The premium has followed a loss through several accounts without making the loss disappear. A household buys protection, an insurer gathers claims, an investor supplies capital and a public institution may stand behind the arrangement when its private resources fail. Each step changes who is exposed and what can be done beforehand. At the end there are still houses to rebuild, incomes to replace and people whose needs cannot wait for a favorable market. The volume's last question concerns the terms on which that burden becomes a common one.
Insurance begins with a gain that need not depend on anyone receiving a subsidy. Imagine one hundred people, each facing an independent one-in-ten chance of a loss of one hundred. Each person's expected loss is ten. An equal sharing of the group's realized losses leaves that expectation unchanged while making each person's contribution less variable. The standard deviation of the individual loss is thirty; the standard deviation of the loss per person in the pooled group is three. The arithmetic has spread an uncertain burden more evenly. It has not reduced the expected resources the accidents consume, and a fund promising fixed premiums would still need resources against departures from the average.
Sharing across different expected risks introduces another dimension. Suppose two equally sized groups face losses of one hundred with probabilities of one tenth and three tenths. Ignoring expenses and capital costs, their separate expected-loss premiums are ten and thirty. A common premium of twenty collects the same total expected funding while transferring ten per person from the lower-risk group to the higher-risk one. This transfer is analytically distinct from the reduction in variability available within either group. A society can favor both, one or neither. The language of a fair premium needs to say which question it is answering: the accuracy of the cost assigned to a class, or the justice of leaving that class to meet the cost.
The time at which a risk is classified can change the meaning of the transfer. A person may want protection while still healthy against a future condition that would make later coverage expensive. If a plan can reprice that person completely once the condition appears, an annual promise to pay specified claims leaves the risk of reclassification uncovered. Arrow observed that a common premium maintained across changing health states can itself be understood as insurance over a longer horizon. What looks like redistribution after the unfavorable information arrives can be the fulfillment of protection people wanted before they knew who would need it.
A common pool nevertheless needs a way to sustain its membership and funding. If the lower-risk group can buy equivalent protection for ten elsewhere, it has a financial reason to leave the plan charging twenty. The arithmetic of the remaining pool then changes. That simple comparison identifies a financing problem, not a theorem that every mixed pool must collapse. Contract terms, information, subsidies and participation rules matter. Rothschild and Stiglitz showed how hidden risk types can make the contracts offered by competing insurers depend on who each contract attracts; in their model a competitive equilibrium can fail to exist, and when a separating equilibrium does exist, its structure is different from a single price at which everyone buys any amount.
The instruments for sustaining solidarity can therefore take different forms. A public contribution can finance the gap between a person's expected cost and the contribution a polity wants that person to bear. Participation rules can keep a designated group within a pool. A publicly financed layer can cover specified losses while other arrangements cover what remains. Each choice has to identify the claim being promised, the resources behind it and the distribution of the funding burden. Calling the undertaking public settles none of those details. Calling it private does not establish that all the costs or all the benefits remain inside the participating accounts.
Incentives belong in this reckoning because protection changes what people can afford to do. Coverage may reduce the private return to preventing an avoidable loss, or alter the amount and kind of service demanded once a loss occurs. It can also make useful action possible for someone who would otherwise avoid it because failure would be ruinous. The welfare question requires the value of the changed action as well as its cost. Arrow's discussion of medical insurance keeps the gains from protection beside the difficulties of incentives, information and institutional control. Those difficulties are reasons to examine the design of the promise, not an arithmetic demonstration that exposure itself is always beneficial.
The pool also has an outer boundary. Change the first example so that one common event either imposes the loss of one hundred on all one hundred people or spares them all. Each still has an expected loss of ten, but equal sharing now leaves each person's loss exactly as variable as before: its standard deviation remains thirty. A larger head count has added no independent exposures. Broader geographical sharing, reserves or arrangements with people outside the affected group may still help, but the common event has to be borne somewhere. A payment from an insurer or a treasury can assign purchasing power; it cannot by itself restore the resources the event destroyed.
Public capacity matters at this boundary because a community can arrange contributions and commitments on terms unavailable to a single voluntary insurer. Its capacity remains a claim on people and resources. Borrowing can change the timing of payments; taxes can change their distribution; preparation can reduce the damage that will need to be repaired. The comparison should therefore include prevention as well as compensation. If a measure costs less than the loss it can credibly avert, its case does not become weaker merely because it produces no dramatic insurance payment afterward. Conversely, a promise of compensation that outruns the resources behind it can leave the very people it was meant to reassure exposed when the claim falls due.
The choice of institutions cannot be read straight from an efficiency result. Arrow begins his inquiry by distinguishing an allocation from its distribution: an outcome that admits no improvement for one person without injury to another can still leave a society wanting a different distribution of burdens. He also observes that nonmarket institutions can develop around risks and information that ordinary exchange handles poorly. Their existence is an invitation to compare what they accomplish, including their costs and failures. Neither the market price nor the public label supplies the complete verdict.
The reckoning asks what kind of security the society intends to provide and what it is willing to require in return. It can preserve a role for prices in identifying costly exposures while declining to make every unlucky person finance those exposures alone. It can require care without assuming that misfortune proves carelessness, and offer protection without pretending that every claim is beyond verification. These are judgments about membership, responsibility and the conditions of a workable promise. The arithmetic makes their consequences clearer. It cannot choose the commitments on the community's behalf.
A common promise earns its meaning when the loss arrives and the resources to meet it arrive too.
The premium began as the price of being spared a peril. It ends as a question about the distribution of the perils that remain. Some can be reduced, some spread more widely, some transferred on terms that help both sides; others require a decision about who will stand behind whom. The fortune records one path through that world, never the whole accounting of it. Beyond this volume stands the spread, the price of those who make transactions possible between people with different positions and different knowledge. The argument remains open.