Stable decision making despite changing outcomes in time series data
Learning Deployable Causal Action Geometry under Temporal Non-Stationarity
Machine Learning
Summary
This paper addresses the problem of making good decisions when the effects of actions change over time and outcomes shift in ways that can confuse simple models. The authors focus on finding the part of the response to actions that stays consistent even as overall outcome levels drift. They develop a new method that carefully separates stable action effects from shifting baselines and variations, enabling reliable decision rules that work well now and in the future. Their approach was tested on simulated data, a retail dataset, and a real-world online shopping platform, showing practical improvements.
What this means in practice
- •For online platform engineers: Deploy decision systems that maintain effectiveness despite shifting user behavior and outcome trends over time.$Commercial implications: Enables e-commerce platforms to reliably improve sales metrics by adapting models to changing customer responses without retraining from scratch.
- •For business intelligence teams: Analyze historical observational time series to extract stable causal effects of actions for better future planning.
Authors
Changjian Liu, Tianyu Wang, Yuwei Xu, Xiaoxuan Deng, Zhilin Zhang, Yong Gao, Chuan Yu, Jian Xu, Bo Zheng
Abstract
Learning a response law for continuous decisions from observational time series is central to many real-world decision systems, yet temporal non-stationarity poses a fundamental challenge: as outcomes drift over time, naively pooled models can conflate action effects with shifting baselines and even reverse action rankings. Even when the one-step response is identifiable, unrestricted cross-time generalization remains fundamentally ill-posed. We therefore seek the part of the response that remains stable for decisions even when outcome levels do not. We propose deployable causal action geometry: the action-response shape that remains after anchoring away action-independent level shifts on a selected link scale and profiling out positive amplitude variation. Building on this target, we develop a two-stage orthogonal estimator that first obtains cross-fitted pilot estimates of anchored responses and then performs honest profiled selection over link, representation, and shape classes. The resulting estimator admits oracle guarantees for the restricted target and, under approximate persistence, produces decision rules that remain near optimal in future periods. Experiments on synthetic data and Dominick's support the proposed target and its transfer behavior. In a randomized Taobao A/B test, the deployed system improves GMV and purchase counts by \(1.16\%\) and \(1.70\%\).