Supply and Demand for Consumers

Supply and demand for consumers is the study of how individual households decide what and how much to buy when faced with prices, their income, and their tastes—and how those individual decisions add up into the market demand curve that, together with supply, sets the price.

From Preferences to Choice

The modern theory begins not with demand curves but with preferences. Following the ordinalist revolution of Vilfredo Pareto (1906) and its consolidation by John Hicks and R.G.D. Allen (Value and Capital, 1934), the consumer is represented by a binary preference relation $\succeq$ over a consumption set $X \subseteq \mathbb{R}^n_+$. If $\succeq$ is complete, transitive, continuous, and locally non-satiated, Debreu's (1954) representation theorem guarantees a continuous utility function $u:X\to\mathbb{R}$ with $x\succeq y \iff u(x)\ge u(y)$, unique only up to strictly increasing transformation. The cardinal utility of Jevons, Menger, and Walras in the 1870s is thus dispensable: choice, not felt intensity, is the primitive.

The consumer solves the utility maximization problem (UMP):

$$\max_{x\in\mathbb{R}^n_+} u(x)\quad\text{s.t.}\quad p\cdot x \le w,$$

where $p\in\mathbb{R}^n_{++}$ is the price vector and $w>0$ is wealth. The solution correspondence $x(p,w)$ is the Marshallian (Walrasian) demand, named for Alfred Marshall's Principles of Economics (1890), which fixed the partial-equilibrium diagram with price on the vertical axis—a transposition, from a mathematician's viewpoint, that still governs textbook convention.

How Consumers Respond to Prices

At an interior optimum with strictly quasiconcave, differentiable $u$, the first-order conditions equate the marginal rate of substitution to the price ratio:

$$\frac{\partial u/\partial x_i}{\partial u/\partial x_j}=\frac{p_i}{p_j}.$$

Marshallian demand is homogeneous of degree zero in $(p,w)$—only relative prices and real wealth matter, ruling out money illusion—and, under local non-satiation, satisfies Walras' Law $p\cdot x(p,w)=w$.

The full price response is decomposed by the Slutsky equation (Eugen Slutsky, 1915):

$$\frac{\partial x_i(p,w)}{\partial p_j}=\underbrace{\frac{\partial h_i(p,u)}{\partial p_j}}_{\text{substitution}}-\underbrace{x_j(p,w)\,\frac{\partial x_i(p,w)}{\partial w}}_{\text{income}},$$

where $h(p,u)$ is Hicksian (compensated) demand solving the dual expenditure minimization problem $\min p\cdot x$ s.t. $u(x)\ge u$. The substitution matrix $S=\big[\partial h_i/\partial p_j\big]$ equals the Hessian of the expenditure function $e(p,u)$; it is symmetric and negative semidefinite, so the compensated own-price effect is never positive, $\partial h_i/\partial p_i\le 0$. This is the theory's hardest empirical content.

The uncompensated effect need not be negative. If a good is inferior ($\partial x_i/\partial w<0$) and absorbs a large budget share, the income term can dominate, producing the Giffen good: $\partial x_i/\partial p_i>0$. Long a textbook curiosity attributed to Marshall's reading of Robert Giffen on Irish potatoes during the 1840s famine, it received controlled empirical support only recently—Robert Jensen and Nolan Miller's (2008) field study of rice and wheat consumption among poor households in Hunan and Gansu, China.

How Consumers Respond to Income

Holding $p$ fixed, the map $w\mapsto x(p,w)$ traces the income expansion path, and for good $i$ the Engel curve (Ernst Engel, 1857). Engel's empirical law—that the budget share of food falls as income rises—remains one of the most robust regularities in economics and anchors modern demand systems. Goods sort by the income elasticity $\varepsilon_{i}^{w}=\tfrac{\partial x_i}{\partial w}\tfrac{w}{x_i}$: normal ($>0$), inferior ($<0$), luxury ($>1$), necessity ($0<\varepsilon<1$). Adding-up ($\sum_i s_i \varepsilon_i^w=1$, with $s_i$ budget shares) forces the average income elasticity, weighted by shares, to unity—no economy can have all goods be luxuries.

The Almost Ideal Demand System of Angus Deaton and John Muellbauer (1980)—flexible, aggregable, and grounded in the PIGLOG cost function—became the workhorse for estimating these responses, later extended to the Quadratic AIDS (Banks, Blundell, Lewbel 1997) to capture non-linear Engel curves. Deaton's Nobel citation (2015) rested substantially on this program of taking demand theory to microdata.

Preferences and Their Restrictions

Preferences enter the demand function through the whole structure of $u$, but empirical work needs restrictions. Homotheticity makes demands linear in wealth and Engel curves rays through the origin—convenient but contradicted by Engel's Law. Quasilinearity, $u(x)=x_1+v(x_2,\dots,x_n)$, removes income effects from all but the numéraire, justifying the use of Marshallian consumer surplus (Marshall's own measure) as an exact welfare index; without it, Hicks's compensating and equivalent variations diverge, and Robert Willig's (1976) bounds tell us by how much.

A deep question is whether preferences are needed at all. Paul Samuelson's revealed preference program (1938, 1948) sought to found demand on observable choices: if bundle $x^0$ is chosen when $x^1$ was affordable, $x^0$ is revealed preferred. The Weak Axiom (WARP) plus budget-exhaustion delivers a negative semidefinite, though not necessarily symmetric, Slutsky matrix. Hendrik Houthakker's (1950) Strong Axiom (SARP) closes the gap, and Sydney Afriat's theorem (1967) gives the finite-data counterpart: a finite dataset is rationalizable by a well-behaved utility function iff it satisfies GARP. This is the bridge between armchair axioms and testable data, and it underlies contemporary nonparametric demand analysis (Blundell, Browning, Crawford).

Aggregating to Market Demand

Market demand is the horizontal sum of individual demands:

$$X(p)=\sum_{k=1}^{K} x^k\!\big(p,w^k\big).$$

Two questions haunt this innocuous sum.

Does market demand inherit individual structure? Generally, no. Market demand satisfies homogeneity and Walras' Law, but the Slutsky restrictions on individuals need not survive aggregation once the wealth distribution can vary. The Sonnenschein–Mantel–Debreu theorem (1973–74) is the notorious result: excess demand functions of a Walrasian economy can be essentially arbitrary continuous functions satisfying only homogeneity and Walras' Law. The "law of demand" is not a property of aggregates. This severely limits what pure theory can say about uniqueness and stability of equilibrium and is a central open problem for general-equilibrium foundations.

When can we ignore the distribution of income? If we want $X$ to depend only on aggregate wealth $\sum_k w^k$, preferences must lie in the Gorman polar form (W.M. Gorman, 1961), with indirect utility $v^k(p,w^k)=a^k(p)+b(p)\,w^k$—identical wealth-slope $b(p)$ across consumers, so that wealth redistribution leaves aggregate demand unchanged (parallel, affine Engel curves). Absent Gorman form, one appeals to Werner Hildenbrand's (1994) distributional approach: aggregation can generate a downward-sloping market demand even when individuals are ill-behaved, provided the density of households across income is sufficiently spread—increasing dispersion averages out perverse individual effects. Whether the law of demand is better seen as a property of individuals or an emergent statistical regularity of populations remains genuinely contested.

Edge Cases and Live Debates

• Corner solutions and discreteness. Kuhn–Tucker conditions replace equalities; for indivisible goods (cars, houses) discrete-choice models (McFadden's random utility, 1974) supplant the smooth calculus, unifying demand estimation with the logit/probit apparatus.
• Non-convex preferences yield demand correspondences, not functions, and discontinuous jumps—relevant to network goods and status consumption.
• Endowment and reference dependence. The neoclassical consumer has no "current holdings"; Kahneman and Tversky's endowment effect and reference-dependent utility (Kőszegi–Rabin, 2006) break the path-independence of compensating/equivalent variation and generate WTA–WTP gaps that violate standard integrability.
• Intertemporal and stochastic extensions. Under uncertainty the object is a von Neumann–Morgenstern expected utility over lotteries; over time it is a lifetime program, connecting consumer demand to the permanent-income and life-cycle hypotheses (Friedman 1957; Modigliani–Brumberg 1954) where "income" becomes human plus financial wealth.
• Bounded rationality. Rational inattention (Sims), sparse maxima (Gabaix), and menu-dependent choice question whether the stable-preference axioms hold at all, motivating "behavioral revealed preference" that tests weaker consistency conditions.

The Consumer as Supplier

The section's title pairs supply with demand deliberately: the household is not only a buyer but a seller—of labor and of goods it is endowed with. Recast the budget as $p\cdot x \le p\cdot \omega$ with endowment $\omega$; the labor–leisure choice makes the wage $w$ both the price of leisure and the source of income. Here the Slutsky decomposition acquires policy bite: a wage rise raises the opportunity cost of leisure (substitution toward work) but also enriches the household (income effect toward leisure), delivering the backward-bending labor supply curve—an endowment-driven Giffen-like reversal, and the microfoundation for debates over income-tax elasticities that run from Frank Ramsey (1927) to the modern optimal-taxation literature of Diamond, Mirrlees, and Saez.