Theorem 10, p. 14 — φ_β(x) = min_α F_β(x, α), argmin = [α_β(x), α⁺_β(x)], F_β(x, ·) finite convex
ProvedRUCVaR.MinFormula.theorem_10Theorem 10 (fundamental minimization formula). Let be a Borel probability measure on , the law of the random vector , and let be the loss of a decision , continuous in on , measurable in , with for each . Let and . Write
for the β-VaR, for the upper β-VaR, and for the β-CVaR, the mean of the β-tail distribution. Then:
- is finite for every , and is convex, hence continuous;
- the minimum of is attained and equals the β-CVaR:
- , and the argmin is the nonempty closed bounded interval
- in particular and . (30)
VaR and CVaR of the loss are thus computed together by minimizing a convex function of one real variable, for any loss distribution, discrete or continuous.
Formalization Note "Finite" is integrability of for every . "min with attainment" is stated as: is the least element of the range of . The argmin is the set of global minimizers. is defined as the mean of the β-tail distribution of Definition 3, not through , so (28) is not true by definition. All standing assumptions of §2 are carried even though only integrability of the loss at the given is used.
import Mathlib import Definitions.Def_MultistageStochastic_RiskFunctional import Definitions.Def_RUCVaR_MinFormula_Setting open MeasureTheory Filter Topology
namespace RUCVaR.MinFormula
/-- Theorem 10 (fundamental minimization formula), p. 14, (28)–(30). -/
theorem theorem_10 {m n : ℕ} (P : Measure (Fin m → ℝ)) [IsProbabilityMeasure P]
(X : Set (Fin n → ℝ)) (f : (Fin n → ℝ) → (Fin m → ℝ) → ℝ)
(hf_cont : ∀ y, ContinuousOn (fun x => f x y) X) (hf_meas : ∀ x, Measurable (f x))
(hf_int : ∀ x ∈ X, Integrable (f x) P) (β : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1)
(x : Fin n → ℝ) (hx : x ∈ X) :
(∀ α : ℝ, Integrable (fun y => max (f x y - α) 0) P) ∧
ConvexOn ℝ Set.univ (Fbeta P f β x) ∧ Continuous (Fbeta P f β x) ∧
IsLeast (Set.range (Fbeta P f β x)) (CVaR P f β x) ∧
(VaR P f β x ≤ VaRPlus P f β x ∧
{α : ℝ | IsMinOn (Fbeta P f β x) Set.univ α} = Set.Icc (VaR P f β x) (VaRPlus P f β x)) ∧
(IsMinOn (Fbeta P f β x) Set.univ (VaR P f β x) ∧
CVaR P f β x = Fbeta P f β x (VaR P f β x)) := by sorry
end RUCVaR.MinFormula
Confirmed by the mission captain (proposal self-audit).
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.