The Revelation

The Axioms put a ranking behind the choice; the next ambition was to recover the ranking from the choice in front of us. The earlier Collected plate, The Inference, turned the direction of explanation around: instead of entering the mind to discover what a person preferred, watch what he takes when he could have taken something else. The movement has an austere appeal. A purchase is a deed, a price can be recorded, a forgone alternative can be inspected. Yet the last of these matters as much as the first. The basket alone says little about the baskets the buyer rejected. To read a preference from a purchase, one must know what was available, on what terms, and what remained the same between purchases. The invisible has acquired an observable test. It has not disappeared from the assumptions that make the test possible.

Revelation begins with an affordable alternative, and even then its first word is modest. If a person selects A while B is available within the same budget, a utility-maximizing account must rank A at least as high as B. At least: the act alone need not establish strict preference, because the person might be indifferent. A later choice of B from a budget that also contains A is therefore not automatically a contradiction. Prices and strict inequalities decide what the record excludes. Nor does choosing a smaller house after a fall in income reveal a newly discovered love of confinement if the larger house is no longer affordable. The theory must carry the opportunity set beside the deed. Without it, necessity can be mistaken for taste, and a movement of the boundary can be mistaken for a movement of the mind.¹

A small invented market gives the test its teeth. There are two divisible goods, and the shopper first buys two units of the first and none of the second. Their prices are two and one, so the chosen basket costs four while two units of the second good cost only two. At the next observation the prices exchange places, the shopper spends four again, and buys two units of the second good; the old basket now costs only two. Each rejected basket lay strictly inside the budget on which the other was selected. Under local nonsatiation, a little room around any basket contains something better; consequently an interior basket cannot tie for the maximum of that budget. A stable ranking must therefore put each chosen basket strictly above the other. It cannot. The failure belongs to that joint account of choices, budgets, goods and stable tastes. It is a substantive failure, not a definition arranged to win.²

For a finite record of purchases, the general condition follows the chains. If A is revealed at least as good as B, perhaps through intervening purchases, B cannot be chosen on a budget with A strictly inside it. This is the Generalized Axiom of Revealed Preference. Afriat's theorem makes its force exact: for the usual linear budgets with positive prices and nonnegative quantities, the condition is equivalent to the existence of a locally nonsatiated utility function rationalizing the observations. Indeed, a continuous, concave, monotonic function can be constructed. The theorem supplies a ranking that could have produced the record without requiring the investigator to guess its formula first. Local nonsatiation does essential work here. A constant utility would call every basket equally good and could accommodate every purchase, making the exercise empty. The theorem's achievement is to retain a restriction that behavior can violate while showing exactly when a coherent representation can be built.³

Existence, however, leaves more than one possible inhabitant behind the door. Suppose, in another invented example, both goods cost one and a budget of two buys two units of the first. The utility formulas 2x + y and 4x + y each select precisely that basket. Now let the first good cost three, the second one, and give the shopper a budget of three. The first formula selects three units of the second good; the second selects one unit of the first. Both accounts fit the observation, yet they disagree about an unobserved opportunity. These are different rankings, not merely different numerical labels for the same ranking. Additional purchases can narrow the disagreement, sometimes decisively. But a passed test does not reconstruct every comparison without remainder. It establishes what the data permit; recovery of the whole order is another claim.⁴

Even a fully specified ranking would leave the earlier plate's hardest question standing: why should the utility that explains a purchase measure the purchaser's good? A person may choose under a mistaken belief about the bill, or may later reject the very conditions under which the choice was made. The mathematical representation alone does not decide which judgment should govern. One can deliberately make respect for choice the basis of welfare analysis, rather than treat choice as a window into an independently measured happiness. Bernheim and Rangel develop such an approach, retaining comparisons on which choice gives unambiguous guidance and allowing ambiguity where choices conflict. But the domain of choices taken as authoritative must still be specified; excluding a confused decision requires a reason. Their construction shows why the alternatives are richer than either declaring every purchase good by definition or declaring behavior useless. The bridge from deed to welfare is something an analyst builds and must defend.⁵

The remainder is therefore neither an embarrassment to conceal nor a license to abandon the record. Revealed preference can expose a contradiction that introspection evades; it can tell us which comparisons survive every admissible reconstruction and which depend on choices not yet observed. What it cannot do merely by succeeding is turn one permitted ranking into the uniquely recovered mind, or turn that mind's choices into a completed account of its interest. The original ambition was to read wanting backward from doing. The durable achievement is more exact: to learn what doing, under stated opportunities and assumptions, obliges us to believe. Beyond that boundary lie further evidence and further judgment. The purchase is a beginning of knowledge. Calling it revelation should not make the remainder sacred.