Total L^p-Impossibility, the Banach-Mazur Price of Time Consistency, and Controlled Spectral Fractional HJB Equations

Kupper and Schachermayer (2009) showed that on L^∞ the law-invariant, strongly time-consistent monetary risk measures form the entropic family ρ_ent^β(X) = (1/β) log E[exp(βX)] together with its limits; imposing coherence leaves only E[X] and ess sup(X) (Shapiro 2012; Pflug and Pichler 2016). This paper extends dynamic risk theory into the heavy-tailed, infinite-variance regime with five results. (A) Elementary classification. An elementary proof on finite partitions, assuming neither convexity, relevance nor the Fatou property, that positive homogeneity leaves only E, ess sup and ess inf on L^∞, only E on L^p (1 ≤ p < ∞), and no measure on L^p (0 < p < 1) or L^0. This rules out VaR_α directly. (B) Total L^p-Impossibility, across and beyond convexity. On L^p (1 ≤ p < ∞), E[X] is the sole finite, convex, law-invariant, strongly time-consistent monetary risk measure. The same collapse holds without convexity for every L^p-continuous strictly monotone measure (with automatic upper semicontinuity at 0 for star-shaped measures). We also give a model-free certainty-equivalent characterization for cash non-additive two-parameter semigroups on general filtrations with jumps. (C) Banach–Mazur price of time consistency. For conditionally law-invariant coherent nested compositions against any spectral/Kusuoka reference ρ_stat over T independent p-stable stages (p ∈ (1,2], symmetric or skewed, including heteroscedastic scales), the multiplicative condition number equals ‖γ‖_q / min_t γ_t ≥ d_BM(ℓ_1^T, ℓ_p^T) = T^{1−1/p}, attained uniquely by uniform weights. This yields a certified T^{1−1/p}-approximation for Bellman policies and quantifies the L^p separation between adapted and terminal law invariance. It extends to a phase transition across p = 1 via VaR_α (p ∈ (0,2]), and fails without conditional law invariance through an L^∞-bounded Radon–Nikodym construction valid for all p ∈ (1,2]. (D) Controlled fractional HJB limit. Extending Stadje (2010) and Pichler and Schlotter (2023) to p ∈ (1,2] and unbounded spectral/Kusuoka measures with distortion g(α) ≤ C α^{1−γ} (γ ∈ [0, 1 − 1/p)), the operator (1 − λ_Δt) E + λ_Δt ρ_0 with λ_Δt = λ_0 (Δt)^{1−1/p} is consistent with a fractional Hamilton–Jacobi–Bellman generator carrying the 1-homogeneous drift λ_0 ρ_0(ζ_1) ‖σ(x,a)ᵀ ∇V‖_p. We give the sharp anomalous rate Θ((Δt)^{1−1/p−γ}) for p < 2, a dimension-optimal remainder constant (1/(K(1)(p−1))) ‖σ‖_{p,p}^p ≈ 2.056/(p−1) · ‖σ‖_{p,p}^p at γ = 0, and the Kusuoka Cauchy pole π^{-1} sup over μ of ∫ u^{-1} μ(du). (E) Controlled Euler–Lévy dynamics. For state- and control-dependent coefficients (f, b, σ) we prove a self-contained comparison principle and locally uniform viscosity convergence (plus a direct Neufeld–Nutz equicontinuity proof in the translation-invariant case). We also explain the VaR versus CVaR/EVaR dynamic-programming asymmetry in Markov decision processes via mixture quasilinearity versus mixture concavity. Relevance. In heavy-tailed settings (such as heavy-tailed gradient noise), time-consistent risk-sensitive control essentially forces the risk-neutral expectation. The paper quantifies exactly what is lost when one nevertheless imposes a time-consistent recursion, via the T^{1−1/p} Banach–Mazur factor.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24418230
Primary Topic
Risk and Portfolio Optimization
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Total L^p-Impossibility, the Banach-Mazur Price of Time Consistency, and Controlled Spectral Fractional HJB Equations

Gabriel Costa de Mello Paiva
Zenodo (CERN European Organization for Nuclear Research)
Risk and Portfolio Optimization
preprint

Total L^p-Impossibility, the Banach-Mazur Price of Time Consistency, and Controlled Spectral Fractional HJB Equations

Gabriel Costa de Mello Paiva
preprint en

Abstract

Kupper and Schachermayer (2009) showed that on L^∞ the law-invariant, strongly time-consistent monetary risk measures form the entropic family ρ_ent^β(X) = (1/β) log E[exp(βX)] together with its limits; imposing coherence leaves only E[X] and ess sup(X) (Shapiro 2012; Pflug and Pichler 2016). This paper extends dynamic risk theory into the heavy-tailed, infinite-variance regime with five results. (A) Elementary classification. An elementary proof on finite partitions, assuming neither convexity, relevance nor the Fatou property, that positive homogeneity leaves only E, ess sup and ess inf on L^∞, only E on L^p (1 ≤ p < ∞), and no measure on L^p (0 < p < 1) or L^0. This rules out VaR_α directly. (B) Total L^p-Impossibility, across and beyond convexity. On L^p (1 ≤ p < ∞), E[X] is the sole finite, convex, law-invariant, strongly time-consistent monetary risk measure. The same collapse holds without convexity for every L^p-continuous strictly monotone measure (with automatic upper semicontinuity at 0 for star-shaped measures). We also give a model-free certainty-equivalent characterization for cash non-additive two-parameter semigroups on general filtrations with jumps. (C) Banach–Mazur price of time consistency. For conditionally law-invariant coherent nested compositions against any spectral/Kusuoka reference ρ_stat over T independent p-stable stages (p ∈ (1,2], symmetric or skewed, including heteroscedastic scales), the multiplicative condition number equals ‖γ‖_q / min_t γ_t ≥ d_BM(ℓ_1^T, ℓ_p^T) = T^{1−1/p}, attained uniquely by uniform weights. This yields a certified T^{1−1/p}-approximation for Bellman policies and quantifies the L^p separation between adapted and terminal law invariance. It extends to a phase transition across p = 1 via VaR_α (p ∈ (0,2]), and fails without conditional law invariance through an L^∞-bounded Radon–Nikodym construction valid for all p ∈ (1,2]. (D) Controlled fractional HJB limit. Extending Stadje (2010) and Pichler and Schlotter (2023) to p ∈ (1,2] and unbounded spectral/Kusuoka measures with distortion g(α) ≤ C α^{1−γ} (γ ∈ [0, 1 − 1/p)), the operator (1 − λ_Δt) E + λ_Δt ρ_0 with λ_Δt = λ_0 (Δt)^{1−1/p} is consistent with a fractional Hamilton–Jacobi–Bellman generator carrying the 1-homogeneous drift λ_0 ρ_0(ζ_1) ‖σ(x,a)ᵀ ∇V‖_p. We give the sharp anomalous rate Θ((Δt)^{1−1/p−γ}) for p < 2, a dimension-optimal remainder constant (1/(K(1)(p−1))) ‖σ‖_{p,p}^p ≈ 2.056/(p−1) · ‖σ‖_{p,p}^p at γ = 0, and the Kusuoka Cauchy pole π^{-1} sup over μ of ∫ u^{-1} μ(du). (E) Controlled Euler–Lévy dynamics. For state- and control-dependent coefficients (f, b, σ) we prove a self-contained comparison principle and locally uniform viscosity convergence (plus a direct Neufeld–Nutz equicontinuity proof in the translation-invariant case). We also explain the VaR versus CVaR/EVaR dynamic-programming asymmetry in Markov decision processes via mixture quasilinearity versus mixture concavity. Relevance. In heavy-tailed settings (such as heavy-tailed gradient noise), time-consistent risk-sensitive control essentially forces the risk-neutral expectation. The paper quantifies exactly what is lost when one nevertheless imposes a time-consistent recursion, via the T^{1−1/p} Banach–Mazur factor.

Zenodo (CERN European Organization for Nuclear Research)
Risk and Portfolio Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon