Proposition 2 (corrected: ) — the relative bias is strictly decreasing in and in
ProvedGamblersFallacy.proposition_2bernoulli-trialsgamblers-fallacyinventoryp2o-correctedp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
A demand arrives at time , and then independently with probability in each of the periods . Let be the expected empirical proportion of arrivals (among those at positions ) that are immediately followed by another arrival. The relative bias is
Then:
- for every , the relative bias is strictly decreasing in on ;
- for every , the relative bias is strictly decreasing in on .
The gambler's fallacy bias of the finite-horizon estimate is therefore largest, relative to , for rare demand and short horizons, which is the regime of long-tail items.
Formalization Note is biasedEstimate n 1 p, the expectation of the positional empirical proportion over the Bernoulli paths of the periods . Both monotonicity claims are StrictAntiOn.
Differences from the source:
- correction: (i) and (ii) "for all " → ; reason: at , , so and the relative bias is for every , so (i) fails at under the strict reading adopted below (a constant is not strictly decreasing), and (ii) fails at under either reading: at the bias rises from at to at
- reading: "decreasing" → strictly decreasing (
StrictAntiOn); reason: the proof shows and concludes - reading: "for all " → ; reason: the paper's arrival probability is (p.5) and the relative bias divides by
- encoding: of (1) → expectation over the paths
Fin n → Boolas a finite weighted sum; reason: the same random variable under the same law
Preamble
import Mathlib import Definitions.Def_GamblersFallacy_biasedEstimate
Formal statement
namespace GamblersFallacy
/-- **Proposition 2 (corrected: `n ≥ 2`) — the relative bias is strictly decreasing in `p` and in
`n`.**
S. Bi, L. He, C.-P. Teo, *Taming the Long Tail: The Gambler's Fallacy in Intermittent Demand
Management*, MSOM (2023), doi:10.1287/msom.2023.1217, Proposition 2, p.14 (proof EC.2.2,
pp.ec3–ec5): "The relative bias between p and p₁(n), defined as (p − p₁(n))/p, is (i) decreasing
in p for all n, and (ii) decreasing in n for all p."
1. For every `n ≥ 2`, `p ↦ (p − p₁(n))/p` is strictly decreasing on `(0, 1)`.
2. For every `p ∈ (0, 1)`, `n ↦ (p − p₁(n))/p` is strictly decreasing on `{n ∈ ℕ | n ≥ 2}`.
**Correction.** "For all n" is corrected to `n ≥ 2`: at `n = 1`, `p̂₁(1) = G(0)`, so `p₁(1) = p`
and the relative bias is `0` for every `p` (not decreasing in `p`), while at `n = 2` it is
`(1 − p)/2 > 0` (so it increases from `n = 1` to `n = 2`).
**Formalization Note.** "Decreasing" is read as strict (the proof shows `f′(p) > 0` and concludes
`(p − p₁(n+1))/p < (p − p₁(n))/p`); `p ∈ (0, 1)` as on p.5. `p₁(n)` is `biasedEstimate n 1 p`,
the expectation of the empirical proportion `p̂₁(n)` over the `2ⁿ` Bernoulli(`p`) paths of the
periods `1, …, n` after the initial arrival at `0`. -/
theorem proposition_2 :
(∀ n : ℕ, 2 ≤ n →
StrictAntiOn (fun p : ℝ => (p - biasedEstimate n 1 p) / p) (Set.Ioo 0 1)) ∧
(∀ p ∈ Set.Ioo (0 : ℝ) 1,
StrictAntiOn (fun n : ℕ => (p - biasedEstimate n 1 p) / p) {n : ℕ | 2 ≤ n}) := by sorry
end GamblersFallacy
Source
Bi–He–Teo, Taming the Long Tail, MSOM (2023), doi:10.1287/msom.2023.1217, Proposition 2, p.14; proof EC.2.2, pp.ec3–ec5
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.