Shadow Prices In Multi-Resource Production

What Does One More Unit of a Resource Buy?

Production decisions are often limited by several resources at once. A factory may have a fixed number of labor hours and a limited supply of material. A profit-maximizing plan uses those resources to choose output levels, but the plan alone does not tell a manager how valuable a small increase in capacity would be.

A Lagrange multiplier attached to a resource constraint provides that local value. It estimates the increase in maximum profit from one additional unit of the resource, provided the optimum changes smoothly and the same constraints remain active. The estimate is marginal: it describes a small change near the current resource levels, not a guaranteed value for any size of expansion.

Profit as a Function of Resource Availability

Let \(x\) and \(y\) denote production quantities, and let \(L\) and \(M\) denote the available amounts of two resources. Suppose profit is \(f(x,y)\), and resource use is described by \(g_1(x,y)\) and \(g_2(x,y)\). The maximum attainable profit at capacities \(L\) and \(M\) is the value function

V(L,M)=\max f(x,y) \quad\text{subject to}\quad g_1(x,y)\le L,\qquad g_2(x,y)\le M.

When \(V\) is differentiable, \(\frac{\partial V}{\partial L}\) measures the approximate change in maximum profit per extra unit of the first resource; \(\frac{\partial V}{\partial M}\) does the same for the second. These derivatives are the economic meaning of the corresponding shadow prices.

Definition: A shadow price for a resource is the local rate of change of optimized profit with respect to that resource’s available amount. Under suitable regularity and stability conditions, it equals the Lagrange multiplier for that resource constraint.

We use the convention that a maximization problem with constraints \(g_i(\mathbf{x})\le b_i\) has Lagrangian

\mathcal{L}(\mathbf{x},\boldsymbol{\lambda}) =f(\mathbf{x})+\sum_i\lambda_i\bigl(b_i-g_i(\mathbf{x})\bigr), \qquad \lambda_i\ge 0.

With this sign convention, a positive multiplier indicates that a small increase in the corresponding capacity can increase maximum profit. Its units are profit per unit of that resource. For example, if profit is measured in thousands of dollars and labor in hours, the labor multiplier is measured in thousands of dollars per hour.

A Two-Resource Production Model

Consider two products, with \(x\) and \(y\) measured in production lots. Profit, in thousands of dollars per planning period, is modeled by

$$ f(x,y)=20x+16y-x^2-y^2. $$

The first resource is labor, measured in hours, with \(x+y\) hours required. The second is material, measured in material units, with \(2x+y\) units required. At capacities \(L=7\) labor hours and \(M=11\) material units, the planning problem is

\max_{x,y\ge 0} \;20x+16y-x^2-y^2 \quad\text{subject to}\quad x+y\le 7,\qquad 2x+y\le 11.

Worked Example: Find the Optimum and Its Shadow Prices

First test the point where both resource constraints bind. Solve \(x+y=7\) and \(2x+y=11\). Subtracting the first equation from the second gives \(x=4\), and then \(y=7-4=3\). Both production quantities are positive.

The gradient of profit at this point is

\nabla f(4,3) = \bigl(20-2(4),\,16-2(3)\bigr) =(12,10).

Let \(\lambda\) be the labor multiplier and \(\mu\) the material multiplier. Stationarity for the two binding resource constraints requires

(12,10)=\lambda(1,1)+\mu(2,1).

Equating components gives \(\lambda+2\mu=12\) and \(\lambda+\mu=10\). Subtracting the second equation from the first yields \(\mu=2\), and then \(\lambda=8\). Both multipliers are nonnegative, as required.

Profit at the candidate point is \(20(4)+16(3)-4^2-3^2=80+48-16-9=103\), or \(103\) thousand dollars. The resource use is \(4+3=7\) labor hours and \(2(4)+3=11\) material units, so both constraints bind exactly. The Hessian of \(f\) is \(\begin{pmatrix}-2&0\\0&-2\end{pmatrix}\), which is negative definite; thus \(f\) is strictly concave. Since the feasible region is convex, this feasible stationary point is the unique global maximum.

The multipliers therefore assign a local value of \(8\) thousand dollars per additional labor hour and \(2\) thousand dollars per additional material unit, as long as the local conditions for this interpretation continue to hold.

Why the Multiplier Gives a Marginal Value

The relationship follows by differentiating profit along a smoothly changing optimal plan. The argument below states the key assumptions: the objective and resource-use functions are differentiable, the optimizer and multipliers vary differentiably with capacities, and the active constraints remain active nearby. A suitable constraint qualification is also needed for the Lagrange multiplier conditions to apply.

Derivation: Suppose the relevant constraints \(g_i(\mathbf{x})\le b_i\) remain binding at an optimum \(\mathbf{x}^*(\mathbf{b})\). Stationarity gives \(\nabla f(\mathbf{x}^*)=\sum_i\lambda_i\nabla g_i(\mathbf{x}^*)\). Along the binding constraints, \(g_i(\mathbf{x}^*(\mathbf{b}))=b_i\), so a small capacity change satisfies \(\nabla g_i(\mathbf{x}^*)\cdot d\mathbf{x}^*=db_i\). Differentiating optimized profit therefore gives

dV =\nabla f(\mathbf{x}^*)\cdot d\mathbf{x}^* =\sum_i\lambda_i\nabla g_i(\mathbf{x}^*)\cdot d\mathbf{x}^* =\sum_i\lambda_i\,db_i.

In particular, changing just capacity \(b_j\) gives \(\frac{\partial V}{\partial b_j}=\lambda_j\). This is a local derivative, not an exact prediction for a finite capacity increase. For a small change \(\Delta b_j\), the first-order estimate is

\Delta V\approx\lambda_j\Delta b_j.

The approximation may become less accurate as the change grows. If the set of binding constraints changes, or if the optimizer switches to a different regime, the value function may have a different derivative or fail to be differentiable at the transition.

Check the Labor Shadow Price Against a Perturbation

Worked Example: Add One-Hundredth of a Labor Hour

Keep material fixed at \(M=11\), and increase labor from \(L=7\) to \(L=7.01\). If both constraints remain binding, solve \(x+y=7.01\) and \(2x+y=11\). Subtraction gives \(x=11-7.01=3.99\), followed by \(y=7.01-3.99=3.02\). The labor use is \(3.99+3.02=7.01\), and material use is \(2(3.99)+3.02=7.98+3.02=11\).

Recalculate the profit:

\begin{aligned} f(3.99,3.02) &=20(3.99)+16(3.02)-(3.99)^2-(3.02)^2\\ &=79.8+48.32-15.9201-9.1204\\ &=103.0795. \end{aligned}

The original maximum was \(103\), so this perturbation increases maximum profit by \(0.0795\) thousand dollars. The multiplier prediction is \(8(0.01)=0.08\) thousand dollars. The exact increase is \(0.0005\) thousand dollars below the first-order prediction; per added labor hour, the observed increase over this small interval is \(0.0795/0.01=7.95\) thousand dollars. This is close to the local shadow price of \(8\).

For this model, the calculation can be checked more generally. With \(M=11\) and labor capacity \(L=7+h\), binding constraints give \(x=4-h\) and \(y=3+2h\). Substitution into the objective yields \(V(7+h,11)=103+8h-5h^2\) near \(h=0\). The derivative at \(h=0\) is \(8\), while the \(-5h^2\) term explains the small discrepancy for a finite change.

Check the Material Shadow Price

Worked Example: Add One-Hundredth of a Material Unit

Now hold labor at \(L=7\) and increase material from \(M=11\) to \(M=11.01\). If both constraints bind, solve \(x+y=7\) and \(2x+y=11.01\). Subtracting gives \(x=4.01\), and then \(y=7-4.01=2.99\). The resource checks are \(4.01+2.99=7\) labor hours and \(2(4.01)+2.99=8.02+2.99=11.01\) material units.

The resulting profit is

\begin{aligned} f(4.01,2.99) &=20(4.01)+16(2.99)-(4.01)^2-(2.99)^2\\ &=80.2+47.84-16.0801-8.9401\\ &=103.0198. \end{aligned}

The increase is \(103.0198-103=0.0198\) thousand dollars. The multiplier predicts \(2(0.01)=0.02\) thousand dollars. Again, the finite change is close to the local prediction. In fact, for this perturbation the change is \(2h-2h^2\), where \(h\) is the material increase, so its derivative at zero is exactly \(2\).

When a Resource Has No Local Scarcity Value

A constraint that is slack at the optimum has multiplier zero under the usual complementary-slackness conditions. If it stays slack after a small capacity change, adding more of that resource does not change the optimum: another condition, or the profit-maximizing balance among products, is limiting production instead. A zero multiplier is a local statement; it does not mean the resource could never become valuable after other capacities or conditions change.

Worked Example: A Slack Material Constraint

Keep labor at \(L=7\), but increase available material to \(M=12\). Along the binding labor constraint \(y=7-x\), profit becomes

$$ f(x,7-x)=20x+16(7-x)-x^2-(7-x)^2 =-2x^2+18x+63. $$

Its derivative with respect to \(x\) is \(-4x+18\), which is zero at \(x=4.5\). Thus \(y=7-4.5=2.5\). The material required is \(2(4.5)+2.5=11.5\), below the available \(12\), so the material constraint is slack. Profit is \(20(4.5)+16(2.5)-(4.5)^2-(2.5)^2=90+40-20.25-6.25=103.5\) thousand dollars.

At this point \(\nabla f=(20-9,16-5)=(11,11)\), so stationarity for the binding labor constraint gives labor multiplier \(11\); the slack material constraint has multiplier \(0\). A small increase in the material limit leaves the optimum unchanged while the constraint remains slack. This verifies the zero local shadow price in this case.

Interpreting Shadow Prices in Decisions

A shadow price is the marginal contribution of extra capacity to optimized profit, before accounting for any price paid to acquire that capacity. If one additional labor hour costs less than its estimated shadow value, purchasing it may improve net profit, provided the estimate remains relevant. The comparison must use consistent units: a multiplier in thousands of dollars per hour must be converted or compared with a labor cost in the same units.

Multipliers also depend on the model. They reflect its profit function, current resource limits, and assumptions about production. They should not be treated automatically as market prices, nor as reliable estimates far from the current optimum. A practical interpretation checks the resource units, verifies which constraints bind, and compares a small predicted change with a recalculated optimum. The perturbation checks here show both the usefulness of the marginal estimate and its expected limitation: it captures the first-order change, while finite changes can include higher-order effects.

Check Your Understanding

Answer these questions using the production model and definitions above.

Under the Lagrangian convention in this tutorial, what does a positive multiplier for a labor constraint estimate?
At \(L=7\) and \(M=11\), what are the units and values of the labor and material shadow prices?
For a labor increase of \(0.01\) hour in the worked calculation, what is the multiplier-based estimate, and what is the actual profit increase?
Why is a shadow-price estimate not generally exact for a finite resource increase?
What does a zero multiplier for a slack resource constraint say about a sufficiently small increase in that resource, while the constraint remains slack?