A man may own the whole harvest while his neighbor goes without, and the allocation may be efficient. The sentence sounds like an accusation against economics. It is, more exactly, a question about what economics has certified. If giving the neighbor grain requires taking grain from its owner, and the owner prefers to keep it, the transfer cannot make someone better off without making someone else worse off. That is the test the allocation has passed. Its passing tells us nothing about the origin of the owner's claim, the urgency of the neighbor's need, or whether the loss imposed by sharing would be justified. A word that sounds like praise has performed a much narrower task.
Pareto efficiency is powerful partly because it asks so little of interpersonal comparison. It compares each person with herself in another arrangement. If somebody can be made better off while nobody is made worse off, an improvement is available; if no feasible improvement of that kind remains, the allocation is efficient. There is no summing of pleasures and no weighing of one person's satisfaction against another's. The gulf between minds can be left uncrossed. The price of that restraint appears when interests conflict. The test cannot say that a large gain to one person warrants a small loss to another, however persuasive the case may be on other grounds. It has excluded precisely the comparison that could decide.
Consider an invented economy stripped down to two people and one hundred divisible units of food. There is no production or spoilage, and each person cares only about her own consumption and strictly prefers more. If forty units are given to each while twenty are wasted, fifty apiece improves both lives. But once all one hundred units are allocated, every division is efficient. One hundred and zero passes the test; fifty and fifty passes it; ninety-nine and one passes it. To increase either person's share is to reduce the other's. The result follows from the assumptions, including the absence of satiation and concern for the other's condition. If the owner would welcome a gift, or food could be produced without anyone's sacrifice, the analysis would change. Efficiency is a conclusion about a specified feasible world, not a blessing conferred on possession wherever it is found.
The first welfare theorem gives this criterion its famous connection to markets. In the standard private-goods model, people choose their preferred affordable bundles, firms maximize profit at prices they take as given, and markets clear. With the relevant regularity conditions, including local nonsatiation of preferences, the resulting competitive equilibrium is Pareto efficient. The shared prices do an extraordinary piece of coordination. A bundle strictly preferred to the one a consumer has chosen must cost more than her budget allows. Under the theorem's conditions, a proposed improvement for everyone weakly and someone strictly would require aggregate resources of greater value than the economy can supply. Each private choice helps establish the impossibility of that collective improvement. Nobody has to decide how many units of one consumer's delight equal a unit of another's.
The budgets remain in the proof from beginning to end. A person chooses what she most wants among what she can afford, and what she can afford depends on her claims to goods, labor income, and profits. The theorem does not establish those claims. In the food economy, a positive price and initial holdings of one hundred and zero sustain the same final allocation: the first person can afford all the food, the second none. Redistribute the initial holdings equally and the equal allocation can also be sustained. Both are competitive in this deliberately spare example; neither is selected over the other by efficiency. A price is a verdict on a hidden contest, mistaken for a fact of nature. Here the hidden contest concerns the means with which people arrive at the market, and the proof faithfully carries its result forward.
The strongest defense of the theorem accepts this limit and insists that its achievement remains real. Waste has victims. A feasible change that helps someone and hurts nobody deserves attention, and institutions that reveal and realize such changes can greatly enlarge what people can do. To say that efficiency does not settle justice is no reason to become careless about the resources justice will have to distribute. Equally, a proposal advertised as a Pareto improvement must count everyone affected. The gain to buyer and seller is not enough if their bargain sends smoke into a neighbor's home. An apparently free improvement can become costly when the excluded person is brought back into the account. The theorem's discipline concerns both the opportunities it reveals and the people a careless application leaves out.
Its second defense is the second welfare theorem. Under additional conditions, notably suitable convexity and the regularity needed for a supporting competitive equilibrium, an efficient allocation chosen on distributive grounds can be sustained by prices after the appropriate redistribution of purchasing power. The mathematical idea is that supporting prices can separate the chosen allocation from infeasible improvements and make individual choices consistent with it. Distribution need not be whatever the unaltered starting claims happen to produce. It can enter as a social choice before decentralized exchange does its work. This is a possibility result with substantial premises. A supporting price system is not a procedure that discovers the just allocation, and the required transfers are part of the construction rather than a gift the market makes afterward.
Between that construction and an actual reform lie the institutions that make transfers possible. A grant fixed independently of subsequent conduct differs from a tax whose amount changes with earnings or a subsidy whose receipt depends on a purchase. The latter can alter the choices they accompany. Administration consumes resources; eligibility rules can exclude people; information can be costly to obtain. These are reasons to investigate the whole proposal, not to declare redistribution inadmissible. Initial property claims also have institutions behind them. The second theorem makes the separation between distribution and allocation intelligible under its assumptions. It does not promise that every practical instrument can carry it out without consequence, nor that the existing instrument has no consequence of its own.
Actual markets can also fail the first theorem's conditions before distribution is even considered. An important risk may lack a contract, a consequence may fall on somebody outside the bargain, or a seller may be able to move the price it is supposed merely to accept. Then the certificate cannot simply be applied. Yet a missing certificate is not a proof that any proposed substitute will work. We must examine the particular failure and the alternatives available. This is what a benchmark is for: to make the comparison more exact. Used as a universal declaration that the market outcome is good, it conceals the assumptions on which its strength depends; used as a universal declaration that markets are useless because they are imperfect, it loses the same strength from the other side.
What, then, is efficiency worth as a measure of the good? It identifies one important consideration without exhausting the considerations that matter. It can show that an opportunity has been missed, and it can expose a claim of necessity that is false. It cannot rank all efficient allocations or tell us which loss it is right to impose when gains and losses conflict. Even the Pareto principle, attractive as it is, presupposes that the preferences being compared are appropriate evidence for the judgment in question; a social assessment may also concern rights, opportunities, or deprivation that those preferences incompletely register. The good does not become immeasurable merely because this measure cannot encompass it. We have reached the boundary of a criterion, not the boundary of reason.
The absence of an unopposed improvement leaves the contested improvements still to be judged.
The man who owns the harvest may therefore possess an efficient allocation and an indefensible share. Calling it efficient neither establishes nor defeats the second charge. The case for giving food to his neighbor must say why the neighbor's gain, claim, or need warrants the owner's loss; it need not pretend that the owner loses nothing. That is a more demanding argument than invoking a theorem, and a more honest one. The next plate takes up the most familiar attempt to make it: the thought that the loss of a dollar to abundance is smaller than its gain to want, and that equality might follow from the arithmetic of diminishing satisfaction.