Shephard's Lemma

Shephard's lemma is a cornerstone duality result in business and economics theory that links the value functions of constrained optimization problems in production and consumption directly to optimal input and commodity demand schedules via first-order partial differentiation. Proved formally by American mathematician and economist Ronald W. Shephard in his seminal 1953 Princeton University Press monograph Cost and Production Functions (and independently applied to consumer expenditure functions by Lionel W. McKenzie in 1957), the lemma states that if a firm's minimum cost function c(w, y) is differentiable at factor price vector w > 0 and output level y, then the unique cost-minimizing conditional factor demand x_i^*(w, y) for input i is equal to the partial derivative of the cost function with respect to that input's price w_i:

In the dual domain of consumer theory, differentiating a household's minimum expenditure function e(p, u) with respect to the price p_i of good i at fixed utility level u yields the Hicksian (compensated) demand function h_i(p, u) = ∂e(p, u) / ∂p_i. Together with Hotelling's lemma (for unconstrained profit functions) and Roy's identity (for indirect utility functions), Shephard's lemma forms the analytical engine of modern microeconomic duality theory, allowing econometricians to estimate systems of input and commodity demand equations without having to solve nonlinear systems of first-order Lagrangian tangency conditions.

The Primal Cost-Minimization Problem

Consider a firm employing n productive inputs x = (x₁, x₂, …, x_n) ∈ ℝ₊ⁿ purchased in competitive factor markets at positive input prices w = (w₁, w₂, …, w_n) ≫ 0 to produce a target scalar output y via a continuous, strictly quasi-concave production function f(x) (such as a Cobb–Douglas production function or Constant Elasticity of Substitution technology). The firm's primal cost-minimization problem is:

Assuming an interior solution x^*(w, y) ≫ 0 where the production constraint binds (f(x^*) = y), the associated Lagrangian function with shadow multiplier λ(w, y) (which represents marginal cost ∂c / ∂y) is:

The first-order necessary conditions for an interior cost minimum require that for every input j ∈ {1, …, n}:

Application of the Envelope Theorem

Substituting the optimal conditional factor demand bundle x^*(w, y) into the objective function defines the minimum cost identity c(w, y) ≡ ∑_j=1ⁿ w_j · x_j^*(w, y). Differentiating c(w, y) partially with respect to the i-th factor price w_i via the chain rule yields both a direct price effect and an indirect factor-reallocation effect:

Substituting the first-order tangency condition w_j = λ · (∂f / ∂x_j) into the summation term gives:

However, because output y is held constant along the isoquant f(x^*(w, y)) ≡ y, differentiating the constraint identity f(x^*(w, y)) = y with respect to w_i implies that the weighted sum of input adjustments along the isoquant is identically zero:

Thus, the indirect factor-substitution term vanishes at the first order (the envelope theorem), leaving Shephard's lemma:

Intuitively, for an infinitesimal increase dw_i in the price of input i, the firm's marginal adjustment of its input mix along the isoquant has zero first-order impact on total cost because at the optimum the marginal cost saved by reducing one input equals the marginal cost incurred by increasing its substitute; hence the rate of increase in minimum total cost equals the quantity x_i^* of input i currently employed.

In consumer choice theory, a household with continuous, locally non-satiated, strictly quasi-concave utility function U(x) facing commodity price vector p = (p₁, …, p_n) ≫ 0 minimizes total monetary expenditure required to attain a target marginal utility and total utility threshold u:

The solution vector h(p, u) = (h₁(p, u), …, h_n(p, u)) is the Hicksian (or utility-compensated) demand bundle, named after Nobel laureate Sir John Hicks. By the exact same envelope-theorem logic, differentiating the expenditure function e(p, u) partially with respect to commodity price p_i yields the Hicksian compensated demand for good i:

This identity provides the cleanest modern proof of the Slutsky equation, which links uncompensated Marshallian demand x_i(p, m) to compensated Hicksian demand via the identity h_i(p, u) ≡ x_i(p, e(p, u)). Differentiating both sides with respect to p_j and substituting ∂e(p, u) / ∂p_j = h_j(p, u) = x_j(p, m) immediately yields the Slutsky decomposition without lengthy bordered-Hessian determinant algebra.

Shephard's lemma reveals why every valid cost function c(w, y) (and expenditure function e(p, u)) must satisfy four structural regularity conditions—independent of whether the underlying production function f(x) is differentiable or even quasi-concave:

1. Homogeneity of Degree 1 in Prices (Euler's Theorem): Scaling all input prices by λ > 0 scales minimum cost by λ: c(λw, y) = λ · c(w, y). By Euler's homogeneous function theorem, differentiating c(w, y) is homogeneous of degree 0 in w—meaning conditional factor demands x_i^*(λw, y) = x_i^*(w, y) depend only on relative factor prices—and ∑_i=1ⁿ w_i (∂c / ∂w_i) = ∑_i=1ⁿ w_i x_i^* = c(w, y).
2. Concavity in Prices and Downward-Sloping Conditional Demand: Consider a firm facing initial input prices w⁰ with optimal input bundle x⁰ = x^*(w⁰, y). If factor prices change to w, the firm could passively keep using the old bundle x⁰ at passive linear cost w · x⁰, or it could re-optimize its input mix along the isoquant to achieve true minimum cost c(w, y) ≤ w · x⁰. Because c(w, y) is the pointwise infimum of a family of linear functions of w (w ↦ w · x for x in the input requirement set V(y)), the cost function c(w, y) is globally concave in w.
3. Negative Semidefiniteness and Symmetry of the Substitution Matrix: Differentiating Shephard's lemma a second time with respect to w_j yields the Hessian matrix of the cost function H_i,j = ∂²c(w, y) / (∂w_i ∂w_j) = ∂x_i^*(w, y) / ∂w_j. By Young's theorem on the equality of mixed partial derivatives, cross-price substitution effects are symmetric (∂x_i^* / ∂w_j = ∂x_j^* / ∂w_i), and because c(w, y) is concave in w, its Hessian matrix is negative semidefinite, guaranteeing that every diagonal own-price conditional demand derivative is non-positive:

Unlike Marshallian consumer demand—which can theoretically slope upward for a Giffen good when a negative income effect outweighs the substitution effect—conditional factor demands and Hicksian compensated demands can never slope upward because output y (or utility u) is held fixed.

In empirical econometrics and general equilibrium theory, researchers rarely specify a restrictive primal production function and solve for input demands analytically. Instead, following Laurits Christensen, Dale Jorgenson, and Lawrence Lau (1973) and Erwin Diewert (1971), econometricians specify a flexible functional form directly for the cost function—such as the Transcendental Logarithmic (Translog) cost function or the Generalized Leontief cost function—and apply Shephard's lemma in logarithmic form to derive linear factor cost-share equations:

For a Translog cost function ln c(w, y) = α₀ + ∑_i α_i ln w_i + ½ ∑_i ∑_j γ_i,j ln w_i ln w_j + α_y ln y, applying logarithmic Shephard's lemma yields a linear system of cost-share equations s_i = α_i + ∑_j γ_i,j ln w_j that can be estimated directly via Zellner's seemingly unrelated regressions (SUR) while testing cross-equation symmetry (γ_i,j = γ_j,i).

What happens to Shephard's lemma when the production function has sharp corners (Leontief technology)?

For a fixed-proportions Leontief production function y = min{x₁ / a₁, x₂ / a₂}, the isoquants have right-angle kinks where the marginal rate of technical substitution is undefined. Nevertheless, the cost function c(w₁, w₂, y) = (a₁w₁ + a₂w₂)y is completely smooth and linear in input prices! Applying Shephard's lemma ∂c / ∂w_i = a_iy recovers the exact conditional factor demand effortlessly despite the non-differentiability of the primal production function. Conversely, if an isoquant has flat linear segments, the cost function has a kink at the price ratio matching that slope, and the subdifferential ∂c / ∂w_i yields the convex set of cost-minimizing input bundles.

How is Roy's identity derived from Shephard's lemma?

Let v(p, m) denote the indirect utility function (maximum utility attainable at prices p and income m). By duality between expenditure minimization and utility maximization, v(p, e(p, u)) ≡ u. Differentiating this identity with respect to p_i while holding u constant gives (∂v / ∂p_i) + (∂v / ∂m) · (∂e / ∂p_i) = 0. Substituting Shephard's lemma ∂e(p, u) / ∂p_i = h_i(p, u) = x_i(p, m) and rearranging yields Roy's identity: x_i(p, m) = −(∂v / ∂p_i) ÷ (∂v / ∂m).

What is the difference between Shephard's lemma and Hotelling's lemma?

Shephard's lemma applies to cost minimization (where output y is held fixed), differentiating c(w, y) with respect to w_i to obtain conditional factor demand x_i^*(w, y). Hotelling's lemma (proved by Harold Hotelling in 1932) applies to unconstrained profit maximization (where output y adjusts endogenously to output price p), differentiating the maximum profit function π(p, w) with respect to output price p to obtain output supply y^*(p, w) = ∂π / ∂p and with respect to input price w_i to obtain unconditional factor demand x_i^*(p, w) = −∂π / ∂w_i.

1. Shephard, R. W. (1953). Cost and Production Functions. Princeton University Press. https://press.princeton.edu/books/hardcover/9780691652252/cost-and-production-functions
2. McKenzie, L. W. (1957). "Demand Theory Without a Utility Index." The Review of Economic Studies, 24(3), 185–189. https://doi.org/10.2307/2296067
3. Diewert, W. E. (1971). "An Application of the Shephard Duality Theorem: A Generalized Leontief Production Function." Journal of Political Economy, 79(3), 481–507. https://doi.org/10.1086/259764
4. Christensen, L. R., Jorgenson, D. W., & Lau, L. J. (1973). "Transcendental Logarithmic Production Frontiers." The Review of Economics and Statistics, 55(1), 28–45. https://doi.org/10.2307/1927992
5. Jorgenson, D. W. (1986). "Econometric Methods for Modeling Producer Behavior." In Z. Griliches & M. D. Intriligator (Eds.), Handbook of Econometrics, Vol. 3 (pp. 1841–1915). Elsevier. https://doi.org/10.1016/S1573-4412(86)03011-803011-8)