variable universal life and annuities split investment risk differently between insurer and buyer.
- Judge short-tailed lines on the combined ratio, long-tailed lines on reserves, catastrophe lines on modeled tail loss.
Insurance splits into several major categories, each with a different risk and financial profile.
| Category | What It Covers | Key Financial Characteristic |
|---|---|---|
| Life Insurance | Pays a benefit on the policyholder’s death, or on survival to a set date for certain products | Very long-dated liabilities, often decades — the defining challenge behind the duration-matching discipline covered in 1.6 |
| Property & Casualty (P&C) | Home, auto, and liability insurance — covers damage to property or legal liability to others | Shorter-tailed claims (typically settled within months to a few years), but exposed to sudden, severe “catastrophe” losses (hurricanes, earthquakes) |
| Health Insurance | Medical expense coverage | High claim frequency, relatively short-tailed, but subject to significant, hard-to-predict medical cost inflation |
| Reinsurance | Insurance purchased by insurance companies themselves — covered fully in 1.7 | Concentrates and redistributes the most extreme tail risk across the global insurance system |
Mortality is the probability of dying within a year at a given age, written q. The pure cost of one year of death coverage is q × face amount. In the Social Security Administration’s 2023 period life table, a US man aged 40 has q = 0.003115 and a man aged 65 has q = 0.016455. One year of $1,000,000 coverage therefore costs 0.003115 × $1,000,000 = $3,115 in expected claims at 40 and 0.016455 × $1,000,000 = $16,455 at 65, about 5.3 times as much. A level-premium policy charges the same amount every year, so early premiums exceed early claims and the excess builds the policy reserve that pays the later, costlier years.
Longevity risk is the mirror image: an annuity writer loses when people live longer than priced. The same table gives a 65-year-old man a life expectancy of 18.12 years. Treating life expectancy as a fixed payment term (a simplification), $10,000 a year at a 4% discount rate is worth $10,000 × (1 − 1.04−18) ÷ 0.04 = $126,593 over 18 years, but $135,903 over 20 years: two extra years of life raise the cost by 7.4%. Because deaths come early for life policies and late for annuities, a company that writes both carries a partial natural hedge.
Lapse is the rate at which policyholders stop paying. Lapses hurt when the insurer has spent acquisition costs it has not yet recovered, and they hurt in a quieter way when healthy people leave and the less healthy stay (selective lapse), pushing the remaining pool’s mortality above the priced level. For products priced on the assumption that many buyers will lapse before claiming, the risk flips: too few lapses are what cost money.
| Life product | What the buyer gets | Who carries the investment risk | Main risk to the insurer |
|---|---|---|---|
| Term life | A death benefit for a fixed period, with no cash value | No savings element | Mortality and selective lapse |
| Whole life | Lifelong coverage with a guaranteed cash value | The insurer, through the guarantee | Interest rates, lapse, mortality |
| Universal life | Flexible premiums; cash value credited at a declared rate above a guaranteed minimum | Mostly the insurer, up to the minimum | Interest-rate guarantees and lapse |
| Variable universal life | Cash value invested in market subaccounts, plus a death benefit | The policyholder | Expense recovery and lapse |
| Life annuity | Income for as long as the annuitant lives | The insurer | Longevity and interest rates |
India’s unit-linked insurance plan (ULIP) is the counterpart of variable universal life; its lock-in rules sit in the India Lens at the end of this Part.
A reading rule for any insurer: classify each line by tail length and correlation, then watch the matching metric. If the line is short-tailed with largely independent claims (personal auto, most health), judge it on the combined ratio (Part 1.3: The Actuarial Function — Pricing Risk Before It Happens). If claims take years to settle or the promise runs for decades (liability, life, annuities), judge it on reserve adequacy (Part 1.4: Reserving — Provisioning for Claims Not Yet Paid) and on how liability values move with discount rates (Part 1.6: The Insurance Balance Sheet — Float and Investment Income). If losses are correlated (coastal property, earthquake), judge it on modeled catastrophe loss and the reinsurance behind it (Part 1.7: Reinsurance — Insuring the Insurers). One headline profit figure blends all three.
Underestimating longevity on a block of annuities. With the simplified annuity values above, pricing 18 years of payments for annuitants who live 20 makes the liability 13.590 ÷ 12.659 − 1 = 7.4% larger than priced: about $74 million on a $1 billion block, surfacing year by year long after the premiums were collected. Annuity writers defend themselves with mortality tables that assume continuing improvement, regular checks of their own experience against the table, and longevity reinsurance.
What is the difference between a life insurer and a property-casualty insurer?
The length and shape of their promises. A life insurer pays fixed or formula-based amounts on death or survival, often decades away, so mortality, longevity and interest rates drive its results. A P&C insurer pays for damage or liability, usually within months to a few years, so frequency, severity, inflation and catastrophes drive its results.
Is term life insurance cheaper than whole life insurance?
Yes, for the same death benefit, because term coverage buys only the mortality risk for a set period, while whole life also funds lifelong coverage and a guaranteed cash value. The gap is widest at young ages, when the annual probability of death is small. Which suits a household is covered in Volume V, Part 8: Operational Risk Governance & Reconciliation.
What is longevity risk?
It is the risk that people live longer than an annuity or pension was priced for, so payments run longer than expected. In the example above, two extra years of life raise the value of a $10,000 annual income by 7.4%. Insurers, pension plans and retirees all carry versions of it.
Life, property-casualty and health insurance differ in the length of their promises and the correlation of their losses: life books run on mortality, longevity and lapse, P&C books on frequency, severity and catastrophes. Read each line with its matching metric: combined ratio for short-tailed lines, reserve adequacy and rate sensitivity for long-dated ones, modeled tail loss for catastrophe lines.
Four questions on this chapter. Decide on your answer first, then click “Reveal Answer.”
1. A 50-year-old has an annual death probability of 0.005. What is the expected claim cost of one year of $500,000 term coverage?
- $5,000
- $25,000
- $2,500
- $250
Reveal Answer
Answer: C. Pure cost of one year of coverage = q × face amount = 0.005 × $500,000 = $2,500.
2. In a variable universal life policy, who mainly carries the investment risk on the cash value?
- The policyholder, through market subaccounts
- The reinsurer, through a quota share treaty
- The insurer, through a guaranteed crediting rate
- The regulator, through the state guaranty fund
Reveal Answer
Answer: A. Variable universal life invests the cash value in subaccounts chosen by the policyholder, so market gains and losses fall on the buyer.
3. Why does selective lapse hurt a life insurer?
- Lapsing policyholders receive more than they paid in premiums
- Healthier people leave, so mortality runs above pricing
- Lapses force the insurer to raise its reported solvency ratio
- Lapses lower the discount rate applied to the remaining policies
Reveal Answer
Answer: B. When healthy policyholders stop paying and less healthy ones stay, the pool’s mortality exceeds the pricing assumption.
4. You are judging the health of a general liability line whose claims take many years to settle. Which measure deserves the most attention?
- Its single-year combined ratio alone
- Its count of policies in force
- Its share of the industry’s premium
- The adequacy of its claims reserves
Reveal Answer
Answer: D. Long-tailed lines report profit long before costs are known, so reserve adequacy decides whether the profit is real.
- Social Security Administration, Actuarial Life Table (2023 period table, 2026 Trustees Report) — Mortality rates at 40 and 65 and life expectancy at 65