RESEARCH DOSSIER

Expenditure Minimization and Consumer Duality

Abstract. This research dossier develops consumer duality as the relationship between preference maximization subject to a budget and expenditure minimization subject to an attainable rank. The analysis derives the inverse value-function identities, Shephard's lemma, Roy's identity, compensated demand restrictions and Slutsky decomposition, distinguishing existence, uniqueness, differentiability and convex recovery. Constructed counterexamples establish the limits of these results. Exact monetary welfare, cost-of-living indexes, finite revealed-preference restrictions and econometric integrability are examined alongside empirical studies of British household expenditures and staple-price responses in China. The original analytical synthesis distinguishes formal validity, empirical identification and welfare interpretation, emphasizing commodity definitions, actual purchasing opportunities and the reference rank. The research uses literature and independently derived examples; it does not report new household microdata estimates or claim novelty for standard results.

Keywords: expenditure function; indirect utility; Hicksian demand; Marshallian demand; consumer duality; Shephard's lemma; Roy's identity; Slutsky equation; integrability; revealed preference; cost-of-living index; compensating variation; equivalent variation; demand systems; JEL D11

Object and historical lineage

The economic object

Expenditure minimization is the problem of finding the least monetary outlay capable of attaining a specified preference level at given commodity prices. Its economic counterpart, utility maximization, selects the highest attainable preference level at a specified monetary budget. Consumer duality is the correspondence between these two descriptions of an individual's opportunities and choices. The expenditure function expresses the monetary requirements of a preference ranking; indirect utility expresses the attainable rank of a monetary opportunity.

The importance of this construction extends well beyond a convenient change of variables. Ordinary market choices reveal responses to prices and resources jointly. The expenditure function separates a change in purchasing opportunities from a change in the relative terms on which goods can be exchanged. This separation supplies the logical foundation of compensated demand, exact cost-of-living comparisons, the restrictions imposed on estimated demand systems, and a large part of applied consumer welfare analysis. It does so conditionally: the price system, commodity space, information, preference ranking and feasible transactions have to be specified.

This dossier develops that conditional structure, investigates what follows mathematically and what requires data, and examines the institutional significance of its assumptions. Its evidence consists of historical records, theoretical articles and notes, selected empirical research, statistical documentation and independently constructed analytical examples. It does not estimate a new household demand system or claim a newly discovered theorem.

Historical development

Slutsky's 1915 study of the consumer budget is the historical point of departure for separating price effects. The original item is bibliographically verified; the historical interpretation is supported by Chipman and Lenfant's (2001) account of its rediscovery. Modern notation and derivations below are an exposition of the subsequent mathematical tradition rather than a claim that Slutsky used the same symbols or assumptions.

The twentieth-century reconstruction of ordinal demand made the consistency of feasible choices central. Samuelson (1938), in A Note on the Pure Theory of Consumer's Behaviour, developed restrictions on observed purchases and their feasibility at alternative budgets. The achievement was to give consumer theory empirical content beyond a chosen numerical utility representation. Hicksian compensation and the later revealed-preference approach share an interest in comparing alternatives within an ordering, but address different questions: one computes the cost of an attainable rank and the other investigates whether observed behavior admits an ordering.

Roy's (1947) La distribution du revenu entre les divers biens supplies the historical attribution for recovering ordinary demand from indirect utility. Shephard's (1970) Theory of Cost and Production Functions develops cost duality in production, building on his 1953 work. The consumer application has a preference-contour constraint in place of an output requirement. This analogy helps explain the common mathematical structure while preserving the economic distinction between the feasibility of a technology and the ranking of consumption.

The subsequent development moved from general restrictions toward estimation and identification. Flexible price-dependent cost functions made it possible to express demand shares in terms of a common expenditure representation. Revealed-preference inequalities supplied procedures that did not require an investigator to select one parametric shape in advance. Both routes remain relevant because the observable price and expenditure data rarely determine the full compensation function by themselves.

Research boundaries

The individual or explicitly modeled household is the analytical unit. Aggregate demand, general-equilibrium feedback, the theory of the firm and social welfare aggregation enter only where they expose the limits of individual consumer duality. The same canonical research object belongs to Microeconomics and Mathematical Economics. Its contribution is a synthesis of optimization, finite-data identification and welfare measurement, with examples designed to distinguish assumptions that are often conflated.

Definitions, assumptions, and institutional scope

Commodities and preferences

Let the commodity space be X = R^n_+, a finite-dimensional set of nonnegative quantities, indexed by economically comparable goods or services. A commodity includes its quality, delivery date and relevant conditions of acquisition. The price vector p lies in R^n_++ and the available budget m is nonnegative. All components are measured in a common monetary unit. The budget set is B(p,m) = {x in X : p·x ≤ m}. A preference relation is assumed complete, transitive and represented by a continuous real-valued utility function U. Local nonsatiation means that each bundle has arbitrarily nearby strictly preferred alternatives within X. Convex preferences mean that upper contour sets are convex; strict convexity of preferences strengthens the uniqueness result for ordinary demand under the usual budget conditions.

Utility is ordinal. If F is strictly increasing on the range of U, F(U(x)) describes exactly the same ranking. An expenditure target must be relabeled by F when its representation changes. The number attached to a utility level has no invariant unit of satisfaction. Monetary expenditure at a given rank, however, is invariant to the relabeling. This is a primary advantage of the dual representation.

A fixed preference ordering is a maintained restriction during a comparison. The availability of goods, household composition, location, access to infrastructure and information can be placed in a state variable z. When these change, the relevant object becomes U(x;z) and possibly X(z). Compensation for a change in p while z remains fixed differs from the compensation for changing the whole state. Price comparisons lose their intended interpretation if underlying quality or access changes unrecorded.

Institutional assumptions

The linear budget assumes a common per-unit marginal price over the feasible quantity range, divisibility, and freedom to select any bundle satisfying the constraint. Fixed charges, minimum orders, rationing, transaction taxes that vary by block, limits on discounts, credit ceilings and a household's inability to trade noncash benefits each modify the feasible set. A price posted online may be irrelevant for a household without the required membership or delivery access.

Money available for the comparison is also an economic choice. Current income, current expenditure, liquid resources and wealth are different objects. In a static demand model, m denotes the resources assigned to the specified commodity budget. Saving and borrowing have to be included as choices or accounted for before treating current income as m. If leisure or home production is excluded, an expenditure function conditional on those opportunities cannot measure all welfare effects of a policy that changes them.

Strictly positive prices and finite attainable targets permit compactness arguments. A zero-priced commodity can make a budget unbounded; an unattainable target gives infinite expenditure by convention. These cases can be treated with a suitably specified constrained model rather than silently importing results proved for positive prices.

Four distinct assumptions

Existence is a question about feasible sets and continuity. Equality of the two optimized descriptions is a question about binding resources and attainment of the relevant rank. Uniqueness is a question about preference geometry and the optimizer. Differentiability is a question about local regularity of value functions and their parameterization. One may hold when another fails. A corner bundle can be unique and yield a differentiable expenditure function; a continuous preference ranking can have a kink in that function at a price tie. A claim about Roy's identity therefore has a different logical burden from a claim about existence of an optimum.

The treatment of the expenditure function here uses linear prices and globally optimized choices as the benchmark. Institutional qualifications are part of the scientific object because a compensation estimate has meaning only relative to the opportunities included in the underlying problem.

Formal consumer duality

Optimization and value functions

Write ordinary or Marshallian demand as the set of maximizers M(p,m) of U(x) on B(p,m), and indirect utility as

v(p,m)=max⁡x≥0{U(x):p⋅x≤m}.v(p,m)=\max_{x\geq0}\{U(x):p\cdot x\leq m\}.

Write H(p,u) for the set of minimizers of expenditure at target u and e for the optimized value:

e(p,u)=min⁡x≥0{p⋅x:U(x)≥u}.e(p,u)=\min_{x\geq0}\{p\cdot x:U(x)\geq u\}.

When there is a single optimizer these correspondences are denoted x(p,m) and h(p,u). Use the correspondence terminology whenever an optimizer is not unique. Levin and Milgrom's (2004) Consumer Theory provides a modern mathematical reference for the standard identities and regularity conditions; the proofs and examples below develop the consequences in the notation of this dossier.

For the utility maximization problem, positive prices bound every quantity by m/p_i; the budget is closed and contains the origin. Continuity gives a maximum. For the expenditure problem, assume a feasible bundle y. Intersect the closed upper contour with the compact set p·x ≤ p·y. The intersection is nonempty and a minimizer of the continuous linear expenditure exists. Every feasible point outside that set is weakly more costly, so the minimum solves the original problem. This proof does not require convex preferences. Convexity is instead important for uniqueness, local optimization and recovery of the feasible contour from supporting linear price systems.

Binding constraints and inverse identities

Local nonsatiation implies budget exhaustion at any ordinary optimum: if expenditure were strictly below m, a small feasible neighborhood would contain a preferred bundle. For a target above U(0), continuity along the segment from zero to a positive expenditure minimizer implies its utility equals the target; a strictly preferred minimizer could otherwise be scaled toward zero until the target is reached at lower cost. The argument requires the commodity set to contain that segment, as X does.

Take u = v(p,m). An ordinary maximizer attains u at cost m, so e(p,u) ≤ m. A cheaper bundle with utility at least u would have slack within the original budget and could be improved through local nonsatiation, contradicting maximality. Consequently e(p,v(p,m)) = m. Conversely, for a feasible u ≥ U(0), a minimizing bundle attains u at its minimum cost, and a strictly preferred bundle within the same cost limit would contradict the scaling argument. Thus v(p,e(p,u)) = u. The equality at the origin follows directly. On this domain the two problems have the same argopt sets at matching parameters:

M(p,m)=H(p,v(p,m)),H(p,u)=M(p,e(p,u)).M(p,m)=H(p,v(p,m)),\qquad H(p,u)=M(p,e(p,u)).

At an unattainable target, e is infinite; at a target below U(0), minimal cost is zero and the inverse equality for the rank fails. If utility is satiated, ordinary expenditure need not exhaust m. It is important to distinguish these domain failures from the preference-convexity conditions used for computational methods.

Properties of expenditure

For a fixed feasible u, e is nonnegative, nondecreasing in each positive price, homogeneous of degree one in p, and concave in p. Homogeneity follows because all feasible bundles have their costs multiplied by the same positive scalar. Monotonicity follows from nonnegative quantities. For any p,q and t in [0,1], every target-feasible bundle has expenditure at tp+(1−t)q at least t e(p,u)+(1−t)e(q,u); minimization proves concavity. This argument requires no differentiability of preferences. A price inequality is therefore a stronger starting point than a system of derivative assumptions.

Expenditure is weakly increasing in u by nested upper contours; in the regular attainable domain above the initial rank it is strictly increasing because equality at two distinct targets would violate the binding utility constraint. For strictly positive prices its finite concave function is continuous on the open positive orthant. A finite e may be locally nondifferentiable when two or more distinct optimal bundles yield the same cost.

Indirect utility is homogeneous of degree zero jointly in p and m because the real budget set is unaltered by scaling both. It is nonincreasing in prices and increasing in m under local nonsatiation and the stated domain. Its lower contour at a fixed target can be described, where the inverse holds, by m ≤ e(p,u); the hypograph of a concave expenditure function is convex. This yields quasiconvexity of indirect utility jointly in prices and budget within that domain. The geometry of the dual objects differs: expenditure is concave in prices while indirect utility does not have to be concave in wealth.

Shephard's lemma

Let h be a minimizer at p. Because the same bundle is feasible at any q,

e(q,u)≤e(p,u)+(q−p)⋅h.e(q,u)\leq e(p,u)+(q-p)\cdot h.

Thus every minimizing bundle is a supergradient of e with respect to p. If e is differentiable at p, the supergradient is the gradient, and all minimizing bundles agree. Hence h_i(p,u) = ∂e(p,u)/∂p_i. This argument is an optimized-value proof without an interior first-order condition. At a kink it provides inequalities and a supergradient correspondence rather than a spurious point derivative. Applying Euler's identity at a differentiable price gives p·h = e, the expenditure accounting equation.

The reverse support argument is useful. For a closed, convex, comprehensive upper contour C(u) in X, a point outside can be separated from C(u) by a nonnegative supporting price vector. In the strictly positive price domain the separating vector may require a limiting argument when some components vanish. With continuity and appropriate closure, the family of inequalities p·x ≥ e(p,u) reconstructs C(u). For a nonconvex upper contour, these linear inequalities generally reconstruct a closed convex comprehensive hull, not every excluded interior point. Hence support-function recovery is stronger than the equality of optimized values at a given budget. Conflating the two is a common source of overly broad claims about duality.

Roy's identity and derivative regularity

Assume v is differentiable near the relevant (p,m), v_m > 0, and both inverse identities apply. Holding u constant and differentiating v(p,e(p,u)) = u yields v_{p_i} + v_m e_{p_i}=0 wherever the derivatives exist. Substituting the expenditure derivative and matching demands gives

xi(p,m)=−vpi(p,m)vm(p,m).x_i(p,m)=-\frac{v_{p_i}(p,m)}{v_m(p,m)}.

A direct envelope argument gives the same result: with a regular multiplier λ on the budget, v_m=λ and v_{p_i}=−λx_i. The inverse derivation clarifies the connection between the two identities. For a differentiable strictly increasing relabeling F with F'(v)>0, both derivatives multiply by F'(v) and the ratio is unchanged. If F'(v)=0 at a point, a numerical utility representation can create a zero denominator while preserving choice. Roy's ratio is then undefined at that point; the preference ordering itself has not failed.

The dual multiplier on the target constraint, when it exists under appropriate smoothness, is e_u. In a smooth inverse relationship e_u=1/v_m at matched rank and prices. Its units are monetary units per unit of an arbitrary utility scale. Its value is therefore representation-dependent even though fixed-rank compensation is not.

Demand restrictions and Slutsky decomposition

Comparative statics

For smooth single-valued demand, h(p,u)=x(p,e(p,u)). Differentiation with respect to p_j at fixed u gives the Slutsky relation

Sij(p,m)=∂hi∂pj(p,v(p,m))=∂xi∂pj(p,m)+xj(p,m)∂xi∂m(p,m).S_{ij}(p,m)=\frac{\partial h_i}{\partial p_j}(p,v(p,m)) =\frac{\partial x_i}{\partial p_j}(p,m)+x_j(p,m)\frac{\partial x_i}{\partial m}(p,m).

In particular ordinary own-price response equals compensated own-price response less the income effect x_i x_{i,m}. When e is twice continuously differentiable, S is its price Hessian. Consequently S is symmetric and negative semidefinite, and S p=0 by degree-one homogeneity of e. The last condition explains why the full substitution matrix cannot be strictly negative definite over all price directions: scaling every price has no effect on compensated quantities.

Budget exhaustion implies p·x=m, so p·x_m=1. Differentiating with respect to p_j at fixed m gives p·x_{p_j}=−x_j. Thus p^T S=0. Degree-zero joint homogeneity of x gives D_p x p + m x_m=0, yielding S p=0 directly. These are useful diagnostics for computational demand systems. A negative diagonal by itself is weaker than negative semidefiniteness, because violations can arise from cross-price terms. Testing eigenvalues on normalized price directions is more informative than inspecting only individual own-price elasticities.

Let w_j=p_j x_j/m be the budget share, η_i the expenditure elasticity and ε^M_{ij}, ε^H_{ij} price elasticities at positive demands. Multiplying the Slutsky equation by p_j/x_i yields ε^H_{ij}=ε^M_{ij}+w_j η_i. Symmetry requires w_i ε^H_{ij}=w_j ε^H_{ji}. Reciprocal compensated derivatives do not imply equal unweighted elasticities because units of goods and expenditure shares differ.

What the compensated law says

For a finite price change p to q at fixed rank, optimality of h(p,u) at p and h(q,u) at q yields

(q−p)⋅(h(q,u)−h(p,u))≤0.(q-p)\cdot(h(q,u)-h(p,u))\leq0.

It follows by adding the two inequalities that each chosen bundle is cheapest at its own prices. If only one price rises, its compensated quantity weakly falls. This finite law is well-defined even when derivatives fail, provided the same target and feasible set are used. Ordinary demand has no corresponding universal own-price sign because the target utility changes with purchasing power.

For a normal good, x_{i,m} ≥ 0, and a positive x_i combines with a nonpositive compensated price response to give a nonpositive ordinary own-price derivative. For an inferior good, x_{i,m}<0, the positive income-effect component can dominate. A Giffen response is a case where it does. Its existence is compatible with a negative semidefinite substitution matrix. Describing an upward-sloping ordinary demand as a violation of all optimization-based demand theory would discard the very decomposition that makes the response possible.

Finite versus differential compensation

For an infinitesimal price change dp, the extra budget needed at fixed rank is h·dp=x·dp at the starting point. For a finite shift, expenditure changes by the integrated compensated demands if differentiability holds along a path. A different construction holds the initial bundle affordable by giving the household (q−p)·x(p,m) in extra budget. This is Slutsky purchasing-power compensation in the finite-change sense. It generally exceeds the exact Hicksian expenditure difference for a price increase experiment because optimization at q can economize on the starting bundle. The two compensation rules agree to first order, but curvature matters for finite changes.

A linear approximation is not a global welfare estimate. When relative prices move sharply, or a household enters or exits a corner, the compensated path can differ substantially from the initial quantity. Empirical analyses should declare whether a derivative, finite substitution experiment or integral of a demand system is being used.

Analytical cases and counterexamples

Cobb–Douglas and exact numerical verification

Let U(x,y)=√(xy), p=(a,b), and m>0. Maximization gives x=m/(2a), y=m/(2b) and v=m/(2√(ab)). Expenditure minimization gives e(a,b,u)=2u√(ab), h_x=u√(b/a), h_y=u√(a/b). The formulas satisfy e(p,v)=m; p·h=e; partial derivatives of e give h; and differentiating v gives the same ordinary demand through Roy's identity.

Choose initial prices (1,1), budget 100, initial bundle (50,50) and rank u_0=50. Increase the first price to 4 while keeping the second at 1. The new ordinary bundle is (12.5,50), u_1=25. At the original rank the new compensated bundle is (25,100) and minimum expenditure is 200. At the new rank the minimum expenditure at old prices is 50. A compensation of 100 after the price change restores the initial rank; a removal of 50 before the change reaches the final rank. Maintaining the initial bundle at new prices would cost 250, requiring an extra 150. The 50-unit difference between this transfer and actual compensation measures the ability to change the consumption mix.

The base-quantity Laspeyres index is 250/100=2.5; the fixed-initial-rank exact index is 200/100=2. The Paasche index computed at the final observed bundle is 100/62.5=1.6. The final-rank exact index also equals 2 because preferences are homothetic. The example checks the inverse identities and bounds without relying on a fitted empirical coefficient.

At the initial point the derivative of h_x with respect to a is −25; that with respect to b is 25. The corresponding rows of the substitution matrix are (−25,25) and (25,−25), with eigenvalues 0 and −50. Ordinary own-price derivative of x is −50; x_m=1/2 and x=50, so S_xx=−50+25=−25. The numerical accounting tests all the key restrictions in a nontrivial two-dimensional instance.

Perfect substitutes and kinks

For U(x,y)=x+y, expenditure e(a,b,u)=u min(a,b). If a<b, the minimizing bundle is (u,0); if a>b, it is (0,u). At equality every nonnegative bundle summing to u is optimal. The value is concave, continuous and homogeneous of degree one. It has no single price gradient at the tie when u>0. A finite compensated comparison remains possible, and the supergradient inequality applies to each minimizing bundle. It is incorrect to replace all of these bundles by a single chosen derivative without specifying a tie-breaking assumption.

A statistical model approximating the kink by a smooth curve may be useful for prediction, but its derivative depends on the smoothing convention. Such a derivative should not be given the same interpretation as an exactly observed unique compensated quantity at the tie.

Perfect complements and corners

For U(x,y)=min(x,y), h=(u,u), e=(a+b)u, ordinary demand x=y=m/(a+b), and v=m/(a+b). The preference representation is nondifferentiable along the optimum, but e and v are smooth in positive prices and m. Every compensated derivative is zero because the target determines the quantities. This is the converse lesson of the substitute example: smooth preferences are not necessary for smooth optimized values.

For a corner with a smooth expenditure representation, take U(x,y)=x+ln(1+y), a,b>0. At a utility target u above ln(a/b), with a>b, the interior solution has y=a/b−1 and x=u−ln(a/b). At a low enough target where unconstrained x would be negative, the minimizer uses only y, y=exp(u)−1. When a≤b, spending on x alone is optimal (for u≥0). The value function is piecewise; first derivatives can agree across a boundary while second derivatives change. The example demonstrates why reported elasticities require local domain checks.

A satiation failure and a nonconvex failure

Let X=R_+, p=1 and U(x)=−(x−1)^2. At m=3, the optimal bundle is x=1; v=0 and e(1,0)=1 rather than 3. The expense minimization problem remains meaningful, but local nonsatiation fails at the bliss point, so expenditure and budget are no longer inverse. The example isolates a resource-exhaustion assumption.

Take U(x,y)=max(x,y) at target 1. The upper contour is the union of x≥1 and y≥1. Its closed convex comprehensive hull also contains (1/2,1/2), which attains only 1/2 in the original ordering. For p>0, e=min(p_x,p_y). Every linear price support inequality for the original contour is satisfied by the midpoint because (p_x+p_y)/2≥min(p_x,p_y). Therefore the complete collection of positive-price expenditures cannot distinguish that nonconvex exclusion from the hull. Convex recovery is not proved by merely exhibiting a dual value function.

Subsistence and income effects

A transparent subsistence specification sets U=(x−a)^α(y−b)^(1−α) on x≥a, y≥b, for a,b≥0 and 0<α<1. The minimum cost of meeting physical commitments is p_x a+p_y b. In the interior above that budget, the residual is allocated in fixed shares. Thus x=a+α(m−p_x a−p_y b)/p_x and y=b+(1−α)(m−p_x a−p_y b)/p_y. Both expenditure derivatives are positive. The model generates changing budget shares and increasing sensitivity near subsistence, but cannot generate an inferior good in that interior domain. Calling every subsistence model a Giffen model is therefore incorrect.

An illustrative local elasticity calculation separates logical possibility from empirical magnitude. Suppose a staple has share 0.7, expenditure elasticity −0.8, and compensated own-price elasticity −0.2 at an interior point. The Slutsky relation gives ordinary own-price elasticity −0.2−0.7(−0.8)=0.36. These numbers are deliberately hypothetical local elasticities, not a fitted global demand system or evidence from China. They show the combination that must be measured; integrability of any proposed complete system would still require the other restrictions.

Exact welfare and price indexes

Monetary valuation of changes

Let p^0,p^1 be two price regimes at the same nominal budget m and unchanged relevant state, with u^0=v(p^0,m), u^1=v(p^1,m). Adopt a loss-positive sign convention: compensating variation for a harmful price change is CV=e(p^1,u^0)−m. Equivalent variation is EV=m−e(p^0,u^1). A price increase in one or more goods has u^1≤u^0 under the benchmark assumptions, giving nonnegative CV and EV. For a beneficial change the signs reverse. Other conventions label gains positively; equations rather than names alone should identify the chosen convention.

CV measures a transfer at the new prices restoring the initial rank. EV measures money removed at the original prices to reach the final rank. Both use the same consumer's ranking, but different reference prices and reference ranks. Their difference is economically informative when the marginal monetary cost of a rank varies with its level and with prices. Neither measures utility in an interpersonal unit.

For a differentiable e, the exact change in expenditure holding u fixed along a smooth price path p(t) is the integral of h(p(t),u)·p'(t) over t. Since h is the gradient of a single potential e, the integral is path-independent within a domain where the same expenditure function applies; in the smooth case the symmetry of the Hessian is its local counterpart. Integrating ordinary demand at fixed budget instead generally changes the rank along the path. Ordinary consumer surplus is an exact money metric in special cases such as an interior quasilinear domain with a fixed numeraire price and relevant income effects absent. Elsewhere it is an approximation whose error has to be assessed.

In the Cobb–Douglas example, ordinary demand for the first good is 50/a. The fixed-budget area under that demand from a=1 to a=4 is 50 ln 4, about 69.31. This differs from CV=100 and EV=50. Thus even simple preferences admit materially different monetary measures of one price change. The numbers are consequences of the explicit model, not survey estimates.

Economic cost-of-living index

A fixed-rank cost-of-living index is I(u)=e(p^1,u)/e(p^0,u) for positive denominator. For an initial bundle x^0 that optimizes at rank u^0, it satisfies

I(u0)≤p1⋅x0p0⋅x0=L.I(u^0)\leq \frac{p^1\cdot x^0}{p^0\cdot x^0}=L.

The minimizing bundle at the new prices costs no more than the feasible old bundle. Similarly for an optimal final bundle x^1 at rank u^1,

P=p1⋅x1p0⋅x1≤I(u1).P=\frac{p^1\cdot x^1}{p^0\cdot x^1}\leq I(u^1).

These inequalities ordinarily concern distinct ranks. A single exact index is not in general bracketed by the two observable formulas unless extra assumptions connect the rank dependence. Under homothetic preferences one can choose a homogeneous representation with e(p,u)=u e(p,1) on positive ranks; then the ratio does not depend on u and P≤I≤L. The Fisher geometric mean √(LP) lies between the indexes when P≤L, but this property alone does not make it exact for every preference system.

The U.S. Bureau of Labor Statistics' CPI concepts handbook explicitly situates the CPI within a conditional cost-of-living objective. Its price index measures covered consumption price movements, not every environmental, time-use or public-service determinant of welfare. It also explains that price-index levels for areas normalized to a base period are not direct spatial comparisons of cost levels. These institutional distinctions are essential when translating theory into a published statistic.

New varieties and changes in access

When a good is introduced or vanishes, a price vector on an identical available commodity space may be an incomplete description. Feenstra's (1994) bibliographically verified study of new import varieties illustrates the relevance of product availability to international price measurement. The analysis of the present dossier is conditional on an explicit choice-set extension: assign an unavailable item zero feasible quantity or represent its absence by an appropriate limiting reservation price if the demand system justifies that representation. An arbitrary high imputed price is not a measurement of a reservation price.

Suppose two goods are perfect substitutes, the existing variety costs 3, and a newly accessible equivalent costs 2. At a fixed amount u of consumption, minimum cost falls from 3u to 2u. If the price collector tracks only the old variety, its observed price has not changed; the accessible cost of the target nevertheless falls by a third. Now suppose preferences assign a positive switching cost or value the old variety's unique features. Perfect substitution ceases to apply. The same new posted price may yield a smaller welfare improvement. This example isolates the identification issue behind new-product valuation without asserting a numerical finding from Feenstra's study.

Distribution and social evaluation

Money-metric welfare changes are not interpersonal utility comparisons by themselves. Two households with identical CV may have very different resources, opportunities and needs. Budget-share heterogeneity changes exposure to a price intervention even when sellers post identical prices. Distributional reporting should provide compensation relative to a declared income or expenditure base and identify the dispersion across relevant groups.

Summing household CV estimates produces a monetary total for the specified population under the model. Interpreting it as social welfare requires a normative weighting rule and the accounting of producer, taxpayer and public-budget effects when relevant. This dossier does not institute such a rule. The monetary representation provides a transparent input to that subsequent evaluation rather than resolving distributional justice within individual consumer theory.

Revealed preference and identification

Recovering preferences from finite observations

Given observations (p^t,x^t) with positive prices and observed expenditure m^t=p^t·x^t, bundle x^t is directly revealed weakly preferred to x^s if p^t·x^s≤m^t. A strict affordability inequality makes the revealed preference strict under local nonsatiation. Construct the transitive closure of the weak relations. The generalized axiom of revealed preference (GARP) forbids an observation x^t revealed weakly preferred through a chain to x^s when x^t was strictly cheaper than the chosen x^s at prices p^s.

Afriat's (1964; journal version 1967) construction of utility functions is the foundational reference. Its operative theorem is independently checked in Varian's (1982) nonparametric demand analysis. Under finite positive-price data, GARP is equivalent to rationalizability by a continuous, monotonic, concave utility function, and to feasibility of the inequalities

Us≤Ut+λtpt⋅(xs−xt),λt>0.U_s\leq U_t+\lambda_t p^t\cdot(x^s-x^t),\qquad \lambda_t>0.

Given feasible numbers, define U(x)=min_t{U_t+λ_t p^t·(x−x^t)}. This is concave as a minimum of affine functions, monotonic because the slopes are positive, and at each observed x^s it takes the assigned U_s by the inequalities and the s term. For each alternative affordable at p^t, its value is bounded above by U_t, because the t affine function has a nonpositive expenditure difference. Hence the observed bundle is optimal. The constructive proof explains why finite compatibility with utility maximization is a testable proposition and why it is an existential one.

A finite-data rejection example

At prices (2,1), suppose x^1=(2,1), costing 5. At prices (1,2), suppose x^2=(1,2), also costing 5. The second bundle costs 4 at the first prices; the first costs 4 at the second prices. The strict revealed choices therefore contradict one another. In the Afriat inequalities, U_2≤U_1−λ_1 and U_1≤U_2−λ_2 imply λ_1+λ_2≤0, impossible with positive multipliers. This diagnostic is independent of any choice of functional form or differentiability.

The result assumes that both observations describe the same relevant ordering and comparable constraints. If the individual moved between regions or faced quantity limits, the affordability matrix should reflect those states. The mathematical rejection is a rejection of the joint observational model; assigning it specifically to cognitive inconsistency requires additional evidence about choices and measurement.

Partial identification and extrapolation

Even when the inequalities are feasible, there are generally many possible utility extensions between and beyond observed bundles. These can yield different expenditure functions for previously unobserved relative prices. The data support a set of counterfactual quantities and compensation values; a unique parametric estimate is obtained only after additional restrictions or normalizations.

Blundell, Browning and Crawford (2003) use nonparametric Engel curves in an application to British household expenditures; their subsequent (2008) bounds on demand responses study exploits expansion paths and variation in relative prices. Such procedures make explicit the information supplied by observations and the range left uncertain. Their application does not eliminate the need for comparability across households or prices. Combining choice observations from distinct individuals requires justified restrictions on heterogeneity, or an explicitly modeled distribution of types.

Integrability of continuous demands

For a smooth ordinary demand defined on an open domain, Walras's law, homogeneity, symmetry and negative semidefiniteness of the associated Slutsky matrix are the characteristic local restrictions of the standard utility-maximizing model. Recovering a global preference ordering from a candidate field requires domain, continuity, regularity and global consistency conditions as well. A matrix passing pointwise symmetry at a handful of estimates does not itself prove a globally valid utility representation.

Formally an expenditure recovery problem can be posed as first-order differential equations ∂e/∂p_i=x_i(p,e), with an initial expenditure anchor at reference prices for each rank. Along a differentiable solution the cross-derivative condition is exactly Slutsky symmetry. It ensures local compatibility under suitable regularity. The stochastic uncertainty of estimated derivatives and the range of prices over which the model is valid determine whether such integration yields an empirically stable welfare estimate.

The use of observed market equilibrium quantities introduces another identification challenge. Market prices may respond to demand shocks. A fitted relationship between price and quantity is then not necessarily an individual compensated or uncompensated structural derivative. Data consistency, exclusion restrictions, experimental design, or justified instruments are needed to support causal counterfactuals.

Econometric demand systems

The expenditure function as a specification device

One route to coherent estimation begins with a parametric expenditure function and derives all quantities from it. Another estimates ordinary demands and imposes the derivative restrictions implied by a common ordering. The first approach makes internal consistency relatively transparent but still has to establish economic regularity over the relevant domain. The second may be convenient computationally, but integrating an inconsistent fitted system can produce path-dependent welfare estimates.

Deaton and Muellbauer's (1980) An Almost Ideal Demand System is a central example. Write its PIGLOG expenditure form as ln e(p,u)=A(p)+u B(p), where A(p)=ln a(p), B(p)=ln b(p)−ln a(p), a and b are positive homogeneous degree-one functions with b>a on the considered domain. Take

A(p)=α0+∑iαiln⁡pi+12∑i,jγijln⁡piln⁡pj,B(p)=β0∏ipiβi.A(p)=\alpha_0+\sum_i\alpha_i\ln p_i+\frac12\sum_{i,j}\gamma_{ij}\ln p_i\ln p_j, \qquad B(p)=\beta_0\prod_i p_i^{\beta_i}.

With symmetric γ, the compensated share is ∂ ln e/∂ ln p_i=α_i+Σ_j γ_ij ln p_j+u β_i B(p). Substituting u=(ln m−A(p))/B(p) gives ordinary shares

wi=αi+∑jγijln⁡pj+βi(ln⁡m−A(p)).w_i=\alpha_i+\sum_j\gamma_{ij}\ln p_j+\beta_i(\ln m-A(p)).

Restrictions Σ_i α_i=1, Σ_i β_i=0, Σ_i γ_ij=0 for each j, symmetry γ_ij=γ_ji and resulting row sums impose adding-up, homogeneity and symmetry. They do not automatically imply positivity of shares or concavity of e everywhere on the price domain. Regularity must be checked at observed and counterfactual values. A globally flexible logarithmic polynomial can fail outside the region of meaningful estimation.

The share derivative with respect to log expenditure is β_i. For positive shares, η_i=1+β_i/w_i by x_i=m w_i/p_i. The further price derivatives require accounting for A(p) at the relevant price. Using an approximate data-weighted price index in place of A(p) alters the empirical specification, so exact welfare properties of the theoretical cost function should not be ascribed without checking the approximation.

Engel curvature and household heterogeneity

The restriction to budget shares linear in log expenditure can be empirically costly. Banks, Blundell and Lewbel (1997) develop Quadratic Engel Curves and Consumer Demand with an integrable quadratic term in logarithmic expenditure. Their British Family Expenditure Survey application draws on 1970–1986 data and connects Engel curvature to welfare consequences of relative-price changes. The empirical lesson is that imposing overly simple income responses may change the pattern of estimated policy losses across the distribution.

In the classical Cobb–Douglas case, a household's constant expenditure shares leave proportional income changes unable to affect its composition. A nonhomothetic system permits budget composition to shift as expenditure changes. Households with children, older residents or different work patterns may also have different implicit price elasticities. A pooled parameter can conceal this variation. One investigator may have enough data to identify group-specific Engel curves but too little independent relative-price variation to estimate all cross-price derivatives precisely. These are different data requirements.

Estimation and validation

A credible study defines the goods, the household population, the observation unit, the resource measure, the price regime and the model's demographic shifters before estimation. Adding-up creates a singular error covariance for exhaustive budget shares; one share equation may be omitted during estimation and recovered through the restrictions. Zero expenditures may indicate censoring, infrequent purchases, unavailability, or corner choice. Each mechanism suggests different likelihood or moment conditions.

One should examine whether budget share estimates remain nonnegative, assess homogeneity and symmetry under alternative specifications, and inspect the normalized compensated substitution matrix. These checks address economic regularity. Out-of-sample fit, sensitivity to price measurement, cluster dependence, instruments, survey weights and confidence intervals address distinct empirical issues. An estimated system can satisfy every theoretical restriction and still forecast poorly; a predictive model can fit observed shares while giving invalid compensating variations.

When exact welfare is a central objective, resampling should propagate uncertainty in coefficients through the expenditure inversion. A policy simulation based on the point estimate alone hides the degree to which the compensation estimate depends on imprecisely measured curvature. If there are multiple acceptable models, report sensitivity across their admissible domains. Precision is an empirical property and cannot be created by an exact theoretical equation.

Comparative evidence

Britain: expenditure composition and welfare

British household expenditure surveys provide repeated observations of consumption composition at different real budgets and dates. The Deaton–Muellbauer application and the later quadratic Engel curve study develop structured representations of those relationships, while the revealed-preference literature offers a way to bound compensation under weaker functional-form commitments. The common scientific question is the expenditure required for an unchanged rank, but the methods extract different information from the data.

A comparison must account for the fact that repeated cross sections observe different households, often at a common set of prices for a locality or year. Changes in the distribution of household types can mimic changes in a representative Engel curve. Price variation over time can coincide with taste shifts, technological changes or revised survey definitions. A structural interpretation of intertemporal demand variation requires these alternative explanations to be addressed. The studies cited here are historical British applications; they are not current estimates for every high-income economy.

China: a sign reversal in ordinary demand

Jensen and Miller (2008), in Giffen Behavior and Subsistence Consumption, examine subsidized staple prices among poor households in Hunan and Gansu. The author-hosted accepted manuscript identifies strong evidence for upward-sloping rice demand in Hunan and less conclusive evidence for wheat in Gansu. The work connects the response to subsistence conditions and variation in the severity of poverty. This dossier uses the directional findings and research design, not newly reproduced coefficients; the published article's indexed abstract corroborates them.

The experimental variation is valuable because price changes in ordinary market records may reflect simultaneous shifts in supply and demand. A targeted subsidy moves the net price faced by selected households and helps distinguish a response to acquisition cost from a simple market correlation. The result concerns ordinary quantity choices under particular subsidy conditions. It does not directly observe compensated quantities or identify the entire Slutsky matrix. In particular, the price intervention may have features of a voucher rather than a freely transferable cash-price experiment; institutional restrictions matter for the equivalence of experimental and theoretical budgets.

The relevant interpretation is conditional. Households close to subsistence may reallocate expenditure toward a cheap calorie source after a loss of purchasing power, whereas those already buying almost nothing else may be forced to reduce the staple itself. The comparison between provinces cautions against labeling a commodity permanently Giffen. The behavioral response depends on access to substitutes, relative prices and available resources. The theoretical possibility and the observed sign reinforce one another without establishing a universal prevalence.

A comparative conclusion

The British evidence illustrates the importance of income-response curvature in estimating compensation; the Chinese experiment illustrates the possibility that income effects dominate substitution. The two are complementary tests of a general mathematical structure rather than replications of the same empirical coefficient. They support a research strategy that treats the entire distribution of resources and the local choice set as relevant to welfare analysis.

A comparison of economies facing food inflation should therefore begin with institution-specific food definitions, access to substitutes, local prices and how budgets are measured. The same national price index can coexist with different compensation requirements across households. The cited empirical cases are geographically restricted and historical. Extrapolating their coefficients to new countries or the present requires new observations and identification.

Measurement and research design

Data scope and comparability

Household data provide purchases, household characteristics and sometimes quantities, locations and prices. Eurostat's Household Budget Survey microdata documentation describes national expenditure surveys and scientific-use files for 2010, 2015 and 2020. Its comparability documentation identifies differences in survey practice, including the treatment of owner-occupied housing imputed rent. These are material differences when a consumption aggregate is used to define a budget or classify real purchasing power.

Deaton and Zaidi's (2002) consumption aggregate guidelines is an established World Bank resource for welfare measurement; its official indexed record was inspected but direct full retrieval was blocked. It is a bibliographic lead, not authority here for unexamined detailed implementation choices. A study that uses national household survey data must retrieve and examine the actual year's metadata and construction methodology before making claims about comparable real consumption.

A household's unit value for a group of products is total spending divided by physical quantity. If households change quality within the group, the unit value changes even at unchanged underlying item prices. Treating it as an exogenous commodity price can confuse quality selection with the price response the model is meant to estimate. Matching store prices to narrowly defined goods may reduce this problem, but discounts, delivery costs and local availability still affect transaction prices.

Resources and consumption

A household budget survey records purchases within a specified recall interval. Purchases of durable goods differ from the consumption services those goods provide. A family that buys a durable this month may have an unusually high measured spending total despite no equivalent jump in its service consumption. Food purchased in bulk may be stored rather than consumed during the observation period. Rent, owner-occupied housing, gifts, own production and transfers in kind require consistent treatment across the measured population.

The cost-of-living object depends on the commodity scope. If a study excludes housing to focus on nondurables, its compensation estimate is conditional on the excluded allocation. If noncash social benefits are counted as unrestricted income although they can be used only for particular goods, linear-budget demand derived from the inflated amount can misstate feasible purchases. Reporting the resource concept is therefore as important as reporting the price index.

A model of household expenditure also imposes assumptions about the household as a decision unit. Shared goods, differing intra-household preferences and bargaining can affect the mapping from household resources to individual welfare. A unitary household model may be a useful aggregate representation without establishing invariant interpersonal individual preference rankings. Comparisons involving individual and household welfare require an explicit equivalence or allocation convention.

Identification protocol

Design begins by specifying the compensation estimand: target household or person, reference price regime, fixed preference rank or state, and eligible transactions. Select a population, survey vintage and sampling design. Identify how prices are observed and whether expenditure represents the specified budget. Check the extent of relative-price variation, the overlap between demographic groups, the presence of corner or censored choices, and whether behavioral responses to an intervention may persist across periods.

Then choose a method consistent with the available variation. For repeated price regimes and sufficient common support, revealed-preference inequalities can supply finite-data restrictions or bounds. For sufficiently informative price and income variation, a demand system can be estimated, regularity checked and expenditure inverted. For an intervention with exogenous net-price variation, causal estimates of ordinary responses may be feasible while compensated responses remain indirect. These designs can answer different questions even on the same dataset.

The current dossier performs analytical verification and literature-based comparison only. It does not possess a licensed household-level file, a survey weight vector, a complete set of matched transaction prices or the code for a new causal estimation. It therefore reports no invented sample statistics or new coefficients. The examples are fully specified and reproducible from their stated assumptions.

Measurement errors and robustness

Measurement error in prices affects both observed affordability and estimated derivatives. Recall error in expenditure can change whether an alternative bundle appears feasible, creating apparent revealed-preference violations. For near-ties, small errors may reverse the sign of an inequality. Sensitivity analysis can perturb observed prices and expenditures within justified error ranges, recompute feasible inequalities, and report whether the conclusions persist. Error ranges should be tied to data collection processes rather than selected to guarantee a desired outcome.

For structural estimates, preserve units, base-date definitions, normalization of prices, and the exact treatment of omitted groups. Test the validity domain of expenditure functions at counterfactual prices, not only at the average household. If a model has a positive own compensated price derivative or a negative budget share in a policy-relevant scenario, the resulting compensation estimate is outside its economically admissible range. Re-estimation or a restricted domain is preferable to silently truncating a computed welfare result.

Reproducibility requires archived data identifiers, permission conditions, variable definitions, estimation code, and results with uncertainty measures. Literature-based synthesis requires a separate record of which source content was actually accessible. A citation cannot compensate for an unavailable original table or unread methodological appendix. The evidence accompanying this dossier distinguishes readable primary texts from catalogue-level verification.

Institutional extensions and limits

Nonlinear budgets and quantity constraints

In many markets a single constant p_i is not the relevant payment schedule. A two-part tariff has a participation charge and a marginal unit price; a capacity limit changes the feasible consumption set. An expenditure problem can instead minimize the actual payment function P(x;z) subject to a rank constraint and institutional eligibility. Under continuity and suitable compactness a minimum can still exist. The familiar concavity in the vector of linear prices must then be proved for the particular parameters; it does not automatically follow with a nonlinear tariff.

Consider a transport subscription with a fee F and per-trip charge c. For a consumer requiring u trips and treating trips as the sole relevant good, cost within the subscription is F+cu for u>0. A per-trip alternative costs d u. The minimum across eligible options is min(F+cu,d u). A fee change can trigger a discrete switch, and the derivative with respect to a fee reflects option participation while the derivative with respect to c reflects trips if the option is chosen. Treating the entire cost as a constant unit price ignores the threshold. The example is not a new duality model; it demonstrates how institutional details mediate the familiar optimized-cost construction.

With a rationed good, a bound x_i≤r removes some bundles from the feasible set. The expenditure to attain a rank becomes a function of both prices and r. A reduction in r may make a target unattainable even if monetary resources rise. Its shadow value measures the local monetary effect of relaxing the constraint under regular conditions. A price-only compensation that leaves the quota unchanged may be impossible. Policy evaluation needs to record the actual feasible set.

Time, uncertainty and information

An intertemporal consumption plan is a bundle of dated goods. If well-defined present-value prices and complete trading opportunities exist, the benchmark duality can be applied to the enlarged space. Borrowing ceilings, income uncertainty and illiquid assets change that space. Holding current expenditure fixed need not hold lifetime resources fixed. An exact expenditure function over dated state-contingent goods is conditional on accessible markets and a specified preference ranking over outcomes.

Information is also part of a decision's environment. If a consumer learns a product's quality between observations, the alternatives being ranked may not be equivalent to their old descriptions. A revealed-preference violation may then identify an omitted information state rather than a logical inconsistency. For uncertain quality or future prices, distinguish the ex ante criterion for choosing from ex post realized welfare. The appropriate compensation concept depends on the question.

Aggregation without false equivalence

Individual expenditure functions can be evaluated at a common price vector and summed at separately specified individual ranks. The derivative of that sum, where differentiable, is the sum of individual compensated quantities holding their ranks fixed. A market ordinary demand response instead holds an income-distribution rule fixed or allows it to change according to an institutional mechanism. In general no single expenditure function of aggregate money and an unspecified representative rank reproduces that entire process.

Even if two households exhibit homothetic demands with different share parameters, changing the distribution of a fixed total income can change aggregate quantities. A representative consumer calibrated at one income distribution may fail at another. This is one mathematical reason aggregate demand and individual monetary welfare should remain separate objects in the scientific catalogue. Special aggregation conditions may give a representative specification; they must be demonstrated for the chosen application rather than assumed from the availability of the individual duality identities.

Original analytical synthesis and research frontiers

The three levels of a compensation claim

The central analytical conclusion of this dossier is that consumer duality establishes a conditional correspondence between economic opportunities and the cost of attaining an ordered outcome. Its empirical use involves three logically distinct levels.

The first level is formal validity: an optimization problem over a specified set establishes the inverse identities, compensation demand and restrictions under the stated assumptions. The second is empirical identification: observed behavior and institutional variation determine how tightly the relevant expenditure function can be recovered. The third is welfare interpretation: the chosen commodity boundaries, eligible transactions and reference rank determine the meaning of the monetary result. Each level has its own evidence and failure conditions.

For example, a model can generate a differentiable, homogeneous and concave expenditure function satisfying all formal checks, yet use prices unrelated to actual transaction opportunities for the poorest households. A revealed-preference test can establish rationalizability without identifying one precise expenditure number for a novel price vector. A precisely identified household monetary compensation can remain an incomplete measure of a policy that also affects environmental quality, leisure or service availability. These are distinct conclusions, not competing definitions.

What the constructed examples establish

The Cobb–Douglas calculation demonstrates that a price increase can require 100 units of monetary compensation while 150 units would keep the old bundle affordable and ordinary consumer surplus changes by about 69.31. The deviation is generated entirely within an internally coherent preference model, so measurement disagreements among these concepts do not constitute evidence of irrationality.

The perfect-substitute tie shows that optimizing choices can be set-valued even though the expenditure function remains continuous and concave. The perfect-complements example shows that nondifferentiability of the direct representation does not force nondifferentiability of the dual value. The nonconvex contour example proves that equality of optimized cost functions does not recover every feature of a nonconvex ordering. These calculations expose the limits of replacing conditional mathematical claims with global statements about behavior.

The finite affordability contradiction demonstrates that a purely observational choice test can reject the maintained model without fitting utility. The hypothetical elasticity calculation demonstrates that a positive ordinary own-price response is compatible with a negative compensated own-price response. In both cases the identity points to specific observable restrictions, whereas the empirical attribution remains conditional on the data. These are original expository calculations and synthesis; no novelty priority is claimed for the mathematical principles used.

Implications for research and policy

An empirical welfare study should explicitly name its compensation concept, target reference prices, household resource measure and the institutional state held fixed. For a long-run policy change, distinguish changes in prices from changes in available quantities and household constraints. When a policy changes feasible consumption in kind as well as in money, the relevant counterfactual includes that change in opportunities.

A cost-of-living analysis should explain whether the rank-independent index property is assumed, whether the price index is only an approximation, and how differences in access, demographic composition and substitution are treated. Theoretical restrictions and data diagnostics should both be reported. A result that depends on extrapolation beyond observed prices should be described as model-dependent and accompanied by appropriate sensitivity or partial-identification analysis.

For economic theory the value of consumer duality lies in transforming an ordering into a family of monetary cost comparisons, whose derivatives and inequalities can be confronted with evidence. It allows a single vocabulary for exact welfare, compensated comparative statics and identification, while retaining the domains in which each is valid.

Open scientific questions

Research remains active on counterfactual demand under limited price variation, nonparametric bounds with heterogeneous preferences, the measurement of price changes when quality and product availability vary, and the identification of household welfare when budgets are nonlinear or constrained. These problems are significant because a formally precise monetary measure can conceal substantial uncertainty about the behavioral object it measures.

A productive empirical program would use matched item-level transaction prices and expenditures, explicit state-dependent eligibility, and sufficient independent price variation. It would estimate or bound preference-consistent demand separately for appropriate household groups, evaluate welfare at the observed reference regimes and at stated counterfactuals, and compare the precision of parametric versus partial-identification results. The dossier specifies this as a research direction; it reports no new empirical findings from a nonexistent dataset.

Future work can also examine how repeated observations of online offers, membership tariffs and delivery restrictions alter the observed budget set. The question is whether a richer accounting of actual opportunities improves the explanatory and counterfactual validity of familiar consumer models. It does not require declaring a new named theory to recognize the significance of these extensions.

Conclusion

Expenditure minimization provides the minimum monetary cost of a preference level. Indirect utility gives the preference level attainable from a monetary budget. Under the regular domain and binding conditions developed here the functions are inverse descriptions, and their optimizing bundles agree at corresponding arguments. Shephard's lemma and Roy's identity make the relationship operational under derivative hypotheses; Slutsky decomposition links compensated movement to ordinary demand through the income effect.

Expenditure concavity, homogeneity and support inequalities explain the restrictions observed in smooth and nonsmooth demand. Revealed-preference inequalities connect optimization to finite observations, while flexible and nonparametric econometric methods address the difficulty of counterfactual identification. Exact price indexes and monetary compensation require an explicit reference rank, price regime and commodity scope. Finite examples establish why an ordinary demand integral or a fixed basket does not generally equal exact welfare compensation.

The historical and comparative research reviewed here shows the continued empirical significance of the two sides of duality. British expenditure studies expose the consequences of Engel curvature; the Chinese staple-price experiment demonstrates the importance of income effects under severe resource constraints. Official statistical documentation clarifies the institutional scope of household expenditures and price indexes. No unreported household data analysis or new named theoretical system is claimed.

The scientific contribution is the integrated examination of mathematical validity, observational identification and welfare interpretation. The value of the benchmark is preserved by identifying exactly the opportunities, preferences and reference conditions to which it applies. Consumer duality thereby provides a rigorous basis for measuring the monetary consequences of economic choice and for specifying the evidence that such measurements require.

References

Afriat, S. N. (1964). The Construction of Utility Functions from Expenditure Data. Cowles Foundation Discussion Paper 177. Journal version: International Economic Review 8(1), 67–77 (1967). Source

Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel Curves and Consumer Demand. Review of Economics and Statistics, 79(4), 527–539. Source

Blundell, R., Browning, M., & Crawford, I. (2003). Nonparametric Engel Curves and Revealed Preference. Econometrica, 71(1), 205–240. DOI: 10.1111/1468-0262.00394. Source

Blundell, R., Browning, M., & Crawford, I. (2008). Best Nonparametric Bounds on Demand Responses. Econometrica, 76(6), 1227–1262. DOI: 10.3982/ECTA6069. Source

Chipman, J. S., & Lenfant, J.-S. (2001). Slutsky's 1915 Article: How It Came to Be Found and Interpreted. Working paper, HAL hal-01771851. Source

Deaton, A., & Muellbauer, J. (1980). An Almost Ideal Demand System. American Economic Review, 70(3), 312–326. Source

Deaton, A., & Zaidi, S. (2002). Guidelines for Constructing Consumption Aggregates for Welfare Analysis. Living Standards Measurement Study Working Paper 135. World Bank. Source

Eurostat. (n.d.). Household Budget Survey: Microdata. European Commission. Accessed November 17, 2026. Source

Eurostat. (n.d.). Household Budget Surveys: Information on Data. European Commission. Accessed November 17, 2026. Source

Feenstra, R. C. (1994). New Product Varieties and the Measurement of International Prices. American Economic Review, 84(1), 157–177. Source

Jensen, R. T., & Miller, N. H. (2008). Giffen Behavior and Subsistence Consumption. American Economic Review, 98(4), 1553–1577. DOI: 10.1257/aer.98.4.1553. Linked January 2008 accepted manuscript. Source

Levin, J., & Milgrom, P. (2004). Consumer Theory. Stanford University, graduate microeconomics lecture notes. Source

Roy, R. (1947). La distribution du revenu entre les divers biens. Econometrica, 15(3), 205–225. DOI: 10.2307/1905479. Source

Samuelson, P. A. (1938). A Note on the Pure Theory of Consumer's Behaviour. Economica, 5(17), 61–71. DOI: 10.2307/2548836. Source

Shephard, R. W. (1970). Theory of Cost and Production Functions. Princeton University Press. ISBN 978-1-4008-7108-7 (electronic reissue). Source

Slutsky, E. (1915). Sulla teoria del bilancio del consumatore. Giornale degli Economisti e Rivista di Statistica, 51(1), 1–26. Source

U.S. Bureau of Labor Statistics. (n.d.). Handbook of Methods: Consumer Price Index, Concepts. Accessed November 17, 2026. Source

Varian, H. R. (1982). The Nonparametric Approach to Demand Analysis. Econometrica, 50(4), 945–973. Source