Non-Identifiability of Loss Distributions under Conditional Value-at-Risk and Lower-Tail Dependence Constraints
Spectrum of Engineering and Management Sciences · 20.11.2026 · 4 (1) · 354-379
DOI: 10.31181/sems412026110 ↗Kurz gefasst
Fachartikel in der Zeitschrift Spectrum of Engineering and Management Sciences, online erschienen am 20.11.2026. Verfasst von Attila Bányai und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Risk summaries support portfolio decisions, but do two matching measures imply the same loss distribution? We investigate this question for 99% portfolio Conditional Value-at-Risk (CVaR) and pairwise lower-tail dependence within a centered, equicorrelated multivariate Student-t family with a fixed asset universe and long-only weights. Analytical transformations preserve both measures, while necessary and sufficient conditions establish portfolio feasibility. We then evaluate distributional differences through the Kolmogorov–Smirnov (KS) distance. Deterministic enumeration of 99,000,000 candidates yields 34,604,311 feasible transformations, including 33,624,311 nonidentity alternatives. The mean KS distances are 0.0159 under equal transformation weighting and 0.0137 under equal weighting of active initial configurations. The maximum, 0.238969, corresponds to a 23.90-percentage-point difference in cumulative loss probabilities. Three contributions emerge. First, a continuous family of distinct, feasible loss distributions can retain both risk measures. Second, at a common CVaR and fixed confidence level, the KS distance depends only on the degrees-of-freedom pair; sensitivity analyses confirm that non-identifiability persists under alternative numerical specifications. Third, a chance-constrained example shows that identical summaries can yield opposite compliance decisions. Thus, additional distributional information is necessary when decisions depend on loss-threshold probabilities.
Zitieren
APA
Bányai, A., Tatay, T., Pataki, L., & Thalmeiner, G. (2026). Non-Identifiability of Loss Distributions under Conditional Value-at-Risk and Lower-Tail Dependence Constraints. Spectrum of Engineering and Management Sciences, 4(1), 354-379. https://doi.org/10.31181/sems412026110
Vancouver
Bányai A, Tatay T, Pataki L, Thalmeiner G. Non-Identifiability of Loss Distributions under Conditional Value-at-Risk and Lower-Tail Dependence Constraints. Spectrum of Engineering and Management Sciences. 2026;4(1):354-379. doi:10.31181/sems412026110
BibTeX
@article{Bnyai2026,
author = {Bányai, Attila and Tatay, Tibor and Pataki, László and Thalmeiner, Gergő},
title = {Non-Identifiability of Loss Distributions under Conditional Value-at-Risk and Lower-Tail Dependence Constraints},
journal = {Spectrum of Engineering and Management Sciences},
year = {2026},
volume = {4},
number = {1},
pages = {354--379},
doi = {10.31181/sems412026110}
}
EUVIMED