Lemma 5, p. 17 — P(D_T ≤ D̂_T) = 1 − β = λ − ψ, λP^M(λZ_T ≤ 1) + P(λZ_T ≥ 1) = 1 − β (38), and P(𝒟) = E[(λZ_T) ∧ ∆] (39)
ProvedProdHedgeCVaR.Joint.lemma_5change-of-measureconditional-probabilitygaussianp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1
Throughout, is a probability space carrying the terminal values of the model of §3: the asset-driven demand component , the asset noise , both measurable with respect to the hedger's information (the paper's ), and independent of . The demand is , the production payoff , and is the density of the martingale measure, . The parameters satisfy and . Fix and a real number . Let be as in (30) (with when and when ), and (37), (29), and . Then
and
These identities convert the closed form (34) into the expression (42) for from which the optimality equations are derived.
Differences from the source:
- generalization: Q the production quantity (Q ≥ 0 in context) → every real Q; reason: the identities (39) hold for every real Q; the paper's case Q ≥ 0 is an instance
- encoding: P(ξ ≤ Φ⁻¹(λZ_T) ∧ (Q − A_T)/(σ̃√T)) with Φ⁻¹ possibly ±∞ → the comparison is made in EReal; reason: Φ⁻¹(λZ_T) is ±∞ at the endpoints; the event is the same
- encoding: D̂_T of (30) with Φ⁻¹(0) = −∞, Φ⁻¹(1) = +∞; E^M, P^M under dP^M/dP = Z_T (26) → D̂_T(λ) in EReal: −∞ if λZ_T ≤ 0, +∞ if λZ_T ≥ 1, else A_T + σ̃√T Φ⁻¹(λZ_T) with Φ⁻¹(u) = sup{x : Φ(x) ≤ u}; E^M[Y] = E[Z_T Y], P^M(S) = E[Z_T 1_S]; reason: Φ is continuous and strictly increasing onto (0, 1), so the order inverse is Φ⁻¹ there; the endpoint values are the paper's conventions (p. 15); E^M and P^M are the expectations under the measure with density Z_T
- generalization: the hedger's information is σ(A_T, X_T), with A_T adapted to the asset filtration F^X and B̃ a Brownian motion independent of B (§3, pp. 10–12) → any sub-σ-algebra G of the probability space such that A_T and B_T are G-measurable and ξ = B̃_T/√T is independent of G; reason: σ(A_T, X_T) = σ(A_T, B_T) (X_T is a strictly increasing function of B_T) satisfies these properties, so the paper's model is the instance G = σ(A_T, X_T); nothing else about G enters the statement
- encoding: the dynamic model: GBM asset (17), demand dD_t = dA_t + σ̃dB̃_t (18)–(19), density process (26) on [0, T] → the terminal values only: A_T ≥ 0, W = B_T with law N(0, T), ξ with law N(0, 1); Z_T = exp(−ηW − η²T/2), D_T = A_T + σ̃√T ξ; reason: every object in the statement is a function of (A_T, B_T, B̃_T); B̃_T has the law of √T ξ
- reading: p is the unit profit, c the net unit cost (p. 5), σ̃ > 0 (p. 10), T > 0 (p. 5), budget C > 0 (p. 12), η ≠ 0 (p. 13) → hypotheses p > 0, c > 0, σ̃ > 0, T > 0, C > 0, η ≠ 0 (Params.Valid); reason: the standing assumptions of §§2–3; c > 0 is needed: with c = 0 the second equation of (45) has no root, since its left-hand side stays below p(1 − α)
Preamble
import Mathlib import Definitions.Def_BlackScholesModel_Core import Definitions.Def_ProdHedgeCVaR_Joint_Model
Formal statement
namespace ProdHedgeCVaR.Joint
open MeasureTheory ProbabilityTheory Filter Topology
/-- Lemma 5, p. 17: (38) and (39), for every λ ≥ 0 and every Q. -/
theorem lemma_5 {Ω : Type*} [MeasurableSpace Ω] (M : Model Ω) (hM : M.Valid)
(lam : ℝ) (hlam : 0 ≤ lam) (Q : ℝ) :
M.P.real {ω | ((M.D ω : ℝ) : EReal) ≤ M.Dhat lam ω} = 1 - M.beta lam ∧
1 - M.beta lam = lam - M.psi lam ∧
lam * M.PM {ω | lam * M.Z ω ≤ 1} + M.P.real {ω | 1 ≤ lam * M.Z ω} = 1 - M.beta lam ∧
M.P.real (M.Dset lam Q) =
M.P.real {ω | ((M.xi ω : ℝ) : EReal) ≤
min (stdNormalInvE (lam * M.Z ω))
(((Q - M.A ω) / (M.θ.sigT * Real.sqrt M.θ.T) : ℝ) : EReal)} ∧
M.P.real {ω | ((M.xi ω : ℝ) : EReal) ≤
min (stdNormalInvE (lam * M.Z ω))
(((Q - M.A ω) / (M.θ.sigT * Real.sqrt M.θ.T) : ℝ) : EReal)} =
∫ ω, min (lam * M.Z ω) (M.Delta Q ω) ∂M.P := by sorry
end ProdHedgeCVaR.Joint
Source
Wang, Yao, Production Planning with Risk Hedging under a CVaR Objective, SSRN 3875464 (April 2020), p. 17, Lemma 5, (37)–(39)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.