RESEARCH DOSSIER

Consumer Demand and Slutsky Decomposition

Abstract. Consumer demand decomposes the local effect of a price change into an adjustment to relative opportunities at constant utility and an adjustment to purchasing power. This monograph reconstructs the Slutsky equation from consumer duality, develops the matrix and elasticity restrictions, examines finite compensation, proves constructive finite-data rationalization, and connects demand regularity to identification, measurement, heterogeneity, and welfare. Synthetic examples clarify uncompensated Giffen behavior, the distinction between accounting and integrability, and the error introduced by aggregating individual income effects. Selected Chinese, UK and Côte d'Ivoire research illustrates the empirical relevance of these distinctions. The study concludes that the interpretation of a compensated response requires an identified decision unit and budget, credible variation, and consistent economic regularity. The empirical discussion synthesizes existing evidence; no new household dataset is estimated.

Keywords: Consumer demand; Slutsky equation; Hicksian demand; Marshallian demand; substitution effect; income effect; integrability; revealed preference; demand systems; Giffen behavior; welfare measurement; household heterogeneity; price elasticity; microeconomics

The research object and historical lineage

Consumer demand is a mapping from economic opportunities to chosen quantities. The Slutsky decomposition describes how the mapping responds to prices through the separate channels of relative trade-offs and purchasing power. This monograph develops the individual demand mechanism, the restrictions that make it coherent with preferences, and the empirical conditions required to recover its components. It treats a demand derivative as a conditional counterfactual: the derivative answers a question about a given decision-maker, a specified feasible set, a specified price intervention, and a specified rule for income.

Slutsky's 1915 article, Sulla teoria del bilancio del consumatore, is the historical source of the decomposition. Its bibliographic record locates the contribution in volume 51 of the Italian journal. The subsequent reconsideration of value theory by Hicks and Allen (1934) belongs to the development of ordinal utility analysis. Samuelson's note on consumer behavior (1938) marks the early revealed-preference approach. The Minneapolis Fed's historical account of demand (2009) describes the delayed recognition of Slutsky's contribution and the difficulty of moving from individual to market behavior. The present reconstruction uses modern notation and independently derives the relationships; historical claims draw on accessible bibliographic records, selected passages of Hicks and Allen and Samuelson, and that historical account.

The scientific significance of the equation is that it joins behavior observed at fixed money income to a hypothetical experiment in which the consumer's utility is fixed. Econometric work typically begins with purchases and expenditures, while welfare analysis needs a monetary expenditure comparison at fixed utility. Their bridge is a mathematically precise restriction, with qualifications imposed by the environment. This relation provides a disciplined language for asking whether households respond to scarcity, shifting relative opportunities, quality selection, or changes in their effective resources.

The dossier belongs to Microeconomics. Its core object is individual demand and the decomposition of price effects. Aggregate demand management, production technology, general equilibrium existence, and the complete economics of welfare taxation are separate research objects. They enter here only when necessary to understand the interpretation of the demand derivatives.

The consumer's feasible set and preference representation

Let q be a nonnegative n-dimensional column vector of quantities, p a strictly positive vector of prices, and m a positive money budget available for the specified consumption period. The basic feasible set is B(p,m) = {q >= 0 : p' q <= m}. All quantities use declared physical or service units; prices are money per unit. The budget is expenditure available to these goods, which may differ from gross earnings. The decision-maker ranks bundles by a utility representation U(q).

A complete and transitive preference relation admits useful numerical representations under familiar continuity conditions. A strictly increasing transformation of U preserves its ordering and the demand selected from each budget. For example, the choice associated with U = q1 q2 can also be represented by log U on the positive orthant. Utility differences across transformations have no intrinsic monetary meaning. The economic content is therefore located in choice and expenditure rather than the numerical level of the representation.

Continuity of U and positive prices give a maximum on the compact budget set. Local nonsatiation implies expenditure of the available budget, because a choice with positive unspent resources has a neighborhood of feasible alternatives. Convex preferences make the set of utility-maximizing bundles convex; strict quasiconcavity provides uniqueness under the usual convex-budget assumptions. Differentiability of the choice map is an additional condition; neither continuity of preferences nor uniqueness at one point guarantees a smooth demand function across all prices.

The benchmark assumes identified commodities, a price-taking agent, a linear tariff, accessible goods, stable tastes during the comparative static, and an expenditure parameter that can be independently specified. Household production, borrowing limits, minimum purchase sizes, nonlinear prices, time constraints, and durable inventories require an explicit extension. If a price move changes the feasible set in another way, the ordinary derivative no longer represents the whole intervention.

The distinction between the physical bundle and the transaction matters. A kilogram of rice at a local shop and a kilogram delivered at home have different service attributes. A good defined as a quality-adjusted service may permit analysis of substitutions within a physical category; a category defined only by observed spending may combine quality, quantity, and availability responses. Specification precedes estimation.

Marshallian demand and indirect utility

The ordinary or Marshallian demand x(p,m) solves max U(q) over B(p,m); the indirect utility v(p,m) is the attained value. Demand is a correspondence where multiple maxima exist. The smooth analysis assumes single-valued demand and sufficient differentiability locally. The budget identity is p' x = m under local nonsatiation. Scaling all prices and money income by a positive scalar leaves the feasible set invariant, so x(tp,tm) = x(p,m). This is homogeneity of degree zero in prices and expenditure together, not homogeneity in prices alone.

At an interior optimum with differentiable U and a regular binding constraint, the Lagrangian is U(q) + lambda(m - p'q). The first-order condition is grad U(x) = lambda p. Ratios of marginal utilities equal ratios of prices. The multiplier lambda is the marginal utility of expenditure in the chosen representation. At a corner, complementary slackness replaces the equalities for absent goods; a good with strictly lower marginal utility per monetary unit than the marginal expenditure value can remain unpurchased.

The envelope identities are v_m = lambda and v_pj = -lambda x_j at regular optima. If v_m is positive, their ratio yields Roy's identity x_j = -v_pj/v_m. The ratio is invariant to smooth strictly increasing transformations when the derivatives exist. Its denominator emphasizes why satiation and discrete changes in feasible behavior deserve attention.

Differentiating the budget with respect to m gives p' x_m = 1. Differentiating with respect to p_j at fixed m gives p' x_pj = -x_j. Differentiating scale invariance at t = 1 gives D_p x p + m x_m = 0. These three restrictions follow from resource accounting and nominal scale invariance, not from a particular utility specification. They are useful diagnostics even before more demanding restrictions of preference consistency are imposed.

An observed fall in purchases need not mean a negative Marshallian own-price derivative. Quantities observed at two dates may embed changing income, expectations, stock, quality, and available substitutes. A comparison at fixed m and stable preferences is a defined experiment rather than an automatic description of two observations.

Hicksian demand and the expenditure function

For a target utility u in the attainable range, define e(p,u) = min {p'q : U(q) >= u, q >= 0}, and let h(p,u) denote cost-minimizing demand. With ordinary regularity and an attained target, e is homogeneous of degree one in prices, nondecreasing in each price for nonnegative goods, and concave in p. Concavity follows immediately because e is an infimum of linear functions of p over a common utility requirement set. For each admissible q and prices p,r, the convex combination of its two costs is at least the corresponding combination of the two minimum costs. Taking the infimum establishes concavity.

At a unique minimizer with differentiability of e, Shephard's lemma gives h_j = e_pj. A direct argument compares expenditure at p and p + d using the old and new minimizers. These choices bound the change in e between the relevant linear cost evaluations; continuity of the minimizers gives the derivative. When minimizers are multiple, any minimizer provides a supergradient of the concave expenditure function. This supports compensated demand analysis even where classical derivatives fail.

Under ordinary duality assumptions and local invertibility in utility, h(p,u) = x(p,e(p,u)) and e(p,v(p,m)) = m. Likewise v(p,e(p,u)) = u for attainable regular utility targets for which the minimum-cost solution achieves the target exactly. These identities describe two optimization problems over the same preferences. Income compensation in the Hicksian sense means that the budget is adjusted to the minimum amount necessary to attain the specified utility after the price change.

Hicksian demand is homogeneous of degree zero in prices at fixed utility. A pure proportional change of the entire price vector scales the monetary cost of the target, leaving the optimal bundle unchanged. This provides an immediate distinction between a rise in the general price level and a change in relative prices: at fixed monetary income both can affect ordinary choices, while compensated proportional price scaling has no substitution effect.

Derivation of the Slutsky equation

The fundamental derivative is obtained from h(p,u) = x(p,e(p,u)). Holding u fixed and applying the chain rule gives, for goods i and j,

∂hi∂pj=∂xi∂pj+∂xi∂mhj.\frac{\partial h_i}{\partial p_j} = \frac{\partial x_i}{\partial p_j} + \frac{\partial x_i}{\partial m}h_j.

At the reference point u = v(p,m), compensated and ordinary demands coincide. Hence

∂xi∂pj=Sij−xj∂xi∂m,Sij=∂hi∂pj∣u=v(p,m).\frac{\partial x_i}{\partial p_j} = S_{ij} - x_j\frac{\partial x_i}{\partial m},\qquad S_{ij}=\left.\frac{\partial h_i}{\partial p_j}\right|_{u=v(p,m)}.

The matrix form is S = D_p x + x_m x'. All derivatives of ordinary demand are evaluated at the same prices and budget. The outer product has row index i and column index j; reversing the vectors gives the wrong result in general. The dimensions also verify the formula: x_j is measured in units of good j, and the income derivative is units of good i per dollar; their product is units of i per dollar-per-unit-of-j.

For an infinitesimal price change dp and simultaneous change dm,

dx=S dp+xm(dm−x′dp).dx=S\,dp+x_m(dm-x'dp).

The scalar in parentheses is the first-order change in purchasing power measured against the original bundle. Compensating with dm = x'dp leaves the pure substitution response. Holding the nominal budget fixed imposes dm = 0. For a price fall in a good purchased initially, the purchasing-power contribution is positive for a normal good and negative for an inferior good. For a price rise the signs reverse.

A useful interpretation follows when the initial expenditure share of a changed good is small: the first-order income effect is proportionately smaller per percentage price change for a given expenditure elasticity. A zero initial quantity removes the conventional local income term through that good, although finite price changes can induce entry. Separating the local derivative and a finite movement is necessary at such boundaries.

The decomposition is a chain-rule identity when the dual mappings exist and are differentiable. It is not an empirical estimate and does not by itself establish that any particular observed price movement is exogenous. Nor is the compensated expenditure change generally an actual cash transfer received by the observed household.

The substitution matrix and integrability

Since S is the Hessian of e with respect to p at fixed u, twice continuous differentiability gives S = S'. Concavity gives z'Sz <= 0 for every real vector z. Thus S is negative semidefinite and its own-price entries are nonpositive. The homogeneity of h implies Sp = 0. Symmetry then gives p'S = 0. A zero eigenvalue induced by scale invariance is expected, so testing strict negative definiteness on the full n-dimensional space is incorrect.

The budget identity can independently establish p'S = 0: p'D_p x = -x' and p'x_m = 1, so p'(D_p x + x_m x') = 0. Homogeneity similarly establishes Sp = 0. The restrictions distinguish derivative changes that merely reallocate cost from changes that affect expenditure. Symmetry implies that compensated cross-price derivatives agree in quantity-per-price units, although the corresponding percentage elasticities need not be equal.

Integrability asks whether a candidate demand system is generated by some preference ordering. For smooth demand on a sufficiently regular domain, nonnegativity, budget exhaustion, homogeneity, a symmetric negative semidefinite Slutsky matrix, and appropriate global/existence regularity provide the standard route to integrability. Local checks at a selected observation do not establish global integrability or an extension to every feasible budget. Singularities, disconnected domains, and boundary behavior can obstruct a naive global conclusion.

The constructive argument is to solve the expenditure differential system e_pj(p,u) = x_j(p,e(p,u)) with a reference expenditure boundary condition. Equality of mixed derivatives requires x_i,pj + x_i,m x_j = x_j,pi + x_j,m x_i. These are the Slutsky symmetry equations. Under regularity they express the compatibility of the partial differential system. Curvature and homogeneity constrain the reconstructed expenditure surface; additional global conditions ensure that it corresponds to well-behaved preferences and can be inverted in utility. The utility label chosen at the boundary is an ordinal normalization.

A locally plausible vector field need not satisfy these conditions. Consider three goods with constant budget shares w1 = 0.4 + a log(p2/p3), w2 = 0.3, w3 = 1 - w1 - w2, where a = 0.05, and demands x_i = m w_i/p_i on a neighborhood of unit prices. This system exhausts the budget and is scale invariant; all shares are positive locally. At p = (1,1,1) and m = 1, S12 = a + w1 w2 = 0.17 while S21 = w2 w1 = 0.12. No smooth scalar expenditure function with these derivatives exists on that neighborhood. The example separates accounting coherence from preference integrability.

Elasticities, shares and economic interpretation

At positive quantities and income define ordinary elasticities E^M_ij = (p_j/x_i)x_i,pj, compensated elasticities E^H_ij = (p_j/x_i)S_ij, and expenditure elasticities eta_i = (m/x_i)x_i,m. Let w_j = p_j x_j/m. Then

EijH=EijM+wjηi.E^H_{ij}=E^M_{ij}+w_j\eta_i.

This representation is dimensionless and useful for comparing commodity categories, as long as each elasticity is computed at an identical reference point. A normal good has eta_i > 0; an inferior good has eta_i < 0. Necessities and luxuries are often classified within the normal range by whether eta is below or above one. These categories are local and depend on the goods and budget definition.

The adding-up restrictions are sum_i w_i eta_i = 1 and sum_i w_i E^M_ij = -w_j. Homogeneity gives sum_j E^M_ij = -eta_i. Compensated homogeneity gives sum_j E^H_ij = 0; compensated adding-up gives sum_i w_i E^H_ij = 0. Symmetry in derivative units gives w_i E^H_ij = w_j E^H_ji. Researchers cannot generally equate unweighted cross elasticities.

For own prices E^M_ii = E^H_ii - w_i eta_i. For a normal good the two effects reinforce each other when its price rises. An inferior good can have a negative ordinary own-price derivative if the negative compensated effect dominates, a zero derivative when effects cancel, or a positive derivative when the income component dominates. A positive ordinary derivative in this setting is Giffen behavior. Inferiority is necessary for Giffen behavior in the differentiable benchmark but is insufficient by itself.

Gross substitutes and complements are defined by the sign of the ordinary cross-price derivative. Net substitutes and complements use the compensated derivative. A pair may be gross substitutes in one direction but not the other. Net substitution is symmetric in derivative units; homogeneity also requires a positive weighted contribution from other goods to offset a strictly negative compensated own effect. With two goods and an interior regular solution, the compensated cross derivative is nonnegative; with more goods some pairs can be net complements.

Finite compensation and a transparent numerical example

Let u0 = v(p0,m0), x0 = x(p0,m0), and x1 = x(p1,m0). The Hicksian intermediate bundle is h(p1,u0); the Slutsky intermediate bundle is x(p1,p1'x0). Each gives a valid exact difference decomposition: first subtract x0 from the intermediate bundle, then subtract the intermediate bundle from x1. The intermediate budgets generally differ. Since the old bundle attains u0, e(p1,u0) <= p1'x0. The difference is the cost saved by optimally reorganizing purchases at new prices. For small changes both compensations agree to first order under differentiability.

For a synthetic Cobb-Douglas example let U = sqrt(q1 q2), m0 = 100, and p0 = (2,1). Demand is x1 = m/(2p1), x2 = m/(2p2). The initial bundle is (25,50) with utility sqrt(1250). Raise the first price to 4 while holding the second price and money budget fixed; the new bundle is (12.5,50). The initial utility could be attained under the new prices at the minimum cost 100 sqrt(2), approximately 141.4214. The compensated bundle is approximately (17.6777,70.7107).

For good 1, the Hicks substitution difference is -7.3223 and the remaining income difference is -5.1777; together they give -12.5. The Slutsky purchasing-power budget is 4(25)+50 = 150, which yields the intermediate bundle (18.75,75). Its substitution difference in good 1 is -6.25 and the income difference is -6.25. Both decompositions reproduce the observed final difference exactly. Their disagreement concerns the choice of intermediate counterfactual, not an algebraic inconsistency.

At the initial point, the Slutsky matrix is [[-6.25,12.5],[12.5,-25]] in quantity-per-price units. It is symmetric, has determinant zero and a negative trace of -31.25, and maps (2,1) to zero. The compensated own elasticity for good 1 is -0.5, the ordinary own elasticity is -1, its expenditure elasticity is 1, and the share is 0.5. This checks the elasticity form without relying on a fitted model.

The compensated law of demand also has a finite counterpart. Optimality at two price vectors with a common utility target gives p0'h0 <= p0'h1 and p1'h1 <= p1'h0. Summing yields (p1-p0)'(h1-h0) <= 0. An isolated rise in one price therefore cannot increase its Hicksian quantity under these cost-minimization assumptions, even if a smooth derivative is unavailable.

Inferiority, Giffen behavior and subsistence

A simple continuous, concave, monotone preference representation supplies a constructive counterexample to the unconditional claim that ordinary demand always slopes downward. Let U(a,b) = min(2a+3b, a+b+10), with a,b >= 0. Set the price of a to 1, the price of b to 1.25, and total expenditure to 8.5. Along the budget, the first linear utility component increases in b and the second decreases; the optimum is at their intersection a+2b=10. Solving gives a=6 and b=2.

For a price q of b near 1.25, the demand at fixed budget is b=(10-m)/(2-q). With m=8.5, the derivative in q is 1.5/(2-q)^2, strictly positive. The derivative in m is -1/(2-q), strictly negative. At the reference point these derivatives are 8/3 and -4/3, respectively. A rise in q to 1.30 increases demand for b to 15/7 and reduces demand for a to 40/7. The new budget identity holds. At the fixed initial utility level the kink remains the unique minimum-cost bundle for these nearby relative prices; its compensated derivative is zero. The Slutsky identity gives 0 - b(-4/3) = 8/3. Giffen behavior therefore emerges in a tractable model with no positive compensated own-price derivative. The example is illustrative, not a newly proposed theory or a behavioral estimate.

Real poor households face more complex subsistence trade-offs. In Giffen Behavior and Subsistence Consumption, Jensen and Miller (2008) study targeted staple-price subsidies among poor households in China. Their reported evidence is stronger for rice in Hunan than for wheat in Gansu. This is evidence of geographically and institutionally bounded behavior, not of upward-sloping demand for staple food everywhere. The study's emphasis on variation with poverty severity is directly relevant to the sign and magnitude of the income effect.

A household relying on a cheap source of calories can increase its use when purchasing power falls because more desirable sources become unaffordable. At another poverty level the feasible set may already be concentrated around the cheapest source, leaving little room for adjustment. In a better-off household the ordinary negative substitution response may dominate. These regimes need not produce a common elasticity.

An observed positive correlation between local market prices and staple purchases is insufficient evidence of a Giffen response. Harvest shocks, food quality, seasonality, a rising local population, or upward-sloping supply can change price and quantity jointly. Credible interpretation requires an intervention or valid design isolating the household's price while accounting for transfers, rationing, and the market's supply response.

Revealed preference and finite data

The finite data counterpart to the differentiable theory begins with observations (p_t,q_t), t=1,...,T, and budgets m_t=p_t'q_t. If q_s was affordable when q_t was selected, q_t is directly revealed at least as good as q_s. Transitive closure defines the revealed-preference relation. The Generalized Axiom of Revealed Preference (GARP) requires that if q_t is revealed at least as good as q_s through this relation, q_t cannot be strictly cheaper than q_s at the latter's prices.

Afriat's 1967 construction of utility functions established a foundation for finite-data rationalization. Varian's nonparametric demand analysis (1982) developed the practical approach to analyzing demand without a predetermined utility form. Tipoe, Adams, and Crawford (2022), in their revealed-preference review, discuss equivalent finite conditions, their computation, and measurement-error interpretations.

The test can be written as a finite system of linear inequalities for levels U_t and positive multipliers lambda_t:

Us≤Ut+λtpt′(qs−qt)U_s\le U_t+\lambda_t p_t'(q_s-q_t)

for every pair t,s. Given a feasible solution, define U(q) = min_t {U_t + lambda_t p_t'(q-q_t)}. This function is concave as a pointwise minimum of affine functions and increasing in goods when all prices and multipliers are positive. At q_s the inequalities and the s-th affine term give U(q_s)=U_s. For a bundle q affordable at budget t, the t-th affine term bounds U(q) above by U_t; therefore q_t maximizes U on its budget. This proves the constructive sufficiency direction directly. Necessity of GARP follows by preference transitivity and local nonsatiation: a strictly cheaper bundle that is ranked at least as high as the selected bundle allows a nearby strictly preferred feasible alternative. The full Afriat equivalence additionally establishes feasibility of the inequalities whenever the finite data obey GARP.

Consider observed bundles q1=(2,1), q2=(1,2) at prices p1=(2,1), p2=(1,2). Each chosen bundle costs 5, while the other bundle costs 4. Each choice rejects a strictly cheaper bundle that is itself selected on another occasion. This violates GARP; no locally nonsatiated stable preference representation can rationalize the two exact choices under the stipulated budgets. Reversing the budgets so that each unchosen alternative costs more produces no such direct two-cycle. Passing a finite test shows compatibility with some preferences; it cannot establish a unique utility function or identify all counterfactual quantities.

Measurement error may produce apparent revealed-preference violations. Near-ties in expenditure comparisons are sensitive to rounding, survey imputation, tariff mismeasurement and quality aggregation. A serious empirical application reports the price and expenditure tolerances, the test's statistical procedure, the number of binding comparisons, and the consequences of alternative data definitions. A permissive error tolerance should be declared independently of the desired rationalization outcome.

Demand systems and empirical regularity

A complete demand system represents the response of all included expenditure shares simultaneously. For the Almost Ideal Demand System associated with Deaton and Muellbauer's 1980 demand-system article, the exact functional form can be written

wi=αi+∑jγijlog⁡pj+βilog⁡(m/P),w_i=\alpha_i+\sum_j\gamma_{ij}\log p_j+\beta_i\log(m/P),

where log P = alpha0 + sum_k alpha_k log p_k + one half sum_k sum_l gamma_kl log p_k log p_l, with symmetric gamma. Adding-up requires sum_i alpha_i=1, sum_i beta_i=0, and sum_i gamma_ij=0 for every j. Homogeneity requires sum_j gamma_ij=0 for every i. Symmetry is represented by gamma_ij=gamma_ji in the exact model. Nonnegative shares and curvature still need evaluation over the actual support. A flexible local demand function can fail global economic regularity.

For an interior quantity x_i=m w_i/p_i, differentiating the share equation with respect to expenditure gives eta_i=1+beta_i/w_i. For the ordinary price elasticity, differentiation gives a term -delta_ij from the price denominator, together with [gamma_ij - beta_i (d log P/d log p_j)]/w_i. Substituting into the Slutsky elasticity identity yields the compensated response. This derivation also explains why a convenient linear price-index approximation may alter derivative expressions: an index used in estimation must match the elasticity formula.

Banks, Blundell, and Lewbel's quadratic Engel curve research (1997) expands the flexibility of the shares with quadratic dependence on log expenditure. Their evidence from UK expenditure surveys highlights the practical importance of allowing curvature when assessing distributional consequences of price changes. Linear expenditure-share relations can be useful approximations at a particular expenditure range and unreliable extrapolations across very different households.

Demand equations share a common budget identity. When the residuals sum to zero, their covariance is singular; an estimation procedure normally removes one equation or uses a suitable constrained method, then reconstructs the omitted equation. The choice of estimation method does not erase the need to check adding-up, homogeneity, symmetry, and curvature in the resulting model. Standard errors should reflect estimation of price indexes, instruments, expenditure elasticities and across-equation covariance.

Conditional demand systems require explicit treatment of the group expenditure budget. Estimating demand for meat varieties at fixed meat expenditure answers a different question from estimating their unconditional response at fixed total consumption expenditure. The latter includes the adjustment of the meat budget to prices and income. Weak separability is a restriction on preferences supporting two-stage budgeting; it must not be silently inferred merely because an analyst has organized data into convenient product groups.

Identification of price and expenditure effects

The decomposition uses structural derivatives. The econometric objective is to estimate them using variation that moves the variables whose derivatives are defined, while holding other determinants appropriately fixed. Write observed quantity as x(p,m,z,theta), where z denotes observed covariates and theta preferences or other unobserved characteristics. In observational data, price, budget, location and quality can be jointly related to theta. A regression of sales on paid price is consequently an association until a credible identification argument is supplied.

Randomized price discounts can identify policy-assignment effects when implemented with suitable compliance and measurement. The effect of an offered discount and the effect of the actual paid price are different estimands if take-up is selective. An instrumental variable strategy needs a relevant instrument, a defensible exclusion restriction, and an interpretation of the population whose response is recovered. Cost shocks may affect both prices and availability; geographical instruments may correlate with tastes and incomes. Household fixed effects remove time-invariant unobservables under an appropriate model but do not eliminate time-varying shocks or anticipatory purchases.

Income effects are equally demanding. Total expenditure is frequently constructed from the same purchases used as dependent variables. An expenditure shock can alter saving, labor supply and borrowing. Instrumenting expenditure requires attention to whether the instrument changes tastes or constraints directly. A temporary cash transfer and a permanent income change may produce different responses because the consumer allocates resources across time. The benchmark income derivative is defined for a particular consumption period and budget; it must not be interpreted automatically as a response to gross annual household income.

Economic restrictions can improve precision and discipline counterfactuals, but imposing symmetry or curvature cannot prove that the price instrument is valid. Conversely a violation of the estimated restrictions can reflect misspecified data, omitted choice margins, preference heterogeneity, sampling noise or substantive failure of the benchmark. A useful workflow separates the design assessment from the economic regularity assessment.

The study by Blundell, Browning, and Crawford (2003) on nonparametric Engel curves links statistical estimates of expansion paths to revealed-preference analysis. It illustrates how changes in expenditure and relative prices can be combined in a disciplined demand analysis without treating a single parametric system as the sole representation of consumer choice. The exact transportability of estimated expansion paths still depends on the underlying populations and commodity definitions.

Measurement, unit values and international data

A unit value is recorded spending divided by recorded quantity. A market price is the marginal cost of a specified unit with specified quality and transaction conditions. They coincide only under compatible definitions. Households may choose cheaper varieties or lower service quality when their budget contracts; an analyst treating their falling unit values as an externally imposed price decline reverses the direction of the economic mechanism.

Deaton's quality and spatial price study (1988) addresses this problem through geographically clustered household information and adjustments for quality choice and measurement error. The reported application uses a 1979 Côte d'Ivoire household survey. The contribution is methodological as well as empirical: using the spatial structure of survey data requires a credible assumption about prices within and between clusters and a defensible separation of household quality variation from market-price variation. The dossier draws no new numerical elasticities from that study.

The European Household Budget Survey documentation identifies national expenditure surveys and scientific-use releases, while acknowledging residual methodological variation across countries. The US Consumer Expenditure microdata documentation identifies income, spending and demographic variables as well as distinct Interview and Diary files. These institutional sources establish possible research infrastructure, not a ready-made dataset that by itself identifies compensated responses.

A reproducible international comparison first defines the consumption period and unit, records the sampling frame, survey years, weights and stratification, and documents outlier handling, nonresponse, imputations, purchase zeros and top coding. It then maps goods across consistent classification versions and establishes how own-production, in-kind transfers, imputed housing services and durable purchases enter expenditure. Finally, it attaches prices at the quality, location and time at which the household could actually purchase the good. Deflating nominal spending with a national general price index does not create individual relative prices.

Pooling countries adds another identification requirement. A difference in expenditure shares may arise from preferences, housing institutions, transport technology, taxes, climate or local availability. Purchasing-power conversion assists comparisons of monetary levels but cannot remove differences in feasible commodity sets. Cross-country evidence is most informative when each setting defines a common economic mechanism with explicitly separate institutional conditions.

No household-level dataset was downloaded or re-estimated for this monograph. The cross-country material consists of evaluated research reports and an audit of institutional data availability. Specific datasets require approved access, versioned codebooks, code, and a sampling/identification plan before new empirical conclusions can be asserted.

Aggregation, household heterogeneity and market behavior

For H households facing common prices p and individually fixed budgets m_h, market demand is X(p,m_1,...,m_H)=sum_h x_h(p,m_h). Differentiating gives D_p X = sum_h S_h - sum_h x_{h,m} x_h'. The second sum contains household-level products of the expenditure derivative and quantities. It cannot generally be recovered from the product of average quantities and average expenditure derivatives.

Suppose total income M changes with an allocation rule m_h(M) and derivative shares a_h = dm_h/dM. Then X_M = sum_h a_h x_{h,m}. The representative-consumer compensation term X_M X' corresponds to the household sum only under additional restrictions on the distribution rule and individual demands. This is why knowledge of aggregate expenditure alone is usually insufficient for welfare inference. Changing the distribution of income with constant M can change aggregate demand, even under common prices and individually consistent choices.

A synthetic scalar illustration makes the covariance issue explicit. If two households buy 2 and 8 units of a good and have income derivatives 0.1 and 0.5 units per dollar, their average product is (0.2+4)/2=2.1 units squared per dollar. The product of the averages is 5(0.3)=1.5. The difference 0.6 is the population covariance in this two-household example. Using the product of averages in the Slutsky adjustment would therefore omit a quantitatively material term. The example concerns a single column-row product, not a claim about a complete observed demand system.

Hausman and Newey's individual heterogeneity and average welfare (2016) explains why one cross-sectional observation per consumer generally fails to identify average exact surplus under sufficiently general heterogeneous demand. Observing the marginal distribution of choices at several price-income points is different from knowing how the same person's choices would move across those points. Bounds on income effects can give bounds on welfare instead of a falsely precise point estimate.

Household aggregation also occurs within the household. A unitary household model represents a joint observed bundle with one ordering. A collective model instead allows multiple individuals, bargaining, public goods and allocation rules. Changes in prices or earnings can affect bargaining power as well as the shared budget. The resulting household demand may be coherent under one behavioral environment but fail the restrictions associated with a fixed household-level utility representation across environments. Interpretation must follow the decision unit and its institutional organization.

Exact welfare and the choice of compensation

For a price increase from p0 to p1 at fixed m, let u0=v(p0,m) and u1=v(p1,m). Use a positive-loss convention for compensating variation CV=e(p1,u0)-m and equivalent variation EV=m-e(p0,u1). A pure price increase with positive quantities weakly lowers attainable utility and makes both expressions nonnegative. A change involving both rising and falling prices can have either sign. Stating the sign convention avoids confusion between loss and benefit definitions used across texts.

Under differentiability, the expenditure difference at fixed u is the line integral of h(p,u) along a path of prices. Since h is the gradient of e, the integral is path independent on its appropriate connected domain. For a single price p_i, CV is the integral of compensated demand evaluated at the original utility, from the old to the new price; EV is the same price integral evaluated at final utility. The two targets correspond to different welfare levels of the same decision-maker under unchanged preferences.

Marshallian consumer surplus for one price change is the corresponding integral of ordinary demand at fixed money budget. It coincides with exact money-metric welfare in important special cases, such as a quasilinear interior model in which ordinary demand for the changed nonnumeraire good is independent of income. With material income effects, the area is generally an approximation and the direction of its error depends on the variation definition and normality properties. For a normal good across the path of a pure price increase, CV is at least ordinary surplus loss and EV is at most that loss, under the relevant global conditions.

In the Cobb-Douglas example above, CV is 100(sqrt(2)-1), about 41.4214. EV is 100(1-1/sqrt(2)), about 29.2893. Ordinary surplus loss is 50 log(2), about 34.6574. Their ordering verifies the interpretation under normality. These are hypothetical currency units, not data. They also show why a doubling of a price should not be evaluated simply by multiplying the initial quantity by the price increase: that gives 50, the old-bundle cost increase, which exceeds minimum compensation.

For a policy analyst, household variation in money-metric loss provides information about incidence. Summing losses gives a monetary aggregate; it does not by itself settle distributional welfare judgments, which require assumptions about social weights and the use of public revenues. Identification of individual compensation and normative aggregation are logically separate operations.

Endowments, nonlinear tariffs and dynamics

The ordinary derivation holds expenditure fixed while differentiating price. If a household owns an endowment omega and has m(p)=p'omega plus independently specified nonasset resources, differentiation of the total demand under the same differentiable preferences yields dx/dp_j=S_col_j + x_m(omega_j-x_j). The income term now reflects net purchases. A net seller may gain purchasing power from a price increase even as a net buyer loses it. This distinction is especially important for farm households consuming goods they also produce.

Income can depend on prices through wages, profits, indexed transfers or rent payments rather than through a fixed physical endowment. In such a case replace omega_j by the relevant derivative dm/dp_j. The partial consumer response remains useful when preferences and the feasible set are well specified; sector-wide or general equilibrium effects additionally require an account of the income-generating process.

Nonlinear tariffs create budgets that need not be summarized by a single price vector. For example, a utility customer facing an increasing block tariff may move across a kink at which the marginal price changes. An equivalent virtual income representation may characterize an interior segment, but it can fail at the kink and must be recomputed if the segment changes. A subsidy may relax a quantity cap without changing a posted price. Under rationing, the value of relaxing the cap is a different margin from the response to a linear price.

With durable goods, inventories and time allocation, measured purchases can differ from consumption. A price discount may induce stockpiling, so a large immediate purchase elasticity reflects timing and substitution over purchase dates. The relevant commodity in an intertemporal model is often a dated or state-contingent service. Demand derivatives must state the horizon and whether future prices, wealth, expectations and credit limits remain fixed.

At a corner or discrete purchase threshold, a classical derivative may be zero on one side and undefined at a jump. Finite compensated inequalities, value-function comparisons and discrete-choice models can retain economic content. The analyst should not interpolate differentiability solely to manufacture an elasticity for an outcome defined on a discrete support.

High-dimensional and partially identified demand

Modern scanner and transaction records can contain many potentially substitutable goods and numerous zero purchases. A household may buy from a small subset because preferences favor it, because a purchase occurs only occasionally, or because other options are unavailable. Zero observations are therefore part of the outcome and the opportunity set must be reconstructed as far as possible. Computing logs only among positive purchases changes the estimand to a selected group.

A high-dimensional price vector makes fully flexible cross-price response estimation difficult. Demand Analysis with Many Prices by Chernozhukov, Hausman, and Newey (2019 working-paper version) explores regularized demand estimation and bounds on consumer surplus. The reported scanner-data application found substantially different panel and cross-sectional elasticity estimates. The result illustrates a possibility arising from price-preference correlation, not a universal ordering of panel and cross-sectional estimates.

Regularization trades variance against bias and depends on support, sparsity or other structure. Economic restrictions can constrain feasible high-dimensional predictions and reduce incoherent counterfactual outcomes. Inference after variable selection and regularization requires a method suitable to the target parameter. A precise-looking matrix of cross-price effects may primarily reflect imposed structure where the data contain little independent price variation.

Partial identification provides a useful alternative. Instead of selecting one utility function from a large rationalization set, estimate the range of responses or monetary compensation compatible with defensible restrictions and observed choices. Such ranges can be decision-relevant when they are narrow enough; wide ranges accurately communicate the insufficiency of available variation. Additional price experiments or panel information are then justified by the counterfactuals they help discriminate.

The research frontier combines richer measurement of actual choice sets with microeconomic regularity and credible identification. These advances permit finer distinctions between product switching, changes in quantity, entry and exit, and quality adjustment. Generalization of the estimates requires attention to how the next population differs in all those margins.

Original analytical synthesis and reproducibility

The central finding of this study is that the usefulness of the Slutsky decomposition rests on a conjunction of three analytically distinct tasks: defining the household's counterfactual budget, identifying the relevant ordinary and expenditure derivatives, and testing whether their combination is consistent with the stipulated preference representation. The first is a problem of economic specification, the second a problem of evidence, and the third a problem of mathematical coherence. The tasks are jointly necessary when a price elasticity is to be interpreted as an exact preference-consistent compensation response.

The formal cases establish why the order matters. In the nonsymmetric three-good example, accounting and scale invariance coexist with failure of preference integrability. In the kinked Giffen example, an upward-sloping ordinary demand coexists with coherent monotone concave preferences and zero compensated substitution. In the two-household example, valid individual derivatives do not justify substitution of average quantities and average expenditure effects into the individual adjustment. These results clarify three distinct ways in which an intuitively plausible interpretation can fail without contradicting the Slutsky identity itself.

An empirical implementation can use the following sequence. Define the decision unit, goods, price concept, time horizon, budget and feasible alternatives. Document observed and unobserved determinants and propose credible variation for price and expenditure. Estimate ordinary and expenditure effects jointly, preserving their covariance and units. Compute compensated derivatives from the outer product and examine the Slutsky matrix on the empirically relevant domain. Compare local predictions with finite compensated choices where large movements or discontinuities are important. Report welfare as an exact estimate only when the maintained structure and evidence identify the necessary expenditure comparison; otherwise report approximations or identified bounds. State the target population and the assumptions governing transfer.

Three forms of uncertainty require separate reporting. Sampling uncertainty concerns variation in a chosen estimate. Identification uncertainty concerns alternative behavioral relationships consistent with the data. Specification uncertainty concerns whether the modeled prices, budget and decision-making unit represent the actual mechanism. A narrow confidence interval addresses the first form under the stated statistical model; it cannot automatically resolve the second or third.

The cross-country evidence supports this interpretation in different ways. The China intervention illustrates a conditional sign reversal, the Côte d'Ivoire study illustrates the difficulty of measuring exogenous prices from unit values, and UK demand-system work illustrates the consequences of inflexible expenditure response. The modern heterogeneity literature demonstrates why individual welfare cannot generally be recovered from aggregate or cross-sectional averages. The cases provide complementary tests of assumptions rather than a common pooled estimate.

A reproducible study should release the versioned source data or an access protocol, codebook, estimation specification, variable construction, sample exclusions, randomization or instrument justification, regularity tests and uncertainty procedure. When restricted data cannot be distributed, synthetic demonstration code and a precise access description can make the method legible, while actual replication remains subject to lawful access. The present dossier contributes independent mathematical derivations and synthetic illustrations, together with an analytical synthesis of existing evidence. It claims no novel named theory, no newly estimated dataset, and no priority for established mathematical relationships.

Conclusion

The Slutsky equation provides a rigorous connection between observed demand under an expenditure budget and the counterfactual demand associated with a fixed utility target. The compensated response is constrained by expenditure minimization, while the ordinary response also depends on how demand changes when purchasing power changes. Own-price negativity belongs to compensated demand; the sign of the ordinary response is conditional. The equation therefore establishes a central organizing result of consumer theory without imposing the same demand curve on every household or institutional setting.

Integrability adds substantive content to accounting identities. Symmetry and negative semidefiniteness connect candidate demand equations to a common expenditure function, provided their local and global assumptions hold. Revealed preference extends the analysis to finite choices without a predetermined utility form. Exact welfare measurement then uses expenditure differences at fixed utility, while partial identification is appropriate when the available data cannot recover these differences uniquely.

Empirical implementation depends on how prices, quantities, spending, quality, and opportunities enter the observations. Household demand may respond to a local market price, a tariff, an anticipated future price, a credit restriction, or an income shock that is correlated with the intervention. These are different counterfactuals. The most useful application of the decomposition preserves those distinctions throughout specification, estimation, economic regularity checks and interpretation.

The dossier's independent synthesis is that credible compensation analysis is a joint achievement of economic specification, identification and integrability. The numerical counterexamples and derivations document why each component is necessary. This conclusion extends the practical interpretation of established consumer theory without claiming a new theoretical system or empirical discovery.

References

  • Afriat, Sydney N. 1967. The Construction of Utility Functions from Expenditure Data. International Economic Review 8 (1): 67–77. DOI: 10.2307/2525382. Source record
  • Banks, James, Richard Blundell, and Arthur Lewbel. 1997. Quadratic Engel Curves and Consumer Demand. Review of Economics and Statistics 79 (4): 527–539. DOI: 10.1162/003465397557015. Source record
  • Blundell, Richard W., Martin Browning, and Ian A. Crawford. 2003. Nonparametric Engel Curves and Revealed Preference. Econometrica 71 (1): 205–240. DOI: 10.1111/1468-0262.00394. Source record
  • Chernozhukov, Victor, Jerry A. Hausman, and Whitney K. Newey. 2019. Demand Analysis with Many Prices. CeMMAP working paper. DOI: 10.1920/wp.cem.2019.5919. Source record
  • Deaton, Angus, and John Muellbauer. 1980. An Almost Ideal Demand System. American Economic Review 70 (3): 312–326. Source record
  • Deaton, Angus. 1988. Quality, Quantity, and Spatial Variation of Price. American Economic Review 78 (3): 418–430. Source record
  • Eurostat. n.d. Household Budget Survey: Microdata. European Commission. Accessed November 17, 2026. Source record
  • Federal Reserve Bank of Minneapolis. 2009. The Mechanics of Demand. Source record
  • Hausman, Jerry A., and Whitney K. Newey. 2016. Individual Heterogeneity and Average Welfare. Econometrica 84 (3): 1225–1248. DOI: 10.3982/ECTA11899. Source record
  • Hicks, J. R., and R. G. D. Allen. 1934. A Reconsideration of the Theory of Value. Part I. Economica 1 (1): 52–76. DOI: 10.2307/2548574. Source record
  • Jensen, Robert T., and Nolan H. Miller. 2008. Giffen Behavior and Subsistence Consumption. American Economic Review 98 (4): 1553–1577. DOI: 10.1257/aer.98.4.1553. Source record
  • Tipoe, Eileen, Abi Adams, and Ian Crawford. 2022. Revealed Preference Analysis and Bounded Rationality. Oxford Economic Papers 74 (2): 313–332. DOI: 10.1093/oep/gpab018. Source record
  • Samuelson, Paul A. 1938. A Note on the Pure Theory of Consumer's Behaviour. Economica 5 (17): 61–71. DOI: 10.2307/2548836. Source record
  • Slutsky, Eugen. 1915. Sulla teoria del bilancio del consumatore. Giornale degli Economisti e Rivista di Statistica 51 (1): 1–26. Source record
  • U.S. Bureau of Labor Statistics. 2026. Public Use Microdata (PUMD): Consumer Expenditure Surveys. Last modified April 9, 2026. Accessed November 17, 2026. Source record
  • Varian, Hal R. 1982. The Nonparametric Approach to Demand Analysis. Econometrica 50 (4): 945–973. DOI: 10.2307/1912771. Source record