The house remains capable of burning after its owner signs the policy. The paper does not wet the timbers, watch the stove or put out a flame. It changes whose resources must answer for the monetary loss. A payment made in the uneventful year buys a claim exercisable in the disastrous one. The owner exchanges a large, uncertain subtraction from wealth for a smaller, certain subtraction. Somewhere behind the signature, another balance sheet accepts the difference. The insurance premium is the price of making that promise stand up.
Bernoulli understood that the same transfer could be worth making from both sides. His 1738 discussion of a merchant's cargo and a prospective insurer gives the parties different resources against which to bear a possible loss. The merchant may pay more than the expected loss to be relieved; the insurer may accept the payment because carrying the exposure is less burdensome in the insurer's position. These are calculations in his argument, not a record of prevailing insurance quotations. Their importance is that insurance need not depend on one side making a mistake. A certain sum can be worth more to the person giving up the risk than the assumption of that risk costs the person receiving it. Pooling will enlarge this possibility, but the underlying exchange already concerns the location of a burden.
A mutual society adds a second operation. Many owners agree to meet the losses of the unfortunate few. Each can contribute a manageable share to a fund large enough to replace an individual loss. The exchange works most readily when the events are sufficiently separate: one member's misfortune need not announce the same misfortune for everyone else. It arranges that a large loss to one person calls upon many resources, while each contributor's share of the collective burden is comparatively stable. The institution turns a sequence of private calamities into a common account, then determines who must replenish that account when experience departs from expectation.
Consider an invented pool of ten thousand houses. Each has a one-percent annual chance of a loss of one hundred thousand pounds; assume the losses are independent and the probabilities known. The expected claim per house is one thousand pounds, so the expected total is ten million. The number of claims has a standard deviation just under ten, making the standard deviation of total claims about £994,987. Divided across ten thousand policies, that is about £99.50 per house. With only one hundred houses, the standard deviation per house would be about £994.99. A hundredfold increase in the number insured has reduced this measure of fluctuation per policy tenfold.
The shrinking quantity is the variation in the average claim. The total claims of the larger pool fluctuate by more pounds than those of the smaller pool. This distinction determines what the insurer must finance. Charging precisely the expected claims does not guarantee enough cash in a particular year; even a very large independent pool can lose more than its mean. Reserves, additional capital, reinsurance or an enforceable right to assess members supply ways of meeting the difference. Each changes the parties ultimately exposed. The law of large numbers makes a certain kind of exposure easier to manage. It is not a certificate that a fund holding its expected annual outlay can honor every promise it has written.
The premium must consequently do more than repeat the average loss. Someone must identify the covered event, collect contributions, investigate claims and maintain the capacity to pay. A loading above expected claims can finance those activities and compensate resources committed to the residual exposure. Its existence does not identify its entire amount as either necessary cost or extraction. The distinction requires an account of the actual undertaking and its alternatives. From the buyer's side, a loaded policy can remain attractive while its price stays below the value of relief from uncertainty. From the insurer's side, an attractive price to the customer does not establish that the promised cover is solvent. The two valuations meet through the contract and the resources supporting it.
A thousand signatures cannot divide a fire that reaches every house.
A common peril changes the arithmetic of aggregation. Suppose, at the opposite extreme of the invented example, that all ten thousand houses suffer their loss together with probability one percent. The expected claim per house is still one thousand pounds. The average claim is now either zero or one hundred thousand, with a standard deviation of about £9,949.87, no matter how many houses are added. Ten thousand names have been written against a single event. Enlarging this pool does not reproduce the protection obtained by adding independent houses. The expected loss survives unchanged while the capacity required to stand behind the promise changes radically.
Between complete independence and a single common loss lies most of the work of diversification. For equally variable losses with a common pairwise correlation ρ, the variance of the average is σ²[(1 − ρ)/N + ρ]. The first part diminishes with the number of policies; a fixed positive common component remains. Independence is a sufficient condition for the simple averaging theorem, not a necessary condition for every useful pool. Dependencies can be weak enough, or the composition sufficiently varied, for averages still to stabilize. The question is how much covariance is added with each new policy. Counting policies without examining their common exposures can turn diversification into the repeated purchase of the same danger.
Nor does dependence make every transfer impossible. An insurer may spread exposure across resources not impaired by the same event, arrange contingent capital, or agree in advance to limits on payment. A reinsurer can take a share of the liability; its usefulness depends on its own capacity when the claim arrives. In a wider community, contributions can distribute the cost of rebuilding among people whose resources remain available. These arrangements change the bearers and the timing of the loss. They cannot restore goods by multiplication of contracts. If the event destroys the resources upon which every promise depends, the relevant limit is the community's remaining ability to provide what has been lost. Monetary claims organize access to that ability; they do not create it by themselves.
Another limit concerns the membership of the pool. Knowing that a group contains high and low risks does not mean an insurer can identify which applicant is which. A common price may attract them in different proportions; the coverage offered can change who buys. Rothschild and Stiglitz made this selection part of a competitive insurance model. With the model's two privately known risk types, any equilibrium must separate them: the high-risk type receives full insurance, while the low-risk type accepts partial insurance that the high-risk type will not prefer. Under some compositions of the population, even that candidate arrangement can be upset by a profitable alternative contract, so an equilibrium of the specified kind does not exist. This is a result about a defined market and information structure. Its force is to show why the price and the extent of coverage cannot always be determined separately from the people they attract.
The act of insuring may also change the loss. A payment arrangement can alter the reward for prevention, the cost of using a covered service, or the incentive to establish that a claim qualifies. The term moral hazard can make this sound like a verdict on character, though the mechanism concerns the incentives supplied by the contract. It does not follow that every insured person becomes careless: bodily harm, inconvenience, attachment to property and uncovered costs can leave powerful reasons for care. It does follow that a probability estimated under one allocation of costs cannot automatically be carried unchanged into another. Coverage may alter the prospect whose expected loss the premium is supposed to recover.
A deductible gives some losses back to the insured, but its possible purposes should remain distinct. It can reduce small claims and their administrative burden, preserve an incentive to avoid loss, or help separate applicants with different risks. It can also arise without hidden behavior or hidden types. Arrow showed that, if a risk-averse buyer may choose indemnities and their price depends only on expected indemnity, an optimal contract covers losses fully above a deductible. For a given expected payment, concentrating protection on the states with the lowest remaining wealth makes the protection more valuable. The result depends on the available contracts and their pricing. It does not make every observed deductible optimal, and it does not establish that a person left unable to meet a loss has received enough protection.
The difficult cases therefore ask for more than a larger list of policyholders. Correlated losses require credible resources beyond the common exposure. Selection requires an arrangement that can endure the choices it induces. Incentives require attention to what the contract changes. A community may also choose to pool people whose expected claims differ, financing part of one group's protection through another's contributions. That choice adds a distributive purpose to the actuarial account. Making the transfer visible allows it to be defended or disputed on its actual terms. Concealing it inside an assertion that all dangers average away leaves both the calculation and the commitment obscure.
The policy joins a price to a promise about a future state. Many independent dangers make one part of that promise easier to support, because the average claim becomes less variable. When dangers are few or connected, the premium must answer to the exposure that remains, the terms on which it is carried, and the resources available when it arrives. Pooling still has work to do, but its success is measured by what happens to the burden rather than by the number of signatures gathered. The house can still burn. The achievement is that the person standing outside it need not stand alone before the whole cost of the fire.