Theorem 8: the heuristic equilibrium of no-search assortment planning is never deeper than optimal,
ProvedAssortSearch.HeurEq.theorem_8A retailer carries the most popular of product variants with preferences (no-purchase preference ), margins with whenever and an operational cost that is concave and increasing on with . Consumers behave according to the independent assortment search model with constant , but the retailer plans with the traditional no-search MNL model, re-estimating preferences from the sales of each assortment it carries.
Let be a heuristic equilibrium of this process: and maximises the predicted profit over the depths . Let be an optimal assortment, , such that every deeper assortment is strictly worse: for . Then
The heuristic equilibrium never contains more variants than the optimal assortment: a retailer that ignores consumer search risks choosing an assortment that is too narrow, never one that is too deep.
Formalization Note Independent assortment model only; the paper dismisses the overlapping case with "the similar process", which would also need the search threshold to be monotone. The paper speaks of "the optimal assortment", presuming uniqueness; the strict-deeper condition on is implied by uniqueness, and without it the conclusion can fail for a shallower optimum. is added (the positivity step needs it). is part of the equilibrium: at nothing is observed. Optimality of is over depths, as in the paper's proof.
import Mathlib import Definitions.Def_AssortSearch_HeurEq_Model
namespace AssortSearch.HeurEq
open RetailVariety.Structure
/-- Theorem 8 (p. 23), independent assortment model: the heuristic equilibrium resulting from
the iterative application of the no-search assortment planning model never contains more variants
than the optimal assortment, `x* ≤ x^o`. Here `x^o` maximises the true profit over the depths
`0, …, n` and every deeper depth is strictly worse (implied by the uniqueness of "the optimal
assortment" that the paper presumes). The heuristic equilibrium has `x* ≥ 1`; at `x* = 0`
nothing is observed and the claim would be trivial. -/
theorem theorem_8 {n : ℕ} (lam : ℝ) (m : Fin n → ℝ) (c : ℝ → ℝ)
(v : Fin n → ℝ) (v0 : ℝ)
(hv : ∀ i, 0 < v i) (hanti : Antitone v) (hv0 : 0 < v0) (hlam : 0 < lam)
(hm : ∀ j k : Fin n, v k ≤ v j → m k ≤ m j)
(hcc : ConcaveOn ℝ (Set.Icc 0 1) c) (hcm : MonotoneOn c (Set.Icc 0 1))
(hc0 : 0 ≤ c 0)
(xstar xo : ℕ) (heq : IsHeuristicEquilibrium lam m c v v0 xstar)
(hopt : IsOptimalDepth lam m c v v0 xo)
(hdeeper : ∀ x : ℕ, xo < x → x ≤ n → trueProfit lam m c v v0 x < trueProfit lam m c v v0 xo) :
xstar ≤ xo := by sorry
end AssortSearch.HeurEq
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.