About

Learning to decide under uncertainty.

I am a Ph.D. candidate in Data Analytics and Decision Science at the Darden School of Business, University of Virginia, advised by Prof. Manel Baucells and Prof. Saša Zorc.

My research develops theory for optimal sequential decision-making under uncertainty — search theory, optimal stopping, Bayesian learning, and multi-armed bandits. A growing focus is mechanism and contract design for the age of AI — how firms should contract on AI use and the cognitive offloading it induces — alongside inference-time computation for large language models and flexible probability modeling with the metalog distribution.

Before my Ph.D., I received an M.A. in Applied Economics from the National University of Singapore and worked as a Product Manager at Tencent and as a Research Associate at CUHK‑Shenzhen.

I will be presenting at the 2026 INFORMS Annual Meeting this year.
Sequential decision-making Search theory & optimal stopping Bayesian learning Multi-armed bandits Mechanism & contract design Inference-time computation for LLMs Cognitive offloading Metalog distributions
Research

Working papers & ongoing projects.

Working Papers
Stephen Z. Xu, Manel Baucells, Saša Zorc
Characterizes the structure of optimal stopping policies for sequential search when past offers cannot be recalled and the decision-maker learns about value through Gaussian signals.
Major Revision · Operations Research
Manel Baucells, Lonnie Chrisman, Thomas Keelin, Stephen Z. Xu
Studies the structural and analytical properties of the metalog family of flexible probability distributions.
BEACON: Bayesian Optimal Stopping for Efficient LLM Sampling
Guangya Wan*, Stephen Z. Xu*, Saša Zorc, Manel Baucells, Mengxuan Hu, Hao Wang, Sheng Li
*Co-first authors with equal contributions
Applies Bayesian optimal stopping to inference-time sampling, deciding adaptively how many samples to draw from a large language model to balance answer quality against compute cost.
Works in Progress
Sequential Search and Pandora’s Box under Conjugate Bayesian Learning
Stephen Z. Xu, Manel Baucells, Saša Zorc
Screening Skill through AI Usage: Optimal Employment Contracts with Endogenous Deskilling
Stephen Z. Xu, Saša Zorc
Characterizes the optimal employment contract when a worker privately observes an evolving skill that AI use itself reshapes, and firms pay per unit of AI compute. The solution reduces to a single token-pricing rule that over-provisions on routine tasks and rations on expert tasks.
Characterizing Tractable Bayesian Experimentation: Exact Gittins and Whittle Indices
Stephen Z. Xu, Manel Baucells, Saša Zorc
Identifies the distributional families for which Bayesian experimentation is tractable, deriving exact Gittins and Whittle indices through a standardized index decomposition for multi-armed and restless bandits.
Cognitive Offloading as an Intertemporal Allocation Problem
Stephen Z. Xu, Emanuele Agrimi, Hami Doan, Linda Fiorini, Fabio Michele Russo, Folco Panizza, Tea Tucić, Alexander J. Gates
Models the decision to offload cognitive effort to AI as an intertemporal trade-off between immediate productivity and the longer-run accumulation of skill and understanding.
Academic Activities

Conferences, teaching & service.

Conferences & Presentations
2026
2026 INFORMS Annual Meeting Presenting
Oct 2026
2026 INFORMS MSOM Conference · Harvard Business School Presented
Jul 2026
2026 Advances in Decision Analysis (ADA) Conference · Duke University Presented
Jun 2026
UVA–IMT AI Workshop
Mar 2026
2025
2025 UVA Quantitative Psychology (DADA) Meeting Presented
Oct 2025
2025 INFORMS Annual Meeting · Atlanta Presented
Oct 2025
2025 Purdue Supply Chain & Operations Management Conference Presented
Aug 2025
2025 INFORMS International Meeting · Singapore Presented
Jul 2025
Games and AI Multidisciplinary Summer School (GAIMSS) & Workshop Presented
Jun 2025
Risk, Uncertainty and Decision (RUD) 2025 Conference
Jun 2025
2024
1st Summer School in Experimental & Behavioral Economics & Workshop
Aug 2024
Paris School of Economics Industrial Organization Summer School & Workshop Presented
Jul 2024
Academic & Teaching Experience
Research Associate 2022–2023
Chinese University of Hong Kong, Shenzhen
  • Collaborated with Prof. Linyi Zhang on empirical research in ESG policy and industrial organization.
  • Led tutorials for the undergraduate course Quantitative Methods for Policy Evaluation.
Referee Service

Journals: Operations Research, Management Science.