Sign up

Research paper

CVaR Deep Optimal Stopping

2 authors
  1. Yechan Jeong
  2. Youngjoon Hong
Venue
arXiv (Cornell University)
Year
2026
Publication date
2026-11-08

Abstract

Held from OpenAlex
Risk-sensitive optimal stopping is useful when poor stopping outcomes carry substantial costs, as in option exercise, sequential medical treatment, and safety-critical control. A natural objective is Conditional Value at Risk (CVaR) of the stopped reward. We develop a theoretical framework and algorithm for CVaR deep optimal stopping. The framework and algorithm address four difficulties in applying risk-sensitive reinforcement learning to optimal stopping: state augmentation, score-function pathology, blindness to success, and dependence on auxiliary networks. We isolate three structural properties of optimal stopping: single-point reward, action-independent dynamics, and sum-zero gradient. We exploit these properties to establish Markov optimality without state augmentation and derive a pathwise gradient estimator. Under bounded policy derivatives, the pathwise estimator has bounded second moments near deterministic policies, avoiding the divergence that can occur with score-function estimators. Building on this framework, we propose DIOS (Distribution-Informed Optimal Stopping) to address the remaining two difficulties. A per-time-step soft penalty combined with tail-level annealing mitigates blindness to success. Estimating the reward quantile directly from each batch removes the need for an auxiliary critic. For a fixed tail level, we establish a finite-time stationarity guarantee under regularity assumptions along the training iterates. Across the 24 main Bermudan max-call configurations, DIOS achieves the highest mean CVaR and the shortest recorded training time among the tested CVaR baselines.

Attribution

Yechan Jeong, Youngjoon Hong — CVaR Deep Optimal Stopping (2026), arXiv (Cornell University). Original: DOI 10.48550/arxiv.2611.11332. Metadata and abstract supplied by OpenAlex (CC0). OpenAlex record: https://openalex.org/W7225706241. Harvested 2026-11-12; record openalex-works-select.v1 + e5164bf53f99; published 2026-11-12.

Identifiers

doi
https://doi.org/10.48550/arxiv.2611.11332
openalex
https://openalex.org/W7225706241

Research topics

Replications

No Quantellect replication yet

Audited strategies and their receipts will appear here after a separate strategy release.

Citation export

Text remains available without JavaScript

Plain text

Yechan Jeong, & Youngjoon Hong (2026). CVaR Deep Optimal Stopping. arXiv (Cornell University).

BibTeX

@article{cvar-deep-optimal-stopping-1808923,
  title = {CVaR Deep Optimal Stopping},
  author = {Yechan Jeong and Youngjoon Hong},
  journal = {arXiv (Cornell University)},
  year = {2026},
  doi = {10.48550/arxiv.2611.11332},
}

RIS

TY  - JOUR
AU  - Yechan Jeong
AU  - Youngjoon Hong
TI  - CVaR Deep Optimal Stopping
JO  - arXiv (Cornell University)
PY  - 2026
DO  - 10.48550/arxiv.2611.11332
UR  - https://doi.org/10.48550/arxiv.2611.11332
ER  -