Insurance and risk pooling explained

Insurance is a pool with rules. A large number of people who each face the same kind of loss pay a known amount into a common fund, and the fund pays whoever the loss actually lands on. Nobody knows in advance who that will be. Everybody knows what their own year is going to cost.

That second sentence is the product. A policy sells two things at once: money if the bad thing happens, and the removal of uncertainty whether it happens or not. The second is why somebody who pays for thirty years and never claims still received what they bought.

It is the opposite of a bet. A bet creates a risk that did not exist until someone placed it. Insurance takes a risk that already exists and moves it from a balance sheet where it is concentrated and could be fatal to one where it is one of a million similar risks and barely registers. No new risk is created. The same possible loss ends up carried by whoever can carry it most cheaply, and the fee for arranging that is what the third section below prices.

Not every risk can be pooled. The usual conditions are that losses are accidental rather than chosen, that there are enough similar exposures for the average to settle down, that a loss can be measured and its chance estimated, that no single event can bankrupt the pool, and that the resulting price is one people will actually pay. Wherever a condition fails you find one of three things: an exclusion buried in the policy, a government scheme standing behind the market, or no cover for sale at any price.

Take a household with a 1 in 200 chance in a year of a fire costing $40,000 to put right. Its expected loss is $0.005 \times 40{,}000 = 200$ dollars a year. That figure is useless to the household, because nobody has a 200 dollar fire. It has a 0.5 percent chance of losing $40,000 and a 99.5 percent chance of losing nothing at all.

The same figure is exactly what the pool needs. Claim counts across many independent households behave like coin tosses. The number of claims expected is $np$ and the spread around that number is $\sqrt{np(1-p)}$, so the spread grows more slowly than the pool does:

The last column is the whole idea. Absolute variation rises as the pool grows, but variation relative to the size of the pool falls in proportion to one over the square root of the membership. Multiply the members by a hundred and the relative uncertainty divides by ten. Individual fires stay completely unpredictable the entire time. Only the average settles, and the average is what a premium has to be quoted from.

The load-bearing word is independent. The arithmetic needs one household's fire to say nothing about its neighbour's. Earthquakes, floods, windstorms and pandemics break that assumption, because one event produces a million correlated claims at once and the average stops settling. That is the insurance version of systematic risk, and insurers answer it with geographic spread, named-peril exclusions, reinsurance, catastrophe bonds and capital held back for the bad year. The identical arithmetic run across holdings rather than across people is diversification.

A pool collecting exactly the expected loss would fail in its first unlucky year, and it would have nothing left over for claims handlers, offices, brokers, taxes or the capital it is required to hold. So the premium is built up rather than copied across. Using the fire above, and rounding hard:

The split is illustrative and the loading varies widely by line of business, but the shape is not in doubt: what you hand over is more than what the pool expects to hand back. Here that gap is $100 a year, and it is the price of the protection rather than a fee hidden inside it. It is also a real cost. If the loading stayed at $100 and you set it aside every year at an assumed 5 percent, it would come to $6,643.88 over thirty years, which the first worked example below sets out. Treat that 5 percent as an illustration rather than a rate on offer anywhere, and read the total as future dollars rather than today's.

So the expected monetary value of nearly every policy is negative for the buyer. Two things sit outside that calculation and explain why people buy anyway.

Money is not equally useful at every level. Losing $40,000 you do not have is not 400 times as bad as losing $100 you do. It is far worse, because the shortfall gets covered by borrowing at whatever rate is on offer, by selling something at whatever price is bid that week, or by not fixing the thing at all. Each of those carries a cost of its own on top of the loss, and none of them appears in the expected value calculation.

And some losses end the game. A person who is wiped out does not get to play enough rounds for the average to arrive. Paying a known amount to make sure the worst case stays survivable is not a bad bet badly priced. It is a different transaction from the one the arithmetic describes, and it is the reason cover exists at all. Run the same logic backwards and you get the corollary most people ignore: where a loss is small enough to absorb, the loading is close to pure cost, which is why the phone screen and the extended warranty are the standard textbook examples of cover that is almost all loading.

A deductible is the first slice of every loss, and it stays with you. Choosing one is not an administrative detail, it is the point where you decide how much of the risk to keep.

Raising it does two useful things. It takes your small claims off the insurer's book entirely, and small claims are the expensive kind per dollar they pay out, because handling one costs roughly the same whether it settles for a few hundred or a few thousand. It also removes the incentive to claim for trivia. Both show up as a lower premium. The schedule below is illustrative and belongs to a broader household policy rather than the single-peril fire example priced earlier at 300 dollars, so do not read the two sets of figures against each other. All amounts are dollars a year:

The last column comes from one division:

$$\text{break-even years} = \frac{\text{extra loss you carry}}{\text{yearly premium saved}}$$

Read it as the question the choice is really asking. If claims come less often than that, the higher deductible wins, and the gap widens the longer the claim-free run goes on. Two things the division quietly assumes. It assumes every claim is large enough to run past the higher deductible, so the extra loss in the third column is the most you could carry rather than what you would typically carry. And it is an average over many years, not a promise about the next one: a single claim in year two reverses the comparison outright. Notice too that the trade gets steadily worse as the deductible climbs. The first step up takes about four claim-free years to pay for itself, the largest one more than twelve. That pattern is the normal one. The first increase strips out the frequent small claims, which is where most of the insurer's cost sits, while later increases only remove claims that were rare to begin with, so the premium saving stops keeping pace with the exposure you are taking on.

A higher deductible is only real if the cash is there on the day. Choosing $2,000 with $500 already set aside means $58.47 a month for two years to stand behind your own decision, which the third worked example prices. That is the same job an emergency fund does, and you can size it with the savings goal calculator.

The deductible is the floor of the cover. The limit is its ceiling, and it is the more dangerous number of the two, because the whole reason you bought the policy lives in the tail above it.

Two problems make insurance harder than the arithmetic suggests. Both come from information the insurer does not have.

Adverse selection happens before the policy is written. People know more about their own risk than the insurer does, and the ones who expect to claim are keenest to buy. If everyone is charged the pool average, the low-risk members are overpaying and start to drop out, which raises the average, which pushes more of them out. In the worst case the process runs to a market with only the highest risks left in it and a price nobody will pay.

Insurers hold it off by asking questions, requiring medical evidence or an inspection, imposing waiting periods, excluding conditions that already exist, and pricing by risk group rather than by one flat rate. The other answer is a pool people do not join one at a time, which is why cover bought for a whole workforce or a whole country prices better than the same cover sold door to door. Where regulators require insurers to accept every applicant, they often pair it with something that keeps low-risk people in, such as fixed enrolment windows. Which of these tools is used, and whether risk-based pricing is allowed at all, differs by country and by line of business, so none of the arrangements named here is universal. In the United States, for instance, individual health cover is shaped by both federal and state rules, and what applies depends on the state.

Moral hazard happens after. Being covered changes behaviour: the insured car gets parked more carelessly, the covered treatment gets used more freely, the loss that would have been absorbed becomes a claim. This is mostly ordinary human response rather than fraud, and it is why cover is never total. Deductibles, co-payments, co-insurance, no-claims discounts, experience rating and outright exclusions all exist to keep some of the loss with the person best placed to prevent it.

One line to keep them apart: adverse selection is about who buys, moral hazard is about what they do once they have.

Sort risks on two axes, how likely and how bad, and the whole subject collapses into four boxes.

• Rare and ruinous: insure it. Liability for harm to other people, the house burning down, long-term disability, income for anyone who depends on you, and medical costs wherever care is billed privately. These are the losses that no savings balance can absorb.
• Common and small: pay for it yourself. Screen repairs, appliance warranties, gadget cover on a travel policy, and the lowest deductible on offer. The chance is high, so the premium is close to the loss plus the loading, and the loss was survivable to begin with.
• Common and ruinous: usually not for sale. Something almost certain to happen and catastrophic when it does is priced out of the market or excluded outright, which is why some perils sit with government schemes.
• Rare and small: ignore it.

Most of the policies people regret buying come from the second box, and most of the gaps that actually hurt sit in the first. Notice that likelihood barely features in the rule. What decides it is what the loss would do to you.

Self-funding a large loss is the honest alternative, and it is slow. Reaching $40,000 in a savings account takes $162.54 a month for fifteen years at an assumed 4 percent, as the second worked example shows, and the loss is uncovered for every one of those years while the balance builds. The pool sells the same $40,000 of protection from the first day for a few hundred dollars a year, because it only ever needs to hold enough for the claims that actually arrive. The two figures are not the straight comparison they look like, though. The savings balance is still yours at the end and the premiums are spent, so what the premium is really buying is cover during the years the fund does not yet exist.

The order of operations usually taught is to cover the ruinous losses first, set each deductible no higher than the cash that could be produced without borrowing, and route the premium saving into the fund that stands behind those deductibles. This page is educational material and not financial advice. What any particular household should carry depends on its dependants, its assets, its tax position and the rules of the country and state it lives in, none of which this page knows.