Proof of Proposition 1, p. 780 —
ProvedCVPricing.CertEquiv.ls_representationleast-squaresp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1statistics
Let be prices with , let be noise values, and let . For every , the least squares estimates computed from satisfy
where are the sample means, and .
This expresses the estimation error of ordinary least squares through the two scalar sums and , which the proof of Proposition 1 then controls.
Formalization Note The paper states the representation in the case of its induction; it holds for every price sequence with (which makes ), and is stated in that generality. The estimate is the referenced lsEstimateOf, the solution of the normal equations (4).
Preamble
import Mathlib import Definitions.Def_CVPricing_CertEquiv_EventA
Formal statement
namespace CVPricing.CertEquiv
open KeskinZeevi.SufficientConditions
theorem ls_representation (a₀ a₁ : ℝ) (p e : ℕ → ℝ) (hp : p 1 ≠ p 2) {t : ℕ} (ht : 2 ≤ t) :
lsEstimateOf p (fun s => a₀ + a₁ * p s + e s) t =
(a₀ + (avgPriceOf e t - avgPriceOf p t * crossSum p e t / infoMetricOf p t),
a₁ + crossSum p e t / infoMetricOf p t) := by sorry
end CVPricing.CertEquiv
Source
den Boer, Zwart, Simultaneously Learning and Optimizing Using Controlled Variance Pricing, Management Science 60(3):770–783 (2014), p. 780 (PDF 12), Appendix, proof of Proposition 1, Case t ≥ 3, the display of the least squares estimates
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.