Theorem 7.1 (risk pooling): the centralized system costs no more than the decentralized one,
ProvedSupplyChainTheory.risk_poolingTheorem 7.1. distribution centers face normally distributed per-period demands with correlation coefficients , each following a base-stock policy with zero lead time, holding cost and backorder cost per unit per period. The optimal expected cost of the decentralized system is the sum of the optimal newsvendor costs, (Eq. 7.1); the centralized system merges the centers into one facing the total demand, normal with mean and variance , and its optimal cost is (Eq. 7.2). Then
This is the risk-pooling effect (Eppen 1979): pooled demand has a smaller standard deviation than the sum of the individual ones, because variances, not standard deviations, add. The book's proof evaluates both sides in closed form, and with , and applies the inequality .
Formalization Note The optimal costs are infima of the newsvendor cost over all
base-stock levels; the Gaussian laws are Mathlib's gaussianReal with variance clamped at
, so a pooled variance that is not positive gives a degenerate demand at the mean. The
correlation matrix is assumed symmetric with unit diagonal and entries at most .
import Definitions.Def_SupplyChainTheory_flexibility
namespace SupplyChainTheory
theorem risk_pooling {N : ℕ} (h p : ℝ) (hh : 0 < h) (hp : 0 < p)
(mu sig : Fin N → ℝ) (rho : Fin N → Fin N → ℝ)
(hsig : ∀ i, 0 < sig i) (hrho : ∀ i j, rho i j ≤ 1) (hrho_diag : ∀ i, rho i i = 1)
(hrho_symm : ∀ i j, rho i j = rho j i) :
optNvCost h p
(ProbabilityTheory.gaussianReal (∑ i, mu i) (Real.toNNReal (pooledVariance sig rho)))
≤ ∑ i, optNvCost h p
(ProbabilityTheory.gaussianReal (mu i) (Real.toNNReal ((sig i) ^ 2))) := by sorry
end SupplyChainTheory
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.