Afriat’s theorem

Overview

Afriat’s theorem is a cornerstone of revealed preference theory. It provides a practical, nonparametric test of whether a finite set of observed consumer choices can be explained by utility maximization, without assuming any particular functional form for preferences. If the data pass the test, the theorem guarantees that there exists a continuous, monotonic, concave utility function that could have generated those choices; if they fail, the choices are inconsistent with standard consumer theory. This matters for executives because it offers a way to validate demand assumptions, build robust bounds on willingness to pay and cost-of-living changes, and analyze pricing experiments using only observed prices and quantities, not detailed psychological models of utility.

Origins and Credit

Paul Samuelson (1938) introduced revealed preference—the idea of inferring preferences from choices. Hendrik Houthakker (1950) added consistency conditions (SARP). Sydney Afriat’s 1967 paper supplied the decisive constructive result now known as Afriat’s theorem, showing when and how a utility function can be built from finite expenditure data. Hal Varian (1982) popularized computational tests and clarified the equivalence with the Generalized Axiom of Revealed Preference (GARP), making the theory operational for applied work.

Core Idea and Mechanics

The data consist of observations t = 1,…,T. In each observation, the consumer faces prices p^t for a vector of goods and chooses bundle x^t, spending p^t·x^t. The budget set at t is all bundles x with p^t·x ≤ p^t·x^t (no separate income is needed; observed expenditure defines affordability).

Revealed preference and GARP. Bundle x^s is directly revealed preferred to x^t if x^s was chosen when x^t was affordable (p^s·x^t ≤ p^s·x^s). One gets indirectly revealed preference by chaining such relations across observations. The Generalized Axiom of Revealed Preference (GARP) requires: if x^s is indirectly revealed preferred to x^t, then x^t cannot be strictly cheaper than x^s at prices p^t. Intuitively, one cannot observe a cycle in which someone repeatedly chooses a bundle that is worse than an affordable alternative.

Afriat’s equivalences. Afriat’s theorem states that the following are equivalent for a finite dataset:

• The data satisfy GARP.
• There exists a continuous, monotonic, concave utility function U(·) such that each x^t solves “maximize U(x) subject to p^t·x ≤ p^t·x^t.”
• There exist numbers U_t and λ_t > 0 (one pair per observation) satisfying the Afriat inequalities for all s,t: U_t ≤ U_s + λ_s · p^s·(x^t − x^s).

If the inequalities are feasible (a linear programming check), one can explicitly construct a piecewise-linear, concave utility function that rationalizes the choices. Passing the test does not identify preferences uniquely; it certifies that some well-behaved utility exists consistent with the data.

Measuring “how close” when violations occur. Real data are noisy. Varian’s critical cost efficiency index (CCEI) scales each budget inward just enough to restore GARP. A CCEI of 0.95, for example, means a 5% tightening of budgets makes the data rationalizable—useful for distinguishing small measurement errors from substantive irrationality.

Key Assumptions and Conditions

• Choices are observed for a finite number of price regimes; each observation reflects a static decision.
• Preferences are monotonic (more of any good is weakly better) and can be represented by a continuous, concave utility function.
• No rationing or quantity constraints; the observed bundle is chosen from the full affordable set.
• Prices and chosen quantities are measured correctly; all relevant goods are included or an “outside good” is consistently treated.
• Price-taking behavior; no strategic or dynamic considerations within an observation.

Implications

• Model-free test of rationality: Firms and policymakers can validate whether observed demand patterns are consistent with utility maximization without committing to a specific functional form.
• Constructive recovery of preferences: When data pass, one can build a rationalizing utility and derive bounds on Hicksian and Marshallian demand responses, willingness to pay, and cost-of-living changes.
• Welfare and policy bounds: Nonparametric methods deliver tight, assumption-light bounds on compensating or equivalent variation for price changes or taxes, crucial in regulatory and antitrust reviews.
• Diagnostics for experiments: A/B pricing tests can be screened for revealed-preference consistency; large GARP violations may signal inventory behavior, stockouts, or measurement error.
• Heterogeneity-friendly: Applied to individual-level panels, revealed preference avoids aggregation biases that can distort parametric demand estimates.

Example in Practice

Retail pricing analytics. A grocery chain runs quarterly price experiments across categories and tracks a panel of households’ weekly baskets. For a given household, each week provides prices p^t and a chosen bundle x^t (quantities by category). Analysts apply Afriat’s test to the household’s set of observations.

Suppose the data pass GARP with a CCEI of 0.98. The team then uses the constructed piecewise-linear utility to bound the household’s cost-of-living change when the chain considers a 10% increase in snack prices and a 5% decrease in produce prices. The nonparametric bounds show the household’s compensating variation lies between $1.10 and $1.60 per week, reflecting substitution toward produce. Aggregating across households (carefully, to respect heterogeneity) yields a distribution of welfare impacts. Merchandising can use these bounds to anticipate churn risk and to design targeted promotions that offset losses for the most adversely affected segments.

If instead the data show large GARP violations, the chain investigates inventory behavior (stockpiling during promotions), coupon application not captured in posted prices, or stockouts. Adjusting for these frictions often restores consistency or clarifies when the standard static model is inappropriate, guiding better experiment design.

Limitations and Common Misunderstandings

• Passing the test is not identification: Many utilities can rationalize the same choices; Afriat’s theorem guarantees existence, not uniqueness or precise elasticities.
• Static and frictionless baseline: Inventory, dynamics, learning, and uncertainty can cause apparent violations even when behavior is sensible; specialized extensions are needed.
• Data requirements: Tests need accurate prices and quantities across multiple regimes for the same decision-maker; aggregated market data can fail GARP due to heterogeneity (the “representative agent” may not exist).
• Scope of prediction: Nonparametric bounds can be wide, especially outside the convex hull of observed price vectors; for forecasting far out of sample, parametric structure may be necessary.
• All-goods issue: Omitting relevant goods (or mismeasuring coupons, quality, or package size) can generate spurious violations; careful data cleaning is essential.

Related Concepts

Revealed preference; GARP; WARP and SARP; Varian’s nonparametric test; Critical cost efficiency index (CCEI); Expenditure minimization and Hicksian demand.