Lemma 1, p. 17 — X′_ε =d X_ε for X′ = max{X, m}, m ∈ [VaR^L_ε(X), VaR^R_ε(X)]
ProvedTailRiskSharing.TailConv.lemma1_tail_maxp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1quantiletail-risk-measurevalue-at-risk
Let be an atomless probability space, , and . Let be any number with
and put . Then the tails of and beyond their -quantiles have the same law:
Raising to a level inside its -quantile interval changes only the body of the distribution, so an -tail risk measure cannot tell and apart. This is the technical lemma behind Theorem 3.
Formalization Note is stated through display (6) as for all .
Preamble
import Mathlib import Definitions.Def_TailRiskSharing_TailConv_Setting open MeasureTheory
Formal statement
namespace TailRiskSharing.TailConv
/-- Lemma 1, p. 17 (proof p. 38): for `X ∈ L^∞`, `ε ∈ (0,1)` and
`m ∈ [VaR^L_ε(X), VaR^R_ε(X)]`, the tails of `X' = max{X, m}` and `X` at level `ε` have the
same law: `X'_ε =d X_ε` (encoded through (6)). -/
theorem lemma1_tail_max {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(hP : TailRiskSharing.VaRConv.IsAtomless P) (ε : ℝ) (hε0 : 0 < ε) (hε1 : ε < 1) (X : Ω → ℝ) (hX : X ∈ Linf P)
(m : ℝ) (hmL : TailRiskSharing.VaRConv.VaRL P ε X ≤ m) (hmR : m ≤ TailRiskSharing.VaRConv.VaRR P ε X) :
TailRiskSharing.VaRTail.TailEquiv P ε (fun ω => max (X ω) m) X := by sorry
end TailRiskSharing.TailConv
Source
Liu, Mao, Wang & Wei, Inf-convolution, Optimal Allocations, and Model Uncertainty for Tail Risk Measures, SSRN 3490348 (version of January 16, 2022), p. 17, Lemma 1; proof in Appendix B, p. 38
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.