Proposition 7.6 — standard deviation, MAD and their one-sided versions are deviation measures
ProvedKallMayerSLP.RiskMeasures.proposition_7_6convex-analysisp2o-gran-per-chapterp2o-plan-bookp2o-v1risk-measuresstochastic-programming
Let be a probability space. The following functionals are deviation measures in the sense of the axioms (7.4):
- (I) on ;
- (J) on ;
- (K) and on ;
- (L) and on , and moreover
That is, each is subadditive, positive homogeneous, strictly positive on non-constant random variables and zero on constants, and invariant under adding a constant.
Formalization Note "Constant" means almost surely constant. On the semideviations in (K) may be , so (K) is stated for functionals with values in .
Differences from the source:
- reading: "ϑ non-constant" / "constant" in (D3) → not almost surely constant / almost surely constant; reason: every functional listed is an expectation and cannot see null sets; under the pointwise reading (D3) fails for each of them
- reading: σ⁺ and σ⁻ of (K) on 𝓛¹₁, as real-valued functions (p.182: "ρ : Υ → ℝ a real–valued function") → σ⁺ and σ⁻ of (K) with values in [0, ∞] on 𝓛¹₁; homogeneity ρ(λϑ) = λρ(ϑ) with λ ≥ 0 cast to [0, ∞]; subadditivity in [0, ∞]; reason: on 𝓛¹₁ the second moment can be infinite (density 2t⁻³ on [1, ∞)), so the printed real-valued reading is ill-posed; the extended-valued reading keeps the printed domain and contains the real-valued statement on 𝓛²₁ (paper.md, reading 2), where the values are finite
- encoding: 𝓛^p₁ (p = 1, 2, ∞) of (1.6) → Lspace ℝ p P: functions Ω → ℝ with MemLp · p P, a submodule of Ω → ℝ; reason: random variables with finite p-th moment (resp. bounded support); same instances
Preamble
import Mathlib import Definitions.Def_KallMayerSLP_RiskMeasures_Lspace import Definitions.Def_KallMayerSLP_RiskMeasures_IsDeviationMeasure import Definitions.Def_KallMayerSLP_RiskMeasures_IsDeviationMeasureENN import Definitions.Def_KallMayerSLP_RiskMeasures_rhoStd import Definitions.Def_KallMayerSLP_RiskMeasures_rhoMAD import Definitions.Def_KallMayerSLP_RiskMeasures_rhoStdPlus import Definitions.Def_KallMayerSLP_RiskMeasures_rhoStdMinus import Definitions.Def_KallMayerSLP_RiskMeasures_rhoMADPlus import Definitions.Def_KallMayerSLP_RiskMeasures_rhoMADMinus open MeasureTheory open scoped ENNReal
Formal statement
namespace KallMayerSLP.RiskMeasures
/-- **Proposition 7.6** (deviation measures).
Kall & Mayer, *Stochastic Linear Programming* (2005), Chapter 2, §7.2, Prop. 7.6, p.188 (PDF 198). The following risk measures are deviation measures
according to the axioms (7.4):
(I) `ρ_Std(ϑ) := σ(ϑ) := √(𝔼[(ϑ − 𝔼[ϑ])²])`, `ϑ ∈ 𝓛²₁`;
(J) `ρ_MAD(ϑ) := 𝔼[|ϑ − 𝔼[ϑ]|]`, `ϑ ∈ 𝓛¹₁`;
(K) `ρ⁺_Std(ϑ) := √(𝔼[((ϑ − 𝔼[ϑ])⁺)²])` and `ρ⁻_Std(ϑ) := √(𝔼[((ϑ − 𝔼[ϑ])⁻)²])`, `ϑ ∈ 𝓛¹₁`;
(L) `ρ⁺_MAD(ϑ) := 𝔼[(ϑ − 𝔼[ϑ])⁺] = ½ρ_MAD(ϑ)` and `ρ⁻_MAD(ϑ) := 𝔼[(ϑ − 𝔼[ϑ])⁻] = ½ρ_MAD(ϑ)`,
`ϑ ∈ 𝓛¹₁`.
**Formalization Note.** `𝓛^p₁` is `Lspace ℝ p P`. "Constant" in (7.4) is "almost surely
constant". On the printed domain `𝓛¹₁` of (K) the semideviations can be `+∞`, so (K) is stated
with values in `[0, ∞]` (`IsDeviationMeasureENN`); on `𝓛²₁` they are finite. The identities
printed in (L) are part of the statement. -/
theorem proposition_7_6 {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P] :
IsDeviationMeasure P (Lspace ℝ 2 P) (rhoStd P) ∧
IsDeviationMeasure P (Lspace ℝ 1 P) (rhoMAD P) ∧
(IsDeviationMeasureENN P (Lspace ℝ 1 P) (rhoStdPlus P) ∧
IsDeviationMeasureENN P (Lspace ℝ 1 P) (rhoStdMinus P)) ∧
(IsDeviationMeasure P (Lspace ℝ 1 P) (rhoMADPlus P) ∧
IsDeviationMeasure P (Lspace ℝ 1 P) (rhoMADMinus P) ∧
∀ ϑ ∈ Lspace ℝ 1 P, rhoMADPlus P ϑ = rhoMAD P ϑ / 2 ∧ rhoMADMinus P ϑ = rhoMAD P ϑ / 2) := by sorry
end KallMayerSLP.RiskMeasures
Source
Kall & Mayer, Stochastic Linear Programming (2005), Chapter 2, §7.2, Proposition 7.6, p.188 (PDF 198); axioms (7.4), p.183 (PDF 193)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.