Iterative assortment planning estimates relative preferences correctly:
ProvedAssortSearch.HeurEq.estimates_ratioassortment-planningconsumer-searchmultinomial-logitp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Consider the independent assortment search model with preferences , and . For an assortment of depth let be the preferences the retailer estimates from the observed demands, normalised by , with depth-test estimates for the variants not carried. Then for all product variants ,
Given that the preference of variant 1 is held fixed, the iterative procedure therefore estimates every product variant's preference correctly under any assortment; only the no-purchase estimate is biased.
Formalization Note 0-based indices: the paper's is i < j in Fin n. The depth satisfies , the range where the estimates are defined.
Preamble
import Mathlib import Definitions.Def_AssortSearch_HeurEq_Model
Formal statement
namespace AssortSearch.HeurEq
open RetailVariety.Structure
/-- Iterative assortment planning estimates the relative preferences of the product variants
correctly under any assortment (p. 21): for every depth `1 ≤ x ≤ n` and all variants
`0 < i < j ≤ n` (0-based: `i < j` in `Fin n`), `v̂_i(x) / v̂_j(x) = v_i / v_j`. Variants outside
`x` carry the depth-test estimates `v̂_i(n)`. -/
theorem estimates_ratio {n : ℕ} (lam : ℝ) (v : Fin n → ℝ) (v0 : ℝ)
(hv : ∀ i, 0 < v i) (hv0 : 0 < v0) (hlam : 0 < lam)
(x : ℕ) (hx1 : 1 ≤ x) (hxn : x ≤ n) (i j : Fin n) (hij : i < j) :
estPref lam v v0 x i / estPref lam v v0 x j = v i / v j := by sorry
end AssortSearch.HeurEq
Source
Cachon, Terwiesch & Xu, Retail Assortment Planning in the Presence of Consumer Search, working paper (Dec. 20, 2002), p. 21 (PDF 23), §5.1, display v̂_i(x)/v̂_j(x) = v_i/v_j
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.