Proof of Theorem 7, p. 8 — for σX, σY > 0, δ* is the unique minimizer of h(δ) = σY f((δ − µY)/σY) + σX f((µX − δ)/σX)
ProvedMeanVarASD.AFSD.deltaStar_minimizes_hconvex-analysisp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-dominance
Let , , and
Then for every , with equality only at .
Since holds exactly when is at least a quantity increasing in , this shows that is the choice of giving the smallest threshold obtainable by this argument.
Formalization Note The positivity of both standard deviations is needed for to be defined; the paper uses it implicitly in dividing by and .
Preamble
import Mathlib import Definitions.Def_MeanVarASD_AFSD_Defs
Formal statement
namespace MeanVarASD.AFSD
/-- Proof of Theorem 7, p. 8: for `σX, σY > 0`, `δ*` is the unique minimizer of `h`. -/
theorem deltaStar_minimizes_h (μX μY σX σY : ℝ) (hσX : 0 < σX) (hσY : 0 < σY) :
∀ δ, hAux μX μY σX σY (deltaStar μX μY σX σY) ≤ hAux μX μY σX σY δ ∧
(hAux μX μY σX σY δ = hAux μX μY σX σY (deltaStar μX μY σX σY) →
δ = deltaStar μX μY σX σY) := by sorry
end MeanVarASD.AFSD
Source
Müller, Scarsini, Tsetlin & Winkler, Ranking distributions when only means and variances are known, INSEAD Working Paper 2020/36/DSC (31 July 2020; SSRN 3440103), p. 8 (PDF 9), proof of Theorem 7, first-order condition h′(δ) = 0 and its unique solution δ*
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.