The policy has been signed, the premium paid, and the insurer has gone home. The owner still decides whether to take the precaution. He bears its inconvenience; the insurer now bears much of the loss it might prevent. Nothing in this arrangement requires a dishonest owner. It requires only an action that matters, a cost of taking it, and a bargain that changes who receives its benefit. The hidden thing is no longer simply a condition the applicant brought to the desk. It is conduct the contract helps to produce after he leaves. This is moral hazard: the difficulty of sharing a consequence while leaving control over its cause in hands one cannot fully observe.
The difference from selection is a difference in the question being asked. The screen tried to distinguish people with different probabilities of loss, holding their conduct fixed. Here we ask how a person’s conduct changes when the terms change. A pool can become more costly because it attracts people with greater existing risks, because its members take different precautions, or because both things happen. A larger claims bill by itself does not tell us which. Nor does a loss prove that care was absent: precautions ordinarily alter probabilities rather than abolish misfortune. To read every accident as shirking is to confuse the result of an action with the action itself, the very confusion a theory of hidden conduct is meant to expose.
Consider an invented precaution that costs six and lowers the probability of a loss of one hundred from one-fifth to one-tenth. For the moment, let the owner be risk-neutral, able to pay, and concerned only with the money in this example. Without insurance, care costs six and saves ten in expected damage, so he takes it. With complete insurance at a premium fixed before he acts, the insurer receives the saving while he still pays the six. He declines the precaution. The physical opportunity has not changed. Its private reward has. Insurance has separated the person who chooses from the account in which part of the choice’s return appears.
A retained loss can restore that return without returning the whole loss. Suppose the owner must bear seventy if the accident occurs, while the insurer pays the remaining thirty. Care then saves him seven in expected personal loss, enough to justify its cost of six. An insurer anticipating that conduct can break even, before expenses, at a premium of three. The owner’s expected outlay with care is sixteen: six for prevention, seven for his expected retained loss, and three for the premium. If he instead omits care after paying the same premium, it is seventeen. Under complete insurance and no precaution, expected damage is twenty and a break-even premium is twenty. The comparison shows how a deductible can sustain a useful action; it does not calculate the best deductible for a person who fears an unaffordable loss.
Returning a stake is not the same as making the stake as large as possible. A deductible puts the unlucky owner back in danger as well as putting prevention back in his interest. If he is risk-averse, that exposure is costly before any accident occurs. If his funds are limited, a large retained loss can undo much of the protection for which he bought the policy. And a circumstance he cannot influence supplies no preventive reason for making him bear it. Holmström’s insurance analysis gives a precise version of the distinction: when an action affects whether an accident occurs but not how large it is once it occurs, the occurrence contains the relevant information about care. Under the model’s conditions, the incentive can be attached to that occurrence while the insurer bears additional variation in the damage.
Observation provides another way to reconnect the accounts. If the precaution can be verified and required, the insurer may make compliance a term of coverage instead of leaving the owner to infer it from a large possible loss. An inspection, a maintenance record, or evidence of a particular failure can inform the arrangement without giving the insurer complete sight of conduct. But the evidence must distinguish something material. Charging the owner for an independent coin toss would add uncertainty without making the precaution more attractive. Even useful monitoring has to earn its cost, including the work of collecting the evidence and deciding what it means. The choice is among instruments that change incentives and distribute risk differently, rather than between vigilance and generosity alone.
The earlier treatment also followed insurance into the decision to consume after a loss. A person facing a lower charge at the point of use may buy more of a covered service. That response is different from taking fewer precautions, though both can change expected claims. Its welfare meaning needs another comparison. A marginal service may cost more resources than the value it creates, but a service previously forgone because the patient could not finance it may be worth more than its cost. One cannot identify the waste by subtracting uninsured use from insured use. The uninsured quantity is an observed choice under particular prices, resources and risks; it is not, by its mere existence, the right quantity for a social account. A co-payment that reduces use must still be judged by what use disappears.
The name can lead the argument astray. An owner who reduces an expensive precaution after its financial reward has changed need not have lied, broken a promise, or become a different person. An employee paid a salary may still work from duty, pride, interest, or concern for a continuing relationship. Those motives belong in the account of what makes an action attractive. If the precaution protects something insurance cannot replace, such as health or an irreplaceable possession, full financial coverage leaves that reason for care intact. The model isolates a displaced incentive; it does not establish that money exhausts motivation, or that every covered person responds in the same direction by the same amount.
The question is which consequence the bargain leaves with the person who can change it, and what it costs him to carry that consequence.
The inquiry also runs in the other direction. A seller who promises to repair a failure gains a reason to prevent it; a contractor who bears the cost of correction gains a reason to avoid a defect. Moving risk can strengthen an incentive when the risk moves toward the party who controls it. This is why “someone else pays” is an incomplete diagnosis. One must identify the action, its owner, the relevant margin, and the person who can prevent the harm most effectively. A contract can improve one party’s incentive while weakening another’s. A warranty may sharpen the maker’s attention to workmanship while dulling the user’s attention to maintenance. The same promise can contain both effects.
The unwatched remain governed by the rewards, obligations, information and relationships that survive being unwatched. Moral hazard names a difficulty in composing those conditions. It does not make risk-sharing a mistake or complete exposure a cure. The useful contract preserves reasons to take care while placing unavoidable misfortune where it can be borne, and it recognizes the occasions on which better evidence can spare both parties a costly guess. When the action is productive effort and the uncertain result is the only readily measured return, that composition becomes the problem of pay. The next plate asks how to reward the work without paying for every favorable accident or charging the worker for every storm.