Learning When to Restart: Nonstationary Newsvendor from Uncensored to Censored Demand

Xin Chen, Jiameng Lyu, Shilin Yuan, Yuan Zhou · 2025

Read Learning When to Restart: Nonstationary Newsvendor from Uncensored to Censored Demand by Xin Chen, Jiameng Lyu, Shilin Yuan, Yuan Zhou (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-18709.txt.

Introduction

The newsvendor problem is a classic: stock enough for demand you cannot foresee, but not so much that the surplus rots. Xin Chen, Jiameng Lyu, Shilin Yuan, and Yuan Zhou study it when demand keeps shifting over time. Their 2025 paper handles both abrupt and gradual change. Demand data may also be censored, showing only sales, not true demand.

Their framework watches the demand distribution for change, then restarts learning when a shift is detected. Two efficient algorithms cover the open-data and censored cases. Neither needs to know how strong the shifts are in advance. The team proves matching upper and lower regret bounds, the theory's gold standard. Tests on nurse staffing and COVID test demand data show strong, robust results. Managers get a deployable method with guarantees, the authors argue.

Catalog record

CREATORXin Chen, Jiameng Lyu, Shilin Yuan, Yuan Zhou
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE153 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-18709

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Learning When to Restart: Nonstationary Newsvendor from Uncensored to Censored Demand Xin Chen, Jiameng Lyu, Shilin Yuan, Yuan Zhou Abstract We study nonstationary newsvendor problems under nonparametric demand models and general distributional measures of nonstationarity, addressing the practical challenges of unknown degree of nonstationarity and demand censoring. We propose a novel distributional-detection-and-restart framework for learning in nonstationary environments, and instantiate it through two efficient algorithms for the uncensored and censored demand settings. The algorithms are fully adaptive, requiring no prior knowledge of the degree and type of nonstationarity, and offer a flexible yet powerful approach to handling both abrupt and gradual changes in nonstationary environments. We establish a comprehensive optimality theory for our algorithms by deriving matching regret upper and lower bounds under both general and refined structural conditions with nontrivial proof techniques that are of independent interest. Numerical experiments using real-world datasets, including nurse staffing data for emergency departments and COVID-19 test demand data, showcase the algorithms’ superior and robust empirical performance. While motivated by the newsvendor problem, the distributional-detection-and-restart framework applies broadly to a wide class of nonstationary stochastic optimization problems. Managerially, our framework provides a practical, easy-to-deploy, and theoretically grounded solution for decision-making under nonstationarity. keywords nonstationarity, newsvendor model, censored demand, sample average approximation, operations management, regret analysis † † runningauthor: Chen, Lyu, Yuan, Zhou † † runningtitle: Nonstationary Newsvendor from Uncensored to Censored Demand † † authors: H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332, USA, [email protected] Department of Management Science, School of Management, Fudan University, Shanghai, 200433, China, [email protected] Department of Decisions, Operations and Technology, Chinese University of Hong Kong, Shatin, Hong Kong, [email protected] Yau Mathematical Sciences Center & Department of Mathematical Sciences, Tsinghua University, Beijing 100084, China, [email protected] † † affiliation: † † affiliation: † † affiliation: † † affiliation: † † affiliation: † † affiliation: † † affiliation: † † affiliation: 1 Introduction The newsvendor model is a fundamental framework in operations management that captures the essence of inventory decisions under demand uncertainty. In this model, the decision-maker must determine the stock level in advance to balance the trade-off between underage cost, arising from insufficient inventory that leads to lost sales or customer dissatisfaction, and overage cost, resulting from excess inventory that incurs holding or disposal costs. This simple yet powerful formulation has been widely applied in various contexts, including retail inventory management for perishable goods and healthcare staffing decisions. While the classic model assumes that the demand distribution is known, real-world applications often face significant uncertainty about this distribution. To address this challenge, recent research has focused on the data-driven newsvendor model , which leverages historical demand data to guide inventory decisions without requiring full knowledge of the underlying demand distribution. A common assumption in most existing literature of data-driven newsvendor problems is stationarity: demand data D_{1},\cdots,D_{T} are drawn from the identical distribution. However, nonstationarity is an intrinsic feature of the real-world operating environment. Stationarity, although analytically convenient, may fail to capture important temporal dynamics, resulting in outdated and suboptimal decisions, as well as substantial operational inefficiencies. In practice, nonstationarity can generally be categorized into two types: abrupt changes and gradual changes. Abrupt changes may result from sudden events such as regulatory shifts or viral marketing campaigns, whereas gradual changes can arise from seasonal trends, market evolution, or long-term economic shifts. For instance, in perishable retail (e.g., supermarket fresh produce), seasonal and climatic trends gradually change the underlying demand pattern over time, and a flash promotion can trigger sudden demand shocks. Similarly, the demand for emergency departments also exhibits nonstationarity. Besides, the timing and degree of these changes are typically unknown a priori. Thus, rather than committing to a fixed adjustment strategy, we should adaptively adjust the data-driven policies online according to real-time environments to mitigate the adverse effects of nonstationarity. A further layer of complexity in data-driven newsvendor problems arises from demand censoring, a pervasive phenomenon in practice. In many settings, when inventories are insufficient, actual demand remains unknown beyond recorded sales, and only censored demand is observable, such as retail stockouts, limited-capacity flash sales, or perishable inventory systems. In nonstationary environments, demand censoring further obscures the true demand pattern, exacerbating the difficulty of demand learning and inventory decision-making. Motivated by the above practical needs, this paper makes several contributions. We introduce a direct and distribution-level characterization of nonstationarity, operating in a fully nonparametric demand model. Based on this characterization, we propose a flexible algorithm design framework tailored to nonstationary environments and develop efficient algorithms for both censored and uncensored demand settings. Moreover, we provide a comprehensive regret analysis that establishes the optimal dynamic regret of the proposed algorithms, under various nonstationarity measures and across different distribution classes for both uncensored and censored demand settings. 1.1 Our Contributions In the following, we present a more detailed discussion of our main contributions. Contribution I: Fully nonparametric modeling and direct characterization of nonstationarity. We introduce a direct, distribution-level characterization of nonstationarity within a fully nonparametric demand model. Existing work on nonstationary newsvendor problems often adopts a parametric perspective, for example, assuming a drifting parameter \{\mu_{t}\}_{t=1}^{T} plus i.i.d. noise. In contrast, we impose no parametric assumptions on the demand distributions. We treat the observed sequence \{D_{t}\}_{t=1}^{T} as draws from an arbitrary sequence of distributions and model nonstationarity directly as distributional change. This approach eliminates restrictive structural assumptions on the noise or the mechanism through which nonstationarity enters the model. We propose two distribution-level measures of nonstationarity on the sequence of distributions that capture both abrupt and gradual changes commonly encountered in real-world settings: 1. The switch number budget : \sum_{t=2}^{T}\mathbb{I}\bigl[D_{t}\overset{d}{\neq}D_{t-1}\bigr]\leq S, which bounds the number of distribution switches. 1 1 1 D_{t}\overset{d}{\neq}D_{t-1} refers to that D_{t} and D_{t-1} have different distributions. \mathbb{I}[\cdot] is the indicator function. 2. The total variation budget : \sum_{t=2}^{T}\|D_{t}-D_{t-1}\|_{\mathrm{TV}}\leq V, which controls the total variation of the distributional shift. 2 2 2 The total variation between two distributions, defined as \|D_{t}-D_{t-1}\|_{\mathrm{TV}}=\frac{1}{2}\int|dD_{t}-dD_{t-1}| , is one of the most common metrics for measuring the distance between distributions. The variation budget V is well suited to environments with gradual change, while the switch budget S is better for settings with a small number of abrupt changes (where V may be large even when S is small). In contrast, prior work has largely focused on parameter-level variation in sequences \{\mu_{t}\}_{t=1}^{T} and has not explicitly accounted for the number of distribution switches. Moreover, our distribution-level measures are more general and natural than parameter-level ones, and are capable of capturing a broader class of nonstationary behaviors. Contribution II: A powerful framework and two efficient algorithms. We propose the new distributional-detection-and-restart framework for learning in nonstationary environments without knowledge of the degree of nonstationarity, and instantiate it through two novel algorithms for the nonstationary newsvendor problems. The core idea of our distributional-detection-and-restart framework is to monitor the stationarity of the distributions and search for the optimal solution based on the data detected to be stationary. This monitoring is carried out through a distributional nonstationarity detection layer. Unlike prior works that detect changes using function values, our method performs distributional detection by comparing empirical distributions across different time intervals. Guided by this framework, we develop two algorithms tailored to different

Read the complete text