Dynamic Wholesale Pricing under Censored-Demand Learning

Michalis Deligiannis, Marco Scarsini, Xavier Venel · 2026

Read Dynamic Wholesale Pricing under Censored-Demand Learning by Michalis Deligiannis, Marco Scarsini, Xavier Venel (2026) 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-2603-13599.txt.

Introduction

Michalis Deligiannis, Marco Scarsini, and Xavier Venel study dynamic wholesale pricing under censored-demand learning. In their two-tier supply chain, firms share POS data and learn demand from censored observations. When stockouts occur, unmet demand is unobserved, so order quantity shapes both profit and future signal quality. The manufacturer can speed or slow learning through wholesale prices, while the retailer weighs inventory effects on future information. They prove a Markov perfect equilibrium exists for Weibull demand, and uniqueness for exponential demand. Simulations show public learning creates conflicting incentives.

Catalog record

CREATORMichalis Deligiannis, Marco Scarsini, Xavier Venel
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67954
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.13599

Browse the full library · About the Tower of Babel

Full text

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

Dynamic Wholesale Pricing under Censored-Demand Learning Michalis Deligiannis, Marco Scarsini, Xavier Venel Abstract We study a finite-horizon dynamic wholesale-price contract between a manufacturer and a retailer, both of whom observe only sales, rather than the true demand. When the retailer stocks out, unmet demand is unobserved, so both parties update a common posterior over the demand distribution from sales data. Each period, the manufacturer sets the wholesale price, the retailer chooses an order quantity, and the public belief state is updated. We characterize MPE as functions of this public belief. Our main results are as follows: for Weibull demand, we extend the well-known scaling approach to this strategic learning setting, prove the existence of an equilibrium, and reduce computation to a standardized one-parameter recursion; for exponential demand, we show that the equilibrium is unique and computable via a simple backward recursion. Keywords: wholesale-price contracts; censored demand; Bayesian learning; Markov perfect equilibrium; dynamic Stackelberg game. 1 Introduction In many retail and supply chain settings, firms observe sales rather than true demand. When stockouts occur, unmet demand remains unobserved, resulting in censored data. Frequent stockouts and short product life cycles further limit the value of historical data, leading firms to rely primarily on recent sales—data that imperfectly reflect underlying demand—for their operational decisions ( Chuang and Kim, 2023 ) . In this paper, we study a dynamic finite-horizon wholesale-price contract between a manufacturer and a retailer who face a random demand whose distribution depends on an unknown parameter. The two agents update their beliefs about this parameter using censored demand observations. The setting is motivated by two common features of contemporary practice. First, in volatile markets, firms rely on short-term B2B agreements to maintain flexibility, which motivates modeling the relationship as a finite-horizon contract ( Feng and Krishnan, 2022 , Chestney, 2023 , Floeck, 2023 ) . Second, advances in digital infrastructure now allow manufacturers and retailers to share real-time POS ( POS ) data, so both parties observe the same sales information. For example, Walmart’s Luminate platform provides manufacturers with granular, real-time sales visibility, which is routinely used to support forecasting and replenishment decisions ( Souza, 2023 , Silberstein, 2023 ) . However, these systems record only realized sales and omit lost demand, meaning that both the manufacturer and the retailer learn from the same censored observations. Although several studies have dealt with either supply chain contracts or censored demand, the interplay between contracts and learning from censored demand remains largely unexplored. To address this gap, we model the interaction between the manufacturer and the retailer as an incomplete-information dynamic Stackelberg game in which, at each period, the manufacturer sets the wholesale price, the retailer responds by choosing an order quantity, and both parties update a shared belief about the unknown demand distribution. This setting extends the classical “selling to the newsvendor” framework of Lariviere and Porteus, 2001 by incorporating strategic dynamic learning from censored data. Censored observations create analytical challenges because many standard demand families do not preserve conjugacy when only sales are observed. To maintain tractability, we employ the newsvendor distribution family introduced by Braden and Freimer, 1991 , which preserves conjugacy under Bayesian updating. Using this framework, we adopt MPE ( MPE ) as the solution concept of our game. Our main contributions are the following: 1. Dynamic contracting with censored-demand learning. We develop a dynamic Stackelberg game under a wholesale-price contract in which the manufacturer’s pricing and the retailer’s ordering decisions are jointly shaped by Bayesian learning from shared, censored data. 2. Existence of an MPE . For Weibull demand, we prove the existence of an MPE using a technique developed by Lariviere and Porteus, 1999 in a non-strategic setting. This technique scales the model so that the conjugate prior depends only on its shape hyperparameter and not on its rate hyperparameter, thereby reducing the dimensionality of the problem. A byproduct of this technique is that, at equilibrium, the manufacturer’s wholesale price depends only on the initial belief and the number of uncensored demand observations. 3. Uniqueness of the MPE . When demand is exponentially distributed, the MPE is unique and can be computed by a simple backward recursion without solving an optimization problem in each period. The remainder of the paper is organized as follows. Section 2 reviews the related literature. Section 3 formulates the dynamic game and introduces the solution concept. Section 4 presents the dimensionality-reduction result, establishes the existence of an MPE, and discusses its implications. Section 5 characterizes the MPE in the exponential-demand case. Section 6 lists some open problems. Finally, Section 7 contains the proofs. 2 Literature Review Our work relates to research on inventory management under censored demand and to dynamic wholesale-price contracts in supply chains. A substantial literature in inventory management examines demand learning in settings where stockouts lead to censored sales. This line of research dates back to Scarf, 1959 , Scarf, 1960 , who shows that, when the demand is observable, the Bayesian dynamic program can be simplified through state-space reduction ( Azoury, 1985 , see also) . Under censored demand, however, conjugacy fails for most standard distributions, and Bayesian updating becomes analytically difficult. To address this challenge, Braden and Freimer, 1991 introduce the newsvendor family, which preserves conjugacy under censoring and enables tractable Bayesian inventory models in finite-horizon settings ( Lariviere and Porteus, 1999 , Bisi et al., 2011 , Mersereau, 2015 , Besbes et al., 2022 , Chuang and Kim, 2023 , Zhang et al., 2025 , see, e.g., ) . Within this family, Lariviere and Porteus, 1999 show that Weibull demand with a gamma prior admits a dimensionality-reduction result, which renders single-firm dynamic inventory problems tractable. We build on this property and extend it from a single decision maker to a strategic two-tier supply chain. Our paper builds on the “selling to the newsvendor” framework of Lariviere and Porteus, 2001 , which studies wholesale-price contracts between a manufacturer and a retailer. More broadly, supply chain contracting in this environment is reviewed by Cachon, 2003 . Several papers study multi-period wholesale pricing or repeated manufacturer–retailer interactions under a known demand distribution or exogenous information updates, and characterize subgame-perfect equilibria ( Anand et al., 2008 , Erhun et al., 2008 , Martínez-de Albéniz and Simchi-Levi, 2013 , Sheng et al., 2017 , e.g., ) . Related work examines how exogenous forecast updates affect pricing and ordering decisions ( Özer et al., 2007 , e.g., ) . We refer to Shen et al., 2019 for a broader review of supply chain contracting with information updating. This stream of literature assumes that demand information is known or evolves exogenously, rather than being learned from censored demand observations. Han, 2016 studies a two-period supplier–retailer Stackelberg newsvendor model with an ex ante constant wholesale price; learning under censored sales is made tractable via a binary Bernoulli–Beta demand model (myopic vs. forward-looking retailer), with a Weibull–Gamma extension treated numerically. A related body of work studies dynamic contracting with information asymmetry about inventory and demand ( Zhang et al., 2010 , Lobel and Xiao, 2017 , Bensoussan et al., 2025 , see, e.g., ) . Kadiyala et al., 2020 study a long-term vendor-managed inventory arrangement in which the manufacturer observes the retailer’s POS

Read the complete text