Afriat's Theorem
Afriat's theorem is a foundational result in microeconomic revealed preference theory proving that a finite dataset of prices and chosen bundles is rationalizable by a well-behaved utility function if and only if it satisfies GARP.
Afriat's theorem—proved by economist Sydney N. Afriat in 1967 and refined by Hal Varian in 1982—is the central constructive theorem of non-parametric revealed preference theory. It answers a fundamental empirical question: given a finite dataset of observed market prices pt and chosen consumption bundles xt across t = 1, ..., T observations, when is the observed behavior consistent with a consumer maximizing a continuous, locally non-satiated utility function subject to a linear budget constraint? Afriat proved that four seemingly distinct conditions—ranging from finite combinatorial axioms to concave utility construction—are mathematically equivalent.
The Four Equivalent Conditions of Afriat's Theorem
For any finite dataset {(pt, xt)}t=1T with pt ∈ R++n and xt ∈ R+n, the following four statements are equivalent:
- Rationalizability by a Non-Satiated Utility Function: There exists a locally non-satiated utility function U(x) such that for every observation t, *U(xt) ≥ U(x)* for all x satisfying pt · x ≤ pt · xt.
- Generalized Axiom of Revealed Preference (GARP): There is no cycle of revealed preferences containing a strict revealed preference. Formally, if xt is indirectly revealed preferred to xs (written xt R xs via a chain of budget-feasible choices pt · xt ≥ pt · xk, ..., ≥ pr · xs), then it cannot be that ps · xs > ps · xt (xs is formally preferred over xt).
- Existence of Afriat Numbers (Linear Inequalities): There exist formally positive real numbers Ut and λt > 0 (representing utility levels and marginal utilities of income) for t = 1, ..., T satisfying the Afriat inequalities:
*Us ≤ Ut + λt pt · (xs − xt)* for all s, t ∈ {1, ..., T}.
- Rationalizability by a Well-Behaved Concave Utility Function: There exists a continuous, formally increasing, and concave utility function U(x) that rationalizes the entire dataset, explicitly constructed as the lower envelope of hyperplanes:
*U(x) = mint [Ut + λt pt · (x − xt)]*.
Significance: No Continuity or Strict Concavity Testable in Finite Data
Afriat's theorem carries a deep methodological implication for econometrics: on any finite dataset, concavity, continuity, and strict monotonicity impose zero additional testable restrictions beyond basic utility maximization (GARP). Whenever a consumer's choices satisfy GARP, one can always construct a piecewise-linear, formally increasing, concave utility function that explains every choice without assuming parametric forms (such as Cobb–Douglas, CES, or translog).
Moreover, condition (3) reduces testing consumer rationality to a linear programming feasibility problem (or graph shortest-path / Floyd–Warshall cycle check for GARP in O(T3) time), enabling precise empirical testing on household scanner panels and laboratory choice experiments.
Extensions: Afriat Efficiency Index and Collective Household Models
Because real experimental and household datasets occasionally exhibit minor GARP violations from small optimization mistakes or measurement noise, Afriat (1973) introduced the Key Cost Efficiency Index (CCEI)—the largest scalar e ∈ [0, 1] such that shrinking every budget line to *e (pt · xt)* eliminates all GARP cycles—measuring how close a decision-maker is to precise rationality. Modern revealed-preference research extends Afriat's theorem to collective household bargaining, intertemporal discounting, revealed stochastic choice, and production theory.
Computational Complexity and Empirical Applications
Implementing Afriat's theorem on real consumption panels involves constructing the binary direct revealed preference matrix R0 (where R0st = 1 if ps · xs ≥ ps · xt), computing its transitive closure R via the Floyd–Warshall algorithm in O(T3) operations, and verifying that no pair (s, t) with s R t also satisfies strict affordability pt · xt > pt · xs. Because household datasets often contain T = 10 to 100 observations per household, this combinatorial check runs in milliseconds.
Applied economists use GARP and the Key Cost Efficiency Index (CCEI)—benchmarked against Bronars' power test (which measures how often random budget-line choices would accidentally pass GARP)—to evaluate rationality in laboratory dictator games, childhood decision-making studies, and scanner-data inflation index construction.
Recent algorithmic refinements extend Afriat's linear-programming characterization to incomplete price–quantity panels, quasilinear utility tests, and stochastic choice datasets, broadening non-parametric rationality testing across behavioral economics and industrial organization.
Frequently Asked Questions
How does GARP differ from SARP and WARP?
WARP (Weak Axiom of Revealed Preference) checks only pairs of observations, whereas SARP (Strong Axiom) forbids any revealed-preference cycle among distinct bundles (requiring unique demand). GARP (Generalized Axiom) allows flat indifference curves where two distinct bundles can be chosen at the same cost, forbidding only cycles containing a strict preference step.
Can Afriat's theorem reject utility maximization?
Yes. If a dataset exhibits a GARP violation—for example, bundle A was affordable when B was chosen with leftover money, yet B was affordable when A was chosen—no locally non-satiated utility function can rationalize those choices.
References
- Afriat, S. N. (2014). Revealed Preference Revealed. The Index Number Problem. https://doi.org/10.1093/acprof:oso/9780199670581.003.0012
- Afriat, S. N. (1987). Algebra of Revealed Preference. Logic of Choice and Economic Theory. https://doi.org/10.1093/0198284616.003.0008
- Freer, M., & Martinelli, C. (2021). A utility representation theorem for general revealed preference. Mathematical Social Sciences. https://doi.org/10.1016/j.mathsocsci.2021.03.018