Revealed Preference

Revealed preference is a foundational methodology in microeconomic consumer theory and business and economics theory that reconstructs a decision-maker's underlying preference ordering—and tests whether observed behavior is consistent with rational utility maximization—directly from observable market choices at varying prices and budgets. Introduced by American economist Paul A. Samuelson in his seminal 1938 article "A Note on the Pure Theory of Consumer's Behaviour" and expanded in 1948, revealed preference theory was originally conceived as a behaviorist alternative to ordinal utility and indifference-curve analysis, which relied on unobservable mental introspection. Through subsequent contributions by Hendrick S. Houthakker (1950), Sydney N. Afriat (1967), and Hal R. Varian (1982), revealed preference evolved from a philosophical critique of utility into the non-parametric empirical foundation of neoclassical demand theory, proving that consistency axioms on finite price–quantity data are logically equivalent to the existence of a well-behaved utility function.

Budget Sets and the Fundamental Observation Principle

Consider a consumer observed across T distinct market purchasing occasions, indexed by t, s ∈ {1, 2, …, T}. At observation t, the consumer faces a positive vector of K commodity prices p^t = (p₁^t, …, p_K^t) ≫ 0 and chooses a non-negative consumption bundle x^t = (x₁^t, …, x_K^t) ∈ ℝ₊^K, spending total expenditure m^t = p^t · x^t = ∑_{k = 1}^K p_k^t x_k^t.

Samuelson's core behavioral principle states that if the consumer purchases bundle x^t at prices p^t when an alternative bundle x^s costs no more than x^t at those same prices (p^t · x^s ≤ p^t · x^t), then the consumer has revealed a preference for x^t over x^s:

1. Direct Weak Revealed Preference (x^t R^D x^s): Bundle x^t is directly revealed preferred to bundle x^s if x^t was chosen when x^s was affordable:

1. Direct Strict Revealed Preference (x^t P^D x^s): Bundle x^t is directly revealed preferred in the strict sense to x^s if x^s was unambiguously cheaper than x^t at prices p^t (or if x^s ≠ x^t under single-valued demand functions):

1. Indirect (Transitive) Revealed Preference (x^t R x^s): The revealed preference relation R is the transitive closure of R^D; that is, x^t R x^s holds if there exists a finite chain of observations r₁, r₂, …, r_m such that:

The Weak Axiom of Revealed Preference (WARP)

Formulated by Samuelson (1938) for single-valued demand functions x(p, m), the Weak Axiom of Revealed Preference (WARP) requires pairwise consistency between any two distinct bundles x^t ≠ x^s: if x^t is directly revealed preferred to x^s, then x^s can never be directly revealed preferred to x^t:

Geometrically, WARP forbids two budget lines from crossing in such a way that each chosen bundle lies inside the affordable triangle of the other budget line. Moreover, WARP alone implies the Compensated Law of Demand and guarantees that the Slutsky substitution term in the Slutsky equation has non-positive diagonal entries: for any price change Δp = p^s − p^t that holds the initial bundle just affordable (p^s · x^t = m^s), WARP implies (p^s − p^t) · (x^s − x^t) ≤ 0, or Δp · Δx^comp ≤ 0.

Houthakker's Strong Axiom (SARP) and Varian's Generalized Axiom (GARP)

While WARP prevents 2-cycles (x¹ R^D x² R^D x¹), when there are K ≥ 3 commodities WARP does not rule out 3-cycles (x¹ R^D x² R^D x³ R^D x¹) and therefore fails to imply a symmetric Slutsky substitution matrix (as demonstrated by Gale's 1960 counterexample). In 1950, Hendrick Houthakker closed the integrability problem by introducing the Strong Axiom of Revealed Preference (SARP), which extends acyclicity to transitive chains of arbitrary length for single-valued demand:

To accommodate multi-valued demand correspondences (flat segments on indifference curves where distinct bundles yield equal utility), Hal Varian (1982) formulated the Generalized Axiom of Revealed Preference (GARP):

The Equivalence of GARP and Utility Maximization

A landmark question in empirical microeconomics is how to test whether a finite dataset D = {(p^t, x^t)}_{t = 1}^T was generated by a utility-maximizing consumer without imposing an arbitrary parametric functional form (such as Cobb–Douglas or CES). We say that a utility function U : ℝ₊^K → ℝ rationalizes the dataset D if, for every observation t ∈ {1, …, T}:

In 1967, Sydney Afriat proved Afriat's Theorem (later streamlined by Diewert, 1973, and Varian, 1982), establishing that for any finite dataset D = {(p^t, x^t)}_{t = 1}^T, the following four statements are logically equivalent:

1. The data can be rationalized by a locally non-satiated utility function U(x).
2. The data satisfy the Generalized Axiom of Revealed Preference (GARP).
3. There exist positive numbers (Afriat numbers) u^t > 0 (utility levels) and λ^t > 0 (marginal utilities of income) for t = 1, …, T that satisfy the system of linear Afriat inequalities:

1. The data can be rationalized by a continuous, monotonically increasing, concave, piecewise-linear utility function explicitly constructed from the Afriat numbers as the lower envelope of supporting hyperplanes:

Afriat's Theorem has two remarkable implications. First, the hypothesis of concavity (diminishing marginal rates of substitution) is empirically untestable on finite price–quantity data: if any locally non-satiated utility function rationalizes the data, then a concave utility function also rationalizes the same data. Second, testing GARP on T observations is computationally efficient in computation and algorithms: one simply constructs the T × T adjacency matrix of the direct relation R^D, computes its transitive closure R in O(T³) operations via Floyd–Warshall algorithm, and verifies that p^s · x^s ≤ p^s · x^t whenever x^t R x^s.

The Afriat Cost Efficiency Index (CCEI)

When real household panel data exhibit minor violations of GARP due to measurement error, inattention, or taste shocks, Afriat (1972) introduced the Cost Efficiency Index (CCEI) (also called Afriat's efficiency index), e ∈ [0, 1]. For a given efficiency level e ∈ (0, 1], the relaxed direct revealed preference relation R^D(e) is defined by e(p^t · x^t) ≥ p^t · x^s. The CCEI is the largest value of e ≤ 1 for which the transitive relation R(e) contains no GARP violations—measuring the smallest fraction (1 − e) of budget expenditure that would have been wasted if the consumer were a utility maximizer subject to optimization errors.

Why are WARP and SARP equivalent when there are only two goods, but different when there are three or more goods?

In a two-commodity consumption space (K = 2), any budget line p₁x₁ + p₂x₂ = m is a straight line in a two-dimensional plane, where the indifference curve's tangent line and the budget line have only one degree of rotational freedom (so curl vanishes identically and the integrability condition holds automatically). As Rolf Mantel and Hal Varian showed, in 2D space it is geometrically impossible to arrange three budget segments to form an intransitive 3-cycle (x¹ R^D x² R^D x³ P^D x¹) without at least one pair already violating 2-cycle WARP. In three or more dimensions (K ≥ 3), budget hyperplanes can twist around one another like a spiral staircase, satisfying WARP pairwise while violating transitivity (SARP).

How is revealed preference used to measure cost-of-living price indices?

Suppose a consumer purchases x⁰ at base-period prices p⁰ and x¹ at comparison-period prices p¹. If the Laspeyres price index L_P = (p¹ · x⁰) / (p⁰ · x⁰) is less than the expenditure ratio (p¹ · x¹) / (p⁰ · x⁰), then p¹ · x¹ > p¹ · x⁰, which means the base bundle x⁰ was unambiguously affordable in period 1 when x¹ was chosen (x¹ P^D x⁰)—proving non-parametrically that the consumer is unambiguously better off in period 1 than in period 0 (and bounding the true Konüs cost-of-living index derived from Shephard's lemma and general equilibrium theory).

1. Samuelson, P. A. (1938). "A Note on the Pure Theory of Consumer's Behaviour." Economica, 5(17), 61–71. https://doi.org/10.2307/2548836
2. Houthakker, H. S. (1950). "Revealed Preference and the Utility Function." Economica, 17(66), 159–174. https://doi.org/10.2307/2549382
3. Afriat, S. N. (1967). "The Construction of Utility Functions from Expenditure Data." International Economic Review, 8(1), 67–77. https://doi.org/10.2307/2525382
4. Varian, H. R. (1982). "The Nonparametric Approach to Demand Analysis." Econometrica, 50(4), 945–973. https://doi.org/10.2307/1912771
5. Chambers, C. P., & Echenique, F. (2016). Revealed Preference Theory. Cambridge University Press. https://doi.org/10.1017/CBO9781316104293