Stable Policy Learning
Abstract
In evidence-based policymaking, typically one experimental sample is observed, then a learned policy recommendation is implemented at scale. Policies learned from the experimental data can perform well in expected welfare, yet random sampling in the experiment can produce recommendations with poor welfare outcomes. In this paper, we ask: how should policy learning algorithms balance expected welfare against sampling risk? Our main contribution is to show that algorithmic stability plays a central role in characterizing and navigating the tradeoff. Intuitively, if a policy learning algorithm's recommendation remains stable when one experimental unit is replaced, then that algorithm has limited sampling risk. We propose a method for policy learning called policy-vote bagging, which learns treatment decisions on many subsamples then averages their votes into treatment probabilities. Relative to using one subsample, averaging across subsamples preserves expected welfare and improves expected utility for a risk-averse researcher. We derive sharp bounds linking estimation accuracy, subsample size, and welfare variation, including an exact guarantee under CARA utility.
Related
- Reinforcement Learning Under Runtime Confounding: Policy Learning and Off-Policy Evaluation
- Policy Learning with Weak Signals
- Orthogonal Policy Learning with Ordinal Outcomes
- PhiBE-Q-Learning: Bridging Off-Policy Reinforcement Learning and Continuous-Time Control
- Resolving the Mobility-Equity-Emissions Trilemma: A Causal Machine Learning Framework for Dynamic Tolling Policy
BibTeX
@misc{barnhard2026stable,
title = {{Stable Policy Learning}},
author = {Harvey Barnhard and Giacomo Opocher and Rahul Singh},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.19418},
eprint = {2609.19418},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.19418},
}
RIS
TY - UNPB
TI - Stable Policy Learning
AU - Harvey Barnhard
AU - Giacomo Opocher
AU - Rahul Singh
PY - 2026
DA - 2026/09/16
PB - arXiv
DO - 10.48550/arxiv.2609.19418
UR - https://arxiv.org/abs/2609.19418
AB - In evidence-based policymaking, typically one experimental sample is observed, then a learned policy recommendation is implemented at scale. Policies learned from the experimental data can perform well in expected welfare, yet random sampling in the experiment can produce recommendations with poor welfare outcomes. In this paper, we ask: how should policy learning algorithms balance expected welfare against sampling risk? Our main contribution is to show that algorithmic stability plays a central role in characterizing and navigating the tradeoff. Intuitively, if a policy learning algorithm's recommendation remains stable when one experimental unit is replaced, then that algorithm has limited sampling risk. We propose a method for policy learning called policy-vote bagging, which learns treatment decisions on many subsamples then averages their votes into treatment probabilities. Relative to using one subsample, averaging across subsamples preserves expected welfare and improves expected utility for a risk-averse researcher. We derive sharp bounds linking estimation accuracy, subsample size, and welfare variation, including an exact guarantee under CARA utility.
ER -