GARP (generalized axiom of revealed preference)

Overview

The generalized axiom of revealed preference (GARP) is a consistency test for observed choices. It asks whether a decision maker’s purchases across different price and income situations could have come from maximizing a stable preference (or utility) subject to a budget. If a dataset of purchases satisfies GARP, then there exists a well-behaved utility function that could rationalize those choices. If it fails, the choices display a “preference cycle” that is inconsistent with standard consumer theory.

GARP matters because it lets analysts assess rationality and discipline demand analysis without assuming a specific utility function or estimating elasticities. It underpins nonparametric tools used to audit procurement practices, evaluate consumer behavior, and bound welfare impacts and cost-of-living changes.

Origins and Credit

Revealed preference was introduced by Paul Samuelson in the late 1930s as a way to infer preferences from choices. Hendrik Houthakker later proposed the Strong Axiom of Revealed Preference (SARP). The modern, practical foundation came with Sidney Afriat (1967), who proved that certain inequalities are both necessary and sufficient for utility maximization with well-behaved preferences (Afriat’s Theorem). Hal Varian (1982) popularized the nonparametric approach and formalized the use of GARP as the testable condition equivalent to Afriat’s inequalities for finite datasets.

Core Idea and Mechanics

Consider observations t = 1, 2, …, T. At observation t, prices are p^t and the chosen bundle of goods is x^t. A bundle y is affordable at t if its cost at those prices does not exceed the cost of what was chosen: p^t·y ≤ p^t·x^t. The choice x^t is said to be directly revealed preferred to y if y was affordable but not chosen; equivalently, p^t·x^t ≥ p^t·y.

Build the revealed preference relation by chaining these direct comparisons: if x^t is directly revealed preferred to some y, and y is directly revealed preferred to z, then x^t is (indirectly) revealed preferred to z.

GARP requires: if x^t is revealed preferred (directly or indirectly) to x^s, then x^s must not be strictly cheaper than x^t at prices p^s. Formally, it cannot be that p^s·x^s < p^s·x^t. Economically, there should be no cycle of choices that implies “x is at least as good as y” while, at the prices where y was actually purchased, y would have cost strictly less than x—an inconsistency with utility maximization.

Afriat’s Theorem links this test to optimization: a finite dataset satisfies GARP if and only if it can be rationalized by a monotone (more is not worse), locally non-satiated (slightly more of some good is strictly better), and concave (exhibits diminishing marginal rates of substitution) utility function. No functional-form assumptions are required.

Key Assumptions and Conditions

• Stable preferences: No shift in tastes across observations.
• Local non-satiation: For any bundle, a slightly larger amount of at least one good is strictly preferred.
• Linear prices and budgets: No quantity discounts, fixed fees, or nonlinear tariffs unless modeled explicitly.
• Feasible choice sets: Observed bundles are chosen from the full budget set; no unobserved rationing or binding constraints beyond the budget.
• Accurate data: Correct prices and quantities; finite number of observations.

Implications

• Test of rationality: Passing GARP indicates choices are consistent with utility maximization; failing suggests inconsistency relative to the standard static model.
• Nonparametric demand analysis: Enables bounds on substitution patterns and counterfactual choices without specifying a utility form.
• Welfare bounds and price indices: Supports tight bounds on consumer welfare changes and cost-of-living indices when prices change.
• Diagnostics and benchmarking: The Afriat Efficiency Index quantifies the minimal expenditure reduction needed to eliminate violations, useful for auditing and performance measurement.

Example in Practice

Suppose a company’s facilities team purchases monthly bundles of cleaning supplies under different supplier price lists. For three months, prices vary and the team buys different mixes of disinfectant and paper products. An analyst checks, for each month, which alternative bundles were affordable at that month’s prices. If the bundle chosen in Month 1 is revealed preferred to an affordable alternative (say, a “balanced” mix), and Month 2’s choice is revealed preferred to another feasible mix, and this chain ultimately implies that Month 3’s chosen bundle was affordable—but at Month 3’s prices that earlier bundle would have cost strictly less—then the data violate GARP.

What does that mean operationally? It does not prove waste or error, but it indicates that the observed sequence of purchases cannot be rationalized by a single, stable objective of “getting the most from the budget” over the period, given the measured prices. Possible responses include:

• Investigate unobserved constraints (storage limits, minimum order quantities, contractual obligations).
• Check for data issues (misrecorded prices or quantities).
• If no explanation arises, consider revising procurement guidelines; compute the Afriat Efficiency Index to estimate potential cost savings.

Limitations and Common Misunderstandings

• Sufficiency, not uniqueness: Passing GARP guarantees at least one rationalizing utility exists, but it does not identify it or provide elasticities.
• Noise sensitivity: Small measurement errors can create apparent violations; robust or “perturbed” tests may be needed.
• Model scope: GARP is a static, single-period test with linear prices. Inventory, dynamics, uncertainty, and nonlinear pricing schemes require extensions.
• Preference shifts: If tastes change across observations, GARP may fail even when each month’s choice is reasonable at the time.
• Not a proof of optimality: Passing GARP does not show behavior is “optimal” in a broader sense—only that it is consistent with some well-behaved utility under the assumed setting.

Related Concepts (names only)

Revealed Preference; Weak Axiom of Revealed Preference (WARP); Strong Axiom of Revealed Preference (SARP); Afriat’s Theorem; Afriat Efficiency Index; Nonparametric Demand Analysis.